d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
f820e31fb2
freebsd-skq/sys/dev/ath/if_ath_tx.c

5543 lines
148 KiB
C
Raw Normal View History

Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*-
* Copyright (c) 2002-2009 Sam Leffler, Errno Consulting
Take credit for the work I've done in this source file.
2012-09-17 03:17:42 +00:00
* Copyright (c) 2010-2012 Adrian Chadd, Xenion Pty Ltd
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer,
* without modification.
* 2. Redistributions in binary form must reproduce at minimum a disclaimer
* similar to the "NO WARRANTY" disclaimer below ("Disclaimer") and any
* redistribution must be conditioned upon including a substantially
* similar Disclaimer requirement for further binary redistribution.
*
* NO WARRANTY
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF NONINFRINGEMENT, MERCHANTIBILITY
* AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
* THE COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY,
* OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
* IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGES.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
* Driver for the Atheros Wireless LAN controller.
*
* This software is derived from work of Atsushi Onoe; his contribution
* is greatly appreciated.
*/
#include "opt_inet.h"
#include "opt_ath.h"
#include "opt_wlan.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysctl.h>
#include <sys/mbuf.h>
#include <sys/malloc.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/kernel.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/errno.h>
#include <sys/callout.h>
#include <sys/bus.h>
#include <sys/endian.h>
#include <sys/kthread.h>
#include <sys/taskqueue.h>
#include <sys/priv.h>
#include <machine/bus.h>
#include <net/if.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#include <net/if_types.h>
#include <net/if_arp.h>
#include <net/ethernet.h>
#include <net/if_llc.h>
#include <net80211/ieee80211_var.h>
#include <net80211/ieee80211_regdomain.h>
#ifdef IEEE80211_SUPPORT_SUPERG
#include <net80211/ieee80211_superg.h>
#endif
#ifdef IEEE80211_SUPPORT_TDMA
#include <net80211/ieee80211_tdma.h>
#endif
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
#include <net80211/ieee80211_ht.h>
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
#include <net/bpf.h>
#ifdef INET
#include <netinet/in.h>
#include <netinet/if_ether.h>
#endif
#include <dev/ath/if_athvar.h>
#include <dev/ath/ath_hal/ah_devid.h> /* XXX for softled */
#include <dev/ath/ath_hal/ah_diagcodes.h>
#include <dev/ath/if_ath_debug.h>
#ifdef ATH_TX99_DIAG
#include <dev/ath/ath_tx99/ath_tx99.h>
#endif
#include <dev/ath/if_ath_misc.h>
#include <dev/ath/if_ath_tx.h>
Modify the TX path to set and use the 11n rate scenario bits. This isn't strictly required to TX (at least non-agg and non-HT40, non-short-GI) frames; but as it needs to be done anyway, just get it done. Linux ath9k uses the rate scenario style path for -all- packets, legacy or otherwise. This code does much the same. Beacon TX still uses the legacy, non-rate-scenario TX descriptor setup. Ath9k also does this. This 11n rate scenario path is only called for chips in the AR5416 HAL; legacy chips use the previous interface for TX'ing.
2011-02-03 20:30:17 +00:00
#include <dev/ath/if_ath_tx_ht.h>
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Add some hooks into the driver to attach, detach and record EDMA descriptor events. This is primarily for the TX EDMA and TX EDMA completion. I haven't yet tied it into the EDMA RX path or the legacy TX/RX path. Things that I don't quite like: * Make the pointer type 'void' in ath_softc and have if_ath_alq*() return a malloc'ed buffer. That would remove the need to include if_ath_alq.h in if_athvar.h. * The sysctl setup needs to be cleaned up.
2012-11-08 18:11:31 +00:00
#ifdef ATH_DEBUG_ALQ
#include <dev/ath/if_ath_alq.h>
#endif
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* How many retries to perform in software
*/
#define SWMAX_RETRIES 10
Map the non-QoS TID to the voice queue, in order to ensure important things like EAPOL frames make it out. After a whole bunch of hacking/testing, I discovered that they weren't being early-dropped by the stack (but I should look at ensuring that later..) but were even making to the hardware transmit queue. They were mostly even being received by the remote end. However, the remote end was completely ignoring them. This didn't happen under 150-170MBit TCP tests as I'm guessing the TX queue stayed very busy and the STA didn't do any scanning. However, when doing 100Mbit/s of TCP traffic, the STA would do background scanning - which involves it coming in and out of powersave mode with the AP. Now, this is a total and utter hack around the real problems, which are: * I need to implement proper power save handling and integrate it into the filtered frames support, so the driver/stack doesn't send frames whilst the station is actually in sleep; * .. but frames were actually making it to the STA (macbook pro) and the AP did receive an ACK; but a tcpdump on the receiving side showed the EAPOL frame never made it. So the stack was dropping it for some reason; * Importantly - the EAPOL frames are currently going into the non-QoS TID, which maps to the BE queue and is susceptible to that queue being busy doing other things, but; * There's other traffic going on in the non-QoS TID from other contexts when scanning is going on and it's possible there's some races causing sequence number/IV issues, but; * Importantly importantlly, I think the interaction with TID 16 multicast traffic in power save mode is causing issues - since I -believe- the sequence number space being used by the EAPOL frames on TID 16 overlaps with the multicast frames that have sequence numbers allocated and are then stuffed on the cabq. Since with EAPOL frames being in TID 16 and queued to the BE queue, it's going to be waiting to be serviced with all of the aggregate traffic going on - and if the CABQ gets emptied beforehand, those TID 16 multicast frames with sequence numbers will go out beforehand. Now, there's quite likely a bunch of "stuff happening slightly out of sequence" going on due to the nature of the TX path (read: lots of overlapping and concurrent ath_start() and ath_raw_xmit() calls going on, sigh) but I thought I had caught them all and stuffed each TID TX behind a lock (that lasted as long as it needed to in order to get the frame onto the relevant destination queue - thus keeping things in order.) Unfortunately the last problem is the big one and I'm going to stare at it some more. If it _is_ So this is a work around for now to ensure that EAPOL frames actually make it out before any other stuff in the non-QoS TID and HOPEFULLY before the CABQ gets active. I'm now going to spend a little time in the TX path figuring out exactly why the sender is rejecting things. There's two (well, three if you count EAPOL contents invalid) possibilities: * The sequence number is out of order (ie, something else like the multicast traffic on CABQ) is going out first on TID 16; * The CCMP IV is out of order (similar to above - but less likely, as the TX key for multicast traffic is different to unicast traffic); * EAPOL contents strangely invalid. AP: Ubiquiti RSPRO, AR9160/AR9220 NICs STA: Macbook Pro, Broadcom 11n NIC
2012-09-26 03:45:42 +00:00
/*
* What queue to throw the non-QoS TID traffic into
*/
#define ATH_NONQOS_TID_AC WME_AC_VO
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
#if 0
static int ath_tx_node_is_asleep(struct ath_softc *sc, struct ath_node *an);
#endif
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
static int ath_tx_ampdu_pending(struct ath_softc *sc, struct ath_node *an,
int tid);
static int ath_tx_ampdu_running(struct ath_softc *sc, struct ath_node *an,
int tid);
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
static ieee80211_seq ath_tx_tid_seqno_assign(struct ath_softc *sc,
struct ieee80211_node *ni, struct ath_buf *bf, struct mbuf *m0);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
static int ath_tx_action_frame_override_queue(struct ath_softc *sc,
struct ieee80211_node *ni, struct mbuf *m0, int *tid);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
static struct ath_buf *
ath_tx_retry_clone(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid, struct ath_buf *bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ALQ logging enhancements: * upon setup, tell the alq code what the chip information is. * add TX/RX path logging for legacy chips. * populate the tx/rx descriptor length fields with a best-estimate. It's overly big (96 bytes when AH_SUPPORT_AR5416 is enabled) but it'll do for now. Whilst I'm here, add CURVNET_RESTORE() here during probe/attach as a partial solution to fixing crashes during attach when the attach fails. There are other attach failures that I have to deal with; those'll come later.
2012-11-16 19:57:16 +00:00
#ifdef ATH_DEBUG_ALQ
void
ath_tx_alq_post(struct ath_softc *sc, struct ath_buf *bf_first)
{
struct ath_buf *bf;
int i, n;
const char *ds;
/* XXX we should skip out early if debugging isn't enabled! */
bf = bf_first;
while (bf != NULL) {
/* XXX should ensure bf_nseg > 0! */
if (bf->bf_nseg == 0)
break;
n = ((bf->bf_nseg - 1) / sc->sc_tx_nmaps) + 1;
for (i = 0, ds = (const char *) bf->bf_desc;
i < n;
i++, ds += sc->sc_tx_desclen) {
if_ath_alq_post(&sc->sc_alq,
ATH_ALQ_EDMA_TXDESC,
sc->sc_tx_desclen,
ds);
}
bf = bf->bf_next;
}
}
#endif /* ATH_DEBUG_ALQ */
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
/*
* Whether to use the 11n rate scenario functions or not
*/
static inline int
ath_tx_is_11n(struct ath_softc *sc)
{
Add the AR9300 HAL ID in to the 11n check routine. I was having TX hang issues, which I root caused to having the legacy ath_hal_setupxtxdesc() called, rather than the 11n rate scenario setup code. This meant that rate control information wasn't being put into frames, causing the MAC to stall/hang.
2012-08-11 22:25:28 +00:00
return ((sc->sc_ah->ah_magic == 0x20065416) ||
(sc->sc_ah->ah_magic == 0x19741014));
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Obtain the current TID from the given frame.
*
* Non-QoS frames need to go into TID 16 (IEEE80211_NONQOS_TID.)
* This has implications for which AC/priority the packet is placed
* in.
*/
static int
ath_tx_gettid(struct ath_softc *sc, const struct mbuf *m0)
{
const struct ieee80211_frame *wh;
int pri = M_WME_GETAC(m0);
wh = mtod(m0, const struct ieee80211_frame *);
if (! IEEE80211_QOS_HAS_SEQ(wh))
return IEEE80211_NONQOS_TID;
else
return WME_AC_TO_TID(pri);
}
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
static void
ath_tx_set_retry(struct ath_softc *sc, struct ath_buf *bf)
{
struct ieee80211_frame *wh;
wh = mtod(bf->bf_m, struct ieee80211_frame *);
/* Only update/resync if needed */
if (bf->bf_state.bfs_isretried == 0) {
wh->i_fc[1] |= IEEE80211_FC1_RETRY;
bus_dmamap_sync(sc->sc_dmat, bf->bf_dmamap,
BUS_DMASYNC_PREWRITE);
}
bf->bf_state.bfs_isretried = 1;
bf->bf_state.bfs_retries ++;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Determine what the correct AC queue for the given frame
* should be.
*
* This code assumes that the TIDs map consistently to
* the underlying hardware (or software) ath_txq.
* Since the sender may try to set an AC which is
* arbitrary, non-QoS TIDs may end up being put on
* completely different ACs. There's no way to put a
* TID into multiple ath_txq's for scheduling, so
* for now we override the AC/TXQ selection and set
* non-QOS TID frames into the BE queue.
*
* This may be completely incorrect - specifically,
* some management frames may end up out of order
* compared to the QoS traffic they're controlling.
* I'll look into this later.
*/
static int
ath_tx_getac(struct ath_softc *sc, const struct mbuf *m0)
{
const struct ieee80211_frame *wh;
int pri = M_WME_GETAC(m0);
wh = mtod(m0, const struct ieee80211_frame *);
if (IEEE80211_QOS_HAS_SEQ(wh))
return pri;
Map the non-QoS TID to the voice queue, in order to ensure important things like EAPOL frames make it out. After a whole bunch of hacking/testing, I discovered that they weren't being early-dropped by the stack (but I should look at ensuring that later..) but were even making to the hardware transmit queue. They were mostly even being received by the remote end. However, the remote end was completely ignoring them. This didn't happen under 150-170MBit TCP tests as I'm guessing the TX queue stayed very busy and the STA didn't do any scanning. However, when doing 100Mbit/s of TCP traffic, the STA would do background scanning - which involves it coming in and out of powersave mode with the AP. Now, this is a total and utter hack around the real problems, which are: * I need to implement proper power save handling and integrate it into the filtered frames support, so the driver/stack doesn't send frames whilst the station is actually in sleep; * .. but frames were actually making it to the STA (macbook pro) and the AP did receive an ACK; but a tcpdump on the receiving side showed the EAPOL frame never made it. So the stack was dropping it for some reason; * Importantly - the EAPOL frames are currently going into the non-QoS TID, which maps to the BE queue and is susceptible to that queue being busy doing other things, but; * There's other traffic going on in the non-QoS TID from other contexts when scanning is going on and it's possible there's some races causing sequence number/IV issues, but; * Importantly importantlly, I think the interaction with TID 16 multicast traffic in power save mode is causing issues - since I -believe- the sequence number space being used by the EAPOL frames on TID 16 overlaps with the multicast frames that have sequence numbers allocated and are then stuffed on the cabq. Since with EAPOL frames being in TID 16 and queued to the BE queue, it's going to be waiting to be serviced with all of the aggregate traffic going on - and if the CABQ gets emptied beforehand, those TID 16 multicast frames with sequence numbers will go out beforehand. Now, there's quite likely a bunch of "stuff happening slightly out of sequence" going on due to the nature of the TX path (read: lots of overlapping and concurrent ath_start() and ath_raw_xmit() calls going on, sigh) but I thought I had caught them all and stuffed each TID TX behind a lock (that lasted as long as it needed to in order to get the frame onto the relevant destination queue - thus keeping things in order.) Unfortunately the last problem is the big one and I'm going to stare at it some more. If it _is_ So this is a work around for now to ensure that EAPOL frames actually make it out before any other stuff in the non-QoS TID and HOPEFULLY before the CABQ gets active. I'm now going to spend a little time in the TX path figuring out exactly why the sender is rejecting things. There's two (well, three if you count EAPOL contents invalid) possibilities: * The sequence number is out of order (ie, something else like the multicast traffic on CABQ) is going out first on TID 16; * The CCMP IV is out of order (similar to above - but less likely, as the TX key for multicast traffic is different to unicast traffic); * EAPOL contents strangely invalid. AP: Ubiquiti RSPRO, AR9160/AR9220 NICs STA: Macbook Pro, Broadcom 11n NIC
2012-09-26 03:45:42 +00:00
return ATH_NONQOS_TID_AC;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
void
ath_txfrag_cleanup(struct ath_softc *sc,
ath_bufhead *frags, struct ieee80211_node *ni)
{
struct ath_buf *bf, *next;
ATH_TXBUF_LOCK_ASSERT(sc);
Migrate the STAILQ lists to TAILQs. A bunch of the 11n TX aggregation logic wants to traverse lists of buffers in various ways. In order to provide O(1) behaviour in this instance, use TAILQs. This does blow out the memory footprint and CPU cycles slightly for some of these operations. I may convert some of these back to STAILQs once the rest of the software transmit queue handling has been stabilised. Sponsored by: Hobnob, Inc.
2011-11-08 17:08:12 +00:00
TAILQ_FOREACH_SAFE(bf, frags, bf_list, next) {
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/* NB: bf assumed clean */
Migrate the STAILQ lists to TAILQs. A bunch of the 11n TX aggregation logic wants to traverse lists of buffers in various ways. In order to provide O(1) behaviour in this instance, use TAILQs. This does blow out the memory footprint and CPU cycles slightly for some of these operations. I may convert some of these back to STAILQs once the rest of the software transmit queue handling has been stabilised. Sponsored by: Hobnob, Inc.
2011-11-08 17:08:12 +00:00
TAILQ_REMOVE(frags, bf, bf_list);
Replace the direct sc_txbuf manipulation with a pair of functions. This is preparation work for having a separate ath_buf queue for management traffic. PR: kern/168170
2012-06-13 05:39:16 +00:00
ath_returnbuf_head(sc, bf);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
ieee80211_node_decref(ni);
}
}
/*
* Setup xmit of a fragmented frame. Allocate a buffer
* for each frag and bump the node reference count to
* reflect the held reference to be setup by ath_tx_start.
*/
int
ath_txfrag_setup(struct ath_softc *sc, ath_bufhead *frags,
struct mbuf *m0, struct ieee80211_node *ni)
{
struct mbuf *m;
struct ath_buf *bf;
ATH_TXBUF_LOCK(sc);
for (m = m0->m_nextpkt; m != NULL; m = m->m_nextpkt) {
Implement a separate, smaller pool of ath_buf entries for use by management traffic. * Create sc_mgmt_txbuf and sc_mgmt_txdesc, initialise/free them appropriately. * Create an enum to represent buffer types in the API. * Extend ath_getbuf() and _ath_getbuf_locked() to take the above enum. * Right now anything sent via ic_raw_xmit() allocates via ATH_BUFTYPE_MGMT. This may not be very useful. * Add ATH_BUF_MGMT flag (ath_buf.bf_flags) which indicates the current buffer is a mgmt buffer and should go back onto the mgmt free list. * Extend 'txagg' to include debugging output for both normal and mgmt txbufs. * When checking/clearing ATH_BUF_BUSY, do it on both TX pools. Tested: * STA mode, with heavy UDP injection via iperf. This filled the TX queue however BARs were still going out successfully. TODO: * Initialise the mgmt buffers with ATH_BUF_MGMT and then ensure the right type is being allocated and freed on the appropriate list. That'd save a write operation (to bf->bf_flags) on each buffer alloc/free. * Test on AP mode, ensure that BAR TX and probe responses go out nicely when the main TX queue is filled (eg with paused traffic to a TID, awaiting a BAR to complete.) PR: kern/168170
2012-06-13 06:57:55 +00:00
/* XXX non-management? */
bf = _ath_getbuf_locked(sc, ATH_BUFTYPE_NORMAL);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if (bf == NULL) { /* out of buffers, cleanup */
After reviewing the mcast/sleep station code a little, undo some brain damage which I committed when I had less clue about such things. Don't ever put normal data frames on the mcast software queue. Just put mcast frames there if needed. Pass the txq decision into ath_tx_normal_setup(), as we've already made the decision. Don't re-do it. Whilst i'm here, add another random debugging statement.
2012-04-08 00:40:16 +00:00
device_printf(sc->sc_dev, "%s: no buffer?\n",
__func__);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
ath_txfrag_cleanup(sc, frags, ni);
break;
}
ieee80211_node_incref(ni);
Migrate the STAILQ lists to TAILQs. A bunch of the 11n TX aggregation logic wants to traverse lists of buffers in various ways. In order to provide O(1) behaviour in this instance, use TAILQs. This does blow out the memory footprint and CPU cycles slightly for some of these operations. I may convert some of these back to STAILQs once the rest of the software transmit queue handling has been stabilised. Sponsored by: Hobnob, Inc.
2011-11-08 17:08:12 +00:00
TAILQ_INSERT_TAIL(frags, bf, bf_list);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
ATH_TXBUF_UNLOCK(sc);
Migrate the STAILQ lists to TAILQs. A bunch of the 11n TX aggregation logic wants to traverse lists of buffers in various ways. In order to provide O(1) behaviour in this instance, use TAILQs. This does blow out the memory footprint and CPU cycles slightly for some of these operations. I may convert some of these back to STAILQs once the rest of the software transmit queue handling has been stabilised. Sponsored by: Hobnob, Inc.
2011-11-08 17:08:12 +00:00
return !TAILQ_EMPTY(frags);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
/*
* Reclaim mbuf resources. For fragmented frames we
* need to claim each frag chained with m_nextpkt.
*/
void
ath_freetx(struct mbuf *m)
{
struct mbuf *next;
do {
next = m->m_nextpkt;
m->m_nextpkt = NULL;
m_freem(m);
} while ((m = next) != NULL);
}
static int
ath_tx_dmasetup(struct ath_softc *sc, struct ath_buf *bf, struct mbuf *m0)
{
struct mbuf *m;
int error;
/*
* Load the DMA map so any coalescing is done. This
* also calculates the number of descriptors we need.
*/
error = bus_dmamap_load_mbuf_sg(sc->sc_dmat, bf->bf_dmamap, m0,
bf->bf_segs, &bf->bf_nseg,
BUS_DMA_NOWAIT);
if (error == EFBIG) {
/* XXX packet requires too many descriptors */
bf->bf_nseg = ATH_TXDESC+1;
} else if (error != 0) {
sc->sc_stats.ast_tx_busdma++;
ath_freetx(m0);
return error;
}
/*
* Discard null packets and check for packets that
* require too many TX descriptors. We try to convert
* the latter to a cluster.
*/
if (bf->bf_nseg > ATH_TXDESC) { /* too many desc's, linearize */
sc->sc_stats.ast_tx_linear++;
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_collapse(m0, M_NOWAIT, ATH_TXDESC);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if (m == NULL) {
ath_freetx(m0);
sc->sc_stats.ast_tx_nombuf++;
return ENOMEM;
}
m0 = m;
error = bus_dmamap_load_mbuf_sg(sc->sc_dmat, bf->bf_dmamap, m0,
bf->bf_segs, &bf->bf_nseg,
BUS_DMA_NOWAIT);
if (error != 0) {
sc->sc_stats.ast_tx_busdma++;
ath_freetx(m0);
return error;
}
KASSERT(bf->bf_nseg <= ATH_TXDESC,
("too many segments after defrag; nseg %u", bf->bf_nseg));
} else if (bf->bf_nseg == 0) { /* null packet, discard */
sc->sc_stats.ast_tx_nodata++;
ath_freetx(m0);
return EIO;
}
DPRINTF(sc, ATH_DEBUG_XMIT, "%s: m %p len %u\n",
__func__, m0, m0->m_pkthdr.len);
bus_dmamap_sync(sc->sc_dmat, bf->bf_dmamap, BUS_DMASYNC_PREWRITE);
bf->bf_m = m0;
return 0;
}
Change the descriptor logic to use bf_lastds to point to the last descriptor, rather than using the maths involving bf_desc[bf_nseg - 1]. When doing TX aggregation, the status will be updated in the -final- descriptor of the -final- subframe in an aggregate. Thus bf_lastds may point to the last descriptor in a completely different ath_buf. Sponsored by: Hobnob, Inc.
2011-11-08 21:25:36 +00:00
/*
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
* Chain together segments+descriptors for a frame - 11n or otherwise.
*
* For aggregates, this is called on each frame in the aggregate.
Change the descriptor logic to use bf_lastds to point to the last descriptor, rather than using the maths involving bf_desc[bf_nseg - 1]. When doing TX aggregation, the status will be updated in the -final- descriptor of the -final- subframe in an aggregate. Thus bf_lastds may point to the last descriptor in a completely different ath_buf. Sponsored by: Hobnob, Inc.
2011-11-08 21:25:36 +00:00
*/
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
static void
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
ath_tx_chaindesclist(struct ath_softc *sc, struct ath_desc *ds0,
struct ath_buf *bf, int is_aggr, int is_first_subframe,
int is_last_subframe)
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
{
struct ath_hal *ah = sc->sc_ah;
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
char *ds;
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
int i, bp, dsp;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
HAL_DMA_ADDR bufAddrList[4];
uint32_t segLenList[4];
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
int numTxMaps = 1;
When assembling the descriptor list, make sure that the "first" descriptor is marked correctly. The existing logic assumed that the first descriptor is i == 0, which doesn't hold for EDMA TX. In this instance, the first time filltxdesc() is called can be up to i == 3. So for a two-buffer descriptor: * firstSeg is set to 0; * lastSeg is set to 1; * the ath_hal_filltxdesc() code will treat it as the last segment in a descriptor chain and blank some of the descriptor fields, causing the TX to stop. When firstSeg is set to 1 (regardless of lastSeg), it overrides the lastSeg setting. Thus, ath_hal_filltxdesc() won't blank out these fields. Tested: AR9380, STA mode. With this, association is successful.
2012-08-19 02:16:22 +00:00
int isFirstDesc = 1;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
Begin modifying the descriptor allocation functions to support a variable sized TX descriptor. This is required for the AR93xx EDMA support which requires 128 byte TX descriptors (which is significantly larger than the earlier hardware.)
2012-07-23 02:26:33 +00:00
/*
* XXX There's txdma and txdma_mgmt; the descriptor
* sizes must match.
*/
struct ath_descdma *dd = &sc->sc_txdma;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
* Fillin the remainder of the descriptor info.
*/
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
/*
Now that the tx map field is correctly populated for both edma and legacy chips, just use that.
2013-03-19 17:54:37 +00:00
* We need the number of TX data pointers in each descriptor.
* EDMA and later chips support 4 TX buffers per descriptor;
* previous chips just support one.
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
*/
Now that the tx map field is correctly populated for both edma and legacy chips, just use that.
2013-03-19 17:54:37 +00:00
numTxMaps = sc->sc_tx_nmaps;
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
/*
* For EDMA and later chips ensure the TX map is fully populated
* before advancing to the next descriptor.
*/
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
ds = (char *) bf->bf_desc;
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
bp = dsp = 0;
bzero(bufAddrList, sizeof(bufAddrList));
bzero(segLenList, sizeof(segLenList));
for (i = 0; i < bf->bf_nseg; i++) {
bufAddrList[bp] = bf->bf_segs[i].ds_addr;
segLenList[bp] = bf->bf_segs[i].ds_len;
bp++;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
/*
* Go to the next segment if this isn't the last segment
* and there's space in the current TX map.
*/
if ((i != bf->bf_nseg - 1) && (bp < numTxMaps))
continue;
/*
* Last segment or we're out of buffer pointers.
*/
bp = 0;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if (i == bf->bf_nseg - 1)
Advance the descriptor pointer by sc->sc_tx_desclen bytes, rather than sizeof(struct ath_desc). This isn't correct for EDMA TX descriptors. This popped up during iperf tests. Ping tests never created frames that had enough segments to overflow into a second descriptor. However, an iperf TCP test would do that after a few seconds; the second descriptor would almost always certainly have garbage. Tested: * AR9380, STA mode * AR9280, STA mode (802.11n TX, legacy TX)
2012-08-20 06:02:09 +00:00
ath_hal_settxdesclink(ah, (struct ath_desc *) ds, 0);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
else
Advance the descriptor pointer by sc->sc_tx_desclen bytes, rather than sizeof(struct ath_desc). This isn't correct for EDMA TX descriptors. This popped up during iperf tests. Ping tests never created frames that had enough segments to overflow into a second descriptor. However, an iperf TCP test would do that after a few seconds; the second descriptor would almost always certainly have garbage. Tested: * AR9380, STA mode * AR9280, STA mode (802.11n TX, legacy TX)
2012-08-20 06:02:09 +00:00
ath_hal_settxdesclink(ah, (struct ath_desc *) ds,
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
bf->bf_daddr + dd->dd_descsize * (dsp + 1));
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
/*
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
* XXX This assumes that bfs_txq is the actual destination
* hardware queue at this point. It may not have been
* assigned, it may actually be pointing to the multicast
* software TXQ id. These must be fixed!
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
*/
Advance the descriptor pointer by sc->sc_tx_desclen bytes, rather than sizeof(struct ath_desc). This isn't correct for EDMA TX descriptors. This popped up during iperf tests. Ping tests never created frames that had enough segments to overflow into a second descriptor. However, an iperf TCP test would do that after a few seconds; the second descriptor would almost always certainly have garbage. Tested: * AR9380, STA mode * AR9280, STA mode (802.11n TX, legacy TX)
2012-08-20 06:02:09 +00:00
ath_hal_filltxdesc(ah, (struct ath_desc *) ds
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
, bufAddrList
, segLenList
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
, bf->bf_descid /* XXX desc id */
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
, bf->bf_state.bfs_tx_queue
When assembling the descriptor list, make sure that the "first" descriptor is marked correctly. The existing logic assumed that the first descriptor is i == 0, which doesn't hold for EDMA TX. In this instance, the first time filltxdesc() is called can be up to i == 3. So for a two-buffer descriptor: * firstSeg is set to 0; * lastSeg is set to 1; * the ath_hal_filltxdesc() code will treat it as the last segment in a descriptor chain and blank some of the descriptor fields, causing the TX to stop. When firstSeg is set to 1 (regardless of lastSeg), it overrides the lastSeg setting. Thus, ath_hal_filltxdesc() won't blank out these fields. Tested: AR9380, STA mode. With this, association is successful.
2012-08-19 02:16:22 +00:00
, isFirstDesc /* first segment */
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
, i == bf->bf_nseg - 1 /* last segment */
Advance the descriptor pointer by sc->sc_tx_desclen bytes, rather than sizeof(struct ath_desc). This isn't correct for EDMA TX descriptors. This popped up during iperf tests. Ping tests never created frames that had enough segments to overflow into a second descriptor. However, an iperf TCP test would do that after a few seconds; the second descriptor would almost always certainly have garbage. Tested: * AR9380, STA mode * AR9280, STA mode (802.11n TX, legacy TX)
2012-08-20 06:02:09 +00:00
, (struct ath_desc *) ds0 /* first descriptor */
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
);
Make sure the aggregate fields are properly cleared - both in the ath_buf and when forming a non-aggregate frame. The non-11n setds function is called when TXing aggregate frames (and yes, I should fix this!) and the non-11n TX aggregation code doesn't clear the delimiter field. I figure it's nicer to do that.
2012-09-09 05:06:16 +00:00
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
/*
* Make sure the 11n aggregate fields are cleared.
*
* XXX TODO: this doesn't need to be called for
* aggregate frames; as it'll be called on all
* sub-frames. Since the descriptors are in
* non-cacheable memory, this leads to some
* rather slow writes on MIPS/ARM platforms.
*/
Make sure the aggregate fields are properly cleared - both in the ath_buf and when forming a non-aggregate frame. The non-11n setds function is called when TXing aggregate frames (and yes, I should fix this!) and the non-11n TX aggregation code doesn't clear the delimiter field. I figure it's nicer to do that.
2012-09-09 05:06:16 +00:00
if (ath_tx_is_11n(sc))
Clear the correct descriptor when going through the chained together gather DMA descriptor list. Pointy hat to: adrian@, for even USING bf->bf_desc here instead of 'ds'.
2012-09-11 04:11:42 +00:00
ath_hal_clr11n_aggr(sc->sc_ah, (struct ath_desc *) ds);
Make sure the aggregate fields are properly cleared - both in the ath_buf and when forming a non-aggregate frame. The non-11n setds function is called when TXing aggregate frames (and yes, I should fix this!) and the non-11n TX aggregation code doesn't clear the delimiter field. I figure it's nicer to do that.
2012-09-09 05:06:16 +00:00
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
/*
* If 11n is enabled, set it up as if it's an aggregate
* frame.
*/
if (is_last_subframe) {
ath_hal_set11n_aggr_last(sc->sc_ah,
(struct ath_desc *) ds);
} else if (is_aggr) {
/*
* This clears the aggrlen field; so
* the caller needs to call set_aggr_first()!
*
* XXX TODO: don't call this for the first
* descriptor in the first frame in an
* aggregate!
*/
ath_hal_set11n_aggr_middle(sc->sc_ah,
(struct ath_desc *) ds,
bf->bf_state.bfs_ndelim);
}
When assembling the descriptor list, make sure that the "first" descriptor is marked correctly. The existing logic assumed that the first descriptor is i == 0, which doesn't hold for EDMA TX. In this instance, the first time filltxdesc() is called can be up to i == 3. So for a two-buffer descriptor: * firstSeg is set to 0; * lastSeg is set to 1; * the ath_hal_filltxdesc() code will treat it as the last segment in a descriptor chain and blank some of the descriptor fields, causing the TX to stop. When firstSeg is set to 1 (regardless of lastSeg), it overrides the lastSeg setting. Thus, ath_hal_filltxdesc() won't blank out these fields. Tested: AR9380, STA mode. With this, association is successful.
2012-08-19 02:16:22 +00:00
isFirstDesc = 0;
Advance the descriptor pointer by sc->sc_tx_desclen bytes, rather than sizeof(struct ath_desc). This isn't correct for EDMA TX descriptors. This popped up during iperf tests. Ping tests never created frames that had enough segments to overflow into a second descriptor. However, an iperf TCP test would do that after a few seconds; the second descriptor would almost always certainly have garbage. Tested: * AR9380, STA mode * AR9280, STA mode (802.11n TX, legacy TX)
2012-08-20 06:02:09 +00:00
bf->bf_lastds = (struct ath_desc *) ds;
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
/*
* Don't forget to skip to the next descriptor.
*/
Advance the descriptor pointer by sc->sc_tx_desclen bytes, rather than sizeof(struct ath_desc). This isn't correct for EDMA TX descriptors. This popped up during iperf tests. Ping tests never created frames that had enough segments to overflow into a second descriptor. However, an iperf TCP test would do that after a few seconds; the second descriptor would almost always certainly have garbage. Tested: * AR9380, STA mode * AR9280, STA mode (802.11n TX, legacy TX)
2012-08-20 06:02:09 +00:00
ds += sc->sc_tx_desclen;
Extend the non-aggregate TX descriptor chain routine to be aware of: * the descriptor ID, and * the multi-buffer support that the EDMA chips support. This is required for successful MAC transmission of multi-descriptor frames. The MAC simply hangs if there are NULL buffers + 0 length pointers, but the descriptor did have TxMore set. This won't be done for the 11n aggregate path, as that will be modified to use the newer API (ie, ath_hal_filltxdesc() and then set first|middle| last_aggr), which will deprecate some of the current code. TODO: * Populate the numTxMaps field in the HAL, then make sure that's fetched by the driver. Then I can undo that hack. Tested: * AR9380, AP mode, TX'ing non-aggregate 802.11n frames; * AR9280, STA/AP mode, doing aggregate and non-aggregate traffic.
2012-08-15 08:14:16 +00:00
dsp++;
/*
* .. and don't forget to blank these out!
*/
bzero(bufAddrList, sizeof(bufAddrList));
bzero(segLenList, sizeof(segLenList));
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
Do a dma sync before the descriptors are chained together. I need to find a better place to do this..
2012-04-07 05:51:43 +00:00
bus_dmamap_sync(sc->sc_dmat, bf->bf_dmamap, BUS_DMASYNC_PREWRITE);
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
}
Push the rate control and descriptor chaining into the descriptor "set" functions, for both legacy and 802.11n. This will simplify supporting the EDMA chipsets as these two descriptor setup functions can just be overridden in their entirety, hiding all of the subtle differences in setting things up. It's not a permanent solution, as eventually the AR5416 HAL should grow similar versions of the 11n descriptor functions and then those can be used. TODO: * Push the "clr11naggr" call into the legacy setds, just to ensure that retried frames don't end up with the aggregate bits set inappropriately; * Remove the "setlasttxdesc" call from the 11n TX path and push it into setds_11n. * Ensure that setds_11n will work correctly for non-aggregate frames; * .. and then when it does, just unconditionally call "setds_11n" for 11n NICs and "setds" for non-11n NICs.
2012-07-31 16:41:09 +00:00
/*
* Set the rate control fields in the given descriptor based on
* the bf_state fields and node state.
*
* The bfs fields should already be set with the relevant rate
* control information, including whether MRR is to be enabled.
*
* Since the FreeBSD HAL currently sets up the first TX rate
* in ath_hal_setuptxdesc(), this will setup the MRR
* conditionally for the pre-11n chips, and call ath_buf_set_rate
* unconditionally for 11n chips. These require the 11n rate
* scenario to be set if MCS rates are enabled, so it's easier
* to just always call it. The caller can then only set rates 2, 3
* and 4 if multi-rate retry is needed.
*/
static void
ath_tx_set_ratectrl(struct ath_softc *sc, struct ieee80211_node *ni,
struct ath_buf *bf)
{
struct ath_rc_series *rc = bf->bf_state.bfs_rc;
/* If mrr is disabled, blank tries 1, 2, 3 */
if (! bf->bf_state.bfs_ismrr)
rc[1].tries = rc[2].tries = rc[3].tries = 0;
Until I figure out what to do here, remind myself that this needs some rate control 'adjustment' when NOACK is set.
2012-11-28 06:55:34 +00:00
#if 0
/*
* If NOACK is set, just set ntries=1.
*/
else if (bf->bf_state.bfs_txflags & HAL_TXDESC_NOACK) {
rc[1].tries = rc[2].tries = rc[3].tries = 0;
rc[0].tries = 1;
}
#endif
Push the rate control and descriptor chaining into the descriptor "set" functions, for both legacy and 802.11n. This will simplify supporting the EDMA chipsets as these two descriptor setup functions can just be overridden in their entirety, hiding all of the subtle differences in setting things up. It's not a permanent solution, as eventually the AR5416 HAL should grow similar versions of the 11n descriptor functions and then those can be used. TODO: * Push the "clr11naggr" call into the legacy setds, just to ensure that retried frames don't end up with the aggregate bits set inappropriately; * Remove the "setlasttxdesc" call from the 11n TX path and push it into setds_11n. * Ensure that setds_11n will work correctly for non-aggregate frames; * .. and then when it does, just unconditionally call "setds_11n" for 11n NICs and "setds" for non-11n NICs.
2012-07-31 16:41:09 +00:00
/*
* Always call - that way a retried descriptor will
* have the MRR fields overwritten.
*
* XXX TODO: see if this is really needed - setting up
* the first descriptor should set the MRR fields to 0
* for us anyway.
*/
if (ath_tx_is_11n(sc)) {
ath_buf_set_rate(sc, ni, bf);
} else {
ath_hal_setupxtxdesc(sc->sc_ah, bf->bf_desc
, rc[1].ratecode, rc[1].tries
, rc[2].ratecode, rc[2].tries
, rc[3].ratecode, rc[3].tries
);
}
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Setup segments+descriptors for an 11n aggregate.
* bf_first is the first buffer in the aggregate.
* The descriptor list must already been linked together using
* bf->bf_next.
*/
static void
ath_tx_setds_11n(struct ath_softc *sc, struct ath_buf *bf_first)
{
struct ath_buf *bf, *bf_prev = NULL;
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
struct ath_desc *ds0 = bf_first->bf_desc;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR, "%s: nframes=%d, al=%d\n",
__func__, bf_first->bf_state.bfs_nframes,
bf_first->bf_state.bfs_al);
Add some debugging to try and catch an invalid TX rate (0x0) that is being reported.
2012-11-13 06:28:57 +00:00
bf = bf_first;
if (bf->bf_state.bfs_txrate0 == 0)
device_printf(sc->sc_dev, "%s: bf=%p, txrate0=%d\n",
__func__, bf, 0);
if (bf->bf_state.bfs_rc[0].ratecode == 0)
device_printf(sc->sc_dev, "%s: bf=%p, rix0=%d\n",
__func__, bf, 0);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
* Setup all descriptors of all subframes - this will
* call ath_hal_set11naggrmiddle() on every frame.
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*/
while (bf != NULL) {
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
"%s: bf=%p, nseg=%d, pktlen=%d, seqno=%d\n",
__func__, bf, bf->bf_nseg, bf->bf_state.bfs_pktlen,
SEQNO(bf->bf_state.bfs_seqno));
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
/*
* Setup the initial fields for the first descriptor - all
* the non-11n specific stuff.
*/
ath_hal_setuptxdesc(sc->sc_ah, bf->bf_desc
, bf->bf_state.bfs_pktlen /* packet length */
, bf->bf_state.bfs_hdrlen /* header length */
, bf->bf_state.bfs_atype /* Atheros packet type */
, bf->bf_state.bfs_txpower /* txpower */
, bf->bf_state.bfs_txrate0
, bf->bf_state.bfs_try0 /* series 0 rate/tries */
, bf->bf_state.bfs_keyix /* key cache index */
, bf->bf_state.bfs_txantenna /* antenna mode */
, bf->bf_state.bfs_txflags | HAL_TXDESC_INTREQ /* flags */
, bf->bf_state.bfs_ctsrate /* rts/cts rate */
, bf->bf_state.bfs_ctsduration /* rts/cts duration */
);
/*
* First descriptor? Setup the rate control and initial
* aggregate header information.
*/
if (bf == bf_first) {
/*
* setup first desc with rate and aggr info
*/
ath_tx_set_ratectrl(sc, bf->bf_node, bf);
}
/*
* Setup the descriptors for a multi-descriptor frame.
* This is both aggregate and non-aggregate aware.
*/
ath_tx_chaindesclist(sc, ds0, bf,
1, /* is_aggr */
!! (bf == bf_first), /* is_first_subframe */
!! (bf->bf_next == NULL) /* is_last_subframe */
);
if (bf == bf_first) {
/*
* Initialise the first 11n aggregate with the
* aggregate length and aggregate enable bits.
*/
ath_hal_set11n_aggr_first(sc->sc_ah,
ds0,
bf->bf_state.bfs_al,
bf->bf_state.bfs_ndelim);
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Link the last descriptor of the previous frame
* to the beginning descriptor of this frame.
*/
if (bf_prev != NULL)
Convert the TX path to use the new HAL methods for accessing the TX descriptor link pointers. This is required for the AR93xx and later chipsets. The RX path is slightly different - the legacy RX path directly accesses ath_desc->ds_link for now, however this isn't at all done for EDMA (FIFO) RX. Now, for those performing a little software archeology here: This is all a bit sub-optimal. "struct ath_desc" is only really relevant for the pre-AR93xx NICs - where ds_link and ds_data is always in the same location. The AR93xx and later NICs have different descriptor layouts altogether. Now, for AR93xx and later NICs, you should never directly reference ds_link and ds_data, as: * the RX descriptors don't have either - the data is _after_ the RX descriptor. They're just one large buffer. There's also no need for a per-descriptor RX buffer size as they're all fixed sizes. * the TX descriptors have 4 buffer and 4 length fields _and_ a link pointer. Each frame takes up one TX FIFO pointer, but it can contain multiple subframes (either multiple frames in a buffer, and/or multiple frames in an aggregate/RIFS burst.) * .. so, when TX frames are queued to a hardware queue, the link pointer is ONLY for buffers in that frame/aggregate. The next frame starts in a new FIFO pointer. * Finally, descriptor completion status is in a different ring. I'll write something up about that when its time to do so. This was inspired by Linux ath9k and the reference driver but is a reimplementation. Obtained from: Linux ath9k, Qualcomm Atheros
2012-07-19 03:51:16 +00:00
ath_hal_settxdesclink(sc->sc_ah, bf_prev->bf_lastds,
bf->bf_daddr);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Save a copy so we can link the next descriptor in */
bf_prev = bf;
bf = bf->bf_next;
}
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Set the first descriptor bf_lastds field to point to
* the last descriptor in the last subframe, that's where
* the status update will occur.
*/
bf_first->bf_lastds = bf_prev->bf_lastds;
/*
* And bf_last in the first descriptor points to the end of
* the aggregate list.
*/
bf_first->bf_last = bf_prev;
Make sure the final descriptor in an aggregate has rate control information. This was broken by me when merging the 802.11n aggregate descriptor chain setup with the default descriptor chain setup, in preparation for supporting AR9380 NICs. The corner case here is quite specific - if you queue an aggregate frame with >1 frames in it, and the last subframe has only one descriptor making it up, then that descriptor won't have the rate control information copied into it. Look at what happens inside ar5416FillTxDesc() if both firstSeg and lastSeg are set to 1. Then when ar5416ProcTxDesc() goes to fill out ts_rate based on the transmit index, it looks at the rate control fields in that descriptor and dutifully sets it to be 0. It doesn't happen for non-aggregate frames - if they have one descriptor, the first descriptor already has rate control info. I removed the call to ath_hal_setuplasttxdesc() when I migrated the code to use the "new" style aggregate chain routines from the HAL. But I missed this particular corner case. This is a bit inefficient with MIPS boards as it involves a few redundant writes into non-cachable memory. I'll chase that up when it matters. Tested: * AR9280 STA mode, TCP iperf traffic * Rui Paulo <rpaulo@> first reported this and has verified it on his AR9160 based AP. PR: kern/173636
2012-11-15 03:00:49 +00:00
/*
* For non-AR9300 NICs, which require the rate control
* in the final descriptor - let's set that up now.
*
* This is because the filltxdesc() HAL call doesn't
* populate the last segment with rate control information
* if firstSeg is also true. For non-aggregate frames
* that is fine, as the first frame already has rate control
* info. But if the last frame in an aggregate has one
* descriptor, both firstseg and lastseg will be true and
* the rate info isn't copied.
*
* This is inefficient on MIPS/ARM platforms that have
* non-cachable memory for TX descriptors, but we'll just
* make do for now.
*
* As to why the rate table is stashed in the last descriptor
* rather than the first descriptor? Because proctxdesc()
* is called on the final descriptor in an MPDU or A-MPDU -
* ie, the one that gets updated by the hardware upon
* completion. That way proctxdesc() doesn't need to know
* about the first _and_ last TX descriptor.
*/
ath_hal_setuplasttxdesc(sc->sc_ah, bf_prev->bf_lastds, ds0);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR, "%s: end\n", __func__);
}
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
/*
* Hand-off a frame to the multicast TX queue.
*
* This is a software TXQ which will be appended to the CAB queue
* during the beacon setup code.
*
* XXX TODO: since the AR9300 EDMA TX queue support wants the QCU ID
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
* as part of the TX descriptor, bf_state.bfs_tx_queue must be updated
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
* with the actual hardware txq, or all of this will fall apart.
*
* XXX It may not be a bad idea to just stuff the QCU ID into bf_state
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
* and retire bfs_tx_queue; then make sure the CABQ QCU ID is populated
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
* correctly.
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
static void
ath_tx_handoff_mcast(struct ath_softc *sc, struct ath_txq *txq,
struct ath_buf *bf)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
KASSERT((bf->bf_flags & ATH_BUF_BUSY) == 0,
("%s: busy status 0x%x", __func__, bf->bf_flags));
if (txq->axq_link != NULL) {
struct ath_buf *last = ATH_TXQ_LAST(txq, axq_q_s);
struct ieee80211_frame *wh;
/* mark previous frame */
wh = mtod(last->bf_m, struct ieee80211_frame *);
wh->i_fc[1] |= IEEE80211_FC1_MORE_DATA;
bus_dmamap_sync(sc->sc_dmat, last->bf_dmamap,
BUS_DMASYNC_PREWRITE);
/* link descriptor */
*txq->axq_link = bf->bf_daddr;
}
ATH_TXQ_INSERT_TAIL(txq, bf, bf_list);
Convert the TX path to use the new HAL methods for accessing the TX descriptor link pointers. This is required for the AR93xx and later chipsets. The RX path is slightly different - the legacy RX path directly accesses ath_desc->ds_link for now, however this isn't at all done for EDMA (FIFO) RX. Now, for those performing a little software archeology here: This is all a bit sub-optimal. "struct ath_desc" is only really relevant for the pre-AR93xx NICs - where ds_link and ds_data is always in the same location. The AR93xx and later NICs have different descriptor layouts altogether. Now, for AR93xx and later NICs, you should never directly reference ds_link and ds_data, as: * the RX descriptors don't have either - the data is _after_ the RX descriptor. They're just one large buffer. There's also no need for a per-descriptor RX buffer size as they're all fixed sizes. * the TX descriptors have 4 buffer and 4 length fields _and_ a link pointer. Each frame takes up one TX FIFO pointer, but it can contain multiple subframes (either multiple frames in a buffer, and/or multiple frames in an aggregate/RIFS burst.) * .. so, when TX frames are queued to a hardware queue, the link pointer is ONLY for buffers in that frame/aggregate. The next frame starts in a new FIFO pointer. * Finally, descriptor completion status is in a different ring. I'll write something up about that when its time to do so. This was inspired by Linux ath9k and the reference driver but is a reimplementation. Obtained from: Linux ath9k, Qualcomm Atheros
2012-07-19 03:51:16 +00:00
ath_hal_gettxdesclinkptr(sc->sc_ah, bf->bf_lastds, &txq->axq_link);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
/*
* Hand-off packet to a hardware queue.
*/
static void
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
ath_tx_handoff_hw(struct ath_softc *sc, struct ath_txq *txq,
struct ath_buf *bf)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
struct ath_hal *ah = sc->sc_ah;
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
* Insert the frame on the outbound list and pass it on
* to the hardware. Multicast frames buffered for power
* save stations and transmit from the CAB queue are stored
* on a s/w only queue and loaded on to the CAB queue in
* the SWBA handler since frames only go out on DTIM and
* to avoid possible races.
*/
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
KASSERT((bf->bf_flags & ATH_BUF_BUSY) == 0,
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
("%s: busy status 0x%x", __func__, bf->bf_flags));
KASSERT(txq->axq_qnum != ATH_TXQ_SWQ,
("ath_tx_handoff_hw called for mcast queue"));
Flesh out some slightly dirty reset/channel change serialisation code for the ath(4) driver. Currently, there's nothing stopping reset, channel change and general TX/RX from overlapping with each other. This wasn't a big deal with pre-11n traffic as it just results in some dropped frames. It's possible this may have also caused some inconsistencies and badly-setup hardware. Since locks can't be held across all of this (the Linux solution) due to LORs with the network stack locks, some state counter variables are used to track what parts of the code the driver is currently in. When the hardware is being reset, it disables the taskqueue and waits for pending interrupts, tx, rx and tx completion before it begins the reset or channel change. TX and RX both abort if called during an active reset or channel change. Finally, the reset path now doesn't flush frames if ATH_RESET_NOLOSS is set. Instead, completed TX and RX frames are passed back up to net80211 before the reset occurs. This is not without problems: * Raw frame xmit are just dropped, rather than placed on a queue. The net80211 stack should be the one which queues these frames rather than the driver. * It's all very messy. It'd be better if these hardware operations were serialised on some kind of work queue, rather than hoping they can be run in parallel. * The taskqueue block/unblock may occur in parallel with the newstate() function - which shuts down the taskqueue and restarts it once the new state is known. It's likely these operations should be refcounted so the taskqueue is restored once no other areas in the code wish to suspend operations. * .. interrupt disable/enable should likely be refcounted as well. With this work, the driver does not drop frames during stuck beacon or fatal errors and thus 11n traffic continues to run correctly. Default and full resets however do still drop frames and it's possible this may occur, causing traffic loss and session stalls. Sponsored by: Hobnob, Inc.
2011-11-18 05:06:30 +00:00
#if 0
/*
* This causes a LOR. Find out where the PCU lock is being
* held whilst the TXQ lock is grabbed - that shouldn't
* be occuring.
*/
ATH_PCU_LOCK(sc);
if (sc->sc_inreset_cnt) {
ATH_PCU_UNLOCK(sc);
DPRINTF(sc, ATH_DEBUG_RESET,
"%s: called with sc_in_reset != 0\n",
__func__);
DPRINTF(sc, ATH_DEBUG_XMIT,
"%s: queued: TXDP[%u] = %p (%p) depth %d\n",
__func__, txq->axq_qnum,
(caddr_t)bf->bf_daddr, bf->bf_desc,
txq->axq_depth);
ATH_TXQ_INSERT_TAIL(txq, bf, bf_list);
if (bf->bf_state.bfs_aggr)
txq->axq_aggr_depth++;
/*
* There's no need to update axq_link; the hardware
* is in reset and once the reset is complete, any
* non-empty queues will simply have DMA restarted.
*/
return;
}
ATH_PCU_UNLOCK(sc);
#endif
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* For now, so not to generate whitespace diffs */
if (1) {
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
#ifdef IEEE80211_SUPPORT_TDMA
int qbusy;
ATH_TXQ_INSERT_TAIL(txq, bf, bf_list);
qbusy = ath_hal_txqenabled(ah, txq->axq_qnum);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 4,
"ath_tx_handoff: txq=%u, add bf=%p, qbusy=%d, depth=%d",
txq->axq_qnum, bf, qbusy, txq->axq_depth);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if (txq->axq_link == NULL) {
/*
* Be careful writing the address to TXDP. If
* the tx q is enabled then this write will be
* ignored. Normally this is not an issue but
* when tdma is in use and the q is beacon gated
* this race can occur. If the q is busy then
* defer the work to later--either when another
* packet comes along or when we prepare a beacon
* frame at SWBA.
*/
if (!qbusy) {
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
ath_hal_puttxbuf(ah, txq->axq_qnum,
bf->bf_daddr);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
txq->axq_flags &= ~ATH_TXQ_PUTPENDING;
DPRINTF(sc, ATH_DEBUG_XMIT,
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
"%s: TXDP[%u] = %p (%p) lastds=%p depth %d\n",
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
__func__, txq->axq_qnum,
(caddr_t)bf->bf_daddr, bf->bf_desc,
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
bf->bf_lastds,
txq->axq_depth);
ATH_KTR(sc, ATH_KTR_TX, 5,
"ath_tx_handoff: TXDP[%u] = %p (%p) "
"lastds=%p depth %d",
txq->axq_qnum,
(caddr_t)bf->bf_daddr, bf->bf_desc,
bf->bf_lastds,
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
txq->axq_depth);
} else {
txq->axq_flags |= ATH_TXQ_PUTPENDING;
DPRINTF(sc, ATH_DEBUG_TDMA | ATH_DEBUG_XMIT,
"%s: Q%u busy, defer enable\n", __func__,
txq->axq_qnum);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 0, "defer enable");
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
} else {
*txq->axq_link = bf->bf_daddr;
DPRINTF(sc, ATH_DEBUG_XMIT,
"%s: link[%u](%p)=%p (%p) depth %d\n", __func__,
txq->axq_qnum, txq->axq_link,
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
(caddr_t)bf->bf_daddr, bf->bf_desc,
txq->axq_depth);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 5,
"ath_tx_handoff: link[%u](%p)=%p (%p) lastds=%p",
txq->axq_qnum, txq->axq_link,
(caddr_t)bf->bf_daddr, bf->bf_desc,
bf->bf_lastds);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if ((txq->axq_flags & ATH_TXQ_PUTPENDING) && !qbusy) {
/*
* The q was busy when we previously tried
* to write the address of the first buffer
* in the chain. Since it's not busy now
* handle this chore. We are certain the
* buffer at the front is the right one since
* axq_link is NULL only when the buffer list
* is/was empty.
*/
ath_hal_puttxbuf(ah, txq->axq_qnum,
Migrate the STAILQ lists to TAILQs. A bunch of the 11n TX aggregation logic wants to traverse lists of buffers in various ways. In order to provide O(1) behaviour in this instance, use TAILQs. This does blow out the memory footprint and CPU cycles slightly for some of these operations. I may convert some of these back to STAILQs once the rest of the software transmit queue handling has been stabilised. Sponsored by: Hobnob, Inc.
2011-11-08 17:08:12 +00:00
TAILQ_FIRST(&txq->axq_q)->bf_daddr);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
txq->axq_flags &= ~ATH_TXQ_PUTPENDING;
DPRINTF(sc, ATH_DEBUG_TDMA | ATH_DEBUG_XMIT,
"%s: Q%u restarted\n", __func__,
txq->axq_qnum);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 4,
"ath_tx_handoff: txq[%d] restarted, bf=%p "
"daddr=%p ds=%p",
txq->axq_qnum,
bf,
(caddr_t)bf->bf_daddr,
bf->bf_desc);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
}
#else
ATH_TXQ_INSERT_TAIL(txq, bf, bf_list);
Fix the non-TDMA build.
2012-10-13 06:27:34 +00:00
ATH_KTR(sc, ATH_KTR_TX, 3,
"ath_tx_handoff: non-tdma: txq=%u, add bf=%p "
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
"depth=%d",
txq->axq_qnum,
bf,
txq->axq_depth);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if (txq->axq_link == NULL) {
ath_hal_puttxbuf(ah, txq->axq_qnum, bf->bf_daddr);
DPRINTF(sc, ATH_DEBUG_XMIT,
"%s: TXDP[%u] = %p (%p) depth %d\n",
__func__, txq->axq_qnum,
(caddr_t)bf->bf_daddr, bf->bf_desc,
txq->axq_depth);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 5,
"ath_tx_handoff: non-tdma: TXDP[%u] = %p (%p) "
"lastds=%p depth %d",
txq->axq_qnum,
(caddr_t)bf->bf_daddr, bf->bf_desc,
bf->bf_lastds,
txq->axq_depth);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
} else {
*txq->axq_link = bf->bf_daddr;
DPRINTF(sc, ATH_DEBUG_XMIT,
"%s: link[%u](%p)=%p (%p) depth %d\n", __func__,
txq->axq_qnum, txq->axq_link,
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
(caddr_t)bf->bf_daddr, bf->bf_desc,
txq->axq_depth);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 5,
"ath_tx_handoff: non-tdma: link[%u](%p)=%p (%p) "
"lastds=%d",
txq->axq_qnum, txq->axq_link,
(caddr_t)bf->bf_daddr, bf->bf_desc,
bf->bf_lastds);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
#endif /* IEEE80211_SUPPORT_TDMA */
Change the descriptor logic to use bf_lastds to point to the last descriptor, rather than using the maths involving bf_desc[bf_nseg - 1]. When doing TX aggregation, the status will be updated in the -final- descriptor of the -final- subframe in an aggregate. Thus bf_lastds may point to the last descriptor in a completely different ath_buf. Sponsored by: Hobnob, Inc.
2011-11-08 21:25:36 +00:00
if (bf->bf_state.bfs_aggr)
txq->axq_aggr_depth++;
Convert the TX path to use the new HAL methods for accessing the TX descriptor link pointers. This is required for the AR93xx and later chipsets. The RX path is slightly different - the legacy RX path directly accesses ath_desc->ds_link for now, however this isn't at all done for EDMA (FIFO) RX. Now, for those performing a little software archeology here: This is all a bit sub-optimal. "struct ath_desc" is only really relevant for the pre-AR93xx NICs - where ds_link and ds_data is always in the same location. The AR93xx and later NICs have different descriptor layouts altogether. Now, for AR93xx and later NICs, you should never directly reference ds_link and ds_data, as: * the RX descriptors don't have either - the data is _after_ the RX descriptor. They're just one large buffer. There's also no need for a per-descriptor RX buffer size as they're all fixed sizes. * the TX descriptors have 4 buffer and 4 length fields _and_ a link pointer. Each frame takes up one TX FIFO pointer, but it can contain multiple subframes (either multiple frames in a buffer, and/or multiple frames in an aggregate/RIFS burst.) * .. so, when TX frames are queued to a hardware queue, the link pointer is ONLY for buffers in that frame/aggregate. The next frame starts in a new FIFO pointer. * Finally, descriptor completion status is in a different ring. I'll write something up about that when its time to do so. This was inspired by Linux ath9k and the reference driver but is a reimplementation. Obtained from: Linux ath9k, Qualcomm Atheros
2012-07-19 03:51:16 +00:00
ath_hal_gettxdesclinkptr(ah, bf->bf_lastds, &txq->axq_link);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
ath_hal_txstart(ah, txq->axq_qnum);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 1,
"ath_tx_handoff: txq=%u, txstart", txq->axq_qnum);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
}
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Restart TX DMA for the given TXQ.
*
* This must be called whether the queue is empty or not.
*/
Break out the hardware handoff and TX DMA restart code into methods. These (and a few others) will differ based on the underlying DMA implementation. For the EDMA NICs, simply stub them out in a fashion which will let me focus on implementing the necessary descriptor API changes.
2012-07-31 02:28:32 +00:00
static void
ath_legacy_tx_dma_restart(struct ath_softc *sc, struct ath_txq *txq)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
struct ath_hal *ah = sc->sc_ah;
Correctly initialise the TXQ link pointer to the last descriptor in the last buffer in the list. The current behaviour (due to me, so pointy hat is firmly on my head here) was incorrect - it was setting the link pointer to the last descriptor of the _first_ buffer in the TXQ. Instead, it should have set it to the last descriptor in the _last_ buffer in the TXQ. This showed up as occasional TX stalls with frames in the TXQ but no TX progress being made. Further inspection showed the TXQ looked like it contained multiple "lists" of frames - there'd be a list of correct frames, then a NULL link pointer, but there'd be a next buffer in the list. Since this code is only called upon an interface reset, it's likely this only began showing up when I started doing stress testing in environments which annoy the radios enough to cause lockups. I've not yet any TX stalls with this patch applied. PR: kern/165866
2012-03-08 23:53:38 +00:00
struct ath_buf *bf, *bf_last;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* This is always going to be cleared, empty or not */
txq->axq_flags &= ~ATH_TXQ_PUTPENDING;
Correctly initialise the TXQ link pointer to the last descriptor in the last buffer in the list. The current behaviour (due to me, so pointy hat is firmly on my head here) was incorrect - it was setting the link pointer to the last descriptor of the _first_ buffer in the TXQ. Instead, it should have set it to the last descriptor in the _last_ buffer in the TXQ. This showed up as occasional TX stalls with frames in the TXQ but no TX progress being made. Further inspection showed the TXQ looked like it contained multiple "lists" of frames - there'd be a list of correct frames, then a NULL link pointer, but there'd be a next buffer in the list. Since this code is only called upon an interface reset, it's likely this only began showing up when I started doing stress testing in environments which annoy the radios enough to cause lockups. I've not yet any TX stalls with this patch applied. PR: kern/165866
2012-03-08 23:53:38 +00:00
/* XXX make this ATH_TXQ_FIRST */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf = TAILQ_FIRST(&txq->axq_q);
Correctly initialise the TXQ link pointer to the last descriptor in the last buffer in the list. The current behaviour (due to me, so pointy hat is firmly on my head here) was incorrect - it was setting the link pointer to the last descriptor of the _first_ buffer in the TXQ. Instead, it should have set it to the last descriptor in the _last_ buffer in the TXQ. This showed up as occasional TX stalls with frames in the TXQ but no TX progress being made. Further inspection showed the TXQ looked like it contained multiple "lists" of frames - there'd be a list of correct frames, then a NULL link pointer, but there'd be a next buffer in the list. Since this code is only called upon an interface reset, it's likely this only began showing up when I started doing stress testing in environments which annoy the radios enough to cause lockups. I've not yet any TX stalls with this patch applied. PR: kern/165866
2012-03-08 23:53:38 +00:00
bf_last = ATH_TXQ_LAST(txq, axq_q_s);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (bf == NULL)
return;
ath_hal_puttxbuf(ah, txq->axq_qnum, bf->bf_daddr);
Correct re-initialise the link pointer to be the final descriptor in the last buffer. This fixes traffic stalls that were occuring with stuck beacon events. PR: kern/170433
2012-08-07 00:42:46 +00:00
ath_hal_gettxdesclinkptr(ah, bf_last->bf_lastds, &txq->axq_link);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_hal_txstart(ah, txq->axq_qnum);
}
/*
* Hand off a packet to the hardware (or mcast queue.)
*
* The relevant hardware txq should be locked.
*/
static void
Break out the hardware handoff and TX DMA restart code into methods. These (and a few others) will differ based on the underlying DMA implementation. For the EDMA NICs, simply stub them out in a fashion which will let me focus on implementing the necessary descriptor API changes.
2012-07-31 02:28:32 +00:00
ath_legacy_xmit_handoff(struct ath_softc *sc, struct ath_txq *txq,
struct ath_buf *bf)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ALQ logging enhancements: * upon setup, tell the alq code what the chip information is. * add TX/RX path logging for legacy chips. * populate the tx/rx descriptor length fields with a best-estimate. It's overly big (96 bytes when AH_SUPPORT_AR5416 is enabled) but it'll do for now. Whilst I'm here, add CURVNET_RESTORE() here during probe/attach as a partial solution to fixing crashes during attach when the attach fails. There are other attach failures that I have to deal with; those'll come later.
2012-11-16 19:57:16 +00:00
#ifdef ATH_DEBUG_ALQ
if (if_ath_alq_checkdebug(&sc->sc_alq, ATH_ALQ_EDMA_TXDESC))
ath_tx_alq_post(sc, bf);
#endif
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (txq->axq_qnum == ATH_TXQ_SWQ)
ath_tx_handoff_mcast(sc, txq, bf);
else
ath_tx_handoff_hw(sc, txq, bf);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
static int
ath_tx_tag_crypto(struct ath_softc *sc, struct ieee80211_node *ni,
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
struct mbuf *m0, int iswep, int isfrag, int *hdrlen, int *pktlen,
int *keyix)
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
{
Add some further debugging to try and aid tracking down what the state of things were just before a full software queue is drained.
2012-03-22 21:48:36 +00:00
DPRINTF(sc, ATH_DEBUG_XMIT,
"%s: hdrlen=%d, pktlen=%d, isfrag=%d, iswep=%d, m0=%p\n",
__func__,
*hdrlen,
*pktlen,
isfrag,
iswep,
m0);
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
if (iswep) {
const struct ieee80211_cipher *cip;
struct ieee80211_key *k;
/*
* Construct the 802.11 header+trailer for an encrypted
* frame. The only reason this can fail is because of an
* unknown or unsupported cipher/key type.
*/
k = ieee80211_crypto_encap(ni, m0);
if (k == NULL) {
/*
* This can happen when the key is yanked after the
* frame was queued. Just discard the frame; the
* 802.11 layer counts failures and provides
* debugging/diagnostics.
*/
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
return (0);
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
}
/*
* Adjust the packet + header lengths for the crypto
* additions and calculate the h/w key index. When
* a s/w mic is done the frame will have had any mic
* added to it prior to entry so m0->m_pkthdr.len will
* account for it. Otherwise we need to add it to the
* packet length.
*/
cip = k->wk_cipher;
(*hdrlen) += cip->ic_header;
(*pktlen) += cip->ic_header + cip->ic_trailer;
/* NB: frags always have any TKIP MIC done in s/w */
if ((k->wk_flags & IEEE80211_KEY_SWMIC) == 0 && !isfrag)
(*pktlen) += cip->ic_miclen;
(*keyix) = k->wk_keyix;
} else if (ni->ni_ucastkey.wk_cipher == &ieee80211_cipher_none) {
/*
* Use station key cache slot, if assigned.
*/
(*keyix) = ni->ni_ucastkey.wk_keyix;
if ((*keyix) == IEEE80211_KEYIX_NONE)
(*keyix) = HAL_TXKEYIX_INVALID;
} else
(*keyix) = HAL_TXKEYIX_INVALID;
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
return (1);
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
}
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
/*
* Calculate whether interoperability protection is required for
* this frame.
*
* This requires the rate control information be filled in,
* as the protection requirement depends upon the current
* operating mode / PHY.
*/
static void
ath_tx_calc_protection(struct ath_softc *sc, struct ath_buf *bf)
{
struct ieee80211_frame *wh;
uint8_t rix;
uint16_t flags;
int shortPreamble;
const HAL_RATE_TABLE *rt = sc->sc_currates;
struct ifnet *ifp = sc->sc_ifp;
struct ieee80211com *ic = ifp->if_l2com;
flags = bf->bf_state.bfs_txflags;
rix = bf->bf_state.bfs_rc[0].rix;
shortPreamble = bf->bf_state.bfs_shpream;
wh = mtod(bf->bf_m, struct ieee80211_frame *);
/*
* If 802.11g protection is enabled, determine whether
* to use RTS/CTS or just CTS. Note that this is only
* done for OFDM unicast frames.
*/
if ((ic->ic_flags & IEEE80211_F_USEPROT) &&
rt->info[rix].phy == IEEE80211_T_OFDM &&
(flags & HAL_TXDESC_NOACK) == 0) {
bf->bf_state.bfs_doprot = 1;
/* XXX fragments must use CCK rates w/ protection */
if (ic->ic_protmode == IEEE80211_PROT_RTSCTS) {
flags |= HAL_TXDESC_RTSENA;
} else if (ic->ic_protmode == IEEE80211_PROT_CTSONLY) {
flags |= HAL_TXDESC_CTSENA;
}
/*
* For frags it would be desirable to use the
* highest CCK rate for RTS/CTS. But stations
* farther away may detect it at a lower CCK rate
* so use the configured protection rate instead
* (for now).
*/
sc->sc_stats.ast_tx_protect++;
}
/*
* If 11n protection is enabled and it's a HT frame,
* enable RTS.
*
* XXX ic_htprotmode or ic_curhtprotmode?
* XXX should it_htprotmode only matter if ic_curhtprotmode
* XXX indicates it's not a HT pure environment?
*/
if ((ic->ic_htprotmode == IEEE80211_PROT_RTSCTS) &&
rt->info[rix].phy == IEEE80211_T_HT &&
(flags & HAL_TXDESC_NOACK) == 0) {
flags |= HAL_TXDESC_RTSENA;
sc->sc_stats.ast_tx_htprotect++;
}
bf->bf_state.bfs_txflags = flags;
}
/*
* Update the frame duration given the currently selected rate.
*
* This also updates the frame duration value, so it will require
* a DMA flush.
*/
static void
ath_tx_calc_duration(struct ath_softc *sc, struct ath_buf *bf)
{
struct ieee80211_frame *wh;
uint8_t rix;
uint16_t flags;
int shortPreamble;
struct ath_hal *ah = sc->sc_ah;
const HAL_RATE_TABLE *rt = sc->sc_currates;
int isfrag = bf->bf_m->m_flags & M_FRAG;
flags = bf->bf_state.bfs_txflags;
rix = bf->bf_state.bfs_rc[0].rix;
shortPreamble = bf->bf_state.bfs_shpream;
wh = mtod(bf->bf_m, struct ieee80211_frame *);
/*
* Calculate duration. This logically belongs in the 802.11
* layer but it lacks sufficient information to calculate it.
*/
if ((flags & HAL_TXDESC_NOACK) == 0 &&
(wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK) != IEEE80211_FC0_TYPE_CTL) {
u_int16_t dur;
if (shortPreamble)
dur = rt->info[rix].spAckDuration;
else
dur = rt->info[rix].lpAckDuration;
if (wh->i_fc[1] & IEEE80211_FC1_MORE_FRAG) {
dur += dur; /* additional SIFS+ACK */
Pull out the if_transmit() work and revert back to ath_start(). My changed had some rather significant behavioural changes to throughput. The two issues I noticed: * With if_start and the ifnet mbuf queue, any temporary latency would get eaten up by some mbufs being queued. With ath_transmit() queuing things to ath_buf's, I'd only get 512 TX buffers before I couldn't queue any further frames. * There's also some non-zero latency involved with TX being pushed into a taskqueue via direct dispatch. Any time the scheduler didn't immediately schedule the ath TX task would cause extra latency. Various 1ge/10ge drivers implement both direct dispatch (if the TX lock can be acquired) and deferred task transmission (if the TX lock can't be acquired), with frames being pushed into a drbd queue. I'll have to do this at some point, but until I figure out how to deal with 802.11 fragments, I'll have to wait a while longer. So what I saw: * lots of extra latency, specially under load - if the taskqueue wasn't immediately scheduled, things went pear shaped; * any extra latency would result in TX ath_buf's taking their sweet time being replenished, so any further calls to ath_transmit() would drop mbufs. * .. yes, there's no explicit backpressure here - things are just dropped. Eek. With this, the general performance has gone up, but those subtle if_start() related race conditions are back. For some reason, this is doubly-obvious with the AR5416 NIC and I don't quite understand why yet. There's an unrelated issue with AR5416 performance in STA mode (it's fine in AP mode when bridging frames, weirdly..) that requires a little further investigation. Specifically - it works fine on a Lenovo T40 (single core CPU) running a March 2012 9-STABLE kernel, but a Lenovo T60 (dual core) running an early November 2012 kernel behaves very poorly. The same hardware with an AR9160 or AR9280 behaves perfectly.
2013-02-13 05:32:19 +00:00
KASSERT(bf->bf_m->m_nextpkt != NULL, ("no fragment"));
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
/*
* Include the size of next fragment so NAV is
* updated properly. The last fragment uses only
* the ACK duration
Since it's not immediately obvious whether the current TX path handles fragment rate lookups correctly, add a comment describing exactly that. The assumption in the fragment duration code is the duration of the next fragment will match the rate used by the current fragment. But I think a rate lookup is being done for _each_ fragment. For older pre-sample rate control this would almost always be the case, but for sample it may be incorrect more often then correct.
2012-10-26 16:31:12 +00:00
*
* XXX TODO: ensure that the rate lookup for each
* fragment is the same as the rate used by the
* first fragment!
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
*/
dur += ath_hal_computetxtime(ah, rt,
Pull out the if_transmit() work and revert back to ath_start(). My changed had some rather significant behavioural changes to throughput. The two issues I noticed: * With if_start and the ifnet mbuf queue, any temporary latency would get eaten up by some mbufs being queued. With ath_transmit() queuing things to ath_buf's, I'd only get 512 TX buffers before I couldn't queue any further frames. * There's also some non-zero latency involved with TX being pushed into a taskqueue via direct dispatch. Any time the scheduler didn't immediately schedule the ath TX task would cause extra latency. Various 1ge/10ge drivers implement both direct dispatch (if the TX lock can be acquired) and deferred task transmission (if the TX lock can't be acquired), with frames being pushed into a drbd queue. I'll have to do this at some point, but until I figure out how to deal with 802.11 fragments, I'll have to wait a while longer. So what I saw: * lots of extra latency, specially under load - if the taskqueue wasn't immediately scheduled, things went pear shaped; * any extra latency would result in TX ath_buf's taking their sweet time being replenished, so any further calls to ath_transmit() would drop mbufs. * .. yes, there's no explicit backpressure here - things are just dropped. Eek. With this, the general performance has gone up, but those subtle if_start() related race conditions are back. For some reason, this is doubly-obvious with the AR5416 NIC and I don't quite understand why yet. There's an unrelated issue with AR5416 performance in STA mode (it's fine in AP mode when bridging frames, weirdly..) that requires a little further investigation. Specifically - it works fine on a Lenovo T40 (single core CPU) running a March 2012 9-STABLE kernel, but a Lenovo T60 (dual core) running an early November 2012 kernel behaves very poorly. The same hardware with an AR9160 or AR9280 behaves perfectly.
2013-02-13 05:32:19 +00:00
bf->bf_m->m_nextpkt->m_pkthdr.len,
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
rix, shortPreamble);
}
if (isfrag) {
/*
* Force hardware to use computed duration for next
* fragment by disabling multi-rate retry which updates
* duration based on the multi-rate duration table.
*/
bf->bf_state.bfs_ismrr = 0;
bf->bf_state.bfs_try0 = ATH_TXMGTTRY;
/* XXX update bfs_rc[0].try? */
}
/* Update the duration field itself */
*(u_int16_t *)wh->i_dur = htole16(dur);
}
}
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
static uint8_t
ath_tx_get_rtscts_rate(struct ath_hal *ah, const HAL_RATE_TABLE *rt,
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
int cix, int shortPreamble)
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
{
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
uint8_t ctsrate;
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
/*
* CTS transmit rate is derived from the transmit rate
* by looking in the h/w rate table. We must also factor
* in whether or not a short preamble is to be used.
*/
/* NB: cix is set above where RTS/CTS is enabled */
KASSERT(cix != 0xff, ("cix not setup"));
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
ctsrate = rt->info[cix].rateCode;
/* XXX this should only matter for legacy rates */
if (shortPreamble)
ctsrate |= rt->info[cix].shortPreamble;
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
return (ctsrate);
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
}
/*
* Calculate the RTS/CTS duration for legacy frames.
*/
static int
ath_tx_calc_ctsduration(struct ath_hal *ah, int rix, int cix,
int shortPreamble, int pktlen, const HAL_RATE_TABLE *rt,
int flags)
{
int ctsduration = 0;
/* This mustn't be called for HT modes */
if (rt->info[cix].phy == IEEE80211_T_HT) {
printf("%s: HT rate where it shouldn't be (0x%x)\n",
__func__, rt->info[cix].rateCode);
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
return (-1);
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
}
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
/*
* Compute the transmit duration based on the frame
* size and the size of an ACK frame. We call into the
* HAL to do the computation since it depends on the
* characteristics of the actual PHY being used.
*
* NB: CTS is assumed the same size as an ACK so we can
* use the precalculated ACK durations.
*/
if (shortPreamble) {
if (flags & HAL_TXDESC_RTSENA) /* SIFS + CTS */
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
ctsduration += rt->info[cix].spAckDuration;
ctsduration += ath_hal_computetxtime(ah,
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
rt, pktlen, rix, AH_TRUE);
if ((flags & HAL_TXDESC_NOACK) == 0) /* SIFS + ACK */
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
ctsduration += rt->info[rix].spAckDuration;
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
} else {
if (flags & HAL_TXDESC_RTSENA) /* SIFS + CTS */
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
ctsduration += rt->info[cix].lpAckDuration;
ctsduration += ath_hal_computetxtime(ah,
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
rt, pktlen, rix, AH_FALSE);
if ((flags & HAL_TXDESC_NOACK) == 0) /* SIFS + ACK */
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
ctsduration += rt->info[rix].lpAckDuration;
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
}
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
return (ctsduration);
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Update the given ath_buf with updated rts/cts setup and duration
* values.
*
* To support rate lookups for each software retry, the rts/cts rate
* and cts duration must be re-calculated.
*
* This function assumes the RTS/CTS flags have been set as needed;
* mrr has been disabled; and the rate control lookup has been done.
*
* XXX TODO: MRR need only be disabled for the pre-11n NICs.
* XXX The 11n NICs support per-rate RTS/CTS configuration.
*/
static void
ath_tx_set_rtscts(struct ath_softc *sc, struct ath_buf *bf)
{
uint16_t ctsduration = 0;
uint8_t ctsrate = 0;
uint8_t rix = bf->bf_state.bfs_rc[0].rix;
uint8_t cix = 0;
const HAL_RATE_TABLE *rt = sc->sc_currates;
/*
* No RTS/CTS enabled? Don't bother.
*/
Remove duplicate txflags field from ath_buf. rename bf_state.bfs_flags to bf_state.bfs_txflags, as that is what it effectively is.
2012-04-07 02:01:26 +00:00
if ((bf->bf_state.bfs_txflags &
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
(HAL_TXDESC_RTSENA | HAL_TXDESC_CTSENA)) == 0) {
/* XXX is this really needed? */
bf->bf_state.bfs_ctsrate = 0;
bf->bf_state.bfs_ctsduration = 0;
return;
}
/*
* If protection is enabled, use the protection rix control
* rate. Otherwise use the rate0 control rate.
*/
if (bf->bf_state.bfs_doprot)
rix = sc->sc_protrix;
else
rix = bf->bf_state.bfs_rc[0].rix;
/*
* If the raw path has hard-coded ctsrate0 to something,
* use it.
*/
if (bf->bf_state.bfs_ctsrate0 != 0)
cix = ath_tx_findrix(sc, bf->bf_state.bfs_ctsrate0);
else
/* Control rate from above */
cix = rt->info[rix].controlRate;
/* Calculate the rtscts rate for the given cix */
ctsrate = ath_tx_get_rtscts_rate(sc->sc_ah, rt, cix,
bf->bf_state.bfs_shpream);
/* The 11n chipsets do ctsduration calculations for you */
if (! ath_tx_is_11n(sc))
ctsduration = ath_tx_calc_ctsduration(sc->sc_ah, rix, cix,
bf->bf_state.bfs_shpream, bf->bf_state.bfs_pktlen,
Remove duplicate txflags field from ath_buf. rename bf_state.bfs_flags to bf_state.bfs_txflags, as that is what it effectively is.
2012-04-07 02:01:26 +00:00
rt, bf->bf_state.bfs_txflags);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Squirrel away in ath_buf */
bf->bf_state.bfs_ctsrate = ctsrate;
bf->bf_state.bfs_ctsduration = ctsduration;
/*
* Must disable multi-rate retry when using RTS/CTS.
*/
Allow 802.11n hardware to support multi-rate retry when RTS/CTS is enabled. The legacy (pre-802.11n) hardware doesn't support this - although the AR5212 era hardware supports MRR, it doesn't have all the bits needed to support MRR + RTS/CTS. The AR5416 and later support a packet duration and RTS/CTS flags per rate scenario, so we should support it. Tested: * AR9280, STA PR: kern/170302
2012-07-31 23:54:15 +00:00
if (!sc->sc_mrrprot) {
bf->bf_state.bfs_ismrr = 0;
bf->bf_state.bfs_try0 =
bf->bf_state.bfs_rc[0].tries = ATH_TXMGTTRY; /* XXX ew */
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
/*
* Setup the descriptor chain for a normal or fast-frame
* frame.
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
*
* XXX TODO: extend to include the destination hardware QCU ID.
* Make sure that is correct. Make sure that when being added
* to the mcastq, the CABQ QCUID is set or things will get a bit
* odd.
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*/
static void
ath_tx_setds(struct ath_softc *sc, struct ath_buf *bf)
{
struct ath_desc *ds = bf->bf_desc;
struct ath_hal *ah = sc->sc_ah;
Add some debugging to try and catch an invalid TX rate (0x0) that is being reported.
2012-11-13 06:28:57 +00:00
if (bf->bf_state.bfs_txrate0 == 0)
device_printf(sc->sc_dev, "%s: bf=%p, txrate0=%d\n",
__func__, bf, 0);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_hal_setuptxdesc(ah, ds
, bf->bf_state.bfs_pktlen /* packet length */
, bf->bf_state.bfs_hdrlen /* header length */
, bf->bf_state.bfs_atype /* Atheros packet type */
, bf->bf_state.bfs_txpower /* txpower */
, bf->bf_state.bfs_txrate0
, bf->bf_state.bfs_try0 /* series 0 rate/tries */
, bf->bf_state.bfs_keyix /* key cache index */
, bf->bf_state.bfs_txantenna /* antenna mode */
Remove duplicate txflags field from ath_buf. rename bf_state.bfs_flags to bf_state.bfs_txflags, as that is what it effectively is.
2012-04-07 02:01:26 +00:00
, bf->bf_state.bfs_txflags /* flags */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
, bf->bf_state.bfs_ctsrate /* rts/cts rate */
, bf->bf_state.bfs_ctsduration /* rts/cts duration */
);
/*
* This will be overriden when the descriptor chain is written.
*/
bf->bf_lastds = ds;
bf->bf_last = bf;
Push the rate control and descriptor chaining into the descriptor "set" functions, for both legacy and 802.11n. This will simplify supporting the EDMA chipsets as these two descriptor setup functions can just be overridden in their entirety, hiding all of the subtle differences in setting things up. It's not a permanent solution, as eventually the AR5416 HAL should grow similar versions of the 11n descriptor functions and then those can be used. TODO: * Push the "clr11naggr" call into the legacy setds, just to ensure that retried frames don't end up with the aggregate bits set inappropriately; * Remove the "setlasttxdesc" call from the 11n TX path and push it into setds_11n. * Ensure that setds_11n will work correctly for non-aggregate frames; * .. and then when it does, just unconditionally call "setds_11n" for 11n NICs and "setds" for non-11n NICs.
2012-07-31 16:41:09 +00:00
/* Set rate control and descriptor chain for this frame */
ath_tx_set_ratectrl(sc, bf->bf_node, bf);
Convert the aggregate descriptor path over to use the same API as the non-aggregate path. I "cheated" by using some TX setup code in our HAL that isn't present in the atheros HAL (or Linux ath9k.) The old path for forming aggregates was: * setup the rate control in the first descriptor; * call chaintxdesc() on all the frames; * call setupfirsttxdesc() on the first descrpitor in the first frame; * call setuplasttxdesc() on the last descriptor in the last frame. The new path for forming aggregates looks like the non-aggregate path: * call setuptxdesc() on the first descriptor in the first frame; * setup the rate control in the first descriptor; * call filltxdesc() on each descriptor in the frame; * if it's an aggregate - call set11n_aggr_{first, middle, last} as appropriate (see the code for a description of what is "appropriate".) Now, this is done primarily for the AR9300 HAL - it doesn't implement the first set of aggregate functions. It just has the older methods and the "first/middle/last" aggregate methods. So, let's convert the code to use these. Note: the AR5416 HAL in FreeBSD had that code (from me, a while ago) and a previous commit brought it up to behave the same as the AR9300 HAL routines. There's some further tidyups to be done - specifically, avoid doing multiple calls to the 11n descriptor functions. I shouldn't call clr11n_aggr(), then set11n_aggr_middle(), then also set11n_aggr_first(). On (at least MIPS) the TX descriptors are in non-cachable memory and this will cause multiple slow writes. I'll debug/tidy that up in a future commit. Tested: * AR9280, STA * AR9280/AR9160, AP * AR9380, STA (using a local, closed source HAL, sorry!)
2012-11-06 06:19:11 +00:00
ath_tx_chaindesclist(sc, ds, bf, 0, 0, 0);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
/*
* Do a rate lookup.
*
* This performs a rate lookup for the given ath_buf only if it's required.
* Non-data frames and raw frames don't require it.
*
* This populates the primary and MRR entries; MRR values are
* then disabled later on if something requires it (eg RTS/CTS on
* pre-11n chipsets.
*
* This needs to be done before the RTS/CTS fields are calculated
* as they may depend upon the rate chosen.
*/
static void
ath_tx_do_ratelookup(struct ath_softc *sc, struct ath_buf *bf)
{
uint8_t rate, rix;
int try0;
if (! bf->bf_state.bfs_doratelookup)
return;
/* Get rid of any previous state */
bzero(bf->bf_state.bfs_rc, sizeof(bf->bf_state.bfs_rc));
ATH_NODE_LOCK(ATH_NODE(bf->bf_node));
ath_rate_findrate(sc, ATH_NODE(bf->bf_node), bf->bf_state.bfs_shpream,
bf->bf_state.bfs_pktlen, &rix, &try0, &rate);
/* In case MRR is disabled, make sure rc[0] is setup correctly */
bf->bf_state.bfs_rc[0].rix = rix;
bf->bf_state.bfs_rc[0].ratecode = rate;
bf->bf_state.bfs_rc[0].tries = try0;
if (bf->bf_state.bfs_ismrr && try0 != ATH_TXMAXTRY)
ath_rate_getxtxrates(sc, ATH_NODE(bf->bf_node), rix,
bf->bf_state.bfs_rc);
ATH_NODE_UNLOCK(ATH_NODE(bf->bf_node));
sc->sc_txrix = rix; /* for LED blinking */
sc->sc_lastdatarix = rix; /* for fast frames */
bf->bf_state.bfs_try0 = try0;
bf->bf_state.bfs_txrate0 = rate;
}
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
/*
* Update the CLRDMASK bit in the ath_buf if it needs to be set.
*/
static void
ath_tx_update_clrdmask(struct ath_softc *sc, struct ath_tid *tid,
struct ath_buf *bf)
{
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
struct ath_node *an = ATH_NODE(bf->bf_node);
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
if (an->clrdmask == 1) {
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
an->clrdmask = 0;
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
}
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Transmit the given frame to the hardware.
*
* The frame must already be setup; rate control must already have
* been done.
*
* XXX since the TXQ lock is being held here (and I dislike holding
* it for this long when not doing software aggregation), later on
* break this function into "setup_normal" and "xmit_normal". The
* lock only needs to be held for the ath_tx_handoff call.
*/
static void
ath_tx_xmit_normal(struct ath_softc *sc, struct ath_txq *txq,
struct ath_buf *bf)
{
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
struct ath_node *an = ATH_NODE(bf->bf_node);
struct ath_tid *tid = &an->an_tid[bf->bf_state.bfs_tid];
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
/*
* For now, just enable CLRDMASK. ath_tx_xmit_normal() does
* set a completion handler however it doesn't (yet) properly
* handle the strict ordering requirements needed for normal,
* non-aggregate session frames.
*
* Once this is implemented, only set CLRDMASK like this for
* frames that must go out - eg management/raw frames.
*/
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Setup the descriptor before handoff */
ath_tx_do_ratelookup(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_calc_duration(sc, bf);
ath_tx_calc_protection(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_set_rtscts(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_rate_fill_rcflags(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_setds(sc, bf);
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
/* Track per-TID hardware queue depth correctly */
tid->hwq_depth++;
/* Assign the completion handler */
bf->bf_comp = ath_tx_normal_comp;
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Hand off to hardware */
ath_tx_handoff(sc, txq, bf);
}
Place the comment where it should be.
2012-09-20 03:04:19 +00:00
/*
* Do the basic frame setup stuff that's required before the frame
* is added to a software queue.
*
* All frames get mostly the same treatment and it's done once.
* Retransmits fiddle with things like the rate control setup,
* setting the retransmit bit in the packet; doing relevant DMA/bus
* syncing and relinking it (back) into the hardware TX queue.
*
* Note that this may cause the mbuf to be reallocated, so
* m0 may not be valid.
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
static int
ath_tx_normal_setup(struct ath_softc *sc, struct ieee80211_node *ni,
After reviewing the mcast/sleep station code a little, undo some brain damage which I committed when I had less clue about such things. Don't ever put normal data frames on the mcast software queue. Just put mcast frames there if needed. Pass the txq decision into ath_tx_normal_setup(), as we've already made the decision. Don't re-do it. Whilst i'm here, add another random debugging statement.
2012-04-08 00:40:16 +00:00
struct ath_buf *bf, struct mbuf *m0, struct ath_txq *txq)
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
{
struct ieee80211vap *vap = ni->ni_vap;
struct ath_hal *ah = sc->sc_ah;
struct ifnet *ifp = sc->sc_ifp;
struct ieee80211com *ic = ifp->if_l2com;
const struct chanAccParams *cap = &ic->ic_wme.wme_chanParams;
int error, iswep, ismcast, isfrag, ismrr;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
int keyix, hdrlen, pktlen, try0 = 0;
u_int8_t rix = 0, txrate = 0;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
struct ath_desc *ds;
struct ieee80211_frame *wh;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
u_int subtype, flags;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
HAL_PKT_TYPE atype;
const HAL_RATE_TABLE *rt;
HAL_BOOL shortPreamble;
struct ath_node *an;
u_int pri;
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
/*
* To ensure that both sequence numbers and the CCMP PN handling
* is "correct", make sure that the relevant TID queue is locked.
* Otherwise the CCMP PN and seqno may appear out of order, causing
* re-ordered frames to have out of order CCMP PN's, resulting
* in many, many frame drops.
*/
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
wh = mtod(m0, struct ieee80211_frame *);
iswep = wh->i_fc[1] & IEEE80211_FC1_WEP;
ismcast = IEEE80211_IS_MULTICAST(wh->i_addr1);
isfrag = m0->m_flags & M_FRAG;
hdrlen = ieee80211_anyhdrsize(wh);
/*
* Packet length must not include any
* pad bytes; deduct them here.
*/
pktlen = m0->m_pkthdr.len - (hdrlen & 3);
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
/* Handle encryption twiddling if needed */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (! ath_tx_tag_crypto(sc, ni, m0, iswep, isfrag, &hdrlen,
&pktlen, &keyix)) {
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
ath_freetx(m0);
return EIO;
}
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
/* packet header may have moved, reset our local pointer */
wh = mtod(m0, struct ieee80211_frame *);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
pktlen += IEEE80211_CRC_LEN;
/*
* Load the DMA map so any coalescing is done. This
* also calculates the number of descriptors we need.
*/
error = ath_tx_dmasetup(sc, bf, m0);
if (error != 0)
return error;
bf->bf_node = ni; /* NB: held reference */
m0 = bf->bf_m; /* NB: may have changed */
wh = mtod(m0, struct ieee80211_frame *);
/* setup descriptors */
ds = bf->bf_desc;
rt = sc->sc_currates;
KASSERT(rt != NULL, ("no rate table, mode %u", sc->sc_curmode));
/*
* NB: the 802.11 layer marks whether or not we should
* use short preamble based on the current mode and
* negotiated parameters.
*/
if ((ic->ic_flags & IEEE80211_F_SHPREAMBLE) &&
(ni->ni_capinfo & IEEE80211_CAPINFO_SHORT_PREAMBLE)) {
shortPreamble = AH_TRUE;
sc->sc_stats.ast_tx_shortpre++;
} else {
shortPreamble = AH_FALSE;
}
an = ATH_NODE(ni);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
//flags = HAL_TXDESC_CLRDMASK; /* XXX needed for crypto errs */
flags = 0;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
ismrr = 0; /* default no multi-rate retry*/
pri = M_WME_GETAC(m0); /* honor classification */
/* XXX use txparams instead of fixed values */
/*
* Calculate Atheros packet type from IEEE80211 packet header,
* setup for rate calculations, and select h/w transmit queue.
*/
switch (wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK) {
case IEEE80211_FC0_TYPE_MGT:
subtype = wh->i_fc[0] & IEEE80211_FC0_SUBTYPE_MASK;
if (subtype == IEEE80211_FC0_SUBTYPE_BEACON)
atype = HAL_PKT_TYPE_BEACON;
else if (subtype == IEEE80211_FC0_SUBTYPE_PROBE_RESP)
atype = HAL_PKT_TYPE_PROBE_RESP;
else if (subtype == IEEE80211_FC0_SUBTYPE_ATIM)
atype = HAL_PKT_TYPE_ATIM;
else
atype = HAL_PKT_TYPE_NORMAL; /* XXX */
rix = an->an_mgmtrix;
txrate = rt->info[rix].rateCode;
if (shortPreamble)
txrate |= rt->info[rix].shortPreamble;
try0 = ATH_TXMGTTRY;
flags |= HAL_TXDESC_INTREQ; /* force interrupt */
break;
case IEEE80211_FC0_TYPE_CTL:
atype = HAL_PKT_TYPE_PSPOLL; /* stop setting of duration */
rix = an->an_mgmtrix;
txrate = rt->info[rix].rateCode;
if (shortPreamble)
txrate |= rt->info[rix].shortPreamble;
try0 = ATH_TXMGTTRY;
flags |= HAL_TXDESC_INTREQ; /* force interrupt */
break;
case IEEE80211_FC0_TYPE_DATA:
atype = HAL_PKT_TYPE_NORMAL; /* default */
/*
* Data frames: multicast frames go out at a fixed rate,
* EAPOL frames use the mgmt frame rate; otherwise consult
* the rate control module for the rate to use.
*/
if (ismcast) {
rix = an->an_mcastrix;
txrate = rt->info[rix].rateCode;
if (shortPreamble)
txrate |= rt->info[rix].shortPreamble;
try0 = 1;
} else if (m0->m_flags & M_EAPOL) {
/* XXX? maybe always use long preamble? */
rix = an->an_mgmtrix;
txrate = rt->info[rix].rateCode;
if (shortPreamble)
txrate |= rt->info[rix].shortPreamble;
try0 = ATH_TXMAXTRY; /* XXX?too many? */
} else {
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Do rate lookup on each TX, rather than using
* the hard-coded TX information decided here.
*/
ismrr = 1;
bf->bf_state.bfs_doratelookup = 1;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
if (cap->cap_wmeParams[pri].wmep_noackPolicy)
flags |= HAL_TXDESC_NOACK;
break;
default:
if_printf(ifp, "bogus frame type 0x%x (%s)\n",
wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK, __func__);
/* XXX statistic */
ath_freetx(m0);
return EIO;
}
Disable this warning debug for now, as I'm now aware of the particular situation where it's occuring. Whilst I'm here, flesh out a more descriptive description.
2012-06-14 04:01:25 +00:00
/*
* There are two known scenarios where the frame AC doesn't match
* what the destination TXQ is.
*
* + non-QoS frames (eg management?) that the net80211 stack has
* assigned a higher AC to, but since it's a non-QoS TID, it's
* being thrown into TID 16. TID 16 gets the AC_BE queue.
* It's quite possible that management frames should just be
* direct dispatched to hardware rather than go via the software
* queue; that should be investigated in the future. There are
* some specific scenarios where this doesn't make sense, mostly
* surrounding ADDBA request/response - hence why that is special
* cased.
*
* + Multicast frames going into the VAP mcast queue. That shows up
* as "TXQ 11".
*
* This driver should eventually support separate TID and TXQ locking,
* allowing for arbitrary AC frames to appear on arbitrary software
* queues, being queued to the "correct" hardware queue when needed.
*/
#if 0
For now, add a quick debugging patch to log when the hw TXQ != the TID/AC.
2012-05-21 22:43:38 +00:00
if (txq != sc->sc_ac2q[pri]) {
device_printf(sc->sc_dev,
"%s: txq=%p (%d), pri=%d, pri txq=%p (%d)\n",
__func__,
txq,
txq->axq_qnum,
pri,
sc->sc_ac2q[pri],
sc->sc_ac2q[pri]->axq_qnum);
}
Disable this warning debug for now, as I'm now aware of the particular situation where it's occuring. Whilst I'm here, flesh out a more descriptive description.
2012-06-14 04:01:25 +00:00
#endif
For now, add a quick debugging patch to log when the hw TXQ != the TID/AC.
2012-05-21 22:43:38 +00:00
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
* Calculate miscellaneous flags.
*/
if (ismcast) {
flags |= HAL_TXDESC_NOACK; /* no ack on broad/multicast */
} else if (pktlen > vap->iv_rtsthreshold &&
(ni->ni_ath_flags & IEEE80211_NODE_FF) == 0) {
flags |= HAL_TXDESC_RTSENA; /* RTS based on frame length */
sc->sc_stats.ast_tx_rts++;
}
if (flags & HAL_TXDESC_NOACK) /* NB: avoid double counting */
sc->sc_stats.ast_tx_noack++;
#ifdef IEEE80211_SUPPORT_TDMA
if (sc->sc_tdma && (flags & HAL_TXDESC_NOACK) == 0) {
DPRINTF(sc, ATH_DEBUG_TDMA,
"%s: discard frame, ACK required w/ TDMA\n", __func__);
sc->sc_stats.ast_tdma_ack++;
ath_freetx(m0);
return EIO;
}
#endif
/*
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
* Determine if a tx interrupt should be generated for
* this descriptor. We take a tx interrupt to reap
* descriptors when the h/w hits an EOL condition or
* when the descriptor is specifically marked to generate
* an interrupt. We periodically mark descriptors in this
* way to insure timely replenishing of the supply needed
* for sending frames. Defering interrupts reduces system
* load and potentially allows more concurrent work to be
* done but if done to aggressively can cause senders to
* backup.
*
* NB: use >= to deal with sc_txintrperiod changing
* dynamically through sysctl.
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (flags & HAL_TXDESC_INTREQ) {
txq->axq_intrcnt = 0;
} else if (++txq->axq_intrcnt >= sc->sc_txintrperiod) {
flags |= HAL_TXDESC_INTREQ;
txq->axq_intrcnt = 0;
}
Shuffle around the RTS/CTS rate/duration logic. * Turn ath_tx_calc_ctsduration() into a function that returns the ctsduration, or -1 for HT rates; * add a printf() to ath_tx_calc_ctsduration() which will be very loud if somehow that function is called with an MCS rate; * Add ath_tx_get_rtscts_rate() which returns the RTS/CTS rate to use for the given data rate, incl. the short preamble flag; * Only call ath_tx_calc_ctsduration() for non-11n chipsets; 11n chipsets don't require the rtscts duration to be calculated.
2011-02-22 00:37:53 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* This point forward is actual TX bits */
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
* At this point we are committed to sending the frame
* and we don't need to look at m_nextpkt; clear it in
* case this frame is part of frag chain.
*/
m0->m_nextpkt = NULL;
if (IFF_DUMPPKTS(sc, ATH_DEBUG_XMIT))
ieee80211_dump_pkt(ic, mtod(m0, const uint8_t *), m0->m_len,
sc->sc_hwmap[rix].ieeerate, -1);
if (ieee80211_radiotap_active_vap(vap)) {
u_int64_t tsf = ath_hal_gettsf64(ah);
sc->sc_tx_th.wt_tsf = htole64(tsf);
sc->sc_tx_th.wt_flags = sc->sc_hwmap[rix].txflags;
if (iswep)
sc->sc_tx_th.wt_flags |= IEEE80211_RADIOTAP_F_WEP;
if (isfrag)
sc->sc_tx_th.wt_flags |= IEEE80211_RADIOTAP_F_FRAG;
sc->sc_tx_th.wt_rate = sc->sc_hwmap[rix].ieeerate;
sc->sc_tx_th.wt_txpower = ni->ni_txpower;
sc->sc_tx_th.wt_antenna = sc->sc_txantenna;
ieee80211_radiotap_tx(vap, m0);
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Blank the legacy rate array */
bzero(&bf->bf_state.bfs_rc, sizeof(bf->bf_state.bfs_rc));
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
* ath_buf_set_rate needs at least one rate/try to setup
* the rate scenario.
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_rc[0].rix = rix;
bf->bf_state.bfs_rc[0].tries = try0;
bf->bf_state.bfs_rc[0].ratecode = txrate;
/* Store the decided rate index values away */
bf->bf_state.bfs_pktlen = pktlen;
bf->bf_state.bfs_hdrlen = hdrlen;
bf->bf_state.bfs_atype = atype;
bf->bf_state.bfs_txpower = ni->ni_txpower;
bf->bf_state.bfs_txrate0 = txrate;
bf->bf_state.bfs_try0 = try0;
bf->bf_state.bfs_keyix = keyix;
bf->bf_state.bfs_txantenna = sc->sc_txantenna;
Remove duplicate txflags field from ath_buf. rename bf_state.bfs_flags to bf_state.bfs_txflags, as that is what it effectively is.
2012-04-07 02:01:26 +00:00
bf->bf_state.bfs_txflags = flags;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_shpream = shortPreamble;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* XXX this should be done in ath_tx_setrate() */
bf->bf_state.bfs_ctsrate0 = 0; /* ie, no hard-coded ctsrate */
bf->bf_state.bfs_ctsrate = 0; /* calculated later */
bf->bf_state.bfs_ctsduration = 0;
bf->bf_state.bfs_ismrr = ismrr;
return 0;
}
/*
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
* Queue a frame to the hardware or software queue.
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*
* This can be called by the net80211 code.
*
* XXX what about locking? Or, push the seqno assign into the
* XXX aggregate scheduler so its serialised?
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
*
* XXX When sending management frames via ath_raw_xmit(),
* should CLRDMASK be set unconditionally?
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*/
int
ath_tx_start(struct ath_softc *sc, struct ieee80211_node *ni,
struct ath_buf *bf, struct mbuf *m0)
{
struct ieee80211vap *vap = ni->ni_vap;
struct ath_vap *avp = ATH_VAP(vap);
Don't flood the cabq/mcastq with frames. In a very noisy 2.4GHz environment (with HT/40 enabled, making it worse) I saw the following occur: * the air was considered "busy" a lot of the time; * the cabq time is quite short due to staggered beacons being enabled; * it just wasn't able to keep up TX'ing CABQ frames; * .. and the cabq would swallow up all the TX ath_buf's. This patch introduces a twiddle which allows the maximum cabq depth to be set, forcing further frames to be dropped. It defaults to the TX buffer count at the moment, so the default behaviour isn't changed. I've also started fleshing out a similar setup for the data path, so it doesn't swallow up all the available TX buffers and preventing management frames (such as ADDBA) out. PR: kern/165895
2012-03-10 04:14:04 +00:00
int r = 0;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
u_int pri;
int tid;
struct ath_txq *txq;
int ismcast;
const struct ieee80211_frame *wh;
int is_ampdu, is_ampdu_tx, is_ampdu_pending;
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
ieee80211_seq seqno;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
uint8_t type, subtype;
Modify the TX path to set and use the 11n rate scenario bits. This isn't strictly required to TX (at least non-agg and non-HT40, non-short-GI) frames; but as it needs to be done anyway, just get it done. Linux ath9k uses the rate scenario style path for -all- packets, legacy or otherwise. This code does much the same. Beacon TX still uses the legacy, non-rate-scenario TX descriptor setup. Ath9k also does this. This 11n rate scenario path is only called for chips in the AR5416 HAL; legacy chips use the previous interface for TX'ing.
2011-02-03 20:30:17 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
* Determine the target hardware queue.
*
After reviewing the mcast/sleep station code a little, undo some brain damage which I committed when I had less clue about such things. Don't ever put normal data frames on the mcast software queue. Just put mcast frames there if needed. Pass the txq decision into ath_tx_normal_setup(), as we've already made the decision. Don't re-do it. Whilst i'm here, add another random debugging statement.
2012-04-08 00:40:16 +00:00
* For multicast frames, the txq gets overridden appropriately
* depending upon the state of PS.
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*
* For any other frame, we do a TID/QoS lookup inside the frame
* to see what the TID should be. If it's a non-QoS frame, the
* AC and TID are overridden. The TID/TXQ code assumes the
* TID is on a predictable hardware TXQ, so we don't support
* having a node TID queued to multiple hardware TXQs.
* This may change in the future but would require some locking
* fudgery.
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
pri = ath_tx_getac(sc, m0);
tid = ath_tx_gettid(sc, m0);
txq = sc->sc_ac2q[pri];
wh = mtod(m0, struct ieee80211_frame *);
ismcast = IEEE80211_IS_MULTICAST(wh->i_addr1);
type = wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK;
subtype = wh->i_fc[0] & IEEE80211_FC0_SUBTYPE_MASK;
Don't flood the cabq/mcastq with frames. In a very noisy 2.4GHz environment (with HT/40 enabled, making it worse) I saw the following occur: * the air was considered "busy" a lot of the time; * the cabq time is quite short due to staggered beacons being enabled; * it just wasn't able to keep up TX'ing CABQ frames; * .. and the cabq would swallow up all the TX ath_buf's. This patch introduces a twiddle which allows the maximum cabq depth to be set, forcing further frames to be dropped. It defaults to the TX buffer count at the moment, so the default behaviour isn't changed. I've also started fleshing out a similar setup for the data path, so it doesn't swallow up all the available TX buffers and preventing management frames (such as ADDBA) out. PR: kern/165895
2012-03-10 04:14:04 +00:00
/*
* Enforce how deep the multicast queue can grow.
*
* XXX duplicated in ath_raw_xmit().
*/
if (IEEE80211_IS_MULTICAST(wh->i_addr1)) {
Fix a panic introduced in a previous commit - non-beaconing modes (eg STA) don't setup the avp mcast queue. This is a bit annoying though - it turns out the mcast queue isn't initialised for STA mode but it's then touched to see whether anything is in it. That should be fixed in a subsequent commit. Noticed by: gperez@entel.upc.edu PR: kern/165895
2012-03-10 19:58:23 +00:00
if (sc->sc_cabq->axq_depth > sc->sc_txq_mcastq_maxdepth) {
Don't flood the cabq/mcastq with frames. In a very noisy 2.4GHz environment (with HT/40 enabled, making it worse) I saw the following occur: * the air was considered "busy" a lot of the time; * the cabq time is quite short due to staggered beacons being enabled; * it just wasn't able to keep up TX'ing CABQ frames; * .. and the cabq would swallow up all the TX ath_buf's. This patch introduces a twiddle which allows the maximum cabq depth to be set, forcing further frames to be dropped. It defaults to the TX buffer count at the moment, so the default behaviour isn't changed. I've also started fleshing out a similar setup for the data path, so it doesn't swallow up all the available TX buffers and preventing management frames (such as ADDBA) out. PR: kern/165895
2012-03-10 04:14:04 +00:00
sc->sc_stats.ast_tx_mcastq_overflow++;
r = ENOBUFS;
}
if (r != 0) {
m_freem(m0);
return r;
}
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* A-MPDU TX */
is_ampdu_tx = ath_tx_ampdu_running(sc, ATH_NODE(ni), tid);
is_ampdu_pending = ath_tx_ampdu_pending(sc, ATH_NODE(ni), tid);
is_ampdu = is_ampdu_tx | is_ampdu_pending;
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: tid=%d, ac=%d, is_ampdu=%d\n",
__func__, tid, pri, is_ampdu);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
/* Set local packet state, used to queue packets to hardware */
bf->bf_state.bfs_tid = tid;
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
bf->bf_state.bfs_tx_queue = txq->axq_qnum;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
bf->bf_state.bfs_pri = pri;
Document that we may end up with some suboptimal handling of data frames with stations in power saving mode. I'm not (yet) sure how to handle TX'ing aggregates frames to stations that are in power saving mode, or whether that's even a feasible thing to do. So in order to (mostly) not forget, leave a couple of comments in the code. The code presently assumes that the aggregation TID state for an ath_node is locked not by the ath_node lock or a node+TID lock, but behind the hardware queue said TID maps to. This assumption is going to be incorrect for stations in power saving mode as we'll be TX'ing frames on the multicast queue. In any case, I'm afraid its a "later problem". :/
2012-03-09 22:58:34 +00:00
/*
After reviewing the mcast/sleep station code a little, undo some brain damage which I committed when I had less clue about such things. Don't ever put normal data frames on the mcast software queue. Just put mcast frames there if needed. Pass the txq decision into ath_tx_normal_setup(), as we've already made the decision. Don't re-do it. Whilst i'm here, add another random debugging statement.
2012-04-08 00:40:16 +00:00
* When servicing one or more stations in power-save mode
* (or) if there is some mcast data waiting on the mcast
* queue (to prevent out of order delivery) multicast frames
* must be bufferd until after the beacon.
*
* TODO: we should lock the mcastq before we check the length.
Document that we may end up with some suboptimal handling of data frames with stations in power saving mode. I'm not (yet) sure how to handle TX'ing aggregates frames to stations that are in power saving mode, or whether that's even a feasible thing to do. So in order to (mostly) not forget, leave a couple of comments in the code. The code presently assumes that the aggregation TID state for an ath_node is locked not by the ath_node lock or a node+TID lock, but behind the hardware queue said TID maps to. This assumption is going to be incorrect for stations in power saving mode as we'll be TX'ing frames on the multicast queue. In any case, I'm afraid its a "later problem". :/
2012-03-09 22:58:34 +00:00
*/
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
if (ismcast && (vap->iv_ps_sta || avp->av_mcastq.axq_depth)) {
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
txq = &avp->av_mcastq;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
/*
* Mark the frame as eventually belonging on the CAB
* queue, so the descriptor setup functions will
* correctly initialise the descriptor 'qcuId' field.
*/
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
bf->bf_state.bfs_tx_queue = sc->sc_cabq->axq_qnum;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Do the generic frame setup */
/* XXX should just bzero the bf_state? */
bf->bf_state.bfs_dobaw = 0;
/* A-MPDU TX? Manually set sequence number */
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
/*
* Don't do it whilst pending; the net80211 layer still
* assigns them.
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (is_ampdu_tx) {
/*
* Always call; this function will
* handle making sure that null data frames
* don't get a sequence number from the current
* TID and thus mess with the BAW.
*/
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
seqno = ath_tx_tid_seqno_assign(sc, ni, bf, m0);
Finish undoing the previous commit - this part of the code is no longer required. PR: kern/166190
2012-06-11 07:08:40 +00:00
/*
* Don't add QoS NULL frames to the BAW.
*/
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
if (IEEE80211_QOS_HAS_SEQ(wh) &&
subtype != IEEE80211_FC0_SUBTYPE_QOS_NULL) {
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_dobaw = 1;
}
}
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
* If needed, the sequence number has been assigned.
* Squirrel it away somewhere easy to get to.
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
*/
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
bf->bf_state.bfs_seqno = M_SEQNO_GET(m0) << IEEE80211_SEQ_SEQ_SHIFT;
Modify the TX path to set and use the 11n rate scenario bits. This isn't strictly required to TX (at least non-agg and non-HT40, non-short-GI) frames; but as it needs to be done anyway, just get it done. Linux ath9k uses the rate scenario style path for -all- packets, legacy or otherwise. This code does much the same. Beacon TX still uses the legacy, non-rate-scenario TX descriptor setup. Ath9k also does this. This 11n rate scenario path is only called for chips in the AR5416 HAL; legacy chips use the previous interface for TX'ing.
2011-02-03 20:30:17 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Is ampdu pending? fetch the seqno and print it out */
if (is_ampdu_pending)
DPRINTF(sc, ATH_DEBUG_SW_TX,
"%s: tid %d: ampdu pending, seqno %d\n",
__func__, tid, M_SEQNO_GET(m0));
/* This also sets up the DMA map */
After reviewing the mcast/sleep station code a little, undo some brain damage which I committed when I had less clue about such things. Don't ever put normal data frames on the mcast software queue. Just put mcast frames there if needed. Pass the txq decision into ath_tx_normal_setup(), as we've already made the decision. Don't re-do it. Whilst i'm here, add another random debugging statement.
2012-04-08 00:40:16 +00:00
r = ath_tx_normal_setup(sc, ni, bf, m0, txq);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (r != 0)
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
goto done;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* At this point m0 could have changed! */
m0 = bf->bf_m;
#if 1
/*
* If it's a multicast frame, do a direct-dispatch to the
* destination hardware queue. Don't bother software
* queuing it.
*/
/*
* If it's a BAR frame, do a direct dispatch to the
* destination hardware queue. Don't bother software
* queuing it, as the TID will now be paused.
* Sending a BAR frame can occur from the net80211 txa timer
* (ie, retries) or from the ath txtask (completion call.)
* It queues directly to hardware because the TID is paused
* at this point (and won't be unpaused until the BAR has
* either been TXed successfully or max retries has been
* reached.)
*/
if (txq == &avp->av_mcastq) {
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX,
Delay sequence number allocation for A-MPDU until just before the frame is queued to the hardware. Because multiple concurrent paths can execute ath_start(), multiple concurrent paths can push frames into the software/hardware TX queue and since preemption/interrupting can occur, there's the possibility that a gap in time will occur between allocating the sequence number and queuing it to the hardware. Because of this, it's possible that a thread will have allocated a sequence number and then be preempted by another thread doing the same. If the second thread sneaks the frame into the BAW, the (earlier) sequence number of the first frame will be now outside the BAW and will result in the frame being constantly re-added to the tail of the queue. There it will live until the sequence numbers cycle around again. This also creates a hole in the RX BAW tracking which can also cause issues. This patch delays the sequence number allocation to occur only just before the frame is going to be added to the BAW. I've been wanting to do this anyway as part of a general code tidyup but I've not gotten around to it. This fixes the PR. However, it still makes it quite difficult to try and ensure in-order queuing and dequeuing of frames. Since multiple copies of ath_start() can be run at the same time (eg one TXing process thread, one TX completion task/one RX task) the driver may end up having frames dequeued and pushed into the hardware slightly/occasionally out of order. And, to make matters more annoying, net80211 may have the same behaviour - in the non-aggregation case, the TX code allocates sequence numbers before it's thrown to the driver. I'll open another PR to investigate this and potentially introduce some kind of final-pass TX serialisation before frames are thrown to the hardware. It's also very likely worthwhile adding some debugging code into ath(4) and net80211 to catch when/if this does occur. PR: kern/166190
2012-03-20 04:50:25 +00:00
"%s: bf=%p: mcastq: TX'ing\n", __func__, bf);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_xmit_normal(sc, txq, bf);
} else if (type == IEEE80211_FC0_TYPE_CTL &&
subtype == IEEE80211_FC0_SUBTYPE_BAR) {
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX,
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
"%s: BAR: TX'ing direct\n", __func__);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_xmit_normal(sc, txq, bf);
} else {
/* add to software queue */
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX,
Delay sequence number allocation for A-MPDU until just before the frame is queued to the hardware. Because multiple concurrent paths can execute ath_start(), multiple concurrent paths can push frames into the software/hardware TX queue and since preemption/interrupting can occur, there's the possibility that a gap in time will occur between allocating the sequence number and queuing it to the hardware. Because of this, it's possible that a thread will have allocated a sequence number and then be preempted by another thread doing the same. If the second thread sneaks the frame into the BAW, the (earlier) sequence number of the first frame will be now outside the BAW and will result in the frame being constantly re-added to the tail of the queue. There it will live until the sequence numbers cycle around again. This also creates a hole in the RX BAW tracking which can also cause issues. This patch delays the sequence number allocation to occur only just before the frame is going to be added to the BAW. I've been wanting to do this anyway as part of a general code tidyup but I've not gotten around to it. This fixes the PR. However, it still makes it quite difficult to try and ensure in-order queuing and dequeuing of frames. Since multiple copies of ath_start() can be run at the same time (eg one TXing process thread, one TX completion task/one RX task) the driver may end up having frames dequeued and pushed into the hardware slightly/occasionally out of order. And, to make matters more annoying, net80211 may have the same behaviour - in the non-aggregation case, the TX code allocates sequence numbers before it's thrown to the driver. I'll open another PR to investigate this and potentially introduce some kind of final-pass TX serialisation before frames are thrown to the hardware. It's also very likely worthwhile adding some debugging code into ath(4) and net80211 to catch when/if this does occur. PR: kern/166190
2012-03-20 04:50:25 +00:00
"%s: bf=%p: swq: TX'ing\n", __func__, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_swq(sc, ni, txq, bf);
}
#else
/*
* For now, since there's no software queue,
* direct-dispatch to the hardware.
*/
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_xmit_normal(sc, txq, bf);
#endif
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
done:
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
return 0;
}
static int
ath_tx_raw_start(struct ath_softc *sc, struct ieee80211_node *ni,
struct ath_buf *bf, struct mbuf *m0,
const struct ieee80211_bpf_params *params)
{
struct ifnet *ifp = sc->sc_ifp;
struct ieee80211com *ic = ifp->if_l2com;
struct ath_hal *ah = sc->sc_ah;
struct ieee80211vap *vap = ni->ni_vap;
int error, ismcast, ismrr;
int keyix, hdrlen, pktlen, try0, txantenna;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
u_int8_t rix, txrate;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
struct ieee80211_frame *wh;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
u_int flags;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
HAL_PKT_TYPE atype;
const HAL_RATE_TABLE *rt;
struct ath_desc *ds;
u_int pri;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
int o_tid = -1;
int do_override;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
wh = mtod(m0, struct ieee80211_frame *);
ismcast = IEEE80211_IS_MULTICAST(wh->i_addr1);
hdrlen = ieee80211_anyhdrsize(wh);
/*
* Packet length must not include any
* pad bytes; deduct them here.
*/
/* XXX honor IEEE80211_BPF_DATAPAD */
pktlen = m0->m_pkthdr.len - (hdrlen & 3) + IEEE80211_CRC_LEN;
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 2,
"ath_tx_raw_start: ni=%p, bf=%p, raw", ni, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: ismcast=%d\n",
__func__, ismcast);
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
pri = params->ibp_pri & 3;
/* Override pri if the frame isn't a QoS one */
if (! IEEE80211_QOS_HAS_SEQ(wh))
pri = ath_tx_getac(sc, m0);
/* XXX If it's an ADDBA, override the correct queue */
do_override = ath_tx_action_frame_override_queue(sc, ni, m0, &o_tid);
/* Map ADDBA to the correct priority */
if (do_override) {
#if 0
device_printf(sc->sc_dev,
"%s: overriding tid %d pri %d -> %d\n",
__func__, o_tid, pri, TID_TO_WME_AC(o_tid));
#endif
pri = TID_TO_WME_AC(o_tid);
}
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
/* Handle encryption twiddling if needed */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (! ath_tx_tag_crypto(sc, ni,
m0, params->ibp_flags & IEEE80211_BPF_CRYPTO, 0,
&hdrlen, &pktlen, &keyix)) {
* Add a rather hacky "does this speak the 11n TX descriptor format" function; which will be later used by the TX path to determine whether to use the extended features or not. * Break out the descriptor chaining logic into a separate function; again so it can be switched out later on for the 11n version when needed. * Refactor out the encryption-swizzling code that's common in the raw and normal TX path.
2011-02-01 06:59:44 +00:00
ath_freetx(m0);
return EIO;
}
/* packet header may have moved, reset our local pointer */
wh = mtod(m0, struct ieee80211_frame *);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Do the generic frame setup */
/* XXX should just bzero the bf_state? */
bf->bf_state.bfs_dobaw = 0;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
error = ath_tx_dmasetup(sc, bf, m0);
if (error != 0)
return error;
m0 = bf->bf_m; /* NB: may have changed */
wh = mtod(m0, struct ieee80211_frame *);
bf->bf_node = ni; /* NB: held reference */
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/* Always enable CLRDMASK for raw frames for now.. */
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
flags = HAL_TXDESC_CLRDMASK; /* XXX needed for crypto errs */
flags |= HAL_TXDESC_INTREQ; /* force interrupt */
if (params->ibp_flags & IEEE80211_BPF_RTS)
flags |= HAL_TXDESC_RTSENA;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
else if (params->ibp_flags & IEEE80211_BPF_CTS) {
/* XXX assume 11g/11n protection? */
bf->bf_state.bfs_doprot = 1;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
flags |= HAL_TXDESC_CTSENA;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/* XXX leave ismcast to injector? */
if ((params->ibp_flags & IEEE80211_BPF_NOACK) || ismcast)
flags |= HAL_TXDESC_NOACK;
rt = sc->sc_currates;
KASSERT(rt != NULL, ("no rate table, mode %u", sc->sc_curmode));
rix = ath_tx_findrix(sc, params->ibp_rate0);
txrate = rt->info[rix].rateCode;
if (params->ibp_flags & IEEE80211_BPF_SHORTPRE)
txrate |= rt->info[rix].shortPreamble;
sc->sc_txrix = rix;
try0 = params->ibp_try0;
ismrr = (params->ibp_try1 != 0);
txantenna = params->ibp_pri >> 2;
if (txantenna == 0) /* XXX? */
txantenna = sc->sc_txantenna;
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Since ctsrate is fixed, store it away for later
* use when the descriptor fields are being set.
*/
if (flags & (HAL_TXDESC_RTSENA|HAL_TXDESC_CTSENA))
bf->bf_state.bfs_ctsrate0 = params->ibp_ctsrate;
Refator the common code which calculates the 802.11g protection duration.
2011-02-01 07:50:26 +00:00
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
* NB: we mark all packets as type PSPOLL so the h/w won't
* set the sequence number, duration, etc.
*/
atype = HAL_PKT_TYPE_PSPOLL;
if (IFF_DUMPPKTS(sc, ATH_DEBUG_XMIT))
ieee80211_dump_pkt(ic, mtod(m0, caddr_t), m0->m_len,
sc->sc_hwmap[rix].ieeerate, -1);
if (ieee80211_radiotap_active_vap(vap)) {
u_int64_t tsf = ath_hal_gettsf64(ah);
sc->sc_tx_th.wt_tsf = htole64(tsf);
sc->sc_tx_th.wt_flags = sc->sc_hwmap[rix].txflags;
if (wh->i_fc[1] & IEEE80211_FC1_WEP)
sc->sc_tx_th.wt_flags |= IEEE80211_RADIOTAP_F_WEP;
if (m0->m_flags & M_FRAG)
sc->sc_tx_th.wt_flags |= IEEE80211_RADIOTAP_F_FRAG;
sc->sc_tx_th.wt_rate = sc->sc_hwmap[rix].ieeerate;
sc->sc_tx_th.wt_txpower = ni->ni_txpower;
sc->sc_tx_th.wt_antenna = sc->sc_txantenna;
ieee80211_radiotap_tx(vap, m0);
}
/*
* Formulate first tx descriptor with tx controls.
*/
ds = bf->bf_desc;
/* XXX check return value? */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Store the decided rate index values away */
bf->bf_state.bfs_pktlen = pktlen;
bf->bf_state.bfs_hdrlen = hdrlen;
bf->bf_state.bfs_atype = atype;
bf->bf_state.bfs_txpower = params->ibp_power;
bf->bf_state.bfs_txrate0 = txrate;
bf->bf_state.bfs_try0 = try0;
bf->bf_state.bfs_keyix = keyix;
bf->bf_state.bfs_txantenna = txantenna;
Remove duplicate txflags field from ath_buf. rename bf_state.bfs_flags to bf_state.bfs_txflags, as that is what it effectively is.
2012-04-07 02:01:26 +00:00
bf->bf_state.bfs_txflags = flags;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_shpream =
!! (params->ibp_flags & IEEE80211_BPF_SHORTPRE);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
/* Set local packet state, used to queue packets to hardware */
bf->bf_state.bfs_tid = WME_AC_TO_TID(pri);
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
bf->bf_state.bfs_tx_queue = sc->sc_ac2q[pri]->axq_qnum;
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
bf->bf_state.bfs_pri = pri;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* XXX this should be done in ath_tx_setrate() */
bf->bf_state.bfs_ctsrate = 0;
bf->bf_state.bfs_ctsduration = 0;
bf->bf_state.bfs_ismrr = ismrr;
Modify the TX path to set and use the 11n rate scenario bits. This isn't strictly required to TX (at least non-agg and non-HT40, non-short-GI) frames; but as it needs to be done anyway, just get it done. Linux ath9k uses the rate scenario style path for -all- packets, legacy or otherwise. This code does much the same. Beacon TX still uses the legacy, non-rate-scenario TX descriptor setup. Ath9k also does this. This 11n rate scenario path is only called for chips in the AR5416 HAL; legacy chips use the previous interface for TX'ing.
2011-02-03 20:30:17 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Blank the legacy rate array */
bzero(&bf->bf_state.bfs_rc, sizeof(bf->bf_state.bfs_rc));
bf->bf_state.bfs_rc[0].rix =
ath_tx_findrix(sc, params->ibp_rate0);
bf->bf_state.bfs_rc[0].tries = try0;
bf->bf_state.bfs_rc[0].ratecode = txrate;
if (ismrr) {
int rix;
rix = ath_tx_findrix(sc, params->ibp_rate1);
bf->bf_state.bfs_rc[1].rix = rix;
bf->bf_state.bfs_rc[1].tries = params->ibp_try1;
rix = ath_tx_findrix(sc, params->ibp_rate2);
bf->bf_state.bfs_rc[2].rix = rix;
bf->bf_state.bfs_rc[2].tries = params->ibp_try2;
rix = ath_tx_findrix(sc, params->ibp_rate3);
bf->bf_state.bfs_rc[3].rix = rix;
bf->bf_state.bfs_rc[3].tries = params->ibp_try3;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* All the required rate control decisions have been made;
* fill in the rc flags.
*/
ath_tx_rate_fill_rcflags(sc, bf);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/* NB: no buffered multicast in power save support */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* If we're overiding the ADDBA destination, dump directly
* into the hardware queue, right after any pending
* frames to that node are.
*/
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: dooverride=%d\n",
__func__, do_override);
In (eventual) preparation for supporting disabling the whole 11n/software retry path - add some code to make it obvious (to me!) how to disable the software tx path.
2012-09-24 06:00:51 +00:00
#if 1
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (do_override) {
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_xmit_normal(sc, sc->sc_ac2q[pri], bf);
} else {
/* Queue to software queue */
ath_tx_swq(sc, ni, sc->sc_ac2q[pri], bf);
}
In (eventual) preparation for supporting disabling the whole 11n/software retry path - add some code to make it obvious (to me!) how to disable the software tx path.
2012-09-24 06:00:51 +00:00
#else
/* Direct-dispatch to the hardware */
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
ath_tx_xmit_normal(sc, sc->sc_ac2q[pri], bf);
#endif
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
return 0;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Send a raw frame.
*
* This can be called by net80211.
*/
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
int
ath_raw_xmit(struct ieee80211_node *ni, struct mbuf *m,
const struct ieee80211_bpf_params *params)
{
struct ieee80211com *ic = ni->ni_ic;
struct ifnet *ifp = ic->ic_ifp;
struct ath_softc *sc = ifp->if_softc;
struct ath_buf *bf;
Don't flood the cabq/mcastq with frames. In a very noisy 2.4GHz environment (with HT/40 enabled, making it worse) I saw the following occur: * the air was considered "busy" a lot of the time; * the cabq time is quite short due to staggered beacons being enabled; * it just wasn't able to keep up TX'ing CABQ frames; * .. and the cabq would swallow up all the TX ath_buf's. This patch introduces a twiddle which allows the maximum cabq depth to be set, forcing further frames to be dropped. It defaults to the TX buffer count at the moment, so the default behaviour isn't changed. I've also started fleshing out a similar setup for the data path, so it doesn't swallow up all the available TX buffers and preventing management frames (such as ADDBA) out. PR: kern/165895
2012-03-10 04:14:04 +00:00
struct ieee80211_frame *wh = mtod(m, struct ieee80211_frame *);
int error = 0;
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
Flesh out some slightly dirty reset/channel change serialisation code for the ath(4) driver. Currently, there's nothing stopping reset, channel change and general TX/RX from overlapping with each other. This wasn't a big deal with pre-11n traffic as it just results in some dropped frames. It's possible this may have also caused some inconsistencies and badly-setup hardware. Since locks can't be held across all of this (the Linux solution) due to LORs with the network stack locks, some state counter variables are used to track what parts of the code the driver is currently in. When the hardware is being reset, it disables the taskqueue and waits for pending interrupts, tx, rx and tx completion before it begins the reset or channel change. TX and RX both abort if called during an active reset or channel change. Finally, the reset path now doesn't flush frames if ATH_RESET_NOLOSS is set. Instead, completed TX and RX frames are passed back up to net80211 before the reset occurs. This is not without problems: * Raw frame xmit are just dropped, rather than placed on a queue. The net80211 stack should be the one which queues these frames rather than the driver. * It's all very messy. It'd be better if these hardware operations were serialised on some kind of work queue, rather than hoping they can be run in parallel. * The taskqueue block/unblock may occur in parallel with the newstate() function - which shuts down the taskqueue and restarts it once the new state is known. It's likely these operations should be refcounted so the taskqueue is restored once no other areas in the code wish to suspend operations. * .. interrupt disable/enable should likely be refcounted as well. With this work, the driver does not drop frames during stuck beacon or fatal errors and thus 11n traffic continues to run correctly. Default and full resets however do still drop frames and it's possible this may occur, causing traffic loss and session stalls. Sponsored by: Hobnob, Inc.
2011-11-18 05:06:30 +00:00
ATH_PCU_LOCK(sc);
if (sc->sc_inreset_cnt > 0) {
device_printf(sc->sc_dev, "%s: sc_inreset_cnt > 0; bailing\n",
__func__);
error = EIO;
ATH_PCU_UNLOCK(sc);
goto bad0;
}
sc->sc_txstart_cnt++;
ATH_PCU_UNLOCK(sc);
I give up - introduce a TX lock to serialise TX operations. I've tried serialising TX using queues and such but unfortunately due to how this interacts with the locking going on elsewhere in the networking stack, the TX task gets delayed, resulting in quite a noticable throughput loss: * baseline TCP for 2x2 11n HT40 is ~ 170mbit/sec; * TCP for TX task in the ath taskq, with the RX also going on - 80mbit/sec; * TCP for TX task in a separate, second taskq - 100mbit/sec. So for now I'm going with the Linux wireless stack approach - lock tx early. The linux code does in the wireless stack, before the 802.11 state stuff happens and before it's punted to the driver. But TX locking needs to also occur at the driver layer as the TX completion code _also_ begins to drain the ifnet TX queue. Whilst I'm here, add some KTR traces for the TX path. Note: * This really should be done at the net80211 layer (as well, at least.) But that'll have to wait for a little more thought to happen.
2012-10-31 06:27:58 +00:00
ATH_TX_LOCK(sc);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || sc->sc_invalid) {
DPRINTF(sc, ATH_DEBUG_XMIT, "%s: discard frame, %s", __func__,
(ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ?
"!running" : "invalid");
m_freem(m);
error = ENETDOWN;
goto bad;
}
Don't flood the cabq/mcastq with frames. In a very noisy 2.4GHz environment (with HT/40 enabled, making it worse) I saw the following occur: * the air was considered "busy" a lot of the time; * the cabq time is quite short due to staggered beacons being enabled; * it just wasn't able to keep up TX'ing CABQ frames; * .. and the cabq would swallow up all the TX ath_buf's. This patch introduces a twiddle which allows the maximum cabq depth to be set, forcing further frames to be dropped. It defaults to the TX buffer count at the moment, so the default behaviour isn't changed. I've also started fleshing out a similar setup for the data path, so it doesn't swallow up all the available TX buffers and preventing management frames (such as ADDBA) out. PR: kern/165895
2012-03-10 04:14:04 +00:00
/*
* Enforce how deep the multicast queue can grow.
*
* XXX duplicated in ath_tx_start().
*/
if (IEEE80211_IS_MULTICAST(wh->i_addr1)) {
Fix a panic introduced in a previous commit - non-beaconing modes (eg STA) don't setup the avp mcast queue. This is a bit annoying though - it turns out the mcast queue isn't initialised for STA mode but it's then touched to see whether anything is in it. That should be fixed in a subsequent commit. Noticed by: gperez@entel.upc.edu PR: kern/165895
2012-03-10 19:58:23 +00:00
if (sc->sc_cabq->axq_depth > sc->sc_txq_mcastq_maxdepth) {
Don't flood the cabq/mcastq with frames. In a very noisy 2.4GHz environment (with HT/40 enabled, making it worse) I saw the following occur: * the air was considered "busy" a lot of the time; * the cabq time is quite short due to staggered beacons being enabled; * it just wasn't able to keep up TX'ing CABQ frames; * .. and the cabq would swallow up all the TX ath_buf's. This patch introduces a twiddle which allows the maximum cabq depth to be set, forcing further frames to be dropped. It defaults to the TX buffer count at the moment, so the default behaviour isn't changed. I've also started fleshing out a similar setup for the data path, so it doesn't swallow up all the available TX buffers and preventing management frames (such as ADDBA) out. PR: kern/165895
2012-03-10 04:14:04 +00:00
sc->sc_stats.ast_tx_mcastq_overflow++;
error = ENOBUFS;
}
if (error != 0) {
m_freem(m);
goto bad;
}
}
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
/*
* Grab a TX buffer and associated resources.
*/
Implement a separate, smaller pool of ath_buf entries for use by management traffic. * Create sc_mgmt_txbuf and sc_mgmt_txdesc, initialise/free them appropriately. * Create an enum to represent buffer types in the API. * Extend ath_getbuf() and _ath_getbuf_locked() to take the above enum. * Right now anything sent via ic_raw_xmit() allocates via ATH_BUFTYPE_MGMT. This may not be very useful. * Add ATH_BUF_MGMT flag (ath_buf.bf_flags) which indicates the current buffer is a mgmt buffer and should go back onto the mgmt free list. * Extend 'txagg' to include debugging output for both normal and mgmt txbufs. * When checking/clearing ATH_BUF_BUSY, do it on both TX pools. Tested: * STA mode, with heavy UDP injection via iperf. This filled the TX queue however BARs were still going out successfully. TODO: * Initialise the mgmt buffers with ATH_BUF_MGMT and then ensure the right type is being allocated and freed on the appropriate list. That'd save a write operation (to bf->bf_flags) on each buffer alloc/free. * Test on AP mode, ensure that BAR TX and probe responses go out nicely when the main TX queue is filled (eg with paused traffic to a TID, awaiting a BAR to complete.) PR: kern/168170
2012-06-13 06:57:55 +00:00
bf = ath_getbuf(sc, ATH_BUFTYPE_MGMT);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if (bf == NULL) {
sc->sc_stats.ast_tx_nobuf++;
m_freem(m);
error = ENOBUFS;
goto bad;
}
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 3, "ath_raw_xmit: m=%p, params=%p, bf=%p\n",
m, params, bf);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
if (params == NULL) {
/*
* Legacy path; interpret frame contents to decide
* precisely how to send the frame.
*/
if (ath_tx_start(sc, ni, bf, m)) {
error = EIO; /* XXX */
goto bad2;
}
} else {
/*
* Caller supplied explicit parameters to use in
* sending the frame.
*/
if (ath_tx_raw_start(sc, ni, bf, m, params)) {
error = EIO; /* XXX */
goto bad2;
}
}
sc->sc_wd_timer = 5;
ifp->if_opackets++;
sc->sc_stats.ast_tx_raw++;
Begin fleshing out some software queue awareness for TIM handling with the power save queue. * introduce some new ATH_NODE lock protected fields, tracking the net80211 psq and TIM state; * when doing buffer transitions - ie, when sending and completing buffers - check the state of the SWQ and update the TIM appropriately. * when clearing the TIM bit, if the SWQ is not empty then delay clearing it. This is racy, but it's no less racy than the current net80211 power save queue management code. Specifically, with multiple TX threads, it's quite plausible that parallel state updates will race and the TIM will be left in an inconsistent state. I'll address that in a follow-up commit.
2012-10-28 21:13:12 +00:00
/*
* Update the TIM - if there's anything queued to the
* software queue and power save is enabled, we should
* set the TIM.
*/
ath_tx_update_tim(sc, ni, 1);
Don't grab the PCU lock inside the TX lock.
2012-12-02 06:50:27 +00:00
ATH_TX_UNLOCK(sc);
Flesh out some slightly dirty reset/channel change serialisation code for the ath(4) driver. Currently, there's nothing stopping reset, channel change and general TX/RX from overlapping with each other. This wasn't a big deal with pre-11n traffic as it just results in some dropped frames. It's possible this may have also caused some inconsistencies and badly-setup hardware. Since locks can't be held across all of this (the Linux solution) due to LORs with the network stack locks, some state counter variables are used to track what parts of the code the driver is currently in. When the hardware is being reset, it disables the taskqueue and waits for pending interrupts, tx, rx and tx completion before it begins the reset or channel change. TX and RX both abort if called during an active reset or channel change. Finally, the reset path now doesn't flush frames if ATH_RESET_NOLOSS is set. Instead, completed TX and RX frames are passed back up to net80211 before the reset occurs. This is not without problems: * Raw frame xmit are just dropped, rather than placed on a queue. The net80211 stack should be the one which queues these frames rather than the driver. * It's all very messy. It'd be better if these hardware operations were serialised on some kind of work queue, rather than hoping they can be run in parallel. * The taskqueue block/unblock may occur in parallel with the newstate() function - which shuts down the taskqueue and restarts it once the new state is known. It's likely these operations should be refcounted so the taskqueue is restored once no other areas in the code wish to suspend operations. * .. interrupt disable/enable should likely be refcounted as well. With this work, the driver does not drop frames during stuck beacon or fatal errors and thus 11n traffic continues to run correctly. Default and full resets however do still drop frames and it's possible this may occur, causing traffic loss and session stalls. Sponsored by: Hobnob, Inc.
2011-11-18 05:06:30 +00:00
ATH_PCU_LOCK(sc);
sc->sc_txstart_cnt--;
ATH_PCU_UNLOCK(sc);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
return 0;
bad2:
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 3, "ath_raw_xmit: bad2: m=%p, params=%p, "
"bf=%p",
m,
params,
bf);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
ATH_TXBUF_LOCK(sc);
Replace the direct sc_txbuf manipulation with a pair of functions. This is preparation work for having a separate ath_buf queue for management traffic. PR: kern/168170
2012-06-13 05:39:16 +00:00
ath_returnbuf_head(sc, bf);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
ATH_TXBUF_UNLOCK(sc);
bad:
I give up - introduce a TX lock to serialise TX operations. I've tried serialising TX using queues and such but unfortunately due to how this interacts with the locking going on elsewhere in the networking stack, the TX task gets delayed, resulting in quite a noticable throughput loss: * baseline TCP for 2x2 11n HT40 is ~ 170mbit/sec; * TCP for TX task in the ath taskq, with the RX also going on - 80mbit/sec; * TCP for TX task in a separate, second taskq - 100mbit/sec. So for now I'm going with the Linux wireless stack approach - lock tx early. The linux code does in the wireless stack, before the 802.11 state stuff happens and before it's punted to the driver. But TX locking needs to also occur at the driver layer as the TX completion code _also_ begins to drain the ifnet TX queue. Whilst I'm here, add some KTR traces for the TX path. Note: * This really should be done at the net80211 layer (as well, at least.) But that'll have to wait for a little more thought to happen.
2012-10-31 06:27:58 +00:00
ATH_TX_UNLOCK(sc);
Flesh out some slightly dirty reset/channel change serialisation code for the ath(4) driver. Currently, there's nothing stopping reset, channel change and general TX/RX from overlapping with each other. This wasn't a big deal with pre-11n traffic as it just results in some dropped frames. It's possible this may have also caused some inconsistencies and badly-setup hardware. Since locks can't be held across all of this (the Linux solution) due to LORs with the network stack locks, some state counter variables are used to track what parts of the code the driver is currently in. When the hardware is being reset, it disables the taskqueue and waits for pending interrupts, tx, rx and tx completion before it begins the reset or channel change. TX and RX both abort if called during an active reset or channel change. Finally, the reset path now doesn't flush frames if ATH_RESET_NOLOSS is set. Instead, completed TX and RX frames are passed back up to net80211 before the reset occurs. This is not without problems: * Raw frame xmit are just dropped, rather than placed on a queue. The net80211 stack should be the one which queues these frames rather than the driver. * It's all very messy. It'd be better if these hardware operations were serialised on some kind of work queue, rather than hoping they can be run in parallel. * The taskqueue block/unblock may occur in parallel with the newstate() function - which shuts down the taskqueue and restarts it once the new state is known. It's likely these operations should be refcounted so the taskqueue is restored once no other areas in the code wish to suspend operations. * .. interrupt disable/enable should likely be refcounted as well. With this work, the driver does not drop frames during stuck beacon or fatal errors and thus 11n traffic continues to run correctly. Default and full resets however do still drop frames and it's possible this may occur, causing traffic loss and session stalls. Sponsored by: Hobnob, Inc.
2011-11-18 05:06:30 +00:00
ATH_PCU_LOCK(sc);
sc->sc_txstart_cnt--;
ATH_PCU_UNLOCK(sc);
bad0:
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_TX, 2, "ath_raw_xmit: bad0: m=%p, params=%p",
m, params);
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
ifp->if_oerrors++;
sc->sc_stats.ast_tx_raw_fail++;
ieee80211_free_node(ni);
Flesh out some slightly dirty reset/channel change serialisation code for the ath(4) driver. Currently, there's nothing stopping reset, channel change and general TX/RX from overlapping with each other. This wasn't a big deal with pre-11n traffic as it just results in some dropped frames. It's possible this may have also caused some inconsistencies and badly-setup hardware. Since locks can't be held across all of this (the Linux solution) due to LORs with the network stack locks, some state counter variables are used to track what parts of the code the driver is currently in. When the hardware is being reset, it disables the taskqueue and waits for pending interrupts, tx, rx and tx completion before it begins the reset or channel change. TX and RX both abort if called during an active reset or channel change. Finally, the reset path now doesn't flush frames if ATH_RESET_NOLOSS is set. Instead, completed TX and RX frames are passed back up to net80211 before the reset occurs. This is not without problems: * Raw frame xmit are just dropped, rather than placed on a queue. The net80211 stack should be the one which queues these frames rather than the driver. * It's all very messy. It'd be better if these hardware operations were serialised on some kind of work queue, rather than hoping they can be run in parallel. * The taskqueue block/unblock may occur in parallel with the newstate() function - which shuts down the taskqueue and restarts it once the new state is known. It's likely these operations should be refcounted so the taskqueue is restored once no other areas in the code wish to suspend operations. * .. interrupt disable/enable should likely be refcounted as well. With this work, the driver does not drop frames during stuck beacon or fatal errors and thus 11n traffic continues to run correctly. Default and full resets however do still drop frames and it's possible this may occur, causing traffic loss and session stalls. Sponsored by: Hobnob, Inc.
2011-11-18 05:06:30 +00:00
Migrate the TX path code out of if_ath and into a separate source file. There's two reasons for this: * the raw and non-raw TX path shares a lot of duplicate code which should be refactored; * the 11n-ready chip TX path needs a little reworking.
2011-01-29 11:35:23 +00:00
return error;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Some helper functions */
/*
* ADDBA (and potentially others) need to be placed in the same
* hardware queue as the TID/node it's relating to. This is so
* it goes out after any pending non-aggregate frames to the
* same node/TID.
*
* If this isn't done, the ADDBA can go out before the frames
* queued in hardware. Even though these frames have a sequence
* number -earlier- than the ADDBA can be transmitted (but
* no frames whose sequence numbers are after the ADDBA should
* be!) they'll arrive after the ADDBA - and the receiving end
* will simply drop them as being out of the BAW.
*
* The frames can't be appended to the TID software queue - it'll
* never be sent out. So these frames have to be directly
* dispatched to the hardware, rather than queued in software.
* So if this function returns true, the TXQ has to be
* overridden and it has to be directly dispatched.
*
* It's a dirty hack, but someone's gotta do it.
*/
/*
* XXX doesn't belong here!
*/
static int
ieee80211_is_action(struct ieee80211_frame *wh)
{
/* Type: Management frame? */
if ((wh->i_fc[0] & IEEE80211_FC0_TYPE_MASK) !=
IEEE80211_FC0_TYPE_MGT)
return 0;
/* Subtype: Action frame? */
if ((wh->i_fc[0] & IEEE80211_FC0_SUBTYPE_MASK) !=
IEEE80211_FC0_SUBTYPE_ACTION)
return 0;
return 1;
}
#define MS(_v, _f) (((_v) & _f) >> _f##_S)
/*
* Return an alternate TID for ADDBA request frames.
*
* Yes, this likely should be done in the net80211 layer.
*/
static int
ath_tx_action_frame_override_queue(struct ath_softc *sc,
struct ieee80211_node *ni,
struct mbuf *m0, int *tid)
{
struct ieee80211_frame *wh = mtod(m0, struct ieee80211_frame *);
struct ieee80211_action_ba_addbarequest *ia;
uint8_t *frm;
uint16_t baparamset;
/* Not action frame? Bail */
if (! ieee80211_is_action(wh))
return 0;
/* XXX Not needed for frames we send? */
#if 0
/* Correct length? */
if (! ieee80211_parse_action(ni, m))
return 0;
#endif
/* Extract out action frame */
frm = (u_int8_t *)&wh[1];
ia = (struct ieee80211_action_ba_addbarequest *) frm;
/* Not ADDBA? Bail */
if (ia->rq_header.ia_category != IEEE80211_ACTION_CAT_BA)
return 0;
if (ia->rq_header.ia_action != IEEE80211_ACTION_BA_ADDBA_REQUEST)
return 0;
/* Extract TID, return it */
baparamset = le16toh(ia->rq_baparamset);
*tid = (int) MS(baparamset, IEEE80211_BAPS_TID);
return 1;
}
#undef MS
/* Per-node software queue operations */
/*
* Add the current packet to the given BAW.
* It is assumed that the current packet
*
* + fits inside the BAW;
* + already has had a sequence number allocated.
*
* Since the BAW status may be modified by both the ath task and
* the net80211/ifnet contexts, the TID must be locked.
*/
void
ath_tx_addto_baw(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid, struct ath_buf *bf)
{
int index, cindex;
struct ieee80211_tx_ampdu *tap;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (bf->bf_state.bfs_isretried)
return;
Fix uninitialised reference. Noticed by: John Hay <jhay@meraka.org.za>
2012-06-11 12:26:23 +00:00
tap = ath_tx_get_tx_tid(an, tid->tid);
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
if (! bf->bf_state.bfs_dobaw) {
device_printf(sc->sc_dev,
"%s: dobaw=0, seqno=%d, window %d:%d\n",
__func__,
SEQNO(bf->bf_state.bfs_seqno),
tap->txa_start,
tap->txa_wnd);
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
"%s: re-added? tid=%d, seqno %d; window %d:%d; "
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
"baw head=%d tail=%d\n",
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
__func__, tid->tid, SEQNO(bf->bf_state.bfs_seqno),
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
tap->txa_start, tap->txa_wnd, tid->baw_head,
tid->baw_tail);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
/*
* Verify that the given sequence number is not outside of the
* BAW. Complain loudly if that's the case.
*/
if (! BAW_WITHIN(tap->txa_start, tap->txa_wnd,
SEQNO(bf->bf_state.bfs_seqno))) {
device_printf(sc->sc_dev,
"%s: bf=%p: outside of BAW?? tid=%d, seqno %d; window %d:%d; "
"baw head=%d tail=%d\n",
__func__, bf, tid->tid, SEQNO(bf->bf_state.bfs_seqno),
tap->txa_start, tap->txa_wnd, tid->baw_head,
tid->baw_tail);
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* ni->ni_txseqs[] is the currently allocated seqno.
* the txa state contains the current baw start.
*/
index = ATH_BA_INDEX(tap->txa_start, SEQNO(bf->bf_state.bfs_seqno));
cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
DPRINTF(sc, ATH_DEBUG_SW_TX_BAW,
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
"%s: tid=%d, seqno %d; window %d:%d; index=%d cindex=%d "
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
"baw head=%d tail=%d\n",
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
__func__, tid->tid, SEQNO(bf->bf_state.bfs_seqno),
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
tap->txa_start, tap->txa_wnd, index, cindex, tid->baw_head,
tid->baw_tail);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
#if 0
assert(tid->tx_buf[cindex] == NULL);
#endif
if (tid->tx_buf[cindex] != NULL) {
device_printf(sc->sc_dev,
"%s: ba packet dup (index=%d, cindex=%d, "
"head=%d, tail=%d)\n",
__func__, index, cindex, tid->baw_head, tid->baw_tail);
device_printf(sc->sc_dev,
"%s: BA bf: %p; seqno=%d ; new bf: %p; seqno=%d\n",
__func__,
tid->tx_buf[cindex],
SEQNO(tid->tx_buf[cindex]->bf_state.bfs_seqno),
bf,
SEQNO(bf->bf_state.bfs_seqno)
);
}
tid->tx_buf[cindex] = bf;
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
if (index >= ((tid->baw_tail - tid->baw_head) &
(ATH_TID_MAX_BUFS - 1))) {
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tid->baw_tail = cindex;
INCR(tid->baw_tail, ATH_TID_MAX_BUFS);
}
}
If software retransmit occurs with an ath_buf marked ATH_BUF_BUSY, it's cloned and that clone is retransmitted. This means that the ath_buf pointer squirreled away on the baw window array is suddenly wrong and was causing all kinds of console output. This updates the pointer in that particular BAW slot to the new ath_buf after ensuring that: * the new and old buffers have the same seqno; * the current slot pointer matches the old buffer pointer. This quietens the debugging output (again), restoring said debugging to only signify when a broken condition has occured. Sponsored by: Hobnob, Inc.
2011-11-09 18:24:20 +00:00
/*
* Flip the BAW buffer entry over from the existing one to the new one.
*
* When software retransmitting a (sub-)frame, it is entirely possible that
* the frame ath_buf is marked as BUSY and can't be immediately reused.
* In that instance the buffer is cloned and the new buffer is used for
* retransmit. We thus need to update the ath_buf slot in the BAW buf
* tracking array to maintain consistency.
*/
static void
ath_tx_switch_baw_buf(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid, struct ath_buf *old_bf, struct ath_buf *new_bf)
{
int index, cindex;
struct ieee80211_tx_ampdu *tap;
int seqno = SEQNO(old_bf->bf_state.bfs_seqno);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
If software retransmit occurs with an ath_buf marked ATH_BUF_BUSY, it's cloned and that clone is retransmitted. This means that the ath_buf pointer squirreled away on the baw window array is suddenly wrong and was causing all kinds of console output. This updates the pointer in that particular BAW slot to the new ath_buf after ensuring that: * the new and old buffers have the same seqno; * the current slot pointer matches the old buffer pointer. This quietens the debugging output (again), restoring said debugging to only signify when a broken condition has occured. Sponsored by: Hobnob, Inc.
2011-11-09 18:24:20 +00:00
tap = ath_tx_get_tx_tid(an, tid->tid);
index = ATH_BA_INDEX(tap->txa_start, seqno);
cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
/*
* Just warn for now; if it happens then we should find out
* about it. It's highly likely the aggregation session will
* soon hang.
*/
if (old_bf->bf_state.bfs_seqno != new_bf->bf_state.bfs_seqno) {
device_printf(sc->sc_dev, "%s: retransmitted buffer"
" has mismatching seqno's, BA session may hang.\n",
__func__);
device_printf(sc->sc_dev, "%s: old seqno=%d, new_seqno=%d\n",
__func__,
old_bf->bf_state.bfs_seqno,
new_bf->bf_state.bfs_seqno);
}
if (tid->tx_buf[cindex] != old_bf) {
device_printf(sc->sc_dev, "%s: ath_buf pointer incorrect; "
" has m BA session may hang.\n",
__func__);
device_printf(sc->sc_dev, "%s: old bf=%p, new bf=%p\n",
__func__,
old_bf, new_bf);
}
tid->tx_buf[cindex] = new_bf;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* seq_start - left edge of BAW
* seq_next - current/next sequence number to allocate
*
* Since the BAW status may be modified by both the ath task and
* the net80211/ifnet contexts, the TID must be locked.
*/
static void
ath_tx_update_baw(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid, const struct ath_buf *bf)
{
int index, cindex;
struct ieee80211_tx_ampdu *tap;
int seqno = SEQNO(bf->bf_state.bfs_seqno);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tap = ath_tx_get_tx_tid(an, tid->tid);
index = ATH_BA_INDEX(tap->txa_start, seqno);
cindex = (tid->baw_head + index) & (ATH_TID_MAX_BUFS - 1);
DPRINTF(sc, ATH_DEBUG_SW_TX_BAW,
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
"%s: tid=%d, baw=%d:%d, seqno=%d, index=%d, cindex=%d, "
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
"baw head=%d, tail=%d\n",
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
__func__, tid->tid, tap->txa_start, tap->txa_wnd, seqno, index,
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
cindex, tid->baw_head, tid->baw_tail);
/*
* If this occurs then we have a big problem - something else
* has slid tap->txa_start along without updating the BAW
* tracking start/end pointers. Thus the TX BAW state is now
* completely busted.
*
* But for now, since I haven't yet fixed TDMA and buffer cloning,
* it's quite possible that a cloned buffer is making its way
* here and causing it to fire off. Disable TDMA for now.
*/
if (tid->tx_buf[cindex] != bf) {
device_printf(sc->sc_dev,
"%s: comp bf=%p, seq=%d; slot bf=%p, seqno=%d\n",
__func__,
bf, SEQNO(bf->bf_state.bfs_seqno),
tid->tx_buf[cindex],
SEQNO(tid->tx_buf[cindex]->bf_state.bfs_seqno));
}
tid->tx_buf[cindex] = NULL;
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
while (tid->baw_head != tid->baw_tail &&
!tid->tx_buf[tid->baw_head]) {
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
INCR(tap->txa_start, IEEE80211_SEQ_RANGE);
INCR(tid->baw_head, ATH_TID_MAX_BUFS);
}
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAW,
"%s: baw is now %d:%d, baw head=%d\n",
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
__func__, tap->txa_start, tap->txa_wnd, tid->baw_head);
}
/*
* Mark the current node/TID as ready to TX.
*
* This is done to make it easy for the software scheduler to
* find which nodes have data to send.
*
* The TXQ lock must be held.
*/
static void
ath_tx_tid_sched(struct ath_softc *sc, struct ath_tid *tid)
{
struct ath_txq *txq = sc->sc_ac2q[tid->ac];
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (tid->paused)
return; /* paused, can't schedule yet */
if (tid->sched)
return; /* already scheduled */
tid->sched = 1;
TAILQ_INSERT_TAIL(&txq->axq_tidq, tid, axq_qelem);
}
/*
* Mark the current node as no longer needing to be polled for
* TX packets.
*
* The TXQ lock must be held.
*/
static void
ath_tx_tid_unsched(struct ath_softc *sc, struct ath_tid *tid)
{
struct ath_txq *txq = sc->sc_ac2q[tid->ac];
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (tid->sched == 0)
return;
tid->sched = 0;
TAILQ_REMOVE(&txq->axq_tidq, tid, axq_qelem);
}
/*
* Assign a sequence number manually to the given frame.
*
* This should only be called for A-MPDU TX frames.
*/
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
static ieee80211_seq
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_seqno_assign(struct ath_softc *sc, struct ieee80211_node *ni,
struct ath_buf *bf, struct mbuf *m0)
{
struct ieee80211_frame *wh;
int tid, pri;
ieee80211_seq seqno;
uint8_t subtype;
/* TID lookup */
wh = mtod(m0, struct ieee80211_frame *);
pri = M_WME_GETAC(m0); /* honor classification */
tid = WME_AC_TO_TID(pri);
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: pri=%d, tid=%d, qos has seq=%d\n",
__func__, pri, tid, IEEE80211_QOS_HAS_SEQ(wh));
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* XXX Is it a control frame? Ignore */
/* Does the packet require a sequence number? */
if (! IEEE80211_QOS_HAS_SEQ(wh))
return -1;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Is it a QOS NULL Data frame? Give it a sequence number from
* the default TID (IEEE80211_NONQOS_TID.)
*
* The RX path of everything I've looked at doesn't include the NULL
* data frame sequence number in the aggregation state updates, so
* assigning it a sequence number there will cause a BAW hole on the
* RX side.
*/
subtype = wh->i_fc[0] & IEEE80211_FC0_SUBTYPE_MASK;
if (subtype == IEEE80211_FC0_SUBTYPE_QOS_NULL) {
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
/* XXX no locking for this TID? This is a bit of a problem. */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
seqno = ni->ni_txseqs[IEEE80211_NONQOS_TID];
INCR(ni->ni_txseqs[IEEE80211_NONQOS_TID], IEEE80211_SEQ_RANGE);
} else {
/* Manually assign sequence number */
seqno = ni->ni_txseqs[tid];
INCR(ni->ni_txseqs[tid], IEEE80211_SEQ_RANGE);
}
*(uint16_t *)&wh->i_seq[0] = htole16(seqno << IEEE80211_SEQ_SEQ_SHIFT);
M_SEQNO_SET(m0, seqno);
/* Return so caller can do something with it if needed */
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: -> seqno=%d\n", __func__, seqno);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
return seqno;
}
/*
* Attempt to direct dispatch an aggregate frame to hardware.
* If the frame is out of BAW, queue.
* Otherwise, schedule it as a single frame.
*/
static void
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
ath_tx_xmit_aggr(struct ath_softc *sc, struct ath_node *an,
struct ath_txq *txq, struct ath_buf *bf)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
struct ath_tid *tid = &an->an_tid[bf->bf_state.bfs_tid];
struct ieee80211_tx_ampdu *tap;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tap = ath_tx_get_tx_tid(an, tid->tid);
/* paused? queue */
if (tid->paused) {
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_HEAD(tid, bf, bf_list);
Add some more debugging to try and nail down exactly what's going on when I see traffic stalls. It turns out that the bug isn't because the first and last frame in the BAW is in the software queue. It is more likely that it's because the first frame in the BAW is still in the software queue and thus there's no more room to allocate and do subsequent TX. PR: kern/166357
2012-03-25 23:50:34 +00:00
/* XXX don't sched - we're paused! */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
return;
}
/* outside baw? queue */
if (bf->bf_state.bfs_dobaw &&
(! BAW_WITHIN(tap->txa_start, tap->txa_wnd,
SEQNO(bf->bf_state.bfs_seqno)))) {
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_HEAD(tid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_sched(sc, tid);
return;
}
Ensure that single-frame aggregate session frames are retransmitted with the correct configuration. Occasionally an aggregate TX would fail and the first frame would be retransmitted as a non-AMPDU frame. Since bfs_aggr=1 and bfs_nframes > 1 (from the previous AMPDU attempt), the aggr completion function would be called and be very confused about what's going on. Noticed by: Kim <w8hdkim@gmail.com> PR: kern/171394
2012-09-07 00:24:27 +00:00
/*
* This is a temporary check and should be removed once
* all the relevant code paths have been fixed.
*
* During aggregate retries, it's possible that the head
* frame will fail (which has the bfs_aggr and bfs_nframes
* fields set for said aggregate) and will be retried as
* a single frame. In this instance, the values should
* be reset or the completion code will get upset with you.
*/
if (bf->bf_state.bfs_aggr != 0 || bf->bf_state.bfs_nframes > 1) {
device_printf(sc->sc_dev, "%s: bfs_aggr=%d, bfs_nframes=%d\n",
__func__,
bf->bf_state.bfs_aggr,
bf->bf_state.bfs_nframes);
bf->bf_state.bfs_aggr = 0;
bf->bf_state.bfs_nframes = 1;
}
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/* Update CLRDMASK just before this frame is queued */
ath_tx_update_clrdmask(sc, tid, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Direct dispatch to hardware */
ath_tx_do_ratelookup(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_calc_duration(sc, bf);
ath_tx_calc_protection(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_set_rtscts(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_rate_fill_rcflags(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_setds(sc, bf);
/* Statistics */
sc->sc_aggr_stats.aggr_low_hwq_single_pkt++;
/* Track per-TID hardware queue depth correctly */
tid->hwq_depth++;
/* Add to BAW */
if (bf->bf_state.bfs_dobaw) {
ath_tx_addto_baw(sc, an, tid, bf);
bf->bf_state.bfs_addedbaw = 1;
}
/* Set completion handler, multi-frame aggregate or not */
bf->bf_comp = ath_tx_aggr_comp;
/* Hand off to hardware */
ath_tx_handoff(sc, txq, bf);
}
/*
* Attempt to send the packet.
* If the queue isn't busy, direct-dispatch.
* If the queue is busy enough, queue the given packet on the
* relevant software queue.
*/
void
ath_tx_swq(struct ath_softc *sc, struct ieee80211_node *ni, struct ath_txq *txq,
struct ath_buf *bf)
{
struct ath_node *an = ATH_NODE(ni);
struct ieee80211_frame *wh;
struct ath_tid *atid;
int pri, tid;
struct mbuf *m0 = bf->bf_m;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Wrap the whole (software) TX path from ifnet dequeue to software queue (or direct dispatch) behind the TXQ lock (which, remember, is doubling as the TID lock too for now.) This ensures that: (a) the sequence number and the CCMP PN allocation is done together; (b) overlapping transmit paths don't interleave frames, so we don't end up with the original issue that triggered kern/166190. Ie, that we don't end up with seqno A, B in thread 1, C, D in thread 2, and they being queued to the software queue as "A C D B" or similar, leading to the BAW stalls. This has been tested: * both STA and AP modes with INVARIANTS and WITNESS; * TCP and UDP TX; * both STA->AP and AP->STA. STA is a Routerstation Pro (single CPU MIPS) and the AP is a dual-core Centrino. PR: kern/166190
2012-06-11 07:44:16 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Fetch the TID - non-QoS frames get assigned to TID 16 */
wh = mtod(m0, struct ieee80211_frame *);
pri = ath_tx_getac(sc, m0);
tid = ath_tx_gettid(sc, m0);
atid = &an->an_tid[tid];
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: bf=%p, pri=%d, tid=%d, qos=%d\n",
__func__, bf, pri, tid, IEEE80211_QOS_HAS_SEQ(wh));
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Set local packet state, used to queue packets to hardware */
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
/* XXX potentially duplicate info, re-check */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_tid = tid;
There's no need to use a TXQ pointer here; we specifically need the hardware queue ID when queuing to EDMA descriptors. This is a small part of trying to reduce the size of ath_buf entries.
2012-12-11 04:19:51 +00:00
bf->bf_state.bfs_tx_queue = txq->axq_qnum;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_pri = pri;
/*
* If the hardware queue isn't busy, queue it directly.
* If the hardware queue is busy, queue it.
* If the TID is paused or the traffic it outside BAW, software
* queue it.
*/
if (atid->paused) {
/* TID is paused, queue */
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: paused\n", __func__);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_TAIL(atid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
} else if (ath_tx_ampdu_pending(sc, an, tid)) {
/* AMPDU pending; queue */
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: pending\n", __func__);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_TAIL(atid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* XXX sched? */
} else if (ath_tx_ampdu_running(sc, an, tid)) {
/* AMPDU running, attempt direct dispatch if possible */
When scheduling frames in an aggregate session, the frames should be scheduled from the head of the software queue rather than trying to queue the newly given frame. This leads to some rather unfortunate out of order (but still valid as it's inside the BAW) frame TX. This now: * Always queues the frame at the end of the software queue; * Tries to direct dispatch the frame at the head of the software queue, to try and fill up the hardware queue. TODO: * I should likely try to queue as many frames to the hardware as I can at this point, rather than doing one at a time; * ath_tx_xmit_aggr() may fail and this code assumes that it'll schedule the TID. Otherwise TX may stall. PR: kern/166190
2012-06-11 07:29:25 +00:00
/*
* Always queue the frame to the tail of the list.
*/
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_TAIL(atid, bf, bf_list);
When scheduling frames in an aggregate session, the frames should be scheduled from the head of the software queue rather than trying to queue the newly given frame. This leads to some rather unfortunate out of order (but still valid as it's inside the BAW) frame TX. This now: * Always queues the frame at the end of the software queue; * Tries to direct dispatch the frame at the head of the software queue, to try and fill up the hardware queue. TODO: * I should likely try to queue as many frames to the hardware as I can at this point, rather than doing one at a time; * ath_tx_xmit_aggr() may fail and this code assumes that it'll schedule the TID. Otherwise TX may stall. PR: kern/166190
2012-06-11 07:29:25 +00:00
/*
* If the hardware queue isn't busy, direct dispatch
* the head frame in the list. Don't schedule the
* TID - let it build some more frames first?
*
* Otherwise, schedule the TID.
*/
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
if (txq->axq_depth < sc->sc_hwq_limit) {
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
bf = ATH_TID_FIRST(atid);
ATH_TID_REMOVE(atid, bf, bf_list);
Ensure that single-frame aggregate session frames are retransmitted with the correct configuration. Occasionally an aggregate TX would fail and the first frame would be retransmitted as a non-AMPDU frame. Since bfs_aggr=1 and bfs_nframes > 1 (from the previous AMPDU attempt), the aggr completion function would be called and be very confused about what's going on. Noticed by: Kim <w8hdkim@gmail.com> PR: kern/171394
2012-09-07 00:24:27 +00:00
/*
* Ensure it's definitely treated as a non-AMPDU
* frame - this information may have been left
* over from a previous attempt.
*/
bf->bf_state.bfs_aggr = 0;
bf->bf_state.bfs_nframes = 1;
/* Queue to the hardware */
Migrate the ath_hal_filltxdesc() API to take a list of buffer/seglen values. The existing API only exposes 'seglen' (the current buffer (segment) length) with the data buffer pointer set in 'ds_data'. This is fine for the legacy DMA engine but it won't work for the EDMA engines. The EDMA engine has a significantly different TX descriptor layout. * The legacy DMA engine had a ds_data pointer at the same offset in the descriptor for both TX and RX buffers; * The EDMA engine has no ds_data for RX - the data is DMAed after the descriptor; * The EDMA engine has support for 4 TX buffer/segment pairs in the TX DMA descriptor; * The EDMA TX completion is in a different FIFO, and the driver will 'link' the status completion entry to a QCU by a "QCU ID". I don't know why it's just not filled in by the hardware, alas. So given that, here are the changes: * Instead of directly fondling 'ds_data' in ath_desc, change the ath_hal_filltxdesc() to take an array of buffer pointers as well as segment len pointers; * The EDMA TX completion status wants a descriptor and queue id. This (for now) uses bf_state.bfs_txq and will extract the hardware QCU ID from that. * .. and this is ugly and wasteful; it should change to just store the QCU in the bf_state and save 3/7 bytes in the process. Now, the weird crap: * The aggregate TX path was using bf_state->bfs_txq for the TXQ, rather than taking a function argument. I've tidied that up. * The multicast queue frames get put on a software TXQ and then that is appended to the hardware CABQ when appropriate. So for now, make sure that bf_state->bfs_txq points at the CABQ when adding frames to the multicast queue. * .. but the multicast queue TX path for now doesn't use the software queue and instead (a) directly sets up the descriptor contents at that point; (b) the frames on the vap->avp_mcastq are then just appended wholesale to the CABQ. So for now, I don't have to worry about making the multicast path work with aggregation or the per-TID software queue. Phew. What's left to do: * I need to modify the 11n ath_hal_chaintxdesc() API to do the same. I'll do that in a subsequent commit. * Remove bf_state.bfs_txq entirely and store the QCU as appropriate. * .. then do the runtime "is this going on the right HWQ?" checks using that, rather than comparing pointer values. Tested on: * AR9280 STA/AP * AR5416 STA/AP
2012-08-05 10:12:27 +00:00
ath_tx_xmit_aggr(sc, an, txq, bf);
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX,
"%s: xmit_aggr\n",
__func__);
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
} else {
DPRINTF(sc, ATH_DEBUG_SW_TX,
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
"%s: ampdu; swq'ing\n",
__func__);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_sched(sc, atid);
}
} else if (txq->axq_depth < sc->sc_hwq_limit) {
/* AMPDU not running, attempt direct dispatch */
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: xmit_normal\n", __func__);
Oops - don't do the clrdmask check in ath_tx_xmit_normal() - the wrong lock may be held. Kim reported that the TID lock wasn't held when ath_tx_update_clrdmask() was called. Well, the underlying hardware TXQ for that TID. I'm betting it's the cabq stuff. ath_tx_xmit_normal() can be called for both real and software cabq. For software cabq, the real destination txq is different to the txq. So, the lock check will fail. Reported by: Kim Culhan <w8hdkim@gmail.com>
2012-09-25 20:41:43 +00:00
/* See if clrdmask needs to be set */
ath_tx_update_clrdmask(sc, atid, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_xmit_normal(sc, txq, bf);
} else {
/* Busy; queue */
Revert r233227 and followup commits as it breaks CCMP PN replay detection. This showed up when doing heavy UDP throughput on SMP machines. The problem with this is because the 802.11 sequence number is being allocated separately to the CCMP PN replay number (which is assigned during ieee80211_crypto_encap()). Under significant throughput (200+ MBps) the TX path would be stressed enough that frame TX/retry would force sequence number and PN allocation to be out of order. So once the frames were reordered via 802.11 seqnos, the CCMP PN would be far out of order, causing most frames to be discarded by the receiver. I've fixed this in some local work by being forced to: (a) deal with the issues that lead to the parallel TX causing out of order sequence numbers in the first place; (b) fix all the packet queuing issues which lead to strange (but mostly valid) TX. I'll begin fixing these in a subsequent commit or five. PR: kern/166190
2012-06-11 06:59:28 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: swq'ing\n", __func__);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_TAIL(atid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_sched(sc, atid);
}
}
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
/*
* Only set the clrdmask bit if none of the nodes are currently
* filtered.
*
* XXX TODO: go through all the callers and check to see
* which are being called in the context of looping over all
* TIDs (eg, if all tids are being paused, resumed, etc.)
* That'll avoid O(n^2) complexity here.
*/
static void
ath_tx_set_clrdmask(struct ath_softc *sc, struct ath_node *an)
{
int i;
ATH_TX_LOCK_ASSERT(sc);
for (i = 0; i < IEEE80211_TID_SIZE; i++) {
if (an->an_tid[i].isfiltered == 1)
Fix this routine to acutally break out and not set clrdmask if any of the TIDs are currently marked as "filtered."
2013-01-21 07:50:38 +00:00
return;
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
}
an->clrdmask = 1;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Configure the per-TID node state.
*
* This likely belongs in if_ath_node.c but I can't think of anywhere
* else to put it just yet.
*
* This sets up the SLISTs and the mutex as appropriate.
*/
void
ath_tx_tid_init(struct ath_softc *sc, struct ath_node *an)
{
int i, j;
struct ath_tid *atid;
for (i = 0; i < IEEE80211_TID_SIZE; i++) {
atid = &an->an_tid[i];
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/* XXX now with this bzer(), is the field 0'ing needed? */
bzero(atid, sizeof(*atid));
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
TAILQ_INIT(&atid->tid_q);
TAILQ_INIT(&atid->filtq.tid_q);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
atid->tid = i;
atid->an = an;
for (j = 0; j < ATH_TID_MAX_BUFS; j++)
atid->tx_buf[j] = NULL;
atid->baw_head = atid->baw_tail = 0;
atid->paused = 0;
atid->sched = 0;
atid->hwq_depth = 0;
atid->cleanup_inprogress = 0;
if (i == IEEE80211_NONQOS_TID)
Map the non-QoS TID to the voice queue, in order to ensure important things like EAPOL frames make it out. After a whole bunch of hacking/testing, I discovered that they weren't being early-dropped by the stack (but I should look at ensuring that later..) but were even making to the hardware transmit queue. They were mostly even being received by the remote end. However, the remote end was completely ignoring them. This didn't happen under 150-170MBit TCP tests as I'm guessing the TX queue stayed very busy and the STA didn't do any scanning. However, when doing 100Mbit/s of TCP traffic, the STA would do background scanning - which involves it coming in and out of powersave mode with the AP. Now, this is a total and utter hack around the real problems, which are: * I need to implement proper power save handling and integrate it into the filtered frames support, so the driver/stack doesn't send frames whilst the station is actually in sleep; * .. but frames were actually making it to the STA (macbook pro) and the AP did receive an ACK; but a tcpdump on the receiving side showed the EAPOL frame never made it. So the stack was dropping it for some reason; * Importantly - the EAPOL frames are currently going into the non-QoS TID, which maps to the BE queue and is susceptible to that queue being busy doing other things, but; * There's other traffic going on in the non-QoS TID from other contexts when scanning is going on and it's possible there's some races causing sequence number/IV issues, but; * Importantly importantlly, I think the interaction with TID 16 multicast traffic in power save mode is causing issues - since I -believe- the sequence number space being used by the EAPOL frames on TID 16 overlaps with the multicast frames that have sequence numbers allocated and are then stuffed on the cabq. Since with EAPOL frames being in TID 16 and queued to the BE queue, it's going to be waiting to be serviced with all of the aggregate traffic going on - and if the CABQ gets emptied beforehand, those TID 16 multicast frames with sequence numbers will go out beforehand. Now, there's quite likely a bunch of "stuff happening slightly out of sequence" going on due to the nature of the TX path (read: lots of overlapping and concurrent ath_start() and ath_raw_xmit() calls going on, sigh) but I thought I had caught them all and stuffed each TID TX behind a lock (that lasted as long as it needed to in order to get the frame onto the relevant destination queue - thus keeping things in order.) Unfortunately the last problem is the big one and I'm going to stare at it some more. If it _is_ So this is a work around for now to ensure that EAPOL frames actually make it out before any other stuff in the non-QoS TID and HOPEFULLY before the CABQ gets active. I'm now going to spend a little time in the TX path figuring out exactly why the sender is rejecting things. There's two (well, three if you count EAPOL contents invalid) possibilities: * The sequence number is out of order (ie, something else like the multicast traffic on CABQ) is going out first on TID 16; * The CCMP IV is out of order (similar to above - but less likely, as the TX key for multicast traffic is different to unicast traffic); * EAPOL contents strangely invalid. AP: Ubiquiti RSPRO, AR9160/AR9220 NICs STA: Macbook Pro, Broadcom 11n NIC
2012-09-26 03:45:42 +00:00
atid->ac = ATH_NONQOS_TID_AC;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
else
atid->ac = TID_TO_WME_AC(i);
}
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
an->clrdmask = 1; /* Always start by setting this bit */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
/*
* Pause the current TID. This stops packets from being transmitted
* on it.
*
* Since this is also called from upper layers as well as the driver,
* it will get the TID lock.
*/
static void
ath_tx_tid_pause(struct ath_softc *sc, struct ath_tid *tid)
{
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tid->paused++;
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL, "%s: paused = %d\n",
__func__, tid->paused);
}
/*
* Unpause the current TID, and schedule it if needed.
*/
static void
ath_tx_tid_resume(struct ath_softc *sc, struct ath_tid *tid)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tid->paused--;
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL, "%s: unpaused = %d\n",
__func__, tid->paused);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
if (tid->paused)
return;
/*
* Override the clrdmask configuration for the next frame
* from this TID, just to get the ball rolling.
*/
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
ath_tx_set_clrdmask(sc, tid->an);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
if (tid->axq_depth == 0)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
return;
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/* XXX isfiltered shouldn't ever be 0 at this point */
if (tid->isfiltered == 1) {
device_printf(sc->sc_dev, "%s: filtered?!\n", __func__);
return;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_sched(sc, tid);
Methodize the process of adding the software TX queue to the taskqueue. Move it (for now) to the TX taskqueue.
2013-02-07 02:15:25 +00:00
/*
* Queue the software TX scheduler.
*/
ath_tx_swq_kick(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* Add the given ath_buf to the TID filtered frame list.
* This requires the TID be filtered.
*/
static void
ath_tx_tid_filt_addbuf(struct ath_softc *sc, struct ath_tid *tid,
struct ath_buf *bf)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
if (! tid->isfiltered)
device_printf(sc->sc_dev, "%s: not filtered?!\n", __func__);
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT, "%s: bf=%p\n", __func__, bf);
/* Set the retry bit and bump the retry counter */
ath_tx_set_retry(sc, bf);
sc->sc_stats.ast_tx_swfiltered++;
Stop abusing the ATH_TID_*() queue macros for filtered frames and give them their own macro set.
2012-10-14 23:52:30 +00:00
ATH_TID_FILT_INSERT_TAIL(tid, bf, bf_list);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
}
/*
* Handle a completed filtered frame from the given TID.
* This just enables/pauses the filtered frame state if required
* and appends the filtered frame to the filtered queue.
*/
static void
ath_tx_tid_filt_comp_buf(struct ath_softc *sc, struct ath_tid *tid,
struct ath_buf *bf)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
if (! tid->isfiltered) {
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT, "%s: filter transition\n",
__func__);
tid->isfiltered = 1;
ath_tx_tid_pause(sc, tid);
}
/* Add the frame to the filter queue */
ath_tx_tid_filt_addbuf(sc, tid, bf);
}
/*
* Complete the filtered frame TX completion.
*
* If there are no more frames in the hardware queue, unpause/unfilter
* the TID if applicable. Otherwise we will wait for a node PS transition
* to unfilter.
*/
static void
ath_tx_tid_filt_comp_complete(struct ath_softc *sc, struct ath_tid *tid)
{
struct ath_buf *bf;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
if (tid->hwq_depth != 0)
return;
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT, "%s: hwq=0, transition back\n",
__func__);
tid->isfiltered = 0;
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
/* XXX ath_tx_tid_resume() also calls ath_tx_set_clrdmask()! */
ath_tx_set_clrdmask(sc, tid->an);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/* XXX this is really quite inefficient */
Stop abusing the ATH_TID_*() queue macros for filtered frames and give them their own macro set.
2012-10-14 23:52:30 +00:00
while ((bf = ATH_TID_FILT_LAST(tid, ath_bufhead_s)) != NULL) {
ATH_TID_FILT_REMOVE(tid, bf, bf_list);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_HEAD(tid, bf, bf_list);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
}
ath_tx_tid_resume(sc, tid);
}
/*
* Called when a single (aggregate or otherwise) frame is completed.
*
* Returns 1 if the buffer could be added to the filtered list
* (cloned or otherwise), 0 if the buffer couldn't be added to the
* filtered list (failed clone; expired retry) and the caller should
* free it and handle it like a failure (eg by sending a BAR.)
*/
static int
ath_tx_tid_filt_comp_single(struct ath_softc *sc, struct ath_tid *tid,
struct ath_buf *bf)
{
struct ath_buf *nbf;
int retval;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* Don't allow a filtered frame to live forever.
*/
if (bf->bf_state.bfs_retries > SWMAX_RETRIES) {
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
sc->sc_stats.ast_tx_swretrymax++;
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT,
"%s: bf=%p, seqno=%d, exceeded retries\n",
__func__,
bf,
bf->bf_state.bfs_seqno);
return (0);
}
/*
* A busy buffer can't be added to the retry list.
* It needs to be cloned.
*/
if (bf->bf_flags & ATH_BUF_BUSY) {
nbf = ath_tx_retry_clone(sc, tid->an, tid, bf);
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT,
"%s: busy buffer clone: %p -> %p\n",
__func__, bf, nbf);
} else {
nbf = bf;
}
if (nbf == NULL) {
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT,
"%s: busy buffer couldn't be cloned (%p)!\n",
__func__, bf);
retval = 1;
} else {
ath_tx_tid_filt_comp_buf(sc, tid, nbf);
retval = 0;
}
ath_tx_tid_filt_comp_complete(sc, tid);
return (retval);
}
static void
ath_tx_tid_filt_comp_aggr(struct ath_softc *sc, struct ath_tid *tid,
struct ath_buf *bf_first, ath_bufhead *bf_q)
{
struct ath_buf *bf, *bf_next, *nbf;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
bf = bf_first;
while (bf) {
bf_next = bf->bf_next;
bf->bf_next = NULL; /* Remove it from the aggr list */
/*
* Don't allow a filtered frame to live forever.
*/
if (bf->bf_state.bfs_retries > SWMAX_RETRIES) {
A couple of quick tidyups: * Delete this debugging print - I used it when debugging the initial TX descriptor chaining code. It now works, so let's toss it. It just confuses people if they enable TX descriptor debugging as they get two slightly different versions of the same descriptor. * Indenting.
2013-02-20 11:22:44 +00:00
sc->sc_stats.ast_tx_swretrymax++;
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT,
"%s: bf=%p, seqno=%d, exceeded retries\n",
__func__,
bf,
bf->bf_state.bfs_seqno);
TAILQ_INSERT_TAIL(bf_q, bf, bf_list);
goto next;
}
if (bf->bf_flags & ATH_BUF_BUSY) {
nbf = ath_tx_retry_clone(sc, tid->an, tid, bf);
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT,
"%s: busy buffer cloned: %p -> %p",
__func__, bf, nbf);
} else {
nbf = bf;
}
/*
* If the buffer couldn't be cloned, add it to bf_q;
* the caller will free the buffer(s) as required.
*/
if (nbf == NULL) {
DPRINTF(sc, ATH_DEBUG_SW_TX_FILT,
"%s: buffer couldn't be cloned! (%p)\n",
__func__, bf);
TAILQ_INSERT_TAIL(bf_q, bf, bf_list);
} else {
ath_tx_tid_filt_comp_buf(sc, tid, nbf);
}
next:
bf = bf_next;
}
ath_tx_tid_filt_comp_complete(sc, tid);
}
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/*
* Suspend the queue because we need to TX a BAR.
*/
static void
ath_tx_tid_bar_suspend(struct ath_softc *sc, struct ath_tid *tid)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAR,
Migrate the bulk of the RX routines out from if_ath.c to if_ath_rx.[ch]. * migrate the rx processing out into if_ath_rx.c * migrate the TSF functions into if_ath_tsf.h, as inlines This is in prepration for supporting the EDMA RX routines, required to support the AR93xx series NICs. TODO: * ath_start() shouldn't be private, but it's called as part of the RX path. I should likely migrate ath_rx_tasklet() back into if_ath.c and then return this to be 'static'. The RX code really shouldn't need to see TX routines (and vice versa.) * ath_beacon_* should be in if_ath_beacon.[ch]. * ath_tdma_* should be in if_ath_tdma.[ch] ...
2012-05-20 02:05:10 +00:00
"%s: tid=%p, bar_wait=%d, bar_tx=%d, called\n",
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
__func__,
Migrate the bulk of the RX routines out from if_ath.c to if_ath_rx.[ch]. * migrate the rx processing out into if_ath_rx.c * migrate the TSF functions into if_ath_tsf.h, as inlines This is in prepration for supporting the EDMA RX routines, required to support the AR93xx series NICs. TODO: * ath_start() shouldn't be private, but it's called as part of the RX path. I should likely migrate ath_rx_tasklet() back into if_ath.c and then return this to be 'static'. The RX code really shouldn't need to see TX routines (and vice versa.) * ath_beacon_* should be in if_ath_beacon.[ch]. * ath_tdma_* should be in if_ath_tdma.[ch] ...
2012-05-20 02:05:10 +00:00
tid,
tid->bar_wait,
tid->bar_tx);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/* We shouldn't be called when bar_tx is 1 */
if (tid->bar_tx) {
device_printf(sc->sc_dev, "%s: bar_tx is 1?!\n",
__func__);
}
/* If we've already been called, just be patient. */
if (tid->bar_wait)
return;
/* Wait! */
tid->bar_wait = 1;
/* Only one pause, no matter how many frames fail */
ath_tx_tid_pause(sc, tid);
}
/*
* We've finished with BAR handling - either we succeeded or
* failed. Either way, unsuspend TX.
*/
static void
ath_tx_tid_bar_unsuspend(struct ath_softc *sc, struct ath_tid *tid)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAR,
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
"%s: tid=%p, called\n",
__func__,
tid);
if (tid->bar_tx == 0 || tid->bar_wait == 0) {
device_printf(sc->sc_dev, "%s: bar_tx=%d, bar_wait=%d: ?\n",
__func__, tid->bar_tx, tid->bar_wait);
}
tid->bar_tx = tid->bar_wait = 0;
ath_tx_tid_resume(sc, tid);
}
/*
* Return whether we're ready to TX a BAR frame.
*
* Requires the TID lock be held.
*/
static int
ath_tx_tid_bar_tx_ready(struct ath_softc *sc, struct ath_tid *tid)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
if (tid->bar_wait == 0 || tid->hwq_depth > 0)
return (0);
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAR, "%s: tid=%p (%d), bar ready\n",
__func__, tid, tid->tid);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
return (1);
}
/*
* Check whether the current TID is ready to have a BAR
* TXed and if so, do the TX.
*
* Since the TID/TXQ lock can't be held during a call to
* ieee80211_send_bar(), we have to do the dirty thing of unlocking it,
* sending the BAR and locking it again.
*
* Eventually, the code to send the BAR should be broken out
* from this routine so the lock doesn't have to be reacquired
* just to be immediately dropped by the caller.
*/
static void
ath_tx_tid_bar_tx(struct ath_softc *sc, struct ath_tid *tid)
{
struct ieee80211_tx_ampdu *tap;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAR,
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
"%s: tid=%p, called\n",
__func__,
tid);
tap = ath_tx_get_tx_tid(tid->an, tid->tid);
/*
* This is an error condition!
*/
if (tid->bar_wait == 0 || tid->bar_tx == 1) {
device_printf(sc->sc_dev,
"%s: tid=%p, bar_tx=%d, bar_wait=%d: ?\n",
__func__,
tid,
tid->bar_tx,
tid->bar_wait);
return;
}
/* Don't do anything if we still have pending frames */
if (tid->hwq_depth > 0) {
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAR,
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
"%s: tid=%p, hwq_depth=%d, waiting\n",
__func__,
tid,
tid->hwq_depth);
return;
}
/* We're now about to TX */
tid->bar_tx = 1;
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/*
* Override the clrdmask configuration for the next frame,
* just to get the ball rolling.
*/
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
ath_tx_set_clrdmask(sc, tid->an);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/*
* Calculate new BAW left edge, now that all frames have either
* succeeded or failed.
*
* XXX verify this is _actually_ the valid value to begin at!
*/
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAR,
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
"%s: tid=%p, new BAW left edge=%d\n",
__func__,
tid,
tap->txa_start);
/* Try sending the BAR frame */
/* We can't hold the lock here! */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
if (ieee80211_send_bar(&tid->an->an_node, tap, tap->txa_start) == 0) {
/* Success? Now we wait for notification that it's done */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
return;
}
/* Failure? For now, warn loudly and continue */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
device_printf(sc->sc_dev, "%s: tid=%p, failed to TX BAR, continue!\n",
__func__, tid);
ath_tx_tid_bar_unsuspend(sc, tid);
}
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
static void
ath_tx_tid_drain_pkt(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid, ath_bufhead *bf_cq, struct ath_buf *bf)
{
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If the current TID is running AMPDU, update
* the BAW.
*/
if (ath_tx_ampdu_running(sc, an, tid->tid) &&
bf->bf_state.bfs_dobaw) {
/*
* Only remove the frame from the BAW if it's
* been transmitted at least once; this means
* the frame was in the BAW to begin with.
*/
if (bf->bf_state.bfs_retries > 0) {
ath_tx_update_baw(sc, an, tid, bf);
bf->bf_state.bfs_dobaw = 0;
}
Disable debugging entries about BAW issues. I haven't seen any issues to do with BAW tracking in the last 9 months or so.
2013-02-21 21:47:35 +00:00
#if 0
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* This has become a non-fatal error now
*/
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
Disable debugging entries about BAW issues. I haven't seen any issues to do with BAW tracking in the last 9 months or so.
2013-02-21 21:47:35 +00:00
#endif
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
}
TAILQ_INSERT_TAIL(bf_cq, bf, bf_list);
}
static void
ath_tx_tid_drain_print(struct ath_softc *sc, struct ath_node *an,
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
const char *pfx, struct ath_tid *tid, struct ath_buf *bf)
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
{
struct ieee80211_node *ni = &an->an_node;
struct ath_txq *txq = sc->sc_ac2q[tid->ac];
struct ieee80211_tx_ampdu *tap;
tap = ath_tx_get_tx_tid(an, tid->tid);
device_printf(sc->sc_dev,
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
"%s: %s: node %p: bf=%p: addbaw=%d, dobaw=%d, "
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
"seqno=%d, retry=%d\n",
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
__func__, pfx, ni, bf,
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
bf->bf_state.bfs_addedbaw,
bf->bf_state.bfs_dobaw,
SEQNO(bf->bf_state.bfs_seqno),
bf->bf_state.bfs_retries);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
device_printf(sc->sc_dev,
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
"%s: node %p: bf=%p: txq[%d] axq_depth=%d, axq_aggr_depth=%d\n",
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
__func__, ni, bf,
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
txq->axq_qnum,
txq->axq_depth,
txq->axq_aggr_depth);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
device_printf(sc->sc_dev,
"%s: node %p: bf=%p: tid txq_depth=%d hwq_depth=%d, bar_wait=%d, isfiltered=%d\n",
__func__, ni, bf,
tid->axq_depth,
tid->hwq_depth,
tid->bar_wait,
tid->isfiltered);
device_printf(sc->sc_dev,
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
"%s: node %p: tid %d: "
"sched=%d, paused=%d, "
"incomp=%d, baw_head=%d, "
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
"baw_tail=%d txa_start=%d, ni_txseqs=%d\n",
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
__func__, ni, tid->tid,
tid->sched, tid->paused,
tid->incomp, tid->baw_head,
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
tid->baw_tail, tap == NULL ? -1 : tap->txa_start,
ni->ni_txseqs[tid->tid]);
/* XXX Dump the frame, see what it is? */
ieee80211_dump_pkt(ni->ni_ic,
mtod(bf->bf_m, const uint8_t *),
bf->bf_m->m_len, 0, -1);
}
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Free any packets currently pending in the software TX queue.
*
* This will be called when a node is being deleted.
*
* It can also be called on an active node during an interface
* reset or state transition.
*
* (From Linux/reference):
*
* TODO: For frame(s) that are in the retry state, we will reuse the
* sequence number(s) without setting the retry bit. The
* alternative is to give up on these and BAR the receiver's window
* forward.
*/
static void
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
ath_tx_tid_drain(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid, ath_bufhead *bf_cq)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
struct ath_buf *bf;
struct ieee80211_tx_ampdu *tap;
struct ieee80211_node *ni = &an->an_node;
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
int t;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tap = ath_tx_get_tx_tid(an, tid->tid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Walk the queue, free frames */
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
t = 0;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
for (;;) {
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
bf = ATH_TID_FIRST(tid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (bf == NULL) {
break;
}
if (t == 0) {
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ath_tx_tid_drain_print(sc, an, "norm", tid, bf);
Now that I've fixed the BAW TX hangs, disable this verbose debugging again.
2012-04-04 22:22:50 +00:00
t = 1;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_REMOVE(tid, bf, bf_list);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
ath_tx_tid_drain_pkt(sc, an, tid, bf_cq, bf);
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/* And now, drain the filtered frame queue */
t = 0;
for (;;) {
Stop abusing the ATH_TID_*() queue macros for filtered frames and give them their own macro set.
2012-10-14 23:52:30 +00:00
bf = ATH_TID_FILT_FIRST(tid);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
if (bf == NULL)
break;
if (t == 0) {
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ath_tx_tid_drain_print(sc, an, "filt", tid, bf);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
t = 1;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
Stop abusing the ATH_TID_*() queue macros for filtered frames and give them their own macro set.
2012-10-14 23:52:30 +00:00
ATH_TID_FILT_REMOVE(tid, bf, bf_list);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
ath_tx_tid_drain_pkt(sc, an, tid, bf_cq, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/*
* Override the clrdmask configuration for the next frame
* in case there is some future transmission, just to get
* the ball rolling.
*
* This won't hurt things if the TID is about to be freed.
*/
Migrate CLRDMASK to be a per-node flag, rather than a per-TID flag. This is easily possible now that the TX is protected by a single lock, rather than a per-TXQ (and thus per-TID) lock. Only set CLRDMASK if none of the destinations are filtered. This likely will need some tuning when it comes time to do UASPD/PS-POLL TX, however at that point it should be manually set anyway. Tested: * AR9280, STA mode TODO: * More thorough testing in AP mode * test other chipsets, just to be safe/sure.
2013-01-21 04:06:04 +00:00
ath_tx_set_clrdmask(sc, tid->an);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Now that it's completed, grab the TID lock and update
* the sequence number and BAW window.
* Because sequence numbers have been assigned to frames
* that haven't been sent yet, it's entirely possible
* we'll be called with some pending frames that have not
* been transmitted.
*
* The cleaner solution is to do the sequence number allocation
* when the packet is first transmitted - and thus the "retries"
* check above would be enough to update the BAW/seqno.
*/
/* But don't do it for non-QoS TIDs */
if (tap) {
#if 0
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: node %p: TID %d: sliding BAW left edge to %d\n",
__func__, an, tid->tid, tap->txa_start);
#endif
ni->ni_txseqs[tid->tid] = tap->txa_start;
tid->baw_tail = tid->baw_head;
}
}
/*
* Flush all software queued packets for the given node.
*
* This occurs when a completion handler frees the last buffer
* for a node, and the node is thus freed. This causes the node
* to be cleaned up, which ends up calling ath_tx_node_flush.
*/
void
ath_tx_node_flush(struct ath_softc *sc, struct ath_node *an)
{
int tid;
ath_bufhead bf_cq;
struct ath_buf *bf;
TAILQ_INIT(&bf_cq);
Migrate the ath(4) KTR logging to use an ATH_KTR() macro. This should eventually be unified with ATH_DEBUG() so I can get both from one macro; that may take some time. Add some new probes for TX and TX completion.
2012-09-24 20:35:56 +00:00
ATH_KTR(sc, ATH_KTR_NODE, 1, "ath_tx_node_flush: flush node; ni=%p",
&an->an_node);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
for (tid = 0; tid < IEEE80211_TID_SIZE; tid++) {
struct ath_tid *atid = &an->an_tid[tid];
/* Free packets */
ath_tx_tid_drain(sc, an, atid, &bf_cq);
Call ath_tx_tid_unsched() after the node has been flushed, so the state can be printed correctly.
2012-09-25 05:56:59 +00:00
/* Remove this tid from the list of active tids */
ath_tx_tid_unsched(sc, atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Handle completed frames */
while ((bf = TAILQ_FIRST(&bf_cq)) != NULL) {
TAILQ_REMOVE(&bf_cq, bf, bf_list);
ath_tx_default_comp(sc, bf, 0);
}
}
/*
* Drain all the software TXQs currently with traffic queued.
*/
void
ath_tx_txq_drain(struct ath_softc *sc, struct ath_txq *txq)
{
struct ath_tid *tid;
ath_bufhead bf_cq;
struct ath_buf *bf;
TAILQ_INIT(&bf_cq);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Iterate over all active tids for the given txq,
* flushing and unsched'ing them
*/
while (! TAILQ_EMPTY(&txq->axq_tidq)) {
tid = TAILQ_FIRST(&txq->axq_tidq);
ath_tx_tid_drain(sc, tid->an, tid, &bf_cq);
ath_tx_tid_unsched(sc, tid);
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
while ((bf = TAILQ_FIRST(&bf_cq)) != NULL) {
TAILQ_REMOVE(&bf_cq, bf, bf_list);
ath_tx_default_comp(sc, bf, 0);
}
}
/*
* Handle completion of non-aggregate session frames.
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
*
* This (currently) doesn't implement software retransmission of
* non-aggregate frames!
*
* Software retransmission of non-aggregate frames needs to obey
* the strict sequence number ordering, and drop any frames that
* will fail this.
*
* For now, filtered frames and frame transmission will cause
* all kinds of issues. So we don't support them.
*
* So anyone queuing frames via ath_tx_normal_xmit() or
* ath_tx_hw_queue_norm() must override and set CLRDMASK.
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*/
void
ath_tx_normal_comp(struct ath_softc *sc, struct ath_buf *bf, int fail)
{
struct ieee80211_node *ni = bf->bf_node;
struct ath_node *an = ATH_NODE(ni);
int tid = bf->bf_state.bfs_tid;
struct ath_tid *atid = &an->an_tid[tid];
struct ath_tx_status *ts = &bf->bf_status.ds_txstat;
/* The TID state is protected behind the TXQ lock */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: bf=%p: fail=%d, hwq_depth now %d\n",
__func__, bf, fail, atid->hwq_depth - 1);
atid->hwq_depth--;
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
#if 0
/*
* If the frame was filtered, stick it on the filter frame
* queue and complain about it. It shouldn't happen!
*/
if ((ts->ts_status & HAL_TXERR_FILT) ||
(ts->ts_status != 0 && atid->isfiltered)) {
device_printf(sc->sc_dev,
"%s: isfiltered=%d, ts_status=%d: huh?\n",
__func__,
atid->isfiltered,
ts->ts_status);
ath_tx_tid_filt_comp_buf(sc, atid, bf);
}
#endif
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
if (atid->isfiltered)
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
device_printf(sc->sc_dev, "%s: filtered?!\n", __func__);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (atid->hwq_depth < 0)
device_printf(sc->sc_dev, "%s: hwq_depth < 0: %d\n",
__func__, atid->hwq_depth);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If the queue is filtered, potentially mark it as complete
* and reschedule it as needed.
*
* This is required as there may be a subsequent TX descriptor
* for this end-node that has CLRDMASK set, so it's quite possible
* that a filtered frame will be followed by a non-filtered
* (complete or otherwise) frame.
*
* XXX should we do this before we complete the frame?
*/
if (atid->isfiltered)
ath_tx_tid_filt_comp_complete(sc, atid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* punt to rate control if we're not being cleaned up
* during a hw queue drain and the frame wanted an ACK.
*/
Remove duplicate txflags field from ath_buf. rename bf_state.bfs_flags to bf_state.bfs_txflags, as that is what it effectively is.
2012-04-07 02:01:26 +00:00
if (fail == 0 && ((bf->bf_state.bfs_txflags & HAL_TXDESC_NOACK) == 0))
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_update_ratectrl(sc, ni, bf->bf_state.bfs_rc,
ts, bf->bf_state.bfs_pktlen,
1, (ts->ts_status == 0) ? 0 : 1);
ath_tx_default_comp(sc, bf, fail);
}
/*
* Handle cleanup of aggregate session packets that aren't
* an A-MPDU.
*
* There's no need to update the BAW here - the session is being
* torn down.
*/
static void
ath_tx_comp_cleanup_unaggr(struct ath_softc *sc, struct ath_buf *bf)
{
struct ieee80211_node *ni = bf->bf_node;
struct ath_node *an = ATH_NODE(ni);
int tid = bf->bf_state.bfs_tid;
struct ath_tid *atid = &an->an_tid[tid];
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL, "%s: TID %d: incomp=%d\n",
__func__, tid, atid->incomp);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
atid->incomp--;
if (atid->incomp == 0) {
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: TID %d: cleaned up! resume!\n",
__func__, tid);
atid->cleanup_inprogress = 0;
ath_tx_tid_resume(sc, atid);
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_default_comp(sc, bf, 0);
}
/*
* Performs transmit side cleanup when TID changes from aggregated to
* unaggregated.
*
* - Discard all retry frames from the s/w queue.
* - Fix the tx completion function for all buffers in s/w queue.
* - Count the number of unacked frames, and let transmit completion
* handle it later.
*
* The caller is responsible for pausing the TID.
*/
static void
Rename ath_tx_cleanup() -> ath_tx_tid_cleanup() in order to not clash with a symbol in if_ath.c
2012-05-21 22:39:13 +00:00
ath_tx_tid_cleanup(struct ath_softc *sc, struct ath_node *an, int tid)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
struct ath_tid *atid = &an->an_tid[tid];
struct ieee80211_tx_ampdu *tap;
struct ath_buf *bf, *bf_next;
ath_bufhead bf_cq;
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAW,
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
"%s: TID %d: called\n", __func__, tid);
TAILQ_INIT(&bf_cq);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* Move the filtered frames to the TX queue, before
* we run off and discard/process things.
*/
/* XXX this is really quite inefficient */
Stop abusing the ATH_TID_*() queue macros for filtered frames and give them their own macro set.
2012-10-14 23:52:30 +00:00
while ((bf = ATH_TID_FILT_LAST(atid, ath_bufhead_s)) != NULL) {
ATH_TID_FILT_REMOVE(atid, bf, bf_list);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_HEAD(atid, bf, bf_list);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Update the frames in the software TX queue:
*
* + Discard retry frames in the queue
* + Fix the completion function to be non-aggregate
*/
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
bf = ATH_TID_FIRST(atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
while (bf) {
if (bf->bf_state.bfs_isretried) {
bf_next = TAILQ_NEXT(bf, bf_list);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_REMOVE(atid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (bf->bf_state.bfs_dobaw) {
ath_tx_update_baw(sc, an, atid, bf);
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
__func__,
SEQNO(bf->bf_state.bfs_seqno));
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
bf->bf_state.bfs_dobaw = 0;
/*
* Call the default completion handler with "fail" just
* so upper levels are suitably notified about this.
*/
TAILQ_INSERT_TAIL(&bf_cq, bf, bf_list);
bf = bf_next;
continue;
}
/* Give these the default completion handler */
bf->bf_comp = ath_tx_normal_comp;
bf = TAILQ_NEXT(bf, bf_list);
}
/* The caller is required to pause the TID */
#if 0
/* Pause the TID */
ath_tx_tid_pause(sc, atid);
#endif
/*
* Calculate what hardware-queued frames exist based
* on the current BAW size. Ie, what frames have been
* added to the TX hardware queue for this TID but
* not yet ACKed.
*/
tap = ath_tx_get_tx_tid(an, tid);
/* Need the lock - fiddling with BAW */
while (atid->baw_head != atid->baw_tail) {
if (atid->tx_buf[atid->baw_head]) {
atid->incomp++;
atid->cleanup_inprogress = 1;
atid->tx_buf[atid->baw_head] = NULL;
}
INCR(atid->baw_head, ATH_TID_MAX_BUFS);
INCR(tap->txa_start, IEEE80211_SEQ_RANGE);
}
/*
* If cleanup is required, defer TID scheduling
* until all the HW queued packets have been
* sent.
*/
if (! atid->cleanup_inprogress)
ath_tx_tid_resume(sc, atid);
if (atid->cleanup_inprogress)
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: TID %d: cleanup needed: %d packets\n",
__func__, tid, atid->incomp);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Handle completing frames and fail them */
while ((bf = TAILQ_FIRST(&bf_cq)) != NULL) {
TAILQ_REMOVE(&bf_cq, bf, bf_list);
ath_tx_default_comp(sc, bf, 1);
}
}
static struct ath_buf *
If software retransmit occurs with an ath_buf marked ATH_BUF_BUSY, it's cloned and that clone is retransmitted. This means that the ath_buf pointer squirreled away on the baw window array is suddenly wrong and was causing all kinds of console output. This updates the pointer in that particular BAW slot to the new ath_buf after ensuring that: * the new and old buffers have the same seqno; * the current slot pointer matches the old buffer pointer. This quietens the debugging output (again), restoring said debugging to only signify when a broken condition has occured. Sponsored by: Hobnob, Inc.
2011-11-09 18:24:20 +00:00
ath_tx_retry_clone(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid, struct ath_buf *bf)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
struct ath_buf *nbf;
int error;
nbf = ath_buf_clone(sc, bf);
#if 0
device_printf(sc->sc_dev, "%s: ATH_BUF_BUSY; cloning\n",
__func__);
#endif
if (nbf == NULL) {
/* Failed to clone */
device_printf(sc->sc_dev,
"%s: failed to clone a busy buffer\n",
__func__);
return NULL;
}
/* Setup the dma for the new buffer */
error = ath_tx_dmasetup(sc, nbf, nbf->bf_m);
if (error != 0) {
device_printf(sc->sc_dev,
"%s: failed to setup dma for clone\n",
__func__);
/*
* Put this at the head of the list, not tail;
* that way it doesn't interfere with the
* busy buffer logic (which uses the tail of
* the list.)
*/
ATH_TXBUF_LOCK(sc);
Oops, return the newly allocated buffer to the queue, not the completed buffer. PR: kern/168170
2012-06-13 05:41:00 +00:00
ath_returnbuf_head(sc, nbf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ATH_TXBUF_UNLOCK(sc);
return NULL;
}
If software retransmit occurs with an ath_buf marked ATH_BUF_BUSY, it's cloned and that clone is retransmitted. This means that the ath_buf pointer squirreled away on the baw window array is suddenly wrong and was causing all kinds of console output. This updates the pointer in that particular BAW slot to the new ath_buf after ensuring that: * the new and old buffers have the same seqno; * the current slot pointer matches the old buffer pointer. This quietens the debugging output (again), restoring said debugging to only signify when a broken condition has occured. Sponsored by: Hobnob, Inc.
2011-11-09 18:24:20 +00:00
/* Update BAW if required, before we free the original buf */
if (bf->bf_state.bfs_dobaw)
ath_tx_switch_baw_buf(sc, an, tid, bf, nbf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Free current buffer; return the older buffer */
bf->bf_m = NULL;
bf->bf_node = NULL;
ath_freebuf(sc, bf);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
return nbf;
}
/*
* Handle retrying an unaggregate frame in an aggregate
* session.
*
* If too many retries occur, pause the TID, wait for
* any further retransmits (as there's no reason why
* non-aggregate frames in an aggregate session are
* transmitted in-order; they just have to be in-BAW)
* and then queue a BAR.
*/
static void
ath_tx_aggr_retry_unaggr(struct ath_softc *sc, struct ath_buf *bf)
{
struct ieee80211_node *ni = bf->bf_node;
struct ath_node *an = ATH_NODE(ni);
int tid = bf->bf_state.bfs_tid;
struct ath_tid *atid = &an->an_tid[tid];
struct ieee80211_tx_ampdu *tap;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tap = ath_tx_get_tx_tid(an, tid);
/*
* If the buffer is marked as busy, we can't directly
* reuse it. Instead, try to clone the buffer.
* If the clone is successful, recycle the old buffer.
* If the clone is unsuccessful, set bfs_retries to max
* to force the next bit of code to free the buffer
* for us.
*/
if ((bf->bf_state.bfs_retries < SWMAX_RETRIES) &&
(bf->bf_flags & ATH_BUF_BUSY)) {
struct ath_buf *nbf;
If software retransmit occurs with an ath_buf marked ATH_BUF_BUSY, it's cloned and that clone is retransmitted. This means that the ath_buf pointer squirreled away on the baw window array is suddenly wrong and was causing all kinds of console output. This updates the pointer in that particular BAW slot to the new ath_buf after ensuring that: * the new and old buffers have the same seqno; * the current slot pointer matches the old buffer pointer. This quietens the debugging output (again), restoring said debugging to only signify when a broken condition has occured. Sponsored by: Hobnob, Inc.
2011-11-09 18:24:20 +00:00
nbf = ath_tx_retry_clone(sc, an, atid, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (nbf)
/* bf has been freed at this point */
bf = nbf;
else
bf->bf_state.bfs_retries = SWMAX_RETRIES + 1;
}
if (bf->bf_state.bfs_retries >= SWMAX_RETRIES) {
DPRINTF(sc, ATH_DEBUG_SW_TX_RETRIES,
"%s: exceeded retries; seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
sc->sc_stats.ast_tx_swretrymax++;
/* Update BAW anyway */
if (bf->bf_state.bfs_dobaw) {
ath_tx_update_baw(sc, an, atid, bf);
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
}
bf->bf_state.bfs_dobaw = 0;
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/* Suspend the TX queue and get ready to send the BAR */
ath_tx_tid_bar_suspend(sc, atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/* Send the BAR if there are no other frames waiting */
if (ath_tx_tid_bar_tx_ready(sc, atid))
ath_tx_tid_bar_tx(sc, atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Free buffer, bf is free after this call */
ath_tx_default_comp(sc, bf, 0);
return;
}
/*
* This increments the retry counter as well as
* sets the retry flag in the ath_buf and packet
* body.
*/
ath_tx_set_retry(sc, bf);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
sc->sc_stats.ast_tx_swretries++;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Insert this at the head of the queue, so it's
* retried before any current/subsequent frames.
*/
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_HEAD(atid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_sched(sc, atid);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/* Send the BAR if there are no other frames waiting */
if (ath_tx_tid_bar_tx_ready(sc, atid))
ath_tx_tid_bar_tx(sc, atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
/*
* Common code for aggregate excessive retry/subframe retry.
* If retrying, queues buffers to bf_q. If not, frees the
* buffers.
*
* XXX should unify this with ath_tx_aggr_retry_unaggr()
*/
static int
ath_tx_retry_subframe(struct ath_softc *sc, struct ath_buf *bf,
ath_bufhead *bf_q)
{
struct ieee80211_node *ni = bf->bf_node;
struct ath_node *an = ATH_NODE(ni);
int tid = bf->bf_state.bfs_tid;
struct ath_tid *atid = &an->an_tid[tid];
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Make sure the aggregate fields are properly cleared - both in the ath_buf and when forming a non-aggregate frame. The non-11n setds function is called when TXing aggregate frames (and yes, I should fix this!) and the non-11n TX aggregation code doesn't clear the delimiter field. I figure it's nicer to do that.
2012-09-09 05:06:16 +00:00
/* XXX clr11naggr should be done for all subframes */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_hal_clr11n_aggr(sc->sc_ah, bf->bf_desc);
ath_hal_set11nburstduration(sc->sc_ah, bf->bf_desc, 0);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* ath_hal_set11n_virtualmorefrag(sc->sc_ah, bf->bf_desc, 0); */
/*
* If the buffer is marked as busy, we can't directly
* reuse it. Instead, try to clone the buffer.
* If the clone is successful, recycle the old buffer.
* If the clone is unsuccessful, set bfs_retries to max
* to force the next bit of code to free the buffer
* for us.
*/
if ((bf->bf_state.bfs_retries < SWMAX_RETRIES) &&
(bf->bf_flags & ATH_BUF_BUSY)) {
struct ath_buf *nbf;
If software retransmit occurs with an ath_buf marked ATH_BUF_BUSY, it's cloned and that clone is retransmitted. This means that the ath_buf pointer squirreled away on the baw window array is suddenly wrong and was causing all kinds of console output. This updates the pointer in that particular BAW slot to the new ath_buf after ensuring that: * the new and old buffers have the same seqno; * the current slot pointer matches the old buffer pointer. This quietens the debugging output (again), restoring said debugging to only signify when a broken condition has occured. Sponsored by: Hobnob, Inc.
2011-11-09 18:24:20 +00:00
nbf = ath_tx_retry_clone(sc, an, atid, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (nbf)
/* bf has been freed at this point */
bf = nbf;
else
bf->bf_state.bfs_retries = SWMAX_RETRIES + 1;
}
if (bf->bf_state.bfs_retries >= SWMAX_RETRIES) {
sc->sc_stats.ast_tx_swretrymax++;
DPRINTF(sc, ATH_DEBUG_SW_TX_RETRIES,
"%s: max retries: seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
ath_tx_update_baw(sc, an, atid, bf);
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
bf->bf_state.bfs_dobaw = 0;
return 1;
}
ath_tx_set_retry(sc, bf);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
sc->sc_stats.ast_tx_swretries++;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_next = NULL; /* Just to make sure */
Make sure the aggregate fields are properly cleared - both in the ath_buf and when forming a non-aggregate frame. The non-11n setds function is called when TXing aggregate frames (and yes, I should fix this!) and the non-11n TX aggregation code doesn't clear the delimiter field. I figure it's nicer to do that.
2012-09-09 05:06:16 +00:00
/* Clear the aggregate state */
bf->bf_state.bfs_aggr = 0;
bf->bf_state.bfs_ndelim = 0; /* ??? needed? */
bf->bf_state.bfs_nframes = 1;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
TAILQ_INSERT_TAIL(bf_q, bf, bf_list);
return 0;
}
/*
* error pkt completion for an aggregate destination
*/
static void
ath_tx_comp_aggr_error(struct ath_softc *sc, struct ath_buf *bf_first,
struct ath_tid *tid)
{
struct ieee80211_node *ni = bf_first->bf_node;
struct ath_node *an = ATH_NODE(ni);
struct ath_buf *bf_next, *bf;
ath_bufhead bf_q;
int drops = 0;
struct ieee80211_tx_ampdu *tap;
ath_bufhead bf_cq;
TAILQ_INIT(&bf_q);
TAILQ_INIT(&bf_cq);
/*
* Update rate control - all frames have failed.
*
* XXX use the length in the first frame in the series;
* XXX just so things are consistent for now.
*/
ath_tx_update_ratectrl(sc, ni, bf_first->bf_state.bfs_rc,
&bf_first->bf_status.ds_txstat,
bf_first->bf_state.bfs_pktlen,
bf_first->bf_state.bfs_nframes, bf_first->bf_state.bfs_nframes);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tap = ath_tx_get_tx_tid(an, tid->tid);
Flesh out the TX aggregation completion statistics. * Failall is now named just that. * Add TX ok and TX fail, for aggregate frame sub-frames. This will break athstats; a followup commit wil resolve this. Sponsored by: Hobnob, Inc.
2011-11-23 05:00:25 +00:00
sc->sc_stats.ast_tx_aggr_failall++;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Retry all subframes */
bf = bf_first;
while (bf) {
bf_next = bf->bf_next;
bf->bf_next = NULL; /* Remove it from the aggr list */
Flesh out the TX aggregation completion statistics. * Failall is now named just that. * Add TX ok and TX fail, for aggregate frame sub-frames. This will break athstats; a followup commit wil resolve this. Sponsored by: Hobnob, Inc.
2011-11-23 05:00:25 +00:00
sc->sc_stats.ast_tx_aggr_fail++;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (ath_tx_retry_subframe(sc, bf, &bf_q)) {
drops++;
bf->bf_next = NULL;
TAILQ_INSERT_TAIL(&bf_cq, bf, bf_list);
}
bf = bf_next;
}
/* Prepend all frames to the beginning of the queue */
while ((bf = TAILQ_LAST(&bf_q, ath_bufhead_s)) != NULL) {
TAILQ_REMOVE(&bf_q, bf, bf_list);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_HEAD(tid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Remove some of the redundant locking done in the TX completion path, when checking whether BAR frames need to be checked.
2012-04-26 23:57:24 +00:00
/*
* Schedule the TID to be re-tried.
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_sched(sc, tid);
/*
* send bar if we dropped any frames
*
* Keep the txq lock held for now, as we need to ensure
* that ni_txseqs[] is consistent (as it's being updated
* in the ifnet TX context or raw TX context.)
*/
if (drops) {
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/* Suspend the TX queue and get ready to send the BAR */
ath_tx_tid_bar_suspend(sc, tid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/*
* Send BAR if required
*/
if (ath_tx_tid_bar_tx_ready(sc, tid))
ath_tx_tid_bar_tx(sc, tid);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Complete frames which errored out */
while ((bf = TAILQ_FIRST(&bf_cq)) != NULL) {
TAILQ_REMOVE(&bf_cq, bf, bf_list);
ath_tx_default_comp(sc, bf, 0);
}
}
/*
* Handle clean-up of packets from an aggregate list.
*
* There's no need to update the BAW here - the session is being
* torn down.
*/
static void
ath_tx_comp_cleanup_aggr(struct ath_softc *sc, struct ath_buf *bf_first)
{
struct ath_buf *bf, *bf_next;
struct ieee80211_node *ni = bf_first->bf_node;
struct ath_node *an = ATH_NODE(ni);
int tid = bf_first->bf_state.bfs_tid;
struct ath_tid *atid = &an->an_tid[tid];
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* update incomp */
Fix two bugs: * when pulling frames off of the TID queue, the ATH_TID_REMOVE() macro decrements the axq_depth field. So don't do it twice. * in ath_tx_comp_cleanup_aggr(), bf wasn't being reset to bf_first before walking the buffer list to complete buffers; so those buffers will leak.
2013-03-15 20:00:08 +00:00
bf = bf_first;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
while (bf) {
atid->incomp--;
bf = bf->bf_next;
}
if (atid->incomp == 0) {
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: TID %d: cleaned up! resume!\n",
__func__, tid);
atid->cleanup_inprogress = 0;
ath_tx_tid_resume(sc, atid);
}
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/* Send BAR if required */
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/* XXX why would we send a BAR when transitioning to non-aggregation? */
Fix two bugs: * when pulling frames off of the TID queue, the ATH_TID_REMOVE() macro decrements the axq_depth field. So don't do it twice. * in ath_tx_comp_cleanup_aggr(), bf wasn't being reset to bf_first before walking the buffer list to complete buffers; so those buffers will leak.
2013-03-15 20:00:08 +00:00
/*
* XXX TODO: we should likely just tear down the BAR state here,
* rather than sending a BAR.
*/
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
if (ath_tx_tid_bar_tx_ready(sc, atid))
ath_tx_tid_bar_tx(sc, atid);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Handle frame completion */
Fix two bugs: * when pulling frames off of the TID queue, the ATH_TID_REMOVE() macro decrements the axq_depth field. So don't do it twice. * in ath_tx_comp_cleanup_aggr(), bf wasn't being reset to bf_first before walking the buffer list to complete buffers; so those buffers will leak.
2013-03-15 20:00:08 +00:00
bf = bf_first;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
while (bf) {
bf_next = bf->bf_next;
ath_tx_default_comp(sc, bf, 1);
bf = bf_next;
}
}
/*
* Handle completion of an set of aggregate frames.
*
* Note: the completion handler is the last descriptor in the aggregate,
* not the last descriptor in the first frame.
*/
static void
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
ath_tx_aggr_comp_aggr(struct ath_softc *sc, struct ath_buf *bf_first,
int fail)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
{
//struct ath_desc *ds = bf->bf_lastds;
struct ieee80211_node *ni = bf_first->bf_node;
struct ath_node *an = ATH_NODE(ni);
int tid = bf_first->bf_state.bfs_tid;
struct ath_tid *atid = &an->an_tid[tid];
struct ath_tx_status ts;
struct ieee80211_tx_ampdu *tap;
ath_bufhead bf_q;
ath_bufhead bf_cq;
int seq_st, tx_ok;
int hasba, isaggr;
uint32_t ba[2];
struct ath_buf *bf, *bf_next;
int ba_index;
int drops = 0;
int nframes = 0, nbad = 0, nf;
int pktlen;
/* XXX there's too much on the stack? */
Avoid using hard-coded numbers here.
2012-05-26 01:35:11 +00:00
struct ath_rc_series rc[ATH_RC_NUM];
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
int txseq;
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR, "%s: called; hwq_depth=%d\n",
__func__, atid->hwq_depth);
Oops - take a copy of ath_tx_status from the buffer before the TX processing is done. The aggregate path was definitely accessing 'ts' before it was actually being assigned. This had the side effect of over-filtering frames, since occasionally that bit would be '1'. Whilst here, do the same thing in the non-aggregate completion function - as calling the filter function may also invalidate bf. Pointy hat to: adrian, for not noticing this over many, many code reviews.
2012-09-18 20:33:04 +00:00
/*
* Take a copy; this may be needed -after- bf_first
* has been completed and freed.
*/
ts = bf_first->bf_status.ds_txstat;
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
TAILQ_INIT(&bf_q);
TAILQ_INIT(&bf_cq);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* The TID state is kept behind the TXQ lock */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
atid->hwq_depth--;
if (atid->hwq_depth < 0)
device_printf(sc->sc_dev, "%s: hwq_depth < 0: %d\n",
__func__, atid->hwq_depth);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If the TID is filtered, handle completing the filter
* transition before potentially kicking it to the cleanup
* function.
Oops - take a copy of ath_tx_status from the buffer before the TX processing is done. The aggregate path was definitely accessing 'ts' before it was actually being assigned. This had the side effect of over-filtering frames, since occasionally that bit would be '1'. Whilst here, do the same thing in the non-aggregate completion function - as calling the filter function may also invalidate bf. Pointy hat to: adrian, for not noticing this over many, many code reviews.
2012-09-18 20:33:04 +00:00
*
* XXX this is duplicate work, ew.
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
*/
if (atid->isfiltered)
ath_tx_tid_filt_comp_complete(sc, atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Punt cleanup to the relevant function, not our problem now
*/
if (atid->cleanup_inprogress) {
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
if (atid->isfiltered)
device_printf(sc->sc_dev,
"%s: isfiltered=1, normal_comp?\n",
__func__);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_comp_cleanup_aggr(sc, bf_first);
return;
}
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If the frame is filtered, transition to filtered frame
* mode and add this to the filtered frame list.
*
* XXX TODO: figure out how this interoperates with
* BAR, pause and cleanup states.
*/
if ((ts.ts_status & HAL_TXERR_FILT) ||
(ts.ts_status != 0 && atid->isfiltered)) {
if (fail != 0)
device_printf(sc->sc_dev,
"%s: isfiltered=1, fail=%d\n", __func__, fail);
ath_tx_tid_filt_comp_aggr(sc, atid, bf_first, &bf_cq);
/* Remove from BAW */
TAILQ_FOREACH_SAFE(bf, &bf_cq, bf_list, bf_next) {
if (bf->bf_state.bfs_addedbaw)
drops++;
if (bf->bf_state.bfs_dobaw) {
ath_tx_update_baw(sc, an, atid, bf);
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
__func__,
SEQNO(bf->bf_state.bfs_seqno));
}
bf->bf_state.bfs_dobaw = 0;
}
/*
* If any intermediate frames in the BAW were dropped when
* handling filtering things, send a BAR.
*/
if (drops)
ath_tx_tid_bar_suspend(sc, atid);
/*
* Finish up by sending a BAR if required and freeing
* the frames outside of the TX lock.
*/
goto finish_send_bar;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* XXX for now, use the first frame in the aggregate for
* XXX rate control completion; it's at least consistent.
*/
pktlen = bf_first->bf_state.bfs_pktlen;
/*
Handle non-xretry errors the same as xretry errors for now. Although I _should_ handle the other errors in various ways (specifically errors like FILT), treating them as having transmitted successfully is completely wrong. Here, they'd be counted as successful and the BAW would be advanced.. but the RX side wouldn't have received them. The specific errors I've been seeing here are HAL_TXERR_FILT. This patch does fix the issue - I've tested it using -i 0.001 pings (enough to start aggregation) and now the behaviour is correct: * The RX side never sees a "moved window" error, and * The TX side sends BARs as needed, with the RX side correctly handling them. PR: kern/167902
2012-05-15 04:55:15 +00:00
* Handle errors first!
*
* Here, handle _any_ error as a "exceeded retries" error.
* Later on (when filtered frames are to be specially handled)
* it'll have to be expanded.
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*/
Handle non-xretry errors the same as xretry errors for now. Although I _should_ handle the other errors in various ways (specifically errors like FILT), treating them as having transmitted successfully is completely wrong. Here, they'd be counted as successful and the BAW would be advanced.. but the RX side wouldn't have received them. The specific errors I've been seeing here are HAL_TXERR_FILT. This patch does fix the issue - I've tested it using -i 0.001 pings (enough to start aggregation) and now the behaviour is correct: * The RX side never sees a "moved window" error, and * The TX side sends BARs as needed, with the RX side correctly handling them. PR: kern/167902
2012-05-15 04:55:15 +00:00
#if 0
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (ts.ts_status & HAL_TXERR_XRETRY) {
Handle non-xretry errors the same as xretry errors for now. Although I _should_ handle the other errors in various ways (specifically errors like FILT), treating them as having transmitted successfully is completely wrong. Here, they'd be counted as successful and the BAW would be advanced.. but the RX side wouldn't have received them. The specific errors I've been seeing here are HAL_TXERR_FILT. This patch does fix the issue - I've tested it using -i 0.001 pings (enough to start aggregation) and now the behaviour is correct: * The RX side never sees a "moved window" error, and * The TX side sends BARs as needed, with the RX side correctly handling them. PR: kern/167902
2012-05-15 04:55:15 +00:00
#endif
if (ts.ts_status != 0) {
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_comp_aggr_error(sc, bf_first, atid);
return;
}
tap = ath_tx_get_tx_tid(an, tid);
/*
* extract starting sequence and block-ack bitmap
*/
/* XXX endian-ness of seq_st, ba? */
seq_st = ts.ts_seqnum;
hasba = !! (ts.ts_flags & HAL_TX_BA);
tx_ok = (ts.ts_status == 0);
isaggr = bf_first->bf_state.bfs_aggr;
ba[0] = ts.ts_ba_low;
ba[1] = ts.ts_ba_high;
/*
* Copy the TX completion status and the rate control
* series from the first descriptor, as it may be freed
* before the rate control code can get its grubby fingers
* into things.
*/
memcpy(rc, bf_first->bf_state.bfs_rc, sizeof(rc));
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
"%s: txa_start=%d, tx_ok=%d, status=%.8x, flags=%.8x, "
"isaggr=%d, seq_st=%d, hasba=%d, ba=%.8x, %.8x\n",
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
__func__, tap->txa_start, tx_ok, ts.ts_status, ts.ts_flags,
isaggr, seq_st, hasba, ba[0], ba[1]);
Disable the hw TID != buffer TID check. I can 100% reliably trigger this on TID 1 traffic by using iperf -S 32 <client fields> to create traffic that maps to TID 1. The reference driver doesn't do this check.
2013-03-09 08:50:17 +00:00
/*
* The reference driver doesn't do this; it simply ignores
* this check in its entirety.
*
* I've seen this occur when using iperf to send traffic
* out tid 1 - the aggregate frames are all marked as TID 1,
* but the TXSTATUS has TID=0. So, let's just ignore this
* check.
*/
#if 0
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Occasionally, the MAC sends a tx status for the wrong TID. */
if (tid != ts.ts_tid) {
device_printf(sc->sc_dev, "%s: tid %d != hw tid %d\n",
__func__, tid, ts.ts_tid);
tx_ok = 0;
}
Disable the hw TID != buffer TID check. I can 100% reliably trigger this on TID 1 traffic by using iperf -S 32 <client fields> to create traffic that maps to TID 1. The reference driver doesn't do this check.
2013-03-09 08:50:17 +00:00
#endif
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* AR5416 BA bug; this requires an interface reset */
if (isaggr && tx_ok && (! hasba)) {
device_printf(sc->sc_dev,
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
"%s: AR5416 bug: hasba=%d; txok=%d, isaggr=%d, "
"seq_st=%d\n",
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
__func__, hasba, tx_ok, isaggr, seq_st);
/* XXX TODO: schedule an interface reset */
Fix build when ATH_DEBUG is not defined.
2012-07-10 18:57:05 +00:00
#ifdef ATH_DEBUG
Print the TX buffer if this error condition is asserted. I need to figure out why this is occuring. Hopefully I can get enough descriptor dumps to figure it out.
2012-07-10 06:10:49 +00:00
ath_printtxbuf(sc, bf_first,
sc->sc_ac2q[atid->ac]->axq_qnum, 0, 0);
Fix build when ATH_DEBUG is not defined.
2012-07-10 18:57:05 +00:00
#endif
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
/*
* Walk the list of frames, figure out which ones were correctly
* sent and which weren't.
*/
bf = bf_first;
nf = bf_first->bf_state.bfs_nframes;
/* bf_first is going to be invalid once this list is walked */
bf_first = NULL;
/*
* Walk the list of completed frames and determine
* which need to be completed and which need to be
* retransmitted.
*
* For completed frames, the completion functions need
* to be called at the end of this function as the last
* node reference may free the node.
*
* Finally, since the TXQ lock can't be held during the
* completion callback (to avoid lock recursion),
* the completion calls have to be done outside of the
* lock.
*/
while (bf) {
nframes++;
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
ba_index = ATH_BA_INDEX(seq_st,
SEQNO(bf->bf_state.bfs_seqno));
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf_next = bf->bf_next;
bf->bf_next = NULL; /* Remove it from the aggr list */
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
"%s: checking bf=%p seqno=%d; ack=%d\n",
__func__, bf, SEQNO(bf->bf_state.bfs_seqno),
ATH_BA_ISSET(ba, ba_index));
if (tx_ok && ATH_BA_ISSET(ba, ba_index)) {
Flesh out the TX aggregation completion statistics. * Failall is now named just that. * Add TX ok and TX fail, for aggregate frame sub-frames. This will break athstats; a followup commit wil resolve this. Sponsored by: Hobnob, Inc.
2011-11-23 05:00:25 +00:00
sc->sc_stats.ast_tx_aggr_ok++;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_update_baw(sc, an, atid, bf);
bf->bf_state.bfs_dobaw = 0;
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
bf->bf_next = NULL;
TAILQ_INSERT_TAIL(&bf_cq, bf, bf_list);
} else {
Flesh out the TX aggregation completion statistics. * Failall is now named just that. * Add TX ok and TX fail, for aggregate frame sub-frames. This will break athstats; a followup commit wil resolve this. Sponsored by: Hobnob, Inc.
2011-11-23 05:00:25 +00:00
sc->sc_stats.ast_tx_aggr_fail++;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (ath_tx_retry_subframe(sc, bf, &bf_q)) {
drops++;
bf->bf_next = NULL;
TAILQ_INSERT_TAIL(&bf_cq, bf, bf_list);
}
nbad++;
}
bf = bf_next;
}
/*
* Now that the BAW updates have been done, unlock
*
* txseq is grabbed before the lock is released so we
* have a consistent view of what -was- in the BAW.
* Anything after this point will not yet have been
* TXed.
*/
txseq = tap->txa_start;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (nframes != nf)
device_printf(sc->sc_dev,
"%s: num frames seen=%d; bf nframes=%d\n",
__func__, nframes, nf);
/*
* Now we know how many frames were bad, call the rate
* control code.
*/
if (fail == 0)
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
ath_tx_update_ratectrl(sc, ni, rc, &ts, pktlen, nframes,
nbad);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* send bar if we dropped any frames
*/
if (drops) {
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/* Suspend the TX queue and get ready to send the BAR */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
ath_tx_tid_bar_suspend(sc, atid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Remove some of the redundant locking done in the TX completion path, when checking whether BAR frames need to be checked.
2012-04-26 23:57:24 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
"%s: txa_start now %d\n", __func__, tap->txa_start);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Remove some of the redundant locking done in the TX completion path, when checking whether BAR frames need to be checked.
2012-04-26 23:57:24 +00:00
/* Prepend all frames to the beginning of the queue */
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
while ((bf = TAILQ_LAST(&bf_q, ath_bufhead_s)) != NULL) {
TAILQ_REMOVE(&bf_q, bf, bf_list);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_INSERT_HEAD(atid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Remove some of the redundant locking done in the TX completion path, when checking whether BAR frames need to be checked.
2012-04-26 23:57:24 +00:00
/*
* Reschedule to grab some further frames.
*/
ath_tx_tid_sched(sc, atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If the queue is filtered, re-schedule as required.
*
* This is required as there may be a subsequent TX descriptor
* for this end-node that has CLRDMASK set, so it's quite possible
* that a filtered frame will be followed by a non-filtered
* (complete or otherwise) frame.
*
* XXX should we do this before we complete the frame?
*/
if (atid->isfiltered)
ath_tx_tid_filt_comp_complete(sc, atid);
finish_send_bar:
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/*
* Send BAR if required
*/
if (ath_tx_tid_bar_tx_ready(sc, atid))
ath_tx_tid_bar_tx(sc, atid);
Remove some of the redundant locking done in the TX completion path, when checking whether BAR frames need to be checked.
2012-04-26 23:57:24 +00:00
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Do deferred completion */
while ((bf = TAILQ_FIRST(&bf_cq)) != NULL) {
TAILQ_REMOVE(&bf_cq, bf, bf_list);
ath_tx_default_comp(sc, bf, 0);
}
}
/*
* Handle completion of unaggregated frames in an ADDBA
* session.
*
* Fail is set to 1 if the entry is being freed via a call to
* ath_tx_draintxq().
*/
static void
ath_tx_aggr_comp_unaggr(struct ath_softc *sc, struct ath_buf *bf, int fail)
{
struct ieee80211_node *ni = bf->bf_node;
struct ath_node *an = ATH_NODE(ni);
int tid = bf->bf_state.bfs_tid;
struct ath_tid *atid = &an->an_tid[tid];
Oops - take a copy of ath_tx_status from the buffer before the TX processing is done. The aggregate path was definitely accessing 'ts' before it was actually being assigned. This had the side effect of over-filtering frames, since occasionally that bit would be '1'. Whilst here, do the same thing in the non-aggregate completion function - as calling the filter function may also invalidate bf. Pointy hat to: adrian, for not noticing this over many, many code reviews.
2012-09-18 20:33:04 +00:00
struct ath_tx_status ts;
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
int drops = 0;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
Oops - take a copy of ath_tx_status from the buffer before the TX processing is done. The aggregate path was definitely accessing 'ts' before it was actually being assigned. This had the side effect of over-filtering frames, since occasionally that bit would be '1'. Whilst here, do the same thing in the non-aggregate completion function - as calling the filter function may also invalidate bf. Pointy hat to: adrian, for not noticing this over many, many code reviews.
2012-09-18 20:33:04 +00:00
/*
* Take a copy of this; filtering/cloning the frame may free the
* bf pointer.
*/
ts = bf->bf_status.ds_txstat;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Update rate control status here, before we possibly
* punt to retry or cleanup.
*
* Do it outside of the TXQ lock.
*/
Remove duplicate txflags field from ath_buf. rename bf_state.bfs_flags to bf_state.bfs_txflags, as that is what it effectively is.
2012-04-07 02:01:26 +00:00
if (fail == 0 && ((bf->bf_state.bfs_txflags & HAL_TXDESC_NOACK) == 0))
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_update_ratectrl(sc, ni, bf->bf_state.bfs_rc,
&bf->bf_status.ds_txstat,
bf->bf_state.bfs_pktlen,
Oops - take a copy of ath_tx_status from the buffer before the TX processing is done. The aggregate path was definitely accessing 'ts' before it was actually being assigned. This had the side effect of over-filtering frames, since occasionally that bit would be '1'. Whilst here, do the same thing in the non-aggregate completion function - as calling the filter function may also invalidate bf. Pointy hat to: adrian, for not noticing this over many, many code reviews.
2012-09-18 20:33:04 +00:00
1, (ts.ts_status == 0) ? 0 : 1);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* This is called early so atid->hwq_depth can be tracked.
* This unfortunately means that it's released and regrabbed
* during retry and cleanup. That's rather inefficient.
*/
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (tid == IEEE80211_NONQOS_TID)
device_printf(sc->sc_dev, "%s: TID=16!\n", __func__);
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX,
"%s: bf=%p: tid=%d, hwq_depth=%d, seqno=%d\n",
__func__, bf, bf->bf_state.bfs_tid, atid->hwq_depth,
SEQNO(bf->bf_state.bfs_seqno));
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
atid->hwq_depth--;
if (atid->hwq_depth < 0)
device_printf(sc->sc_dev, "%s: hwq_depth < 0: %d\n",
__func__, atid->hwq_depth);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If the TID is filtered, handle completing the filter
* transition before potentially kicking it to the cleanup
* function.
*/
if (atid->isfiltered)
ath_tx_tid_filt_comp_complete(sc, atid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* If a cleanup is in progress, punt to comp_cleanup;
* rather than handling it here. It's thus their
* responsibility to clean up, call the completion
* function in net80211, etc.
*/
if (atid->cleanup_inprogress) {
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
if (atid->isfiltered)
device_printf(sc->sc_dev,
"%s: isfiltered=1, normal_comp?\n",
__func__);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: cleanup_unaggr\n",
__func__);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_comp_cleanup_unaggr(sc, bf);
return;
}
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* XXX TODO: how does cleanup, BAR and filtered frame handling
* overlap?
*
* If the frame is filtered OR if it's any failure but
* the TID is filtered, the frame must be added to the
* filtered frame list.
*
* However - a busy buffer can't be added to the filtered
* list as it will end up being recycled without having
* been made available for the hardware.
*/
Oops - take a copy of ath_tx_status from the buffer before the TX processing is done. The aggregate path was definitely accessing 'ts' before it was actually being assigned. This had the side effect of over-filtering frames, since occasionally that bit would be '1'. Whilst here, do the same thing in the non-aggregate completion function - as calling the filter function may also invalidate bf. Pointy hat to: adrian, for not noticing this over many, many code reviews.
2012-09-18 20:33:04 +00:00
if ((ts.ts_status & HAL_TXERR_FILT) ||
(ts.ts_status != 0 && atid->isfiltered)) {
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
int freeframe;
if (fail != 0)
device_printf(sc->sc_dev,
"%s: isfiltered=1, fail=%d\n",
__func__,
fail);
freeframe = ath_tx_tid_filt_comp_single(sc, atid, bf);
if (freeframe) {
/* Remove from BAW */
if (bf->bf_state.bfs_addedbaw)
drops++;
if (bf->bf_state.bfs_dobaw) {
ath_tx_update_baw(sc, an, atid, bf);
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
}
bf->bf_state.bfs_dobaw = 0;
}
/*
* If the frame couldn't be filtered, treat it as a drop and
* prepare to send a BAR.
*/
if (freeframe && drops)
ath_tx_tid_bar_suspend(sc, atid);
/*
* Send BAR if required
*/
if (ath_tx_tid_bar_tx_ready(sc, atid))
ath_tx_tid_bar_tx(sc, atid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If freeframe is set, then the frame couldn't be
* cloned and bf is still valid. Just complete/free it.
*/
if (freeframe)
ath_tx_default_comp(sc, bf, fail);
return;
}
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Don't bother with the retry check if all frames
* are being failed (eg during queue deletion.)
*/
Handle non-xretry errors the same as xretry errors for now. Although I _should_ handle the other errors in various ways (specifically errors like FILT), treating them as having transmitted successfully is completely wrong. Here, they'd be counted as successful and the BAW would be advanced.. but the RX side wouldn't have received them. The specific errors I've been seeing here are HAL_TXERR_FILT. This patch does fix the issue - I've tested it using -i 0.001 pings (enough to start aggregation) and now the behaviour is correct: * The RX side never sees a "moved window" error, and * The TX side sends BARs as needed, with the RX side correctly handling them. PR: kern/167902
2012-05-15 04:55:15 +00:00
#if 0
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (fail == 0 && ts->ts_status & HAL_TXERR_XRETRY) {
Handle non-xretry errors the same as xretry errors for now. Although I _should_ handle the other errors in various ways (specifically errors like FILT), treating them as having transmitted successfully is completely wrong. Here, they'd be counted as successful and the BAW would be advanced.. but the RX side wouldn't have received them. The specific errors I've been seeing here are HAL_TXERR_FILT. This patch does fix the issue - I've tested it using -i 0.001 pings (enough to start aggregation) and now the behaviour is correct: * The RX side never sees a "moved window" error, and * The TX side sends BARs as needed, with the RX side correctly handling them. PR: kern/167902
2012-05-15 04:55:15 +00:00
#endif
Oops - take a copy of ath_tx_status from the buffer before the TX processing is done. The aggregate path was definitely accessing 'ts' before it was actually being assigned. This had the side effect of over-filtering frames, since occasionally that bit would be '1'. Whilst here, do the same thing in the non-aggregate completion function - as calling the filter function may also invalidate bf. Pointy hat to: adrian, for not noticing this over many, many code reviews.
2012-09-18 20:33:04 +00:00
if (fail == 0 && ts.ts_status != 0) {
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: retry_unaggr\n",
__func__);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_aggr_retry_unaggr(sc, bf);
return;
}
/* Success? Complete */
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: TID=%d, seqno %d\n",
__func__, tid, SEQNO(bf->bf_state.bfs_seqno));
if (bf->bf_state.bfs_dobaw) {
ath_tx_update_baw(sc, an, atid, bf);
bf->bf_state.bfs_dobaw = 0;
if (! bf->bf_state.bfs_addedbaw)
device_printf(sc->sc_dev,
"%s: wasn't added: seqno %d\n",
__func__, SEQNO(bf->bf_state.bfs_seqno));
}
Implement my first cut at filtered frames in aggregation sessions. The hardware can optionally "filter" frames if successive transmissions to a given node (ie, "entry in the keycache") fail. That way the hardware can implement a kind of early abort of all the other frames queued to that destination, rather than simply trying to TX each frame to that destination (and failing.) The background: * If a frame comes back as being filtered, the hardware didn't try to TX it (or it was outside the TX burst opportunity.) So, take it as a hint that some (but not all, see below) frames to the destination may be filtered. * If the CLRDMASK bit is set in a TX descriptor, the "filter to this destination" bit in the keycache entry is cleared and TX to that host will be unconditionally retried. * Right now everything has the CLRDMASK bit set, so filtered frames tend to be aggregates and frames that fall outside of the WME burst window. It was a bit worse in the past as I had messed up the TX flags and CLRDMASK wasn't being set on aggregate frames. The annoying bits: * It's easy (ish) to do for aggregate session frames - firstly, they can be retried in any order as long as they're within the BAW, and there's already a bunch of infrastructure tracking how many frames the TID has queued to the hardware (tid->hwq_depth.) However, for frames that bypassed the software queue, hwq_depth doesn't get incremented. I'll fix that in a subsequent commit. * For non-aggregate session frames, the only retries that can occur are ones for sequence numbers that hvaen't successfully been TXed yet. Since there's no re-ordering going on in non-aggregate sessions, if any subsequent seqno frames make it out, any filtered frames before that seqno need to be dropped. Hence why this initially is just for aggregate session frames. * Since there may be intermediary frames to the destination that have CLRDMASK set - for example, any directly dispatched management frames to that destination - it's possible that there will be some filtered frames followed up by some non filtered frames. Thus, it can't be assumed that once you see a filtered frame for the given destination node, all subsequent frames for all TIDs will be filtered. Ok, with that in mind: * Create a per-TID filtered frame queue for frames that the hardware returns as filtered. * Track filtered frames per-tid, rather than per-node. It just makes the locking much easier. * When a filtered frame appears in the completion function, the node transitions to "filtered", and all subsequent completed error frames (filtered or otherwise) are put on the filtered frame queue. The TID is paused once (during the transition from non-filtered to filtered). * If a filtered frame retry count exceeds SWMAX_RETRIES, a BAR should be sent. * Once all the frames queued to the hardware for the given filtered frame TID, transition back from filtered frame to non-filtered frame, which means pre-pending all the filtered frames onto the head of the software queue, clearing the filtered frame state and unpausing the TID. Things get quite hairy around handling completion (aggr, non-aggr, norm, direct-dispatched frames to a hardware queue); whether it's an "error", "cleanup" or "BAR" state as well as filtered, which order to do things in (eg do filtered BEFORE checking for BAR, as the filter completion may be needed to actually transmit a BAR frame.) This work has definitely reminded me that I have to tidy up all the locking and remove some of the ridiculous lock/unlock/lock/unlock going on in the completion functions. It's also reminded me that I should really split out TID versus hardware TXQ locking, even if the underlying locking is still the destination hardware TXQ. Finally, this is all pre-requisite for working on AP mode power save support (PS-POLL, uAPSD) as well as improving performance to misbehaving nodes (as they can transition into filter mode, stopping any TX until everything has caught up.) Finally (ish) - this should also be done for non-aggregate sessions as there are still plenty of laptops and mobile devices that don't speak 802.11n but do wish for stable, useful power save AP support where packets aren't simply dropped. This requires software retransmission for non-aggregate sessions to be implemented, which includes the caveats I've mentioned above. Finally finally - this doesn't yet do anything about the CLRDMASK bit in the TX descriptor. That's still unconditionally set to 1. I'll debug the current work (mostly ensuring I haven't busted up the hairy transitions between BAR, filtered, error (all frames in an aggregate failing) and cleanup (when transitioning from aggregation -> non-aggregation.)) Finally finally finally - this is all original work by yours truely, rather than ported from the Atheros internal driver codebase or Linux ath9k. Tested: * AR9280, AR5416 in STA mode * AR9280, AR9130 in hostap mode * Lots and lots of iperf testing in very marginal and non-marginal conditions, complete with inducing filtered frames + BAR TX conditions.
2012-09-18 10:14:17 +00:00
/*
* If the queue is filtered, re-schedule as required.
*
* This is required as there may be a subsequent TX descriptor
* for this end-node that has CLRDMASK set, so it's quite possible
* that a filtered frame will be followed by a non-filtered
* (complete or otherwise) frame.
*
* XXX should we do this before we complete the frame?
*/
if (atid->isfiltered)
ath_tx_tid_filt_comp_complete(sc, atid);
Implement BAR TX. A BAR frame must be transmitted when an frame in an A-MPDU session fails to transmit - it's retried too often, or it can't be cloned for re-transmission. The BAR frame tells the remote side to advance the left edge of the block-ack window (BAW) to a new value. In order to do this: * TX for that particular node/TID must be paused; * The existing frames in the hardware queue needs to be completed, whether they're TXed successfully or otherwise; * The new left edge of the BAW is then communicated to the remote side via a BAR frame; * Once the BAR frame has been sucessfully TXed, aggregation can resume; * If the BAR frame can't be successfully TXed, the aggregation session is torn down. This is a first pass that implements the above. What needs to be done/ tested: * What happens during say, a channel reset / stuck beacon _and_ BAR TX. It _should_ be correctly buffered and retried once the reset has completed. But if a bgscan occurs (and they shouldn't, grr) the BAR frame will be forcibly failed and the aggregation session will be torn down. Yes, another reason to disable bgscan until I've figured this out. * There's way too much locking going on here. I'm going to do a couple of further passes of sanitising and refactoring so the (re) locking isn't so heavy. Right now I'm going for correctness, not speed. * The BAR TX can fail if the hardware TX queue is full. Since there's no "free" space kept for management frames, a full TX queue (from eg an iperf test) can race with your ability to allocate ath_buf/mbufs and cause issues. I'll knock this on the head with a subsequent commit. * I need to do some _much_ more thorough testing in hostap mode to ensure that many concurrent traffic streams to different end nodes are correctly handled. I'll find and squish whichever bugs show up here. But, this is an important step to being able to flip on 802.11n by default. The last issue (besides bug fixes, of course) is HT frame protection and I'll address that in a subsequent commit.
2012-04-04 23:45:15 +00:00
/*
* Send BAR if required
*/
if (ath_tx_tid_bar_tx_ready(sc, atid))
ath_tx_tid_bar_tx(sc, atid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_default_comp(sc, bf, fail);
/* bf is freed at this point */
}
void
ath_tx_aggr_comp(struct ath_softc *sc, struct ath_buf *bf, int fail)
{
if (bf->bf_state.bfs_aggr)
ath_tx_aggr_comp_aggr(sc, bf, fail);
else
ath_tx_aggr_comp_unaggr(sc, bf, fail);
}
/*
* Schedule some packets from the given node/TID to the hardware.
*
* This is the aggregate version.
*/
void
ath_tx_tid_hw_queue_aggr(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid)
{
struct ath_buf *bf;
struct ath_txq *txq = sc->sc_ac2q[tid->ac];
struct ieee80211_tx_ampdu *tap;
ATH_AGGR_STATUS status;
ath_bufhead bf_q;
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: tid=%d\n", __func__, tid->tid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
tap = ath_tx_get_tx_tid(an, tid->tid);
if (tid->tid == IEEE80211_NONQOS_TID)
device_printf(sc->sc_dev, "%s: called for TID=NONQOS_TID?\n",
__func__);
for (;;) {
status = ATH_AGGR_DONE;
/*
* If the upper layer has paused the TID, don't
* queue any further packets.
*
* This can also occur from the completion task because
* of packet loss; but as its serialised with this code,
* it won't "appear" half way through queuing packets.
*/
if (tid->paused)
break;
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
bf = ATH_TID_FIRST(tid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (bf == NULL) {
break;
}
/*
* If the packet doesn't fall within the BAW (eg a NULL
* data frame), schedule it directly; continue.
*/
if (! bf->bf_state.bfs_dobaw) {
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
"%s: non-baw packet\n",
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
__func__);
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_REMOVE(tid, bf, bf_list);
Ensure that single-frame aggregate session frames are retransmitted with the correct configuration. Occasionally an aggregate TX would fail and the first frame would be retransmitted as a non-AMPDU frame. Since bfs_aggr=1 and bfs_nframes > 1 (from the previous AMPDU attempt), the aggr completion function would be called and be very confused about what's going on. Noticed by: Kim <w8hdkim@gmail.com> PR: kern/171394
2012-09-07 00:24:27 +00:00
if (bf->bf_state.bfs_nframes > 1)
device_printf(sc->sc_dev,
"%s: aggr=%d, nframes=%d\n",
__func__,
bf->bf_state.bfs_aggr,
bf->bf_state.bfs_nframes);
/*
* This shouldn't happen - such frames shouldn't
* ever have been queued as an aggregate in the
* first place. However, make sure the fields
* are correctly setup just to be totally sure.
*/
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_aggr = 0;
Ensure that single-frame aggregate session frames are retransmitted with the correct configuration. Occasionally an aggregate TX would fail and the first frame would be retransmitted as a non-AMPDU frame. Since bfs_aggr=1 and bfs_nframes > 1 (from the previous AMPDU attempt), the aggr completion function would be called and be very confused about what's going on. Noticed by: Kim <w8hdkim@gmail.com> PR: kern/171394
2012-09-07 00:24:27 +00:00
bf->bf_state.bfs_nframes = 1;
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/* Update CLRDMASK just before this frame is queued */
ath_tx_update_clrdmask(sc, tid, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_do_ratelookup(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_calc_duration(sc, bf);
ath_tx_calc_protection(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_set_rtscts(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_rate_fill_rcflags(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_setds(sc, bf);
ath_hal_clr11n_aggr(sc->sc_ah, bf->bf_desc);
sc->sc_aggr_stats.aggr_nonbaw_pkt++;
/* Queue the packet; continue */
goto queuepkt;
}
TAILQ_INIT(&bf_q);
/*
* Do a rate control lookup on the first frame in the
* list. The rate control code needs that to occur
* before it can determine whether to TX.
* It's inaccurate because the rate control code doesn't
* really "do" aggregate lookups, so it only considers
* the size of the first frame.
*/
ath_tx_do_ratelookup(sc, bf);
bf->bf_state.bfs_rc[3].rix = 0;
bf->bf_state.bfs_rc[3].tries = 0;
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_calc_duration(sc, bf);
ath_tx_calc_protection(sc, bf);
ath_tx_set_rtscts(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_rate_fill_rcflags(sc, bf);
status = ath_tx_form_aggr(sc, an, tid, &bf_q);
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
"%s: ath_tx_form_aggr() status=%d\n", __func__, status);
/*
* No frames to be picked up - out of BAW
*/
if (TAILQ_EMPTY(&bf_q))
break;
/*
* This assumes that the descriptor list in the ath_bufhead
* are already linked together via bf_next pointers.
*/
bf = TAILQ_FIRST(&bf_q);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
if (status == ATH_AGGR_8K_LIMITED)
sc->sc_aggr_stats.aggr_rts_aggr_limited++;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* If it's the only frame send as non-aggregate
* assume that ath_tx_form_aggr() has checked
* whether it's in the BAW and added it appropriately.
*/
if (bf->bf_state.bfs_nframes == 1) {
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
"%s: single-frame aggregate\n", __func__);
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/* Update CLRDMASK just before this frame is queued */
ath_tx_update_clrdmask(sc, tid, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
bf->bf_state.bfs_aggr = 0;
Make sure the aggregate fields are properly cleared - both in the ath_buf and when forming a non-aggregate frame. The non-11n setds function is called when TXing aggregate frames (and yes, I should fix this!) and the non-11n TX aggregation code doesn't clear the delimiter field. I figure it's nicer to do that.
2012-09-09 05:06:16 +00:00
bf->bf_state.bfs_ndelim = 0;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_setds(sc, bf);
ath_hal_clr11n_aggr(sc->sc_ah, bf->bf_desc);
if (status == ATH_AGGR_BAW_CLOSED)
sc->sc_aggr_stats.aggr_baw_closed_single_pkt++;
else
sc->sc_aggr_stats.aggr_single_pkt++;
} else {
DPRINTF(sc, ATH_DEBUG_SW_TX_AGGR,
style(9) changes. This shouldn't change functionality.
2012-01-11 00:16:44 +00:00
"%s: multi-frame aggregate: %d frames, "
"length %d\n",
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
__func__, bf->bf_state.bfs_nframes,
bf->bf_state.bfs_al);
bf->bf_state.bfs_aggr = 1;
sc->sc_aggr_stats.aggr_pkts[bf->bf_state.bfs_nframes]++;
sc->sc_aggr_stats.aggr_aggr_pkt++;
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/* Update CLRDMASK just before this frame is queued */
ath_tx_update_clrdmask(sc, tid, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
/*
* Calculate the duration/protection as required.
*/
ath_tx_calc_duration(sc, bf);
ath_tx_calc_protection(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Update the rate and rtscts information based on the
* rate decision made by the rate control code;
* the first frame in the aggregate needs it.
*/
ath_tx_set_rtscts(sc, bf);
/*
* Setup the relevant descriptor fields
* for aggregation. The first descriptor
* already points to the rest in the chain.
*/
ath_tx_setds_11n(sc, bf);
}
queuepkt:
/* Set completion handler, multi-frame aggregate or not */
bf->bf_comp = ath_tx_aggr_comp;
if (bf->bf_state.bfs_tid == IEEE80211_NONQOS_TID)
device_printf(sc->sc_dev, "%s: TID=16?\n", __func__);
/* Punt to txq */
ath_tx_handoff(sc, txq, bf);
/* Track outstanding buffer count to hardware */
/* aggregates are "one" buffer */
tid->hwq_depth++;
/*
* Break out if ath_tx_form_aggr() indicated
* there can't be any further progress (eg BAW is full.)
* Checking for an empty txq is done above.
*
* XXX locking on txq here?
*/
if (txq->axq_aggr_depth >= sc->sc_hwq_limit ||
status == ATH_AGGR_BAW_CLOSED)
break;
}
}
/*
* Schedule some packets from the given node/TID to the hardware.
*/
void
ath_tx_tid_hw_queue_norm(struct ath_softc *sc, struct ath_node *an,
struct ath_tid *tid)
{
struct ath_buf *bf;
struct ath_txq *txq = sc->sc_ac2q[tid->ac];
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: node %p: TID %d: called\n",
__func__, an, tid->tid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Check - is AMPDU pending or running? then print out something */
if (ath_tx_ampdu_pending(sc, an, tid->tid))
device_printf(sc->sc_dev, "%s: tid=%d, ampdu pending?\n",
__func__, tid->tid);
if (ath_tx_ampdu_running(sc, an, tid->tid))
device_printf(sc->sc_dev, "%s: tid=%d, ampdu running?\n",
__func__, tid->tid);
for (;;) {
/*
* If the upper layers have paused the TID, don't
* queue any further packets.
*/
if (tid->paused)
break;
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
bf = ATH_TID_FIRST(tid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
if (bf == NULL) {
break;
}
Migrate the TID TXQ accesses to a new set of macros, rather than reusing the ATH_TXQ_* macros. * Introduce the new macros; * rename the TID queue and TID filtered frame queue so the compiler tells me I'm using the wrong macro. These should correspond 1:1 to the existing code.
2012-10-07 23:45:19 +00:00
ATH_TID_REMOVE(tid, bf, bf_list);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Sanity check! */
if (tid->tid != bf->bf_state.bfs_tid) {
device_printf(sc->sc_dev, "%s: bfs_tid %d !="
" tid %d\n",
__func__, bf->bf_state.bfs_tid, tid->tid);
}
/* Normal completion handler */
bf->bf_comp = ath_tx_normal_comp;
Prepare for software retransmission of non-aggregate frames but ensure it's disabled. The previous commit to enable CLRDMASK setting didn't do it at all correctly for non-aggregate sessions - so the CLRDMASK bit would be cleared and never re-set. * move ath_tx_update_clrdmask() to be called by functions that setup descriptors and queue frames to the hardware, rather than scattered everywhere. * Force CLRDMASK to be set on all non-aggregate session frames being transmitted. * Use ath_tx_normal_comp() now on non-aggregate sessoin frames that are queued via ath_tx_xmit_normal(). That way the TID hwq is updated and they can trigger (eventual) filter frame queue resets and software retransmits. There's still a bit more work to do in this area to reverse the silly short-sightedness on my part, however it's likely going to be better to fix this now than just reverting the patch. Thanks to people on the freebsd-wireless@ mailing list for promptly pointing this out.
2012-09-24 06:42:20 +00:00
/*
* Override this for now, until the non-aggregate
* completion handler correctly handles software retransmits.
*/
bf->bf_state.bfs_txflags |= HAL_TXDESC_CLRDMASK;
Introduce the CLRDMASK gating based on tid->clrdmask, enabling filtered frames to occur. * Create a new function which will set the bf_flags CLRDMASK bit if required. * For raw frames, always set CLRDMASK. * For BAR, ADDBA frames, always set CLRDMASK. * For everything else, check if CLRDMASK needs to be set before calling tx_setds() or tx_setds11n(). * When unpausing a queue or drain/resetting it, set tid->clrdmask=1 just to ensure traffic starts flowing. What I need to do: * Modify that function to _clear_ the CLRDMASK if it's not required, or retried frames may have CLRDMASK set when they don't need to. (Which isn't a huge deal, but..) Whilst I'm here: * ath_tx_normal_xmit() should really act like the AMPDU session TX functions - any incomplete frames will end up being assigned ath_tx_normal_comp() which will decrement tid->hwq_depth - but that won't have been incremented. So whilst I'm here, add a comment to do that. * Fix the debug print function to be slightly clearer about things; it's not a good sign when I can't interpret my own debugging output. I've done some testing on AR9280/AR5416/AR9160 STA and AP modes.
2012-09-20 03:13:20 +00:00
/* Update CLRDMASK just before this frame is queued */
ath_tx_update_clrdmask(sc, tid, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Program descriptors + rate control */
ath_tx_do_ratelookup(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_calc_duration(sc, bf);
ath_tx_calc_protection(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_set_rtscts(sc, bf);
Break out the legacy duration and protection code into routines, call these after rate control selection is done. The duration/protection code wasn't working - it expected the rix to be valid. Unfortunately after I moved the rate control selection into late in the process, the rix value isn't valid and thus the protection/ duration code would get things wrong. HT frames are now correctly protected with an RTS and for the AR5416, this involves having the aggregate frames be limited to 8K. TODO: * Fix up the DMA sync to occur just before the frame is queued to the hardware. I'm adjusting the duration here but not doing the DMA flush. * Doubly/triply ensure that the aggregate frames are being limited to the correct size, or the AR5416 will get unhappy when TXing RTS-protected aggregates.
2012-04-07 05:48:26 +00:00
ath_tx_rate_fill_rcflags(sc, bf);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_setds(sc, bf);
/* Track outstanding buffer count to hardware */
/* aggregates are "one" buffer */
tid->hwq_depth++;
/* Punt to hardware or software txq */
ath_tx_handoff(sc, txq, bf);
}
}
/*
* Schedule some packets to the given hardware queue.
*
* This function walks the list of TIDs (ie, ath_node TIDs
* with queued traffic) and attempts to schedule traffic
* from them.
*
* TID scheduling is implemented as a FIFO, with TIDs being
* added to the end of the queue after some frames have been
* scheduled.
*/
void
ath_txq_sched(struct ath_softc *sc, struct ath_txq *txq)
{
struct ath_tid *tid, *next, *last;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK_ASSERT(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* Don't schedule if the hardware queue is busy.
* This (hopefully) gives some more time to aggregate
* some packets in the aggregation queue.
*/
if (txq->axq_aggr_depth >= sc->sc_hwq_limit) {
sc->sc_aggr_stats.aggr_sched_nopkt++;
return;
}
last = TAILQ_LAST(&txq->axq_tidq, axq_t_s);
TAILQ_FOREACH_SAFE(tid, &txq->axq_tidq, axq_qelem, next) {
/*
* Suspend paused queues here; they'll be resumed
* once the addba completes or times out.
*/
DPRINTF(sc, ATH_DEBUG_SW_TX, "%s: tid=%d, paused=%d\n",
__func__, tid->tid, tid->paused);
ath_tx_tid_unsched(sc, tid);
if (tid->paused) {
continue;
}
if (ath_tx_ampdu_running(sc, tid->an, tid->tid))
ath_tx_tid_hw_queue_aggr(sc, tid->an, tid);
else
ath_tx_tid_hw_queue_norm(sc, tid->an, tid);
/* Not empty? Re-schedule */
if (tid->axq_depth != 0)
ath_tx_tid_sched(sc, tid);
/* Give the software queue time to aggregate more packets */
if (txq->axq_aggr_depth >= sc->sc_hwq_limit) {
break;
}
/*
* If this was the last entry on the original list, stop.
* Otherwise nodes that have been rescheduled onto the end
* of the TID FIFO list will just keep being rescheduled.
*/
if (tid == last)
break;
}
}
/*
* TX addba handling
*/
/*
* Return net80211 TID struct pointer, or NULL for none
*/
struct ieee80211_tx_ampdu *
ath_tx_get_tx_tid(struct ath_node *an, int tid)
{
struct ieee80211_node *ni = &an->an_node;
struct ieee80211_tx_ampdu *tap;
if (tid == IEEE80211_NONQOS_TID)
return NULL;
Migrate the net80211 TX aggregation state to be from per-AC to per-TID. TODO: * Test mwl(4) more thoroughly! Reviewed by: bschmidt (for iwn)
2012-04-15 20:29:39 +00:00
tap = &ni->ni_tx_ampdu[tid];
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
return tap;
}
/*
* Is AMPDU-TX running?
*/
static int
ath_tx_ampdu_running(struct ath_softc *sc, struct ath_node *an, int tid)
{
struct ieee80211_tx_ampdu *tap;
if (tid == IEEE80211_NONQOS_TID)
return 0;
tap = ath_tx_get_tx_tid(an, tid);
if (tap == NULL)
return 0; /* Not valid; default to not running */
return !! (tap->txa_flags & IEEE80211_AGGR_RUNNING);
}
/*
* Is AMPDU-TX negotiation pending?
*/
static int
ath_tx_ampdu_pending(struct ath_softc *sc, struct ath_node *an, int tid)
{
struct ieee80211_tx_ampdu *tap;
if (tid == IEEE80211_NONQOS_TID)
return 0;
tap = ath_tx_get_tx_tid(an, tid);
if (tap == NULL)
return 0; /* Not valid; default to not pending */
return !! (tap->txa_flags & IEEE80211_AGGR_XCHGPEND);
}
/*
* Is AMPDU-TX pending for the given TID?
*/
/*
* Method to handle sending an ADDBA request.
*
* We tap this so the relevant flags can be set to pause the TID
* whilst waiting for the response.
*
* XXX there's no timeout handler we can override?
*/
int
ath_addba_request(struct ieee80211_node *ni, struct ieee80211_tx_ampdu *tap,
int dialogtoken, int baparamset, int batimeout)
{
struct ath_softc *sc = ni->ni_ic->ic_ifp->if_softc;
Migrate the net80211 TX aggregation state to be from per-AC to per-TID. TODO: * Test mwl(4) more thoroughly! Reviewed by: bschmidt (for iwn)
2012-04-15 20:29:39 +00:00
int tid = tap->txa_tid;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
struct ath_node *an = ATH_NODE(ni);
struct ath_tid *atid = &an->an_tid[tid];
/*
* XXX danger Will Robinson!
*
* Although the taskqueue may be running and scheduling some more
* packets, these should all be _before_ the addba sequence number.
* However, net80211 will keep self-assigning sequence numbers
* until addba has been negotiated.
*
* In the past, these packets would be "paused" (which still works
* fine, as they're being scheduled to the driver in the same
* serialised method which is calling the addba request routine)
* and when the aggregation session begins, they'll be dequeued
* as aggregate packets and added to the BAW. However, now there's
* a "bf->bf_state.bfs_dobaw" flag, and this isn't set for these
* packets. Thus they never get included in the BAW tracking and
* this can cause the initial burst of packets after the addba
* negotiation to "hang", as they quickly fall outside the BAW.
*
* The "eventual" solution should be to tag these packets with
* dobaw. Although net80211 has given us a sequence number,
* it'll be "after" the left edge of the BAW and thus it'll
* fall within it.
*/
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
/*
* This is a bit annoying. Until net80211 HT code inherits some
* (any) locking, we may have this called in parallel BUT only
* one response/timeout will be called. Grr.
*/
if (atid->addba_tx_pending == 0) {
ath_tx_tid_pause(sc, atid);
atid->addba_tx_pending = 1;
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: called; dialogtoken=%d, baparamset=%d, batimeout=%d\n",
__func__, dialogtoken, baparamset, batimeout);
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: txa_start=%d, ni_txseqs=%d\n",
__func__, tap->txa_start, ni->ni_txseqs[tid]);
return sc->sc_addba_request(ni, tap, dialogtoken, baparamset,
batimeout);
}
/*
* Handle an ADDBA response.
*
* We unpause the queue so TX'ing can resume.
*
* Any packets TX'ed from this point should be "aggregate" (whether
* aggregate or not) so the BAW is updated.
*
* Note! net80211 keeps self-assigning sequence numbers until
* ampdu is negotiated. This means the initially-negotiated BAW left
* edge won't match the ni->ni_txseq.
*
* So, being very dirty, the BAW left edge is "slid" here to match
* ni->ni_txseq.
*
* What likely SHOULD happen is that all packets subsequent to the
* addba request should be tagged as aggregate and queued as non-aggregate
* frames; thus updating the BAW. For now though, I'll just slide the
* window.
*/
int
ath_addba_response(struct ieee80211_node *ni, struct ieee80211_tx_ampdu *tap,
int status, int code, int batimeout)
{
struct ath_softc *sc = ni->ni_ic->ic_ifp->if_softc;
Migrate the net80211 TX aggregation state to be from per-AC to per-TID. TODO: * Test mwl(4) more thoroughly! Reviewed by: bschmidt (for iwn)
2012-04-15 20:29:39 +00:00
int tid = tap->txa_tid;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
struct ath_node *an = ATH_NODE(ni);
struct ath_tid *atid = &an->an_tid[tid];
int r;
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: called; status=%d, code=%d, batimeout=%d\n", __func__,
status, code, batimeout);
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: txa_start=%d, ni_txseqs=%d\n",
__func__, tap->txa_start, ni->ni_txseqs[tid]);
/*
* Call this first, so the interface flags get updated
* before the TID is unpaused. Otherwise a race condition
* exists where the unpaused TID still doesn't yet have
* IEEE80211_AGGR_RUNNING set.
*/
r = sc->sc_addba_response(ni, tap, status, code, batimeout);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
atid->addba_tx_pending = 0;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/*
* XXX dirty!
* Slide the BAW left edge to wherever net80211 left it for us.
* Read above for more information.
*/
tap->txa_start = ni->ni_txseqs[tid];
ath_tx_tid_resume(sc, atid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
return r;
}
/*
* Stop ADDBA on a queue.
Make sure the BAR TX session pause is correctly unpaused when a node is reassociating. PR: kern/169432
2012-06-26 07:56:15 +00:00
*
* This can be called whilst BAR TX is currently active on the queue,
* so make sure this is unblocked before continuing.
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*/
void
ath_addba_stop(struct ieee80211_node *ni, struct ieee80211_tx_ampdu *tap)
{
struct ath_softc *sc = ni->ni_ic->ic_ifp->if_softc;
Migrate the net80211 TX aggregation state to be from per-AC to per-TID. TODO: * Test mwl(4) more thoroughly! Reviewed by: bschmidt (for iwn)
2012-04-15 20:29:39 +00:00
int tid = tap->txa_tid;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
struct ath_node *an = ATH_NODE(ni);
struct ath_tid *atid = &an->an_tid[tid];
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL, "%s: called\n", __func__);
Make sure the BAR TX session pause is correctly unpaused when a node is reassociating. PR: kern/169432
2012-06-26 07:56:15 +00:00
/*
* Pause TID traffic early, so there aren't any races
* Unblock the pending BAR held traffic, if it's currently paused.
*/
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_pause(sc, atid);
Make sure the BAR TX session pause is correctly unpaused when a node is reassociating. PR: kern/169432
2012-06-26 07:56:15 +00:00
if (atid->bar_wait) {
/*
* bar_unsuspend() expects bar_tx == 1, as it should be
* called from the TX completion path. This quietens
* the warning. It's cleared for us anyway.
*/
atid->bar_tx = 1;
ath_tx_tid_bar_unsuspend(sc, atid);
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* There's no need to hold the TXQ lock here */
sc->sc_addba_stop(ni, tap);
/*
Rename ath_tx_cleanup() -> ath_tx_tid_cleanup() in order to not clash with a symbol in if_ath.c
2012-05-21 22:39:13 +00:00
* ath_tx_tid_cleanup will resume the TID if possible, otherwise
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
* it'll set the cleanup flag, and it'll be unpaused once
* things have been cleaned up.
*/
Rename ath_tx_cleanup() -> ath_tx_tid_cleanup() in order to not clash with a symbol in if_ath.c
2012-05-21 22:39:13 +00:00
ath_tx_tid_cleanup(sc, an, tid);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
/*
* Note: net80211 bar_timeout() doesn't call this function on BAR failure;
* it simply tears down the aggregation session. Ew.
*
* It however will call ieee80211_ampdu_stop() which will call
* ic->ic_addba_stop().
*
* XXX This uses a hard-coded max BAR count value; the whole
* XXX BAR TX success or failure should be better handled!
*/
void
ath_bar_response(struct ieee80211_node *ni, struct ieee80211_tx_ampdu *tap,
int status)
{
struct ath_softc *sc = ni->ni_ic->ic_ifp->if_softc;
Migrate the net80211 TX aggregation state to be from per-AC to per-TID. TODO: * Test mwl(4) more thoroughly! Reviewed by: bschmidt (for iwn)
2012-04-15 20:29:39 +00:00
int tid = tap->txa_tid;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
struct ath_node *an = ATH_NODE(ni);
struct ath_tid *atid = &an->an_tid[tid];
int attempts = tap->txa_attempts;
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
DPRINTF(sc, ATH_DEBUG_SW_TX_BAR,
Migrate the bulk of the RX routines out from if_ath.c to if_ath_rx.[ch]. * migrate the rx processing out into if_ath_rx.c * migrate the TSF functions into if_ath_tsf.h, as inlines This is in prepration for supporting the EDMA RX routines, required to support the AR93xx series NICs. TODO: * ath_start() shouldn't be private, but it's called as part of the RX path. I should likely migrate ath_rx_tasklet() back into if_ath.c and then return this to be 'static'. The RX code really shouldn't need to see TX routines (and vice versa.) * ath_beacon_* should be in if_ath_beacon.[ch]. * ath_tdma_* should be in if_ath_tdma.[ch] ...
2012-05-20 02:05:10 +00:00
"%s: called; tap=%p, atid=%p, txa_tid=%d, atid->tid=%d, status=%d, attempts=%d\n",
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
__func__,
Migrate the bulk of the RX routines out from if_ath.c to if_ath_rx.[ch]. * migrate the rx processing out into if_ath_rx.c * migrate the TSF functions into if_ath_tsf.h, as inlines This is in prepration for supporting the EDMA RX routines, required to support the AR93xx series NICs. TODO: * ath_start() shouldn't be private, but it's called as part of the RX path. I should likely migrate ath_rx_tasklet() back into if_ath.c and then return this to be 'static'. The RX code really shouldn't need to see TX routines (and vice versa.) * ath_beacon_* should be in if_ath_beacon.[ch]. * ath_tdma_* should be in if_ath_tdma.[ch] ...
2012-05-20 02:05:10 +00:00
tap,
atid,
tap->txa_tid,
atid->tid,
Migrate ath_debug and sc_debug from an int to a uint64_t / QUAD; add some more BAR debugging logic. * Change the definition of ath_debug and ath_softc.sc_debug from int to uint64_t; * Change the relevant sysctls; * Add a new BAR TX debugging field; * Use this in if_ath_tx. This has been tested by using the sysctl program, which happily allows for fields > 32 bits to be configured.
2012-05-15 23:39:37 +00:00
status,
attempts);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Note: This may update the BAW details */
sc->sc_bar_response(ni, tap, status);
/* Unpause the TID */
/*
* XXX if this is attempt=50, the TID will be downgraded
* XXX to a non-aggregate session. So we must unpause the
* XXX TID here or it'll never be done.
Add a work-around for some strange net80211 BAR races in the wireless stack. There are unfortunately quite a few odd cases in BAR TX and BAR TX retransmission that I haven't yet fully diagnosed. So for now, add this work-around so the resume() function isn't called too often, decrementing pause to -1 (and causing things to stay paused.)
2012-09-20 03:03:01 +00:00
*
* Also, don't call it if bar_tx/bar_wait are 0; something
* has beaten us to the punch? (XXX figure out what?)
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
*/
if (status == 0 || attempts == 50) {
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Add a work-around for some strange net80211 BAR races in the wireless stack. There are unfortunately quite a few odd cases in BAR TX and BAR TX retransmission that I haven't yet fully diagnosed. So for now, add this work-around so the resume() function isn't called too often, decrementing pause to -1 (and causing things to stay paused.)
2012-09-20 03:03:01 +00:00
if (atid->bar_tx == 0 || atid->bar_wait == 0)
device_printf(sc->sc_dev,
"%s: huh? bar_tx=%d, bar_wait=%d\n",
__func__,
atid->bar_tx, atid->bar_wait);
else
ath_tx_tid_bar_unsuspend(sc, atid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
}
/*
* This is called whenever the pending ADDBA request times out.
* Unpause and reschedule the TID.
*/
void
ath_addba_response_timeout(struct ieee80211_node *ni,
struct ieee80211_tx_ampdu *tap)
{
struct ath_softc *sc = ni->ni_ic->ic_ifp->if_softc;
Migrate the net80211 TX aggregation state to be from per-AC to per-TID. TODO: * Test mwl(4) more thoroughly! Reviewed by: bschmidt (for iwn)
2012-04-15 20:29:39 +00:00
int tid = tap->txa_tid;
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
struct ath_node *an = ATH_NODE(ni);
struct ath_tid *atid = &an->an_tid[tid];
DPRINTF(sc, ATH_DEBUG_SW_TX_CTRL,
"%s: called; resuming\n", __func__);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
atid->addba_tx_pending = 0;
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Fix up some corner cases with aggregation handling. I've come across a weird scenario in net80211 where two TX streams will happily attempt to setup an aggregation session together. If we're very lucky, it happens concurrently on separate CPUs and the total lack of locking in the net80211 aggregation code causes this stuff to race. Badly. So >1 call would occur to the ath(4) addba start, but only one call would complete to addba complete or timeout. The TID would thus stay paused. The real fix is to implement some proper per-node (or maybe per-TID) locking in net80211, which then could be leveraged by the ath(4) TX aggregation code. Whilst I'm at it, shuffle around the debugging messages a bit. I like to keep people on their toes.
2012-05-22 06:31:03 +00:00
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
/* Note: This updates the aggregate state to (again) pending */
sc->sc_addba_response_timeout(ni, tap);
/* Unpause the TID; which reschedules it */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
ath_tx_tid_resume(sc, atid);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Introduce TX aggregation and software TX queue management for Atheros AR5416 and later wireless devices. This is a very large commit - the complete history can be found in the user/adrian/if_ath_tx branch. Legacy (ie, pre-AR5416) devices also use the per-software TXQ support and (in theory) can support non-aggregation ADDBA sessions. However, the net80211 stack doesn't currently support this. In summary: TX path: * queued frames normally go onto a per-TID, per-node queue * some special frames (eg ADDBA control frames) are thrown directly onto the relevant hardware queue so they can go out before any software queued frames are queued. * Add methods to create, suspend, resume and tear down an aggregation session. * Add in software retransmission of both normal and aggregate frames. * Add in completion handling of aggregate frames, including parsing the block ack bitmap provided by the hardware. * Write an aggregation function which can assemble frames into an aggregate based on the selected rate control and channel configuration. * The per-TID queues are locked based on their target hardware TX queue. This matches what ath9k/atheros does, and thus simplified porting over some of the aggregation logic. * When doing TX aggregation, stick the sequence number allocation in the TX path rather than net80211 TX path, and protect it by the TXQ lock. Rate control: * Delay rate control selection until the frame is about to be queued to the hardware, so retried frames can have their rate control choices changed. Frames with a static rate control selection have that applied before each TX, just to simplify the TX path (ie, not have "static" and "dynamic" rate control special cased.) * Teach ath_rate_sample about aggregates - both completion and errors. * Add an EWMA for tracking what the current "good" MCS rate is based on failure rates. Misc: * Introduce a bunch of dirty hacks and workarounds so TID mapping and net80211 frame inspection can be kept out of the net80211 layer. Because of the way this code works (and it's from Atheros and Linux ath9k), there is a consistent, 1:1 mapping between TID and AC. So we need to ensure that frames going to a specific TID will _always_ end up on the right AC, and vice versa, or the completion/locking will simply get very confused. I plan on addressing this mess in the future. Known issues: * There is no BAR frame transmission just yet. A whole lot of tidying up needs to occur before BAR frame TX can occur in the "correct" place - ie, once the TID TX queue has been drained. * Interface reset/purge/etc results in frames in the TX and RX queues being removed. This creates holes in the sequence numbers being assigned and the TX/RX AMPDU code (on either side) just hangs. * There's no filtered frame support at the present moment, so stations going into power saving mode will simply have a number of frames dropped - likely resulting in a traffic "hang". * Raw frame TX is going to just not function with 11n aggregation. Likely this needs to be modified to always override the sequence number if the frame is going into an aggregation session. However, general raw frame injection currently doesn't work in general in net80211, so let's just ignore this for now until this is sorted out. * HT protection is just not implemented and won't be until the above is sorted out. In addition, the AR5416 has issues RTS protecting large aggregates (anything >8k), so the work around needs to be ported and tested. Thus, this will be put on hold until the above work is complete. * The rate control module 'sample' is the only currently supported module; onoe/amrr haven't been tested and have likely bit rotted a little. I'll follow up with some commits to make them work again for non-11n rates, but they won't be updated to handle 11n and aggregation. If someone wishes to do so then they're welcome to send along patches. * .. and "sample" doesn't really do a good job of 11n TX. Specifically, the metrics used (packet TX time and failure/success rates) isn't as useful for 11n. It's likely that it should be extended to take into account the aggregate throughput possible and then choose a rate which maximises that. Ie, it may be acceptable for a higher MCS rate with a higher failure to be used if it gives a more acceptable throughput/latency then a lower MCS rate @ a lower error rate. Again, patches will be gratefully accepted. Because of this, ATH_ENABLE_11N is still not enabled by default. Sponsored by: Hobnob, Inc. Obtained from: Linux, Atheros
2011-11-08 22:43:13 +00:00
}
Begin separating out the TX DMA setup in preparation for TX EDMA support. * Introduce TX DMA setup/teardown methods, mirroring what's done in the RX path. Although the TX DMA descriptor is setup via ath_desc_alloc() / ath_desc_free(), there TX status descriptor ring will be allocated in this path. * Remove some of the TX EDMA capability probing from the RX path and push it into the new TX EDMA path.
2012-07-23 03:52:18 +00:00
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
/*
* Check if a node is asleep or not.
*/
Begin fleshing out some software queue awareness for TIM handling with the power save queue. * introduce some new ATH_NODE lock protected fields, tracking the net80211 psq and TIM state; * when doing buffer transitions - ie, when sending and completing buffers - check the state of the SWQ and update the TIM appropriately. * when clearing the TIM bit, if the SWQ is not empty then delay clearing it. This is racy, but it's no less racy than the current net80211 power save queue management code. Specifically, with multiple TX threads, it's quite plausible that parallel state updates will race and the TIM will be left in an inconsistent state. I'll address that in a follow-up commit.
2012-10-28 21:13:12 +00:00
int
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
ath_tx_node_is_asleep(struct ath_softc *sc, struct ath_node *an)
{
ATH_NODE_LOCK_ASSERT(an);
return (an->an_is_powersave);
}
/*
* Mark a node as currently "in powersaving."
* This suspends all traffic on the node.
*
* This must be called with the node/tx locks free.
*
* XXX TODO: the locking silliness below is due to how the node
* locking currently works. Right now, the node lock is grabbed
* to do rate control lookups and these are done with the TX
* queue lock held. This means the node lock can't be grabbed
* first here or a LOR will occur.
*
* Eventually (hopefully!) the TX path code will only grab
* the TXQ lock when transmitting and the ath_node lock when
* doing node/TID operations. There are other complications -
* the sched/unsched operations involve walking the per-txq
* 'active tid' list and this requires both locks to be held.
*/
void
ath_tx_node_sleep(struct ath_softc *sc, struct ath_node *an)
{
struct ath_tid *atid;
struct ath_txq *txq;
int tid;
ATH_NODE_UNLOCK_ASSERT(an);
/*
* It's possible that a parallel call to ath_tx_node_wakeup()
* will unpause these queues.
*
* The node lock can't just be grabbed here, as there's places
* in the driver where the node lock is grabbed _within_ a
* TXQ lock.
* So, we do this delicately and unwind state if needed.
*
* + Pause all the queues
* + Grab the node lock
* + If the queue is already asleep, unpause and quit
* + else just mark as asleep.
*
* A parallel sleep() call will just pause and then
* find they're already paused, so undo it.
*
* A parallel wakeup() call will check if asleep is 1
* and if it's not (ie, it's 0), it'll treat it as already
* being awake. If it's 1, it'll mark it as 0 and then
* unpause everything.
*
* (Talk about a delicate hack.)
*/
/* Suspend all traffic on the node */
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
for (tid = 0; tid < IEEE80211_TID_SIZE; tid++) {
atid = &an->an_tid[tid];
txq = sc->sc_ac2q[atid->ac];
ath_tx_tid_pause(sc, atid);
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
ATH_NODE_LOCK(an);
/* In case of concurrency races from net80211.. */
if (an->an_is_powersave == 1) {
ATH_NODE_UNLOCK(an);
device_printf(sc->sc_dev,
"%s: an=%p: node was already asleep\n",
__func__, an);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
for (tid = 0; tid < IEEE80211_TID_SIZE; tid++) {
atid = &an->an_tid[tid];
txq = sc->sc_ac2q[atid->ac];
ath_tx_tid_resume(sc, atid);
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
return;
}
/* Mark node as in powersaving */
an->an_is_powersave = 1;
ATH_NODE_UNLOCK(an);
}
/*
* Mark a node as currently "awake."
* This resumes all traffic to the node.
*/
void
ath_tx_node_wakeup(struct ath_softc *sc, struct ath_node *an)
{
struct ath_tid *atid;
struct ath_txq *txq;
int tid;
ATH_NODE_UNLOCK_ASSERT(an);
ATH_NODE_LOCK(an);
/* In case of concurrency races from net80211.. */
if (an->an_is_powersave == 0) {
ATH_NODE_UNLOCK(an);
device_printf(sc->sc_dev,
"%s: an=%p: node was already awake\n",
__func__, an);
return;
}
/* Mark node as awake */
an->an_is_powersave = 0;
ATH_NODE_UNLOCK(an);
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_LOCK(sc);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
for (tid = 0; tid < IEEE80211_TID_SIZE; tid++) {
atid = &an->an_tid[tid];
txq = sc->sc_ac2q[atid->ac];
ath_tx_tid_resume(sc, atid);
}
Delete the per-TXQ locks and replace them with a single TX lock. I couldn't think of a way to maintain the hardware TXQ locks _and_ layer on top of that per-TXQ software queuing and any other kind of fine-grained locks (eg per-TID, or per-node locks.) So for now, to facilitate some further code refactoring and development as part of the final push to get software queue ps-poll and u-apsd handling into this driver, just do away with them entirely. I may eventually bring them back at some point, when it looks slightly more architectually cleaner to do so. But as it stands at the present, it's not really buying us much: * in order to properly serialise things and not get bitten by scheduling and locking interactions with things higher up in the stack, we need to wrap the whole TX path in a long held lock. Otherwise we can end up being pre-empted during frame handling, resulting in some out of order frame handling between sequence number allocation and encryption handling (ie, the seqno and the CCMP IV get out of sequence); * .. so whilst that's the case, holding the lock for that long means that we're acquiring and releasing the TXQ lock _inside_ that context; * And we also acquire it per-frame during frame completion, but we currently can't hold the lock for the duration of the TX completion as we need to call net80211 layer things with the locks _unheld_ to avoid LOR. * .. the other places were grab that lock are reset/flush, which don't happen often. My eventual aim is to change the TX path so all rejected frame transmissions and all frame completions result in any ieee80211_free_node() calls to occur outside of the TX lock; then I can cut back on the amount of locking that goes on here. There may be some LORs that occur when ieee80211_free_node() is called when the TX queue path fails; I'll begin to address these in follow-up commits.
2012-12-02 06:24:08 +00:00
ATH_TX_UNLOCK(sc);
Pause and unpause the software queues for a given node based on the net80211 node power save state. * Add an ATH_NODE_UNLOCK_ASSERT() check * Add a new node field - an_is_powersave * Pause/unpause the queue based on the node state * Attempt to handle net80211 concurrency issues so the queue doesn't get paused/unpaused more than once at a time from the net80211 power save code. Whilst here (and breaking my usual rule), set CLRDMASK when a queue is unpaused, regardless of whether the queue has some pending traffic. This means the first frame from that TID (now or later) will hvae CLRDMASK set. Also whilst here, bump the swretrymax counters whenever the filtered frames code expires a frame. Again, breaking my rule, but this is just a statistics thing rather than a functional change. This doesn't fix ps-poll (but it doesn't break it too much worse than it is at the present) or correcting the TID updates. That's next on the list. Tested: * AR9220 AP (Atheros AP96 reference design) * Macbook Pro and LG Optimus 1 Android phone, both setting and clearing power save state (but not using PS-POLL.)
2012-10-03 23:23:45 +00:00
}
Begin separating out the TX DMA setup in preparation for TX EDMA support. * Introduce TX DMA setup/teardown methods, mirroring what's done in the RX path. Although the TX DMA descriptor is setup via ath_desc_alloc() / ath_desc_free(), there TX status descriptor ring will be allocated in this path. * Remove some of the TX EDMA capability probing from the RX path and push it into the new TX EDMA path.
2012-07-23 03:52:18 +00:00
static int
ath_legacy_dma_txsetup(struct ath_softc *sc)
{
/* nothing new needed */
return (0);
}
static int
ath_legacy_dma_txteardown(struct ath_softc *sc)
{
/* nothing new needed */
return (0);
}
void
ath_xmit_setup_legacy(struct ath_softc *sc)
{
Modify ath_descdma_setup() to take a descriptor size parameter. The AR9300 and later descriptors are 128 bytes, however I'd like to make sure that isn't used for earlier chips. * Populate the TX descriptor length field in the softc with sizeof(ath_desc) * Use this field when allocating the TX descriptors * Pre-AR93xx TX/RX descriptors will use the ath_desc size; newer ones will query the HAL for these sizes.
2012-07-23 23:40:13 +00:00
/*
* For now, just set the descriptor length to sizeof(ath_desc);
* worry about extracting the real length out of the HAL later.
*/
sc->sc_tx_desclen = sizeof(struct ath_desc);
ALQ logging enhancements: * upon setup, tell the alq code what the chip information is. * add TX/RX path logging for legacy chips. * populate the tx/rx descriptor length fields with a best-estimate. It's overly big (96 bytes when AH_SUPPORT_AR5416 is enabled) but it'll do for now. Whilst I'm here, add CURVNET_RESTORE() here during probe/attach as a partial solution to fixing crashes during attach when the attach fails. There are other attach failures that I have to deal with; those'll come later.
2012-11-16 19:57:16 +00:00
sc->sc_tx_statuslen = sizeof(struct ath_desc);
Modify ath_descdma_setup() to take a descriptor size parameter. The AR9300 and later descriptors are 128 bytes, however I'd like to make sure that isn't used for earlier chips. * Populate the TX descriptor length field in the softc with sizeof(ath_desc) * Use this field when allocating the TX descriptors * Pre-AR93xx TX/RX descriptors will use the ath_desc size; newer ones will query the HAL for these sizes.
2012-07-23 23:40:13 +00:00
sc->sc_tx_nmaps = 1; /* only one buffer per TX desc */
Begin separating out the TX DMA setup in preparation for TX EDMA support. * Introduce TX DMA setup/teardown methods, mirroring what's done in the RX path. Although the TX DMA descriptor is setup via ath_desc_alloc() / ath_desc_free(), there TX status descriptor ring will be allocated in this path. * Remove some of the TX EDMA capability probing from the RX path and push it into the new TX EDMA path.
2012-07-23 03:52:18 +00:00
sc->sc_tx.xmit_setup = ath_legacy_dma_txsetup;
sc->sc_tx.xmit_teardown = ath_legacy_dma_txteardown;
Migrate some more TX side setup routines to be methods.
2012-07-31 03:09:48 +00:00
sc->sc_tx.xmit_attach_comp_func = ath_legacy_attach_comp_func;
Break out the hardware handoff and TX DMA restart code into methods. These (and a few others) will differ based on the underlying DMA implementation. For the EDMA NICs, simply stub them out in a fashion which will let me focus on implementing the necessary descriptor API changes.
2012-07-31 02:28:32 +00:00
sc->sc_tx.xmit_dma_restart = ath_legacy_tx_dma_restart;
sc->sc_tx.xmit_handoff = ath_legacy_xmit_handoff;
Break out ath_draintxq() into a method and un-methodize ath_tx_processq(). Now that I understand what's going on with this, I've realised that it's going to be quite difficult to implement a processq method in the EDMA case. Because there's a separate TX status FIFO, I can't just run processq() on each EDMA TXQ to see what's finished. i have to actually run the TX status queue and handle individual TXQs. So: * unmethodize ath_tx_processq(); * leave ath_tx_draintxq() as a method, as it only uses the completion status for debugging rather than actively completing the frames (ie, all frames here are failed); * Methodize ath_draintxq(). The EDMA ath_draintxq() will have to take care of running the TX completion FIFO before (potentially) freeing frames in the queue. The only two places where ath_tx_draintxq() (on a single TXQ) are used: * ath_draintxq(); and * the CABQ handling in the beacon setup code - it drains the CABQ before populating the CABQ with frames for a new beacon (when doing multi-VAP operation.) So it's quite possible that once I methodize the CABQ and beacon handling, I can just drop ath_tx_draintxq() in its entirety. Finally, it's also quite possible that I can remove ath_tx_draintxq() in the future and just "teach" it to not check the status when doing EDMA.
2012-08-12 00:37:29 +00:00
sc->sc_tx.xmit_drain = ath_legacy_tx_drain;
Begin separating out the TX DMA setup in preparation for TX EDMA support. * Introduce TX DMA setup/teardown methods, mirroring what's done in the RX path. Although the TX DMA descriptor is setup via ath_desc_alloc() / ath_desc_free(), there TX status descriptor ring will be allocated in this path. * Remove some of the TX EDMA capability probing from the RX path and push it into the new TX EDMA path.
2012-07-23 03:52:18 +00:00
}
Reference in New Issue Copy Permalink