related to "shared" CPLs.
a) Combine t4_set_tcb_field and t4_set_tcb_field_rpl into a single
function. Allow callers to direct the response to any iq. Tidy up
set_ulp_mode_iscsi while there to use names from t4_tcb.h instead of
magic constants.
b) Remove all CPL handler tables from struct adapter. This reduces its
size by around 2KB. All handlers are now registered at MOD_LOAD instead
of attach or some kind of initialization/activation. The registration
functions do not need an adapter parameter any more.
c) Add per-iq handlers to deal with CPLs whose destination cannot be
determined solely from the opcode. There are 2 such CPLs in use right
now: SET_TCB_RPL and L2T_WRITE_RPL. The base driver continues to send
filter and L2T_WRITEs over the mgmtq and solicits the reply on fwq.
t4_tom (including the DDP code) now uses the port's ctrlq to send
L2T_WRITEs and SET_TCB_FIELDs and solicits the reply on an ofld_rxq.
fwq and ofld_rxq have different handlers that know what kind of tid to
expect in the reply. Update t4_write_l2e and callers to to support any
wrq/iq combination.
Approved by: re@ (kib@)
Sponsored by: Chelsio Communications
vcxgbe/vcxl interfaces and retire the 'n' interfaces. The main
cxgbe/cxl interfaces and tunables related to them are not affected by
any of this and will continue to operate as usual.
The driver used to create an additional 'n' interface for every
cxgbe/cxl interface if "device netmap" was in the kernel. The 'n'
interface shared the wire with the main interface but was otherwise
autonomous (with its own MAC address, etc.). It did not have normal
tx/rx but had a specialized netmap-only data path. r291665 added
another set of virtual interfaces (the 'v' interfaces) to the driver.
These had normal tx/rx but no netmap support.
This revision consolidates the features of both the interfaces into the
'v' interface which now has a normal data path, TOE support, and native
netmap support. The 'v' interfaces need to be created explicitly with
the hw.cxgbe.num_vis tunable. This means "device netmap" will not
result in the automatic creation of any virtual interfaces.
The following tunables can be used to override the default number of
queues allocated for each 'v' interface. nofld* = 0 will disable TOE on
the virtual interface and nnm* = 0 to will disable native netmap
support.
# number of normal NIC queues
hw.cxgbe.ntxq_vi
hw.cxgbe.nrxq_vi
# number of TOE queues
hw.cxgbe.nofldtxq_vi
hw.cxgbe.nofldrxq_vi
# number of netmap queues
hw.cxgbe.nnmtxq_vi
hw.cxgbe.nnmrxq_vi
hw.cxgbe.nnm{t,r}xq{10,1}g tunables have been removed.
--- tl;dr version ---
The workflow for netmap on cxgbe starting with FreeBSD 11 is:
1) "device netmap" in the kernel config.
2) "hw.cxgbe.num_vis=2" in loader.conf. num_vis > 2 is ok too, you'll
end up with multiple autonomous netmap-capable interfaces for every
port.
3) "dmesg | grep vcxl | grep netmap" to verify that the interface has
netmap queues.
4) Use any of the 'v' interfaces for netmap. pkt-gen -i vcxl<n>... .
One major improvement is that the netmap interface has a normal data
path as expected.
5) Just ignore the cxl interfaces if you want to use netmap only. No
need to bring them up. The vcxl interfaces are completely independent
and everything should just work.
---------------------
Approved by: re@ (gjb@)
Relnotes: Yes
Sponsored by: Chelsio Communications
And factor out tcp_lro_rx_done, which deduplicates the same logic with
netinet/tcp_lro.c
Reviewed by: gallatin (1st version), hps, zbb, np, Dexuan Cui <decui microsoft com>
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D5725
Move the code that reads all the parameters to t4_init_sge_params in the
shared code. Use these per-adapter values instead of globals.
Sponsored by: Chelsio Communications
update to the latest internal shared code.
- Add a chip_params structure to keep track of hardware constants for
all generations of Terminators handled by cxgbe.
- Update t4_hw_pci_read_cfg4 to work with T6.
- Update the hardware debug sysctls (hidden within dev.<tNnex>.<n>.misc.*) to
work with T6. Most of the changes are in the decoders for the CIM
logic analyzer and the MPS TCAM.
- Acquire the regwin lock around indirect register accesses.
Obtained from: Chelsio Communications
Sponsored by: Chelsio Communications
- Add optimizing LRO wrapper which pre-sorts all incoming packets
according to the hash type and flowid. This prevents exhaustion of
the LRO entries due to too many connections at the same time.
Testing using a larger number of higher bandwidth TCP connections
showed that the incoming ACK packet aggregation rate increased from
~1.3:1 to almost 3:1. Another test showed that for a number of TCP
connections greater than 16 per hardware receive ring, where 8 TCP
connections was the LRO active entry limit, there was a significant
improvement in throughput due to being able to fully aggregate more
than 8 TCP stream. For very few very high bandwidth TCP streams, the
optimizing LRO wrapper will add CPU usage instead of reducing CPU
usage. This is expected. Network drivers which want to use the
optimizing LRO wrapper needs to call "tcp_lro_queue_mbuf()" instead
of "tcp_lro_rx()" and "tcp_lro_flush_all()" instead of
"tcp_lro_flush()". Further the LRO control structure must be
initialized using "tcp_lro_init_args()" passing a non-zero number
into the "lro_mbufs" argument.
- Make LRO statistics 64-bit. Previously 32-bit integers were used for
statistics which can be prone to wrap-around. Fix this while at it
and update all SYSCTL's which expose LRO statistics.
- Ensure all data is freed when destroying a LRO control structures,
especially leftover LRO entries.
- Reduce number of memory allocations needed when setting up a LRO
control structure by precomputing the total amount of memory needed.
- Add own memory allocation counter for LRO.
- Bump the FreeBSD version to force recompilation of all KLDs due to
change of the LRO control structure size.
Sponsored by: Mellanox Technologies
Reviewed by: gallatin, sbruno, rrs, gnn, transport
Tested by: Netflix
Differential Revision: https://reviews.freebsd.org/D4914
Each virtual interface has its own MAC address, queues, and statistics.
The dedicated netmap interfaces (ncxgbeX / ncxlX) were already implemented
as additional VIs on each port. This change allows additional non-netmap
interfaces to be configured on each port. Additional virtual interfaces
use the naming scheme vcxgbeX or vcxlX.
Additional VIs are enabled by setting the hw.cxgbe.num_vis tunable to a
value greater than 1 before loading the cxgbe(4) or cxl(4) driver.
NB: The first VI on each port is the "main" interface (cxgbeX or cxlX).
T4/T5 NICs provide a limited number of MAC addresses for each physical port.
As a result, a maximum of six VIs can be configured on each port (including
the "main" interface and the netmap interface when netmap is enabled).
One user-visible result is that when netmap is enabled, packets received
or transmitted via the netmap interface are no longer counted in the stats
for the "main" interface, but are not accounted to the netmap interface.
The netmap interfaces now also have a new-bus device and export various
information sysctl nodes via dev.n(cxgbe|cxl).X.
The cxgbetool 'clearstats' command clears the stats for all VIs on the
specified port along with the port's stats. There is currently no way to
clear the stats of an individual VI.
Reviewed by: np
MFC after: 1 month
Sponsored by: Chelsio
a) Front load as much work as possible in if_transmit, before any driver
lock or software queue has to get involved.
b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This
is specifically for the tx multiqueue model where one of the if_transmit
producer threads becomes the consumer and other producers carry on as
usual. mp_ring is implemented as standalone code and it should be
possible to use it in any driver with tx multiqueue. It also has:
- the ability to enqueue/dequeue multiple items. This might become
significant if packet batching is ever implemented.
- an abdication mechanism to allow a thread to give up writing tx
descriptors and have another if_transmit thread take over. A thread
that's writing tx descriptors can end up doing so for an unbounded
time period if a) there are other if_transmit threads continuously
feeding the sofware queue, and b) the chip keeps up with whatever the
thread is throwing at it.
- accurate statistics about interesting events even when the stats come
at the expense of additional branches/conditional code.
The NIC txq lock is uncontested on the fast path at this point. I've
left it there for synchronization with the control events (interface
up/down, modload/unload).
c) Add support for "type 1" coalescing work request in the normal NIC tx
path. This work request is optimized for frames with a single item in
the DMA gather list. These are very common when forwarding packets.
Note that netmap tx in cxgbe already uses these "type 1" work requests.
d) Do not request automatic cidx updates every 32 descriptors. Instead,
request updates via bits in individual work requests (still every 32
descriptors approximately). Also, request an automatic final update
when the queue idles after activity. This means NIC tx reclaim is still
performed lazily but it will catch up quickly as soon as the queue
idles. This seems to be the best middle ground and I'll probably do
something similar for netmap tx as well.
e) Implement a faster tx path for WRQs (used by TOE tx and control
queues, _not_ by the normal NIC tx). Allow work requests to be written
directly to the hardware descriptor ring if room is available. I will
convert t4_tom and iw_cxgbe modules to this faster style gradually.
MFC after: 2 months
adapters. Set the pack boundary for T5 cards to be the same as the
PCIe max payload size. The chip likes it this way.
In this revision the driver allocate rx buffers that align on both
boundaries. This is not a strict requirement and a followup commit
will switch the driver to a more relaxed allocation strategy.
MFC after: 2 weeks
from the FreeBSD network code. The flag is still kept around in the
"sys/mbuf.h" header file, but does no longer have any users. Instead
the "m_pkthdr.rsstype" field in the mbuf structure is now used to
decide the meaning of the "m_pkthdr.flowid" field. To modify the
"m_pkthdr.rsstype" field please use the existing "M_HASHTYPE_XXX"
macros as defined in the "sys/mbuf.h" header file.
This patch introduces new behaviour in the transmit direction.
Previously network drivers checked if "M_FLOWID" was set in "m_flags"
before using the "m_pkthdr.flowid" field. This check has now now been
replaced by checking if "M_HASHTYPE_GET(m)" is different from
"M_HASHTYPE_NONE". In the future more hashtypes will be added, for
example hashtypes for hardware dedicated flows.
"M_HASHTYPE_OPAQUE" indicates that the "m_pkthdr.flowid" value is
valid and has no particular type. This change removes the need for an
"if" statement in TCP transmit code checking for the presence of a
valid flowid value. The "if" statement mentioned above is now a direct
variable assignment which is then later checked by the respective
network drivers like before.
Additional notes:
- The SCTP code changes will be committed as a separate patch.
- Removal of the "M_FLOWID" flag will also be done separately.
- The FreeBSD version has been bumped.
MFC after: 1 month
Sponsored by: Mellanox Technologies
Reorganize struct sge_iq. Make the iq entry size a compile time
constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly.
MFC after: 2 weeks
custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with
EBUSY if any such cluster has been handed up to the kernel but hasn't
been freed yet. This prevents a panic later when the cluster finally
needs to be freed but rxb_free is gone from the kernel.
MFC after: 1 week
get_scatter_segment() in get_fl_payload() fails. While here,
fix the code to adjust fl_bufs_used when a failure occurs for
any other scatter segment.
MFC after: 3 days
Netmap gets its own hardware-assisted virtual interface and won't take
over or disrupt the "normal" interface in any way. You can use both
simultaneously.
For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface
(note the 'n' prefix) in the hardware to accompany each cxl<N>
interface. These two ifnet's per port share the same wire but really
are separate interfaces in the hardware and software. Each gets its own
L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You
should run netmap on the 'n' interfaces only, that's what they are for.
With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port
of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now.
Single port receive is at 33Mpps but this is very much a work in
progress. I expect it to be closer to 40Mpps once done. In any case
the current effort can already saturate multiple 10G ports of a T5 card
at the smallest legal packet size. T4 gear is totally untested.
trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43🆎cd:ef
881.952141 main [1621] interface is ncxl0
881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0
881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0
881.962540 main [1804] mapped 334980KB at 0x801dff000
Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus.
10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43🆎cd:ef)
881.962562 main [1882] Sending 512 packets every 0.000000000 s
881.962563 main [1884] Wait 2 secs for phy reset
884.088516 main [1886] Ready...
884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1
884.088607 sender_body [996] start
884.093246 sender_body [1064] drop copy
885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec)
886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec)
887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec)
888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec)
889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec)
890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec)
891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec)
892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec)
893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec)
894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec)
895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec)
896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec)
...
Relnotes: Yes
Sponsored by: Chelsio Communications.
- More flexible cluster size selection, including the ability to fall
back to a safe cluster size (PAGE_SIZE from zone_jumbop by default) in
case an allocation of a larger size fails.
- A single get_fl_payload() function that assembles the payload into an
mbuf chain for any kind of freelist. This replaces two variants: one
for freelists with buffer packing enabled and another for those without.
- Buffer packing with any sized cluster. It was limited to 4K clusters
only before this change.
- Enable buffer packing for TOE rx queues as well.
- Statistics and tunables to go with all these changes. The driver's
man page will be updated separately.
MFC after: 5 weeks
them up as part of firmware initialization (which the driver gets to do
only if it's the master driver).
Read the range of tids available for the ETHOFLD functionality if it's
enabled.
New is_ftid() and is_etid() functions to test whether a tid falls within
the range of filter tids or ETHOFLD tids respectively.
MFC after: 2 weeks
cards.
This is a T4 and T5 chip feature which lets the chip deliver multiple
Ethernet frames in a single buffer. This is more efficient within the
chip, in the driver, and reduces wastage of space in rx buffers.
- Always allocate rx buffers from the jumbop zone, no matter what the
MTU is. Do not use the normal cluster refcounting mechanism.
- Reserve space for an mbuf and a refcount in the cluster itself and let
the chip DMA multiple frames in the rest.
- Use the embedded mbuf for the first frame and allocate mbufs on the
fly for any additional frames delivered in the cluster. Each of these
mbufs has a reference on the underlying cluster.
all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and
for general purpose monitoring without tapping any cxgbe or cxl ifnet
directly.
Tracers on the T4/T5 chips provide access to Ethernet frames exactly as
they were received from or transmitted on the wire. On transmit, a
tracer will capture a frame after TSO segmentation, hw VLAN tag
insertion, hw L3 & L4 checksum insertion, etc. It will also capture
frames generated by the TCP offload engine (TOE traffic is normally
invisible to the kernel). On receive, a tracer will capture a frame
before hw VLAN extraction, runt filtering, other badness filtering,
before the steering/drop/L2-rewrite filters or the TOE have had a go at
it, and of course before sw LRO in the driver.
There are 4 tracers on a chip. A tracer can trace only in one direction
(tx or rx). For now cxgbetool will set up tracers to capture the first
128B of every transmitted or received frame on a given port. This is a
small subset of what the hardware can do. A pseudo ifnet with the same
name as the nexus driver (t4nex0 or t5nex0) will be created for tracing.
The data delivered to this ifnet is an additional copy made inside the
chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual.
/* watch cxl0, which is the first port hanging off t5nex0. */
# cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting)
# cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving)
# cxgbetool t5nex0 tracer list
# tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire
If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K
"frames" with no L3/L4 checksum but this will show you the frames that
were actually transmitted.
/* all done */
# cxgbetool t5nex0 tracer 0 disable
# cxgbetool t5nex0 tracer 1 disable
# cxgbetool t5nex0 tracer list
# ifconfig t5nex0 destroy
when the interface is up.
- Add a tunable to control the TOE's rx coalesce feature (enabled by
default as it always has been). Consider the interface MTU or the
coalesce size when deciding which cluster zone to use to fill the
offload rx queue's free list. The tunable is:
dev.{t4nex,t5nex}.<N>.toe.rx_coalesce
MFC after: 1 day
includes support for the NIC and TOE features of the 40G, 10G, and
1G/100M cards based on the T5.
The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4)
has been updated instead of writing a brand new driver. T5 cards will
show up as cxl (short for cxlgb) ports attached to the t5nex bus driver.
Sponsored by: Chelsio
CSUM_TCP too. They are all set explicitly by the kernel usually.
While here, fix an unrelated bug where hardware L4 checksum calculation
was accidentally disabled for some IPv6 packets.
Reported by: alfred@
MFC after: 3 days
- Setup multiple DDP page sizes. When the driver attempts DDP it will
try to combine physically contiguous pages into regions of these sizes.
- Set the indicate size such that the payload carried in the indicate can
be copied in the header mbuf (and the 16K rx buffer can be recycled).
- Set DDP threshold to the max payload that the chip will coalesce and
deliver to the driver (this is ~16K by default, which is also why the
offload rx queue is backed by 16K buffers). If the chip is able to
coalesce up to the max it's allowed to, it's a good sign that the peer
is transmitting in bulk without any TCP PSH.
MFC after: 2 weeks
interface's MTU. Initialize such freelists with correct values.
This wasn't a problem for common MTUs (1500 and 9000) as the buffers (2048
and 9216 in size) happened to have enough spare room. I ran into it when
playing around with unusual MTUs.
MFC after: 2 weeks
values).
- cong_drop specifies what to do on congestion: nothing, backpressure,
or drop.
- fl_pktshift specifies the padding before Ethernet payload.
- fl_pad specifies the boundary upto which to pad Ethernet payload.
- spg_len controls the length of the status page.
MFC after: 2 weeks
- Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs.
These are available as t3_tom and t4_tom modules that augment cxgb(4)
and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as
usual with or without these extra features.
- iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the
works and will follow soon.
Build-tested with make universe.
30s overview
============
What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the
capabilities of an interface:
# ifconfig -m | grep TOE
Enable/disable TCP offload on an interface (just like any other ifnet
capability):
# ifconfig cxgbe0 toe
# ifconfig cxgbe0 -toe
Which connections are offloaded? Look for toe4 and/or toe6 in the
output of netstat and sockstat:
# netstat -np tcp | grep toe
# sockstat -46c | grep toe
Reviewed by: bz, gnn
Sponsored by: Chelsio communications.
MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
- Device configuration via plain text config file. Also able to operate
when not attached to the chip as the master driver.
- Generic "work request" queue that serves as the base for both ctrl and
ofld tx queues.
- Generic interrupt handler routine that can process any event on any
kind of ingress queue (via a dispatch table).
- A couple of new driver ioctls. cxgbetool can now install a firmware
to the card ("loadfw" command) and can read the card's memory
("memdump" and "tcb" commands).
- Lots of assorted information within dev.t4nex.X.misc.* This is
primarily for debugging and won't show up in sysctl -a.
- Code to manage the L2 tables on the chip.
- Updates to cxgbe(4) man page to go with the tunables that have changed.
- Updates to the shared code in common/
- Updates to the driver-firmware interface (now at fw 1.4.16.0)
MFC after: 1 month
queues. Try to have a set of these per port when possible, fall back
to sharing a common pool between all ports otherwise.
- One control queue per port (used to be one per hardware channel).
- t4_eth_rx now handles Ethernet rx only.
- sysctls to display pidx/cidx for some queues.
MFC after: 1 week
Reference code that shows how to get a packet's timestamp out of
cxgbe(4). Disabled by default because we don't have a standard way
today to pass this information up the stack.
The timestamp is 60 bits wide and each increment represents 1 tick of
the T4's core clock. As an example, the timestamp granularity is ~4.4ns
for this card:
# sysctl dev.t4nex.0.core_clock
dev.t4nex.0.core_clock: 228125
MFC after: 1 week
- Enable 5-tuple and every-packet lookup.
- Setup the default filter mode to allow filtering/steering based on IP
protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match
type; all combined together. You can also filter based on MAC index,
Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll
have to modify the default filter mode and exclude some of the
match-fields in it.
IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter
rules.
- Add driver ioctls to get/set the global filter mode.
- Add driver ioctls to program and delete hardware filters. A couple of
the "switch" actions that rewrite Ethernet and VLAN information and
switch the packet out of another port may not work as the L2 code is not
yet in place. Everything else, including all "drop" and "pass" rules
with RSS or absolute qid, should work.
Obtained from: Chelsio Communications
that could have allowed the hardware pidx to reach the cidx even though
the freelist isn't empty. (Haven't actually seen this but it was there
waiting to happen..)
MFC after: 1 week
now a suitable base for all kinds of egress queues.
- Add control queues (sge_ctrlq) and allocate one of these per hardware
channel. They can be used to program filters and steer traffic (and
more).
MFC after: 1 week
down. The ingress queue lock was unused and has been removed as part of
these changes.
- An in-flight egress update from the SGE must be handled before the
queue that requested it is destroyed. Wait for the update to arrive.
- Interrupt handlers must stop processing rx events for a queue before
the queue is destroyed. Events that have not yet been processed
should be ignored once the queue disappears.
MFC after: 1 week
queue has its own interrupt. If the exact number that we need is not a
power of 2 and we're using MSI, then switch to interrupt multiplexing.
While here, replace the magic numbers with something more readable.
MFC after: 3 days
