- remove mbuf iovec - useful, but adds too much complexity when isolated to
the driver
- remove driver private caching - insufficient benefit over UMA to justify
the added complexity and maintenance overhead
- remove separate logic for managing multiple transmit queues, with the
new drbr routines the control flow can be made to much more closely resemble
legacy drivers
- remove dedicated service threads, with per-cpu callouts one can get the same
benefit much more simply by registering a callout 1 tick in the future if there
are still buffered packets
- remove embedded mbuf usage - Jeffr's changes will (I hope) soon be integrated
greatly reducing the overhead of using kernel APIs for reference counting
clusters
- add hysteresis to descriptor coalescing logic
- add coalesce threshold sysctls to allow users to decide at run-time
between optimizing for forwarding / UDP or optimizing for TCP
- add once per second watchdog to effectively close the very rare races
occurring from coalescing
- incorporate Navdeep's changes to the initialization path required to
convert port and adapter locks back to ordinary mutexes (silencing BPF
LOR complaints)
- enable prefetches in get_packet and tx cleaning
Reviewed by: navdeep@
MFC after: 2 weeks
by if_free (w/o doing if_attach); move ifq_attach to if_alloc and
rename ifq_attach/detach to ifq_init/ifq_delete to better identify
their purpose
Reviewed by: jhb, kmacy
the ROUTETABLES kernel option thus there is no need to include opt_route.h
anymore in all consumers of vnet.h and no longer depend on it for module
builds.
Remove the hidden include in flowtable.h as well and leave the two
explicit #includes in ip_input.c and ip_output.c.
- Each socket upcall is now invoked with the appropriate socket buffer
locked. It is not permissible to call soisconnected() with this lock
held; however, so socket upcalls now return an integer value. The two
possible values are SU_OK and SU_ISCONNECTED. If an upcall returns
SU_ISCONNECTED, then the soisconnected() will be invoked on the
socket after the socket buffer lock is dropped.
- A new API is provided for setting and clearing socket upcalls. The
API consists of soupcall_set() and soupcall_clear().
- To simplify locking, each socket buffer now has a separate upcall.
- When a socket upcall returns SU_ISCONNECTED, the upcall is cleared from
the receive socket buffer automatically. Note that a SO_SND upcall
should never return SU_ISCONNECTED.
- All this means that accept filters should now return SU_ISCONNECTED
instead of calling soisconnected() directly. They also no longer need
to explicitly clear the upcall on the new socket.
- The HTTP accept filter still uses soupcall_set() to manage its internal
state machine, but other accept filters no longer have any explicit
knowlege of socket upcall internals aside from their return value.
- The various RPC client upcalls currently drop the socket buffer lock
while invoking soreceive() as a temporary band-aid. The plan for
the future is to add a new flag to allow soreceive() to be called with
the socket buffer locked.
- The AIO callback for socket I/O is now also invoked with the socket
buffer locked. Previously sowakeup() would drop the socket buffer
lock only to call aio_swake() which immediately re-acquired the socket
buffer lock for the duration of the function call.
Discussed with: rwatson, rmacklem
Calculate the exact number of vectors we'll use before calling
pci_alloc_msix. Don't grab nine all the time.
Call cxgb_setup_interrupts once per T3, not once per port. Ditto
for cxgb_teardown_interrupts.
Don't leak resources when interrupt setup fails in the middle.
Obtained from: Navdeep Parhar
MFC after: 10 days
only if prepping the adapter failed.
Slight adjustment to comments.
Fix a bug whereby downing the interface didn't preven it from
processing packets.
Submitted by: Navdeep Parhar
MFC after: 1 week
1) Add a sysctl that will say what type of PHYs exist on the card.
2) Fix a bug that occurs when an AEL 2005 PHY resets without a transciever
in the card.
3) Unify the PHY link detection code.
Obtained from: Navdeep Parhar
MFC after: 10 days
and down more cleanly. This addresses a problem where if we have the
link flap during boot the driver would lock up the system.
Reviewed by: jhb
MFC after: 1 week
previously always pointing to the default vnet context, to a
dynamically changing thread-local one. The currvnet context
should be set on entry to networking code via CURVNET_SET() macros,
and reverted to previous state via CURVNET_RESTORE(). Recursions
on curvnet are permitted, though strongly discuouraged.
This change should have no functional impact on nooptions VIMAGE
kernel builds, where CURVNET_* macros expand to whitespace.
The curthread->td_vnet (aka curvnet) variable's purpose is to be an
indicator of the vnet context in which the current network-related
operation takes place, in case we cannot deduce the current vnet
context from any other source, such as by looking at mbuf's
m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so
far curvnet has turned out to be an invaluable consistency checking
aid: it helps to catch cases when sockets, ifnets or any other
vnet-aware structures may have leaked from one vnet to another.
The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros
was a result of an empirical iterative process, whith an aim to
reduce recursions on CURVNET_SET() to a minimum, while still reducing
the scope of CURVNET_SET() to networking only operations - the
alternative would be calling CURVNET_SET() on each system call entry.
In general, curvnet has to be set in three typicall cases: when
processing socket-related requests from userspace or from within the
kernel; when processing inbound traffic flowing from device drivers
to upper layers of the networking stack, and when executing
timer-driven networking functions.
This change also introduces a DDB subcommand to show the list of all
vnet instances.
Approved by: julian (mentor)
TCPSTAT_INC(), rather than directly manipulating the fields across the
kernel. This will make it easier to change the implementation of
these statistics, such as using per-CPU versions of the data structures.
MFC after: 3 days
The tick routine was not being restarted in the init_locked routine
which could resulted in loss of carrier when updating the MTU.
Submitted by: Navdeep Parhar at Chelsio
MFC after: 3 weeks
Firmware upgraded to 7.1.0 (from 5.0.0).
T3C EEPROM and SRAM added; Code to update eeprom/sram fixed.
fl_empty and rx_fifo_ovfl counters can be observed via sysctl.
Two new cxgbtool commands to get uP logic analyzer info and uP IOQs
Synced up with Chelsio's "common code" (as of 03/03/09)
Submitted by: Navdeep Parhar at Chelsio
Reviewed by: gnn
MFC after: 2 weeks
net/route.h.
Remove the hidden include of opt_route.h and net/route.h from net/vnet.h.
We need to make sure that both opt_route.h and net/route.h are included
before net/vnet.h because of the way MRT figures out the number of FIBs
from the kernel option. If we do not, we end up with the default number
of 1 when including net/vnet.h and array sizes are wrong.
This does not change the list of files which depend on opt_route.h
but we can identify them now more easily.
had been the only flag with random usage patterns.
Switch inc_flags to be used as a real bit field by using
INC_ISIPV6 with bitops to check for the 'isipv6' condition.
While here fix a place or two where in case of v4 inc_flags
were not properly initialized before.[1]
Found by: rwatson during review [1]
Discussed with: rwatson
Reviewed by: rwatson
MFC after: 4 weeks
1. separating L2 tables (ARP, NDP) from the L3 routing tables
2. removing as much locking dependencies among these layers as
possible to allow for some parallelism in the search operations
3. simplify the logic in the routing code,
The most notable end result is the obsolescent of the route
cloning (RTF_CLONING) concept, which translated into code reduction
in both IPv4 ARP and IPv6 NDP related modules, and size reduction in
struct rtentry{}. The change in design obsoletes the semantics of
RTF_CLONING, RTF_WASCLONE and RTF_LLINFO routing flags. The userland
applications such as "arp" and "ndp" have been modified to reflect
those changes. The output from "netstat -r" shows only the routing
entries.
Quite a few developers have contributed to this project in the
past: Glebius Smirnoff, Luigi Rizzo, Alessandro Cerri, and
Andre Oppermann. And most recently:
- Kip Macy revised the locking code completely, thus completing
the last piece of the puzzle, Kip has also been conducting
active functional testing
- Sam Leffler has helped me improving/refactoring the code, and
provided valuable reviews
- Julian Elischer setup the perforce tree for me and has helped
me maintaining that branch before the svn conversion
packet loss, of between 10-30%. The fix is to put the PHY into
and take it out of local loopback mode when resetting the interface.
Obtained from: Chelsio Inc.
MFC after: 3 days
directly include only the header files needed. This reduces the
unneeded spamming of various headers into lots of files.
For now, this leaves us with very few modules including vnet.h
and thus needing to depend on opt_route.h.
Reviewed by: brooks, gnn, des, zec, imp
Sponsored by: The FreeBSD Foundation
when it sees only received packets. In some cases where a device only
recieves data it mistakenly thinks that its transmitting side is broken
and resets the device.
Obtained from: Chelsio Inc.
MFC after: 3 days
- break complex conditionals in to multiple lines to avoid wrapping
- remove copious unused debug statements
- be more aggressive about cleaning in the calling thread
- eliminate usage of ENOSPC
- increase number of iterations that cxgbsp can do
- eliminate "initerr" usage to simplify ENOBUFS handling
- when coalescing pass all packets to BPF
- always set overrun if hardware queue is full
- invert sense of hw.cxgb.singleq tunable to hw.cxgb.multiq
- don't wake up transmitting thread by default
- add per tx queue ifaltq to handle ALTQ
- remove several unused functions in cxgb_multiq.c
- add several sysctls: multiq_tx_enable, coalesce_tx_enable,
and wakeup_tx_thread
- this obsoletes the hw.cxgb.snd_queue_len as ifq is replaced
by a buf_ring
and ifnet functions
- add memory barriers to <machine/atomic.h>
- update drivers to only conditionally define their own
- add lockless producer / consumer ring buffer
- remove ring buffer implementation from cxgb and update its callers
- add if_transmit(struct ifnet *ifp, struct mbuf *m) to ifnet to
allow drivers to efficiently manage multiple hardware queues
(i.e. not serialize all packets through one ifq)
- expose if_qflush to allow drivers to flush any driver managed queues
This work was supported by Bitgravity Inc. and Chelsio Inc.
1) Fix a bug in dealing with the Alerus 1006 PHY which prevented the
device from ever coming back up once it had been set to down.
2) Add a kernel tunable (hw.cxgb.snd_queue_len) which makes it possible
to give the device more than IFQ_MAXLEN entries in its send queue. The
default remains 50.
3) Add code to place the card'd identification and serial number into
its description (%desc) so that users can tell which card they have
installed.
for virtualization.
Instead of initializing the affected global variables at instatiation,
assign initial values to them in initializer functions. As a rule,
initialization at instatiation for such variables should never be
introduced again from now on. Furthermore, enclose all instantiations
of such global variables in #ifdef VIMAGE_GLOBALS blocks.
Essentialy, this change should have zero functional impact. In the next
phase of merging network stack virtualization infrastructure from
p4/vimage branch, the new initialization methology will allow us to
switch between using global variables and their counterparts residing in
virtualization containers with minimum code churn, and in the long run
allow us to intialize multiple instances of such container structures.
Discussed at: devsummit Strassburg
Reviewed by: bz, julian
Approved by: julian (mentor)
Obtained from: //depot/projects/vimage-commit2/...
X-MFC after: never
Sponsored by: NLnet Foundation, The FreeBSD Foundation
