Use MPI2_IOCSTATUS_MASK when checking IOCStatus to mask off the log bit, and
make a few more things endian-safe.
- Fix possible use of invalid pointer.
It was possible to use an invalid pointer to get the target ID value. To fix
this, initialize a local Target ID variable to an invalid value and change that
variable to a valid value only if the pointer to the Target ID is not NULL.
- No need to set the MPSSAS_SHUTDOWN flag because it's never used.
- done_ccb pointer can be used if it is NULL.
To prevent this, move check for done_ccb == NULL to before done_ccb is used in
mpssas_stop_unit_done().
- Disks can go missing until a reboot is done in some cases.
This is due to the DevHandle not being released, which causes the Firmware to
not allow that disk to be re-added.
Reviewed by: ken
Approved by: re (gjb), ken, scottl, ambrisko (mentors)
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D6872
This started showing up when doing lots of aggregate traffic. For TDMA it's
always no-ACK traffic and I didn't notice this, and I didn't notice it
when doing 11abg traffic as it didn't fail enough in a bad way to trigger
this.
This showed up as the fifo depth being < 0.
Eg:
Jun 19 09:23:07 gertrude kernel: ath0: ath_tx_edma_push_staging_list: queued 2 packets; depth=2, fifo depth=1
Jun 19 09:23:07 gertrude kernel: ath0: ath_edma_tx_processq: Q1, bf=0xfffffe000385f068, start=1, end=1
Jun 19 09:23:07 gertrude kernel: ath0: ath_edma_tx_processq: Q1: FIFO depth is now 0 (1)
Jun 19 09:23:07 gertrude kernel: ath0: ath_edma_tx_processq: Q1, bf=0xfffffe0003866fe8, start=0, end=1
Jun 19 09:23:07 gertrude kernel: ath0: ath_edma_tx_processq: Q1: FIFO depth is now -1 (0)
So, clear the flags before adding them to a TX queue, so if they're
re-added for the retransmit path it'll clear whatever they were and
not double-account the FIFOEND flag. Oops.
Tested:
* AR9380, STA mode, 11n iperf testing (~130mbit)
Approved by: re (delphij)
It turns out the frame scheduling policies (eg DBA_GATED) operate on
a single TX FIFO entry. ASAP scheduling is fine; those frames always
go out.
DBA-gated sets the TX queue ready when the DBA timer fires, which triggers
a beacon transmit. Normally this is used for content-after-beacon queue
(CABQ) work, which needs to burst out immediately after a beacon.
(eg broadcast, multicast, etc frames.) This is a general policy that you
can use for any queue, and Sam's TDMA code uses it.
When DBA_GATED is used and something like say, an 11e TX burst window,
it only operates on a single TX FIFO entry. If you have a single frame
per TX FIFO entry and say, a 2.5ms long burst window (eg TDMA!) then it'll
only burst a single frame every 2.5ms. If there's no gating (eg ASAP) then
the burst window is fine, and multiple TX FIFO slots get used.
The CABQ code does pack in a list of frames (ie, the whole cabq) but
up until this commit, the normal TX queues didn't. It showed up when
I started to debug TDMA on the AR9380 and later.
This commit doesn't fix the TDMA case - that's still broken here, because
all I'm doing here is allowing 'some' frames to be bursting, but I'm
certainly not filling the whole TX FIFO slot entry with frames.
Doing that 'properly' kind of requires me to take into account how long
packets should take to transmit and say, doing 1.5 or something times that
per TX FIFO slot, as if you partially transmit a slot, when it's next
gated it'll just finish that TX FIFO slot, then not advance to the next
one.
Now, I /also/ think queuing a new packet restarts DMA, but you have to
push new frames into the TX FIFO. I need to experiment some more with
this because if it's really the case, I will be able to do TDMA support
without the egregious hacks I have in my local tree. Sam's TDMA code
for previous chips would just kick the TXE bit to push along DMA
again, but we can't do that for EDMA chips - we /have/ to push a new
frame into the TX FIFO to restart DMA. Ugh.
Tested:
* AR9380, STA mode
* AR9380, hostap mode
* AR9580, hostap mode
Approved by: re (gjb)
This allows IPv6 link local addresses (and other IPv6 functionality) to work.
PR: 210355
Submitted by: Steve Wahl and David Bright (both at Dell Inc.)
Reviewed by: cem, mav
Tested by: mav (on Intel hardware)
Approved by: re (kib)
MFC after: 5 days
Sponsored by: Dell Inc.
Differential Revision: https://reviews.freebsd.org/D6885
Maps Sonics/OCP per-core address spaces to bcma(4)-compatible port/region
identifiers.
This permits the use of common address map identifiers in bhnd device
drivers, independent of the underlying interconnect type.
Approved by: re (gjb), adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D6850
- Delete all chipc children on attachment failure.
- Added missing bhnd_nexus bhnd_bus_deactivate_resource implementation.
- Drop a CHIPC_UNLOCK() accidentally left behind after lifting
synchronization into the chipc region refcounting API.
- Fix re-allocation of chipc resources. Previously, the resource ID was
reset to -1 on release, preventing later re-allocation.
Approved by: re (gjb), adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D6849
supported, e.g. CPUID or MSR, return ENODEV from the ioctl which needs
that feature.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Approved by: re (hrs)
Inserting a full mbuf with an external cluster into the socket buffer
resulted in sbspace() returning -MLEN. However, since sb_hiwat is
unsigned, the -MLEN value was converted to unsigned in comparisons. As a
result, the socket buffer was never autosized. Note that sb_lowat is signed
to permit direct comparisons with sbspace(), but sb_hiwat is unsigned.
Follow suit with what tcp_output() does and compare the value of sbused()
with sb_hiwat instead.
Approved by: re (gjb)
Sponsored by: Chelsio Communications
This reduces the size of kaiocb slightly. I've also added some generic
fields that other backends can use in place of the BIO-specific fields.
Change the socket and Chelsio DDP backends to use 'backend3' instead of
abusing _aiocb_private.status directly. This confines the use of
_aiocb_private to the AIO internals in vfs_aio.c.
Reviewed by: kib (earlier version)
Approved by: re (gjb)
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D6547
Release the hold on ep->com immediately after sending the RST. This
fixes a bug that sometimes leaves userspace iWARP tools hung when the
user presses ^C.
Submitted by: Krishnamraju Eraparaju @ Chelsio
Approved by: re (gjb@)
Sponsored by: Chelsio Communications
When allocating a new mbuf or bus_dmamap_load()-ing it fails,
we can just keep the old mbuf since we are dropping that packet anyway.
Instead of doing bus_dmamap_create() and bus_dmamap_destroy() all the time,
create an extra bus_dmamap_t which we can use to safely try
bus_dmamap_load()-ing the new mbuf. On success we just swap the spare
bus_dmamap_t with the data->map of that ring entry.
Tested:
Tested with Intel AC7260, verified with vmstat -m that new kernel no
longer visibly leaks memory from the M_DEVBUF malloc type.
Before, leakage was 1KB every few seconds while ping(8)-ing over the wlan
connection.
Submitted by: Imre Vadasz <imre@vdsz.com>
Approved by: re@
Obtained from: DragonflyBSD.git cc440b26818b5dfdd9af504d71c1b0e6522b53ef
Differential Revision: https://reviews.freebsd.org/D6742
For DWC_GMAC_ALT_DESC implementations, the multicast hash table has only
64 entries. Instead of 8 registers starting at 0x500, a pair of registers
at 0x08 and 0x0c are used instead.
Approved by: re (hrs)
Submitted by: Guy Yur <guyyur@gmail.com>
Some later code I'll commit pushes lists of frames into the EDMA TX
FIFO, rather than a single frame at a time. The CABQ code already
pushes frame lists, but it turns out we should actually be doing it
in general or performance tanks. :(
Since key table is cleared on every device shutdown,
static WEP keys (which are set only once) need to be
reinstalled manually every time when device starts running.
Tested with RTL8188EU, STA (all ciphers) / IBSS (WPA-none) modes.
r298930 removed the inittodr call, but it seems like this prevents
"calcru: runtime went backwards ..." messages from occasionally appearing
when resuming from migration.
Reported by: Karl Pielorz <kpielorz@tdx.co.uk>
Sponsored by: Citrix Systems R&D
ticks are signed int and if statistics is not updated for a long time
(more than INT_MAX ticks, but less than UINT_MAX) difference becomes
negative and less than hz for a long time.
Other option to repeat is simply load driver (which initializes
timestamps to 0) when ticks are negative.
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D6777
This fixes a warning that occurs in a number of files that use the
random_harvest_queue function.
Differential Revision: https://reviews.freebsd.org/D4229
Submitted by: stevek@juniper.net
Reviewed by: markm
Approved by: so
Changes:
- Fixed incorrect MIPS74k vendor ID in the bhnd core descriptor tables
- Fixed MIPS core driver's matching against MIPS/MIPS33 cores.
- Improved MIPS3302 core description.
- Enabled BUS_PASS_BUS on the bhnd nexus drivers to allow early probing
of the MIPS core.
- Enabled BUS_PASS_CPU on the MIPS core driver to ensure correct attach
order.
- Disabled matching of the MIPS core driver on non-SoC devices.
Reviewed by: Michael Zhilin <mizhka@gmail.com>
Approved by: adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D6735
This patch adds the missing pieces needed for device setup using the
mlx5en driver inside a virtual machine which is providing hardware
access through SR-IOV.
Sponsored by: Mellanox Technologies
MFC after: 1 week
- Validate the scheduling class against the actual limit (which is chip
specific) instead of a magic number.
- Return an error if an attempt is made to manipulate the tx queues of a
VI that hasn't been initialized.
Sponsored by: Chelsio Communications
The SYSINIT runs at SI_SUB_KICK_SCHEDULER after the scheduler is fully
initialized and timers are working. This fixes booting in the
EARLY_AP_STARTUP case.
A couple of mostly cosmetic fixes for the final initialization of netfront:
- Switch to "connected" state before starting to kick the rings.
- Correctly use "rxq" in the initialization loop (previously rxq was not
updated in the loop, and netfront would kick np->rxq[N] several times).
- Declare and define xn_connect as static, it's not used outside of this
file.
Reviewed by: Wei Liu <wei.liu2@citrix.com>
Sponsored by: Citrix Systems R&D
Differential revision: https://reviews.freebsd.org/D6657
In mediatek etherswitch support, functions mtkswitch_reg_write32_mt7621
and mtkswitch_reg_read32_mt7621 are called without locks held, so
lock assertions fail. Remove the lock assertions.
Sponsored by: Smartcom - Bulgaria AD
This is done because one has no point to have more channels since they
will be unused.
Submitted by: Ivan Malov <Ivan.Malov at oktetlabs.ru>
Reviewed by: gnn
Sponsored by: Solarflare Communications, Inc.
Differential Revision: https://reviews.freebsd.org/D6720
And fix message processing; only channel messages are supported.
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6706
This is a NOP.
The COMPAT_IA32 was renamed in r205014 to COMPAT_FREEBSD32 and
COMPAT_ARCH32 does not seem to have existed. Also remove some
leftovers from the sysent rework in r301404. Include
freebsd32_util.h for the freebsd32_sysent prototype.
X-MFC-With: r301404
Reported by: kib
MFC after: 3 days
Sponsored by: EMC / Isilon Storage Division
This trips me up whenever I'm fooling around with partially supported
NICs that fail to fully attach or initialise - the firmware gets loaded
and references, but something fails - and the firmware references
aren't cleaned up.
After perusing the PHY-LP code (don't ask why; honest) I discovered that
it /has/ 5GHz support - but it's not ever used. I found one NIC - a
BCM4312 w/ pci id 0x4315 - which advertised dual-band PHY-LP support.
Turns out it works.
Whilst here, move up the support bit logging code so I can use it
to debug this.
Tested:
* BCM4312 (pci id 0x4315); 5GHz STA operation
On Medford, licenses are required to enable RX and event cut through and to
disable RX batching. To avoid the need for the driver to make decisions based on
the licensing state, the MC_CMD_INIT_EVQ has been extended to allow us to leave
the decision to the firmware. If the adapter is licensed for low-latency use,
the firmware will choose the optimal settings for latency, otherwise it will use
the best settings for throughput.
For Huntington we still need to choose the settings ourselves.
Submitted by: Mark Spender <mspender at solarflare.com>
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D6717
the graphics drivers can benefit from access to the lid handle for querying and getting notifications
Submitted by: kmacy
Differential Revision: https://reviews.freebsd.org/D6643
Add support for fetching SPROM data from OTP on chipsets with an
IPX OTP controller (including the BCM43225).
This integrates the NVRAM data source into the chipc_caps capability
structure, and adds a sprom_offset field that can be used with OTP
to locate the SPROM image data (found within the general use
region, H/W subregion).
This also removes one of two duplicate parse error messages reported by
both the bhnd_sprom driver and the underlying SPROM parsing API.
Approved by: adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D6729
Now that bhnd(4) provides feature parity with the previous siba/mips
implementation, we can switch sentry5 over and begin lifting common
support code out for use across bhnd(4) embedded targets.
Changes:
- Fixed enumeration of siba(4) per-core address maps, required for
discovery of memory mapped chipc flash region on siba(4) devices.
- Simplified bhnd kernel configuration (dropped 'bhndbus' option).
- Replaced files.broadcom's direct file references with their
corresponding standard kernel options.
- Lifted out common bcma/siba nexus support, inheriting from the new
generic bhnd_nexus driver.
- Dropped now-unused sentry5 siba code.
- Re-integrated BCM into the universe build now that it actually compiles.
Approved by: adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D6712
This adds support for serial (via SPI) and parallel (via CFI) flash
as found on BCM47xx/BCM53xx SoCs.
Submitted by: Michael Zhilin <mizhka@gmail.com>
Approved by: adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D6250
- Correct IRQ lines for UART (to fix IRQ lookup in future)
- Check device unit in resource assignment during chipc_add_child
- If chipc hint->size is RM_MAX_END, resource end should be same as window end
- Clear reference from resource list entry to resource in case of resource release
- Add CHIPC_GET_CAPS implementation
- Correct chipc flash constants (to be unshifted)
- Default implementation of get_attach_type should iterate over device tree
- Add default implementation for BHND_CHIPC_GET_CAPS usable by chipc grandchildren
Submitted by: Michael Zhilin <mizhka@gmail.com>
Approved by: adrian (mentor)
Differential Revision: https://reviews.freebsd.org/D6584
PowerPC64 has two different ABIs, neither of which is elf64_freebsd_sysvec.
Using sysent and freebsd32_sysent achieves the same effect.
X-MFC-With: r301130
MFC after: 1 week
Sponsored by: EMC / Isilon Storage Division
When the low-latency firmware variant is running, it is reported as not
being capable of batching RX events, but it can still do so if the
FORCE_EV_MERGING flag is set on an RXQ. Therefore we need to handle
batched RX events even if the capability isn't set.
If this bug is fixed in the firmware such that the capability is set
even when running the low-latency firmware variant, it will almost
always be reported so I don't think we lose much by removing the check.
Submitted by: Mark Spender <mspender at solarflare.com>
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D6705
This also adjusts the timer values used to match the Linux net
driver implementation:
a) non-zero time intervals should result in at least one quantum
b) timer load/reload values are only zero biased for Falcon/Siena
Submitted by: Andy Moreton <amoreton at solarflare.com>
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D6704
BAR size to 1MB. According to Xeon v3 specifications and my tests, that
size register is write-once and so not writeable after BIOS written it.
Instead of that, make the code work with BAR of any sufficient size,
properly calculating offset within its base. It also simplifies the code.
Discussed with: cem
MFC after: 2 weeks
Sponsored by: iXsystems, Inc.
NTB_MSIX_RECEIVED status, before making upper layers overwrite it.
This is not completely perfect, but now it works better then before.
MFC after: 2 weeks
Sponsored by: iXsystems, Inc.
Improvements after r301220.
Bus space methods are not called so simple pmap_mapdev will suffice.
Use OF_getencprop to get buffer with already converted endianess.
Pointed out by: ian
Submitted by: Michal Stanek <mst@semihalf.com>
Obtained from: Semihalf
The current error path in case of failure during attach/initialization is
not correct and leaves blkback in a stuck state. This is due to blkback
waiting for blkfront to switch to state XenbusStateClosed, but if blkfront
never attached (because the guest is not even started) it cannot possibly
make it to that state.
Instead just wait for the frontend to be in a state different than
XenbusStateConnected in order to proceed with the shutdown. Also, it is
wrong to call xbb_detach directly because it destroys the lock which can
still be used by xbb_frontend_changed.
Sponsored by: Citrix Systems R&D
Hotplug scripts are needed in order to use fancy disk configurations in xl,
like iSCSI disks. The job of hotplug scripts is to locally attach the disk
and present it to blkback as a block device or a regular file.
This change introduces a new xenstore node in the blkback hierarchy, called
"physical-device-path". This is a straigh replacement for the "params" node,
which was used before.
Hotplug scripts will need to read the "params" node, perform whatever
actions are necessary and then write the "physical-device-path" node. The
hotplug script is also in charge of detaching the disk once the domain has
been shutdown.
Sponsored by: Citrix Systems R&D
That check wasn't enough to handle appending a two byte character
following it.
This prevented my T400 (Intel Core 2 Duo P8400) from attaching;
it would panic from a stack overflow detection.
Only HMAC-SHA256 is added as it is the only SHA-2 variant supported by
cryptodev. It is not possible to register hardware support for other
algorithms in the family including regular non-keyed SHA256.
Submitted by: Michal Stanek <mst@semihalf.com>
Obtained from: Semihalf
Sponsored by: Stormshield
Differential revision: https://reviews.freebsd.org/D6219
The output of HMAC was previously truncated to 12 bytes. This was only
correct in case of one particular crypto client - the new version of IPSEC.
Fix by taking into account the cri_mlen field in cryptoini session request
filled in by the client.
Submitted by: Michal Stanek <mst@semihalf.com>
Obtained from: Semihalf
Sponsored by: Stormshield
Differential revision: https://reviews.freebsd.org/D6218
TDMA and CESA registers are placed in different ranges of memory. Split
memory resource in DTS to reflect that. This change is needed to support
multiple CESA nodes as otherwise the ranges of different nodes would
overlap.
In consequence, CESA_WRITE and CESA_READ macros have been split depending
on which range of registers is accessed. Offsets for CESA registers have
been modified as the base address has changed.
Submitted by: Michal Stanek <mst@semihalf.com>
Obtained from: Semihalf
Sponsored by: Stormshield
Differential revision: https://reviews.freebsd.org/D6217
On other platforms with CESA accelerator the SRAM memory is mapped in
early init before driver is attached. This method only works correctly
with mappings no smaller than L1 section size (1MB). There may be more
SRAM blocks and they may have smaller sizes than 1MB as is the case
for Armada38x. Instead, map SRAM memory with bus_space_map() in CESA
driver attach. Note that we can no longer assume that VA == PA for the
SRAM.
Submitted by: Michal Stanek <mst@semihalf.com
Obtained from: Semihalf
Sponsored by: Stormshield
Differential revision: https://reviews.freebsd.org/D6215
The PV backend will only pick the new options when the interface is detached
and reattached again, so perform a full reset when changing options. This is
very fast, and should not be noticeable by the user.
Reviewed by: Wei Liu <wei.liu2@citrix.com>
Sponsored by: Citrix Systems R&D
Differential revision: https://reviews.freebsd.org/D6658
Just calling gnttab_end_foreign_access_ref doesn't free the references,
instead call gnttab_end_foreign_access with a NULL page argument in order to
have the grant references freed. The code that maps the ring
(xenbus_map_ring) already uses gnttab_grant_foreign_access which takes care
of allocating a grant reference.
Reviewed by: Wei Liu <wei.liu2@citrix.com>
Sponsored by: Citrix Systems R&D
Differential revision: https://reviews.freebsd.org/D6608
This patch fixes two issues seen on hot-unplug. The first one is a panic
caused by calling ether_ifdetach after freeing the internal netfront queue
structures. ether_ifdetach will call xn_qflush, and this needs to be done
before freeing the queues. This prevents the following panic:
Fatal trap 9: general protection fault while in kernel mode
cpuid = 2; apic id = 04
instruction pointer = 0x20:0xffffffff80b1687f
stack pointer = 0x28:0xfffffe009239e770
frame pointer = 0x28:0xfffffe009239e780
code segment = base 0x0, limit 0xfffff, type 0x1b
= DPL 0, pres 1, long 1, def32 0, gran 1
processor eflags = interrupt enabled, resume, IOPL = 0
current process = 0 (thread taskq)
[ thread pid 0 tid 100015 ]
Stopped at strlen+0x1f: movq (%rcx),%rax
db> bt
Tracing pid 0 tid 100015 td 0xfffff800038a6000
strlen() at strlen+0x1f/frame 0xfffffe009239e780
kvprintf() at kvprintf+0xfa0/frame 0xfffffe009239e890
vsnprintf() at vsnprintf+0x31/frame 0xfffffe009239e8b0
kassert_panic() at kassert_panic+0x5a/frame 0xfffffe009239e920
__mtx_lock_flags() at __mtx_lock_flags+0x164/frame 0xfffffe009239e970
xn_qflush() at xn_qflush+0x59/frame 0xfffffe009239e9b0
if_detach() at if_detach+0x17e/frame 0xfffffe009239ea10
netif_free() at netif_free+0x97/frame 0xfffffe009239ea30
netfront_detach() at netfront_detach+0x11/frame 0xfffffe009239ea40
[...]
Another panic can be triggered by hot-plugging a NIC:
Fatal trap 18: integer divide fault while in kernel mode
cpuid = 0; apic id = 00
instruction pointer = 0x20:0xffffffff80902203
stack pointer = 0x28:0xfffffe00508d3660
frame pointer = 0x28:0xfffffe00508d36a0
code segment = base 0x0, limit 0xfffff, type 0x1b
= DPL 0, pres 1, long 1, def32 0, gran 1
processor eflags = interrupt enabled, resume, IOPL = 0
current process = 2960 (ifconfig)
[ thread pid 2960 tid 100088 ]
Stopped at xn_txq_mq_start+0x33: divl %esi,%eax
db> bt
Tracing pid 2960 tid 100088 td 0xfffff8000850aa00
xn_txq_mq_start() at xn_txq_mq_start+0x33/frame 0xfffffe00508d36a0
ether_output() at ether_output+0x570/frame 0xfffffe00508d3720
arprequest() at arprequest+0x433/frame 0xfffffe00508d3820
arp_ifinit() at arp_ifinit+0x49/frame 0xfffffe00508d3850
xn_ioctl() at xn_ioctl+0x1a2/frame 0xfffffe00508d3890
in_control() at in_control+0x882/frame 0xfffffe00508d3910
ifioctl() at ifioctl+0xda1/frame 0xfffffe00508d39a0
kern_ioctl() at kern_ioctl+0x246/frame 0xfffffe00508d3a00
sys_ioctl() at sys_ioctl+0x171/frame 0xfffffe00508d3ae0
amd64_syscall() at amd64_syscall+0x2db/frame 0xfffffe00508d3bf0
Xfast_syscall() at Xfast_syscall+0xfb/frame 0xfffffe00508d3bf0
--- syscall (54, FreeBSD ELF64, sys_ioctl), rip = 0x8011e185a, rsp =
0x7fffffffe478, rbp = 0x7fffffffe4c0 ---
This is caused by marking the driver as active before it's fully
initialized, and thus calling xn_txq_mq_start with num_queues set to 0.
Reviewed by: Wei Liu <wei.liu2@citrix.com>
Sponsored by: Citrix Systems R&D
Differential revision: https://reviews.freebsd.org/D6646
In order to use custom taskqueues we would have to mask the interrupt, which
is basically what is already done for an interrupt handler, or else we risk
loosing interrupts. This switches netfront to the same interrupt handling
that was done before multiqueue support was added.
Reviewed by: Wei Liu <wei.liu2@citrix.com>
Sponsored by: Citrix Systems R&D
This is based on Linux commit 1f3c2eba1e2d866ef99bb9b10ade4096e3d7607c from
David Vrabel:
A full Rx ring only requires 1 MiB of memory. This is not enough memory
that it is useful to dynamically scale the number of Rx requests in the ring
based on traffic rates, because:
a) Even the full 1 MiB is a tiny fraction of a typically modern Linux
VM (for example, the AWS micro instance still has 1 GiB of memory).
b) Netfront would have used up to 1 MiB already even with moderate
data rates (there was no adjustment of target based on memory
pressure).
c) Small VMs are going to typically have one VCPU and hence only one
queue.
Keeping the ring full of Rx requests handles bursty traffic better than
trying to converge on an optimal number of requests to keep filled.
Reviewed by: Wei Liu <wei.liu2@citrix.com>
Sponsored by: Citrix Systems R&D
Currently FreeBSD is not properly fetching the TSO information from the Xen
PV ring, and thus the received packets didn't have all the necessary
information, like the segment size or even the TSO flag set.
Sponsored by: Citrix Systems R&D
* The "if (!data->valid_tx_ant || !data->valid_rx_ant) {" check was getting
triggered with a 3165 chipset.
Submitted by: Imre Vadasz <imre@vdsz.com>
Obtained from: DragonflyBSD 3655dfb6fc311fc83e5ce8370dd91b4cd4a37991
Move some declarations to if_iwmreg.h.
Remove iwm_fw_alive(); just call iwm_post_alive() directly.
Simplify iwm_mvm_add_sta().
Return timeout error from iwm_apm_init().
Print a message when init (i.e. boot) firmware fails to load.
Remove some commented-out code which wouldn't compile anyway.
Move iwm_mvm_tx_fifo to if_iwmreg.h to match better where Linux puts it.
Taken-From: OpenBSD (if_iwm.c r1.80 and if_iwmreg.h r1.11)
Submitted by: Imre Vadasz <imre@vdsz.com>
Obtained from: DragonflyBSD 29fcb331e5620ae145a6ab9cdda830e22fff626a
This is the initial framework to call into the MCI HAL routines and drive
the basic state engine.
The MCI bluetooth coex model uses a command channel between wlan and
bluetooth, rather than a 2-wire or 3-wire signaling protocol to control things.
This means the wlan and bluetooth chip exchange a lot more information and
signaling, even at the per-packet level. The NICs in question can share
the input LNA and output PA on the die, so they absolutely can't stomp
on each other in a silly fashion. It also allows for the bluetooth side
to signal when profiles come and go, so the driver can take appropriate
control. There's also the possibility of dynamic bluetooth/wlan duty cycle
control which I haven't yet really played with.
It configures things up with a static "wlan wins everything" coexistence,
configures up the available 2GHz channel map for bluetooth, sets a static
duty cycle for bluetooth/wifi traffic priority and drives the basics needed to
keep the MCI HAL code happy.
It doesn't do any actual coexistence except to default to "wlan wins everything",
which at least demonstrates that things do indeed work. Bluetooth inquiry frames
still trump wifi (including beacons), so that demonstrates things really do
indeed seem to work.
Tested:
* AR9462 (WB222), STA mode + bt
* QCA9565 (WB335), STA mode + bt
TODO:
* .. the rest of coexistence. yes, bluetooth, not people. That stuff's hard.
* It doesn't do the initial BT side calibration, which requires a WLAN chip
reset. I'll fix up the reset path a bit more first before I enable that.
* The 1-ant and 2-ant configuration bits aren't being set correctly in
if_ath_btcoex.c - I'll dig into that and fix it in a subsequent commit.
* It's not enabled by default for WB222/WB225 even though I believe it now
can be - I'll chase that up in a subsequent commit.
Obtained from: Qualcomm Atheros, Linux ath9k
comp_handler_lock but c4iw_destroy_cq has already freed the CQ memory
(which is where the lock resides).
Submitted by: Krishnamraju Eraparaju @ Chelsio
Sponsored by: Chelsio Communications
Do not set HWRTSEN bit when CTS-to-self is used; CTS2SELF bit triggers
CTS frame transmission by itself (and it does not work when HWRTSEN bit
is set).
Tested with:
* RTL8188CUS, HOSTAP mode (11g)
* RTL8188EU, STA mode (11g)
Receive all beacons in HOSTAP mode; they will give more information about
present non-ERP / legacy BSSs (used to choose protection mode).
Tested with RTL8188CUS (HOSTAP, urtwn) + RTL8821AU (HOSTAP, 11b mode).
This simplifies setting an initial interrupt moderation value, and
avoids most calls to evx_ev_qmoderate from contexts where MCDI is
not allowed (MCDI is need for an EVQ timer workaround in a later patch).
Submitted by: Andy Moreton <amoreton at solarflare.com>
Reviewed by: gnn
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D6673
by default. This is a workaround for a too simplistic ICL module
choosing mechanism. To use it, specify offload in ctl.conf
or iscsi.conf.
This fixes a problem where "kldload cxgbei" wedges the iSCSI stack,
if you don't have a Chelsio card installed, or the endpoints of the
iSCSI session are not reachable through addresses configured
on that interface.
Reviewed by: np@
MFC after: 1 month
The buggy code was using the rxq index but should use the evq label
associated with the rxq. It was missed in r298735.
Reviewed by: gnn
Sponsored by: Solarflare Communications, Inc.
MFC after: 2 days
Differential Revision: https://reviews.freebsd.org/D6661
discovery without attaching to the targets ("iscsictl -Ad ... -e off"),
and then attach to selected ones ("iscsictl -Mi ... -e on").
PR: 204129
MFC after: 1 month
Relnotes: yes
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D6633
this to be the case. This will mean we don't try and handle the cache in
bus_dmamap_sync when it is not needed.
Obtained from: ABT Systems Ltd
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D6605
- Avoid unnecessary indirection.
- Avoid bit fields.
- Use __packed.
Reviewed by: Jun Su <junsu microsoft com>
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6636
The legacy bits are just from ah.h; the MCI bits are from the ar9300
HAL "freebsd" extras.
A subsequent commit will include ah_btcoex.h into ah.h and remove
the older defintions.
While I'm here, prefix function names w/ vmbus, since unlike Hyper-V
timecounter, Hyper-V event timer will not work w/o vmbus.
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6598
Some machine BIOSes use the I2C bus and leave it in a state that causes
interrupts to not work properly due to a pending interrupt having been
latched.
Refactor the code a bit to clear pending interrupts when I2C is enabled.
This fixes the primary problem.
Also fix a possible race condition in the interrupt handler where the
interrupt was being cleared after reading the status instead of before.
Reported by: pfg
Reviewed by: jhb
Approved by: jhb
Obtained from: DragonFly BSD
Differential Revision: https://reviews.freebsd.org/D6586
While I'm here remove the event timer's dependency on hv_vmbus_priv.h
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6574
- Make sure that STIMER0 is disabled before writting to it, since
writing to an enabled STIMER will result in undefined behaviour.
- It is unnecessary to reconfigure STIMER0 upon each et_start().
- Make sure that MSR_HV_REF_TIME_COUNT will not return 0, since
writing 0 to STIMER_COUNT will disable the target STIMER.
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6573
When hwpmc stops sampling it will set the pm_state to something other
than PMC_STATE_RUNNING. This means the following sequence can happen:
CPU 0: Enter the interrupt handler
CPU 0: Set the thread TDP_CALLCHAIN pflag
CPU 1: Stop sampling
CPU 0: Call pmc_process_samples, sampling is stopped so clears ps_nsamples
CPU 0: Finishes interrupt processing with the TDP_CALLCHAIN flag set
CPU 0: Call pmc_capture_user_callchain to capture the user call chain
CPU 0: Find all the pmc sample are free so no call chains need to be captured
CPU 0: KASSERT because of this
This fixes the issue by checking if any of the samples have been stopped
and including this in te KASSERT.
PR: 204273
Reviewed by: bz, gnn
Obtained from: ABT Systems Ltd
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D6581
This is like the WB222 coexistence (ie, "MCI", a message bus inside the
chip), and it's currently a cut/paste so I can start using it to flesh
out the differences with WB222.
It doesn't completely /do/ bluetooth coexistence, because it turns out
I need to add some contigmalloc'ed buffers to the btcoex path for this
type of hardware. I'm putting this work in the "people would like
to see functioning-ish btcoex before FreeBSD-11" bucket because I see
this as "broken".
Tested:
* QCA9535 (WB335) NIC, BT + 2GHz STA
avoid panicking debug kernels.
t4_tom does not keep track of a connection once it switches to ULP mode
iWARP. If the connection falls out of ULP mode the driver/hardware seq#
etc. are out of sync. A better fix would be to figure out what the
current seq# are, update the driver's state, and perform all sanity
checks as usual.
This also prevents tracing to a P_INEXEC process since it could race
with other processes attaching to it in filemon_event_process_exec() due
to the filemon_get_proc() race of incrementing ref and then locking the
filemon. With the no-P_INEXEC invariant in place the p_filemon may only
be the same or NULL when trying to drop it in
filemon_event_process_exec().
MFC after: 2 weeks
Sponsored by: EMC / Isilon Storage Division
Differential Revision: https://reviews.freebsd.org/D6545
If the parent is DEAD or connect_request_upcall() fails, the parent
mutex is left locked. This leads to a hang when process_mpa_request()
is called again for another child of the listening endpoint.
Submitted by: Krishnamraju Eraparaju @ Chelsio
Obtained from: upstream iw_cxgb4
Sponsored by: Chelsio Communications
wrap the implementation so that it returns an error if INTRNG support is
not available. It should be possible to write a non-INTRNG implementation
of this function some day. In the meantime, there is code that contains
calls to this function (so the decl is needed), but have runtime checks to
avoid calling it in the non-INTRNG case.
Use for forever loop instead of while.
Found by lint on illumos.
Submitted by: Garrett D'Amore <garrett at damore.org>
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
if_iwm - GC some dead code, left by a partially applied OpenBSD change.
Taken-From: OpenBSD (if_iwm.c r1.69)
Submitted by: Imre Vadasz <imre@vdsz.com>
Obtained from: DragonflyBSD git 07dfed32ea39b980b0b80d27ff938e7c3ca4c0b5
This will be more accurate as the actual name is provided if ran
from an absolute path in do_execve().
MFC after: 2 weeks
Sponsored by: EMC / Isilon Storage Division
Pressing the PEK (power enable key) will shutdown the board.
Some events are reported to devd via system "PMU" and subsystem
"Battery", "AC" and "USB" such as connected/disconnected.
Some sensors values (power source voltage/current) are reported via
sysctl (dev.axp209_pmu.X.)
It also expose a gpioc node usable in kernel and userland. Only 3 of
the 4 GPIO are exposed (The GPIO3 is different and mostly unused on
boards). Most popular boards uses GPIO1 as a sense pin for OTG power.
Add a dtsi file that adds gpio-controller capability to the device as
upstream doesn't defined it and include it in our custom DTS.
Reviewed by: jmcneill
Approved by: cognet (mentor)
Differential Revision: https://reviews.freebsd.org/D6135
- Use device's channel list instead of default one (adds 12, 13 and 14
2GHz channels).
- Add ic_getradiocaps() method.
Differential Revision: https://reviews.freebsd.org/D6171
- Use device's channel list instead of default one (+ 12, 13 and 14
2GHz channels).
- Add ic_getradiocaps() method.
Differential Revision: https://reviews.freebsd.org/D6170
- Use device's channel list instead of default one (from
ieee80211_init_channels()).
- Add ic_getradiocaps() method.
Differential Revision: https://reviews.freebsd.org/D6144
- Use macros to define rate indices; don't hardcode them in code.
- Add method for 'rate' -> 'rate index' conversion
(and array for the opposite).
- Determine if rate is CCK / OFDM via appropriate macro.
Reviewed by: adrian
Differential Revision: https://reviews.freebsd.org/D4837
Previously the USB PHY driver would enable all regulators at attach time.
This prevented boards from booting when powered by the USB OTG port, as
it didn't take VBUS presence into consideration.
in ICL interface.
- the ordering of parameters to icl_conn_task_setup is different, so that
the "cookie" is last.
- the icl_conn_connected() method is gone, replaced by much simpler mechanism.
I'd rather keep the ICL interface as small as possible.
- I don't really like the s/offload/driver/g. The "tcp" is not a driver;
"iser" is not really a driver either. I'd prefer to leave it as it is.
- the check for ic_session_type_discovery() in iser_conn_handoff() is gone,
as handoff cannot happen for discovery sessions.
- ic_session_login_phase() and ic_session_type_discovery() are gone. If you
had your handoff method called - you're no longer in either of those.
- the way maxtags is passed is different; now it's simply ic->ic_maxtags.
It's cleaner, and the old way would cause weird things to happen if
fail_on_disconnection=1 and the user changed the maxtags sysctl before
reconnecting (basically the CAM idea of maxtags would be different from
iSER one).
- icl_hba_misc() is gone; declare support for PIM_UNMAPPED by setting
ic->ic_unmapped flag.
- the way we find the "iser" ICL module is rewritten - we have a flag
for icl_register() that says if the module is iSER-capable or not.
- icl_conn_release() is gone; iser_conn_release() is called from
iser_conn_free() (no functional change in this case) and at the beginning
of icl_conn_connect(), to handle reconnection.
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
written by Sagi Grimberg <sagig at mellanox.com> and Max Gurtovoy
<maxg at mellanox.com>.
This code comes from https://github.com/sagigrimberg/iser-freebsd, branch
iser-rebase-11-current-r291993. It's not connected to the build just yet;
it still needs some tweaks to adapt to my changes to iSCSI infrastructure.
Big thanks to Mellanox for their support for FreeBSD!
Obtained from: Mellanox Technologies
MFC after: 1 month
Relnotes: yes
This is a simple ioctl and mmap API to issue SPI transactions from
userland. It's useful for simple devices (eg spi temperature sensors,
etc) for experimentation.
TODO:
* Write some documentation!
Submitted by: green
This was triggering a panic on detach; the SPROM shadow is now
maintained by the bhnd_sprom_chipc driver, and should be removed
from chipc.
Submitted by: Landon Fuller <landonf@landonf.org>
Differential Revision: https://reviews.freebsd.org/D6548
This adds support for automatically configuring bhnd_chipc bus children
with associated resources, using an internal 'hints' table based directly
on Michael Zhilin's chipc resource mapping work.
The bhnd_sprom_chipc driver has been converted to use DEVICE_IDENTIFY()
with the new resource table.
This should be nearly drop-in compatible with the child device drivers
in D6250.
Submitted by: Landon Fuller <landonf@landonf.org>
Reviewed by: Michael Zhilin <mizhka@gmail.com>
Differential Revision: https://reviews.freebsd.org/D6525
tunable SYSCTL's. Linux module parameters are associated with the
module they belong to. FreeBSD does not share this concept of a parent
module. Instead add macros which define the prefix to use for the
module parameters in the LinuxKPI consumers.
While at it convert all "bool" LinuxKPI module parameters to "byte"
type, because we don't have a "bool" type of SYSCTL in FreeBSD.
Sponsored by: Mellanox Technologies
MFC after: 1 week
Currently all radiotap structures in ral(4) are packed, but are not
aligned, which causes ral based devices to crash when one does
'ifconfig wlan0 up' for a wlan interface with a ral wlandev on arches
that care about structure alignment (e.g., MIPS).
Adding an aligned attribute helps fix this problem and ral devices
can be properly brought up.
Reviewed by: adrian
Sponsored by: Smartcom - Bulgaria AD
And release IDT vector before releasing ISR resources on interrupt
teardown path. We still have some work to do on the interrupt tearing
down path.
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6519
Though it is highly unlikely this function would fail w/ BUS_DMA_WAITOK,
we had better to check its return value; better safe then sorry here.
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6518
This was found while reworking the device tree nodes for dtsec to match the
Linux device tree. Instead of waiting and expecting later code to call
dpaa_portal_map_registers(), do the equivalent immediately upon mapping.
Otherwise, it's possible to access the pages before that function is called, and
hang the CPU.
This unifies handling of core, chip, and board-level device
matching, and adds support for matching device drivers
against the bus attach type (e.g. SoC vs WiFi adapter).
Core-level quirks on Broadcom's chipsets generally are specific
to some combination of chip model, core revision, chip
package (e.g. 12x9 SMT package), SROM revision, etc.
Unifying the match APIs for all three attribute types (core, chip,
board/srom) allows defining a single device quirk table that
matches across all of those attributes.
Submitted by: Landon Fuller <landonf@landonf.org>
Differential Revision: https://reviews.freebsd.org/D6515
than sc->areq. This is a bounds check to ensure we're not just cramming
arbitrarily sized nonsense into the driver and overflowing the heap.
PR: 209545
Submitted by: cturt@hardenedbsd.org
MFC after: 2 weeks
Option EFSYS_OPT_ALLOW_UNCONFIGURED_NIC disables check that the adapter
MAC address is not a local address (beginning 02).
Submitted by: Laurence Evans <levans at solarflare.com>
Reviewed by: gnn
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D6508
And
- Move message and event flags to vmbus_softc per-cpu data.
- Get rid of hv_setup_arg, which serves no purpose now.
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6502
* convert phy_getinfo() to take a "gmode" flag, rather than the siba
TGSHIGH flags and then check for 2GHz. This should ensure that
gmode is set correctly even on DUALPHY NICs.
* move the siba_powerup() call and the TGSHIGH decoding into a
call to bwn_is_bus_siba(), and return an error if it's called
on anything else. We don't yet do anything else, but when we do..
Tested:
* BCM4322, 11a STA
This is all for the bhnd(4) work in progress. It's enough to probe/attach
all the bhnd internals, but we're missing OTP support and some cleanup
code. And, well, all the rest of the bhnd(4) migration.
So no, this won't give you BCM43225 support. Sorry!
This patchset adds support to bhnd_chipc for sharing SYS_RES_MEMORY
resources with its children, allowing us to hang devices off of
bhnd_chipc that rely on access to a subset of the device register space
that bhnd_chipc itself must also allocate.
We could avoid most of this heavy lifting if RF_SHAREABLE+SYS_RES_MEMORY
wasn't limited to use with allocations at the same size/offset.
As a work-around, I implemented something similar to vga_pci.c, which
implements similar reference counting of of PCI BAR resources for its
children.
With these changes, chipc will use reference counting of SYS_RES_MEMORY
allocation/activation requests, to decide when to allocate/activate/
deactivate/release resources from the parent bhnd(4) bus.
The requesting child device is allocated a new resource from chipc's
rman, pointing to (possibly a subregion of) the refcounted bhnd resources
allocated by chipc.
Other resource types are just passed directly to the parent bhnd bus;
RF_SHAREABLE works just fine with IRQs.
I also lifted the SPROM device code out into a common driver, since this
now allows me to hang simple subclasses off of a common driver off of both
bhndb_pci and bhnd_chipc.
Tested:
* (landonf) Tested against BCM4331 and BCM4312, confirmed that SPROM still
attaches and can be queried.
Submitted by: Landon Fuller <landonf@landonf.org>
Reviewed by: mizkha@gmail.com
Differential Revision: https://reviews.freebsd.org/D6471
The EFSYS_PREEMPT_DISABLE() and EFSYS_PREEMPT_ENABLE() macros
were used to ensure correct timing of I2C operations. The APIs
for I2C operations have been removed, so these macros have no
callers.
Submitted by: Andy Moreton <amoreton at solarflare.com>
Sponsored by: Solarflare Communications, Inc.
MFC after: 1 week
It does not belong to the vmbus.
While I'm here rework the Hypercall setup, e.g. use busdma(9)
and avoid bit fields.
Discussed with: Jun Su <junsu microsoft com>
MFC after: 1 week
Sponsored by: Microsoft OSTC
Differential Revision: https://reviews.freebsd.org/D6445
The variable "size" stores number of words (4bytes). But the loop over
memory uses size as number of bytes to scan memory. As result it fetches
only 1/4th of memory.
This patch solves this problem and nvram2env fetches all NVRAM variables.
Test plan:
Pre-requisites: any MIPS board with ASCII-based NVRAM mapped into memory
* Add "device nvram2env" into kernel configuration
* Specify hints: base is mandatory (according to nvram2env(4))
hint.nvram.0.base=0x1c7f8000 (it's valid address for Asus RT-N53 with
flags = 0x4)
* Build & load kernel with bootverbose
Actual result: only part of nvram variables are found
Expected result: all variables are found
Submitted by: Michael Zhilin <mizhka@gmail.com>
Differential Revision: https://reviews.freebsd.org/D6466
This adds a BHND_BUS_GET_ATTACH_TYPE(); the primary use-case is to let
chipc make a coarse-grained determination as to whether UART, SPI, etc
drivers ought to be attached, and on fullmac devices, whether a real
CPU driver ought to be skipped for the ARM core, etc.
Tested:
* BCM4331 (BHND)
* BCM4312 (SIBA)
Submitted by: Landon Fuller <landonf@landonf.org>
Differential Revision: https://reviews.freebsd.org/D6492
Override global retry limit (which is set in R92C_RL) via per-frame
TX descriptor field. Obsoletes D3840 (should work better with 2+ vaps).
Tested with RTL8188EU and RTL8192CUS in STA mode (maxretry = [3-9]).
After r285994, sysctl(8) was fixed to use 273.15 instead of 273.20 as 0C
reference and as result, the temperature read in sysctl(8) now exibits a
+0.1C difference.
This commit fix the kernel references to match the reference value used in
sysctl(8) after r285994.
Sponsored by: Rubicon Communications (Netgate)
The GPIO hardware should not be owned by a single device, this defeats any
chance of use of the GPIO controller as an interrupt source.
ow(4) is now the only consumer of this 'feature' before we can remove it
for good.
Discussed with: ian, bsdimp
Fix panics which were present when BGX and PF module were unloaded.
Reviewed by: zbb
Obtained from: Semihalf
Sponsored by: Cavium
Differential Revision: https://reviews.freebsd.org/D6346
Add directory structure and fix dependencies to be able to
build and use Cavium VNIC driver as a module.
Reviewed by: zbb
Obtained from: Semihalf
Sponsored by: Cavium
Differential Revision: https://reviews.freebsd.org/D6345
utilizing previously unused arg field of struct ccb_notify_acknowledge.
This makes new QUERY TASK, QUERY TASK SET and QUERY ASYNC EVENT requests
really functional for CAM target mode drivers.
tested on the Pass 1.1 and 2.0 ThunderX machines in the Netperf cluster.
Reviewed by: jhb
Obtained from: ABT Systems Ltd
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D6453
Store the last doorbell write in the mlx5e_sq structure and write the
doorbell to the hardware when the transmit routine finishes
transmitting all queued mbufs.
Sponsored by: Mellanox Technologies
Tested by: Netflix
MFC after: 1 week
This patch implements a sysctl which allows setting a factor, N, for
how many work queue elements can be generated before requiring a
completion event. When a completion event happens the code simulates N
completion events instead of only one. When draining a transmit queue,
N-1 NOPs are transmitted at most, to force generation of the final
completion event. Further a timer is running every HZ ticks to flush
any remaining data off the transmit queue when the tx_completion_fact
> 1.
The goal of this feature is to reduce the PCI bandwidth needed when
transmitting data.
Sponsored by: Mellanox Technologies
Tested by: Netflix
MFC after: 1 week