This fixes issue in nvmecontrol(8), where clang throws a cast-align
warning when casting a __packed structure pointer to a uint32_t
pointer as part of printing raw hex output.
Reported by: dhw
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.
support to FreeBSD. A full description of the overall functionality
being added is below. nvmexpress.org defines NVM Express as "an optimized
register interface, command set and feature set fo PCI Express (PCIe)-based
Solid-State Drives (SSDs)."
This commit adds nvme(4) and nvd(4) driver source code and Makefiles
to the tree.
Full NVMe functionality description:
Add nvme(4) and nvd(4) drivers and nvmecontrol(8) for NVM Express (NVMe)
device support.
There will continue to be ongoing work on NVM Express support, but there
is more than enough to allow for evaluation of pre-production NVM Express
devices as well as soliciting feedback. Questions and feedback are welcome.
nvme(4) implements NVMe hardware abstraction and is a provider of NVMe
namespaces. The closest equivalent of an NVMe namespace is a SCSI LUN.
nvd(4) is an NVMe consumer, surfacing NVMe namespaces as GEOM disks.
nvmecontrol(8) is used for NVMe configuration and management.
The following are currently supported:
nvme(4)
- full mandatory NVM command set support
- per-CPU IO queues (enabled by default but configurable)
- per-queue sysctls for statistics and full command/completion queue
dumps for debugging
- registration API for NVMe namespace consumers
- I/O error handling (except for timeoutsee below)
- compilation switches for support back to stable-7
nvd(4)
- BIO_DELETE and BIO_FLUSH (if supported by controller)
- proper BIO_ORDERED handling
nvmecontrol(8)
- devlist: list NVMe controllers and their namespaces
- identify: display controller or namespace identify data in
human-readable or hex format
- perftest: quick and dirty performance test to measure raw
performance of NVMe device without userspace/physio/GEOM
overhead
The following are still work in progress and will be completed over the
next 3-6 months in rough priority order:
- complete man pages
- firmware download and activation
- asynchronous error requests
- command timeout error handling
- controller resets
- nvmecontrol(8) log page retrieval
This has been primarily tested on amd64, with light testing on i386. I
would be happy to provide assistance to anyone interested in porting
this to other architectures, but am not currently planning to do this
work myself. Big-endian and dmamap sync for command/completion queues
are the main areas that would need to be addressed.
The nvme(4) driver currently has references to Chatham, which is an
Intel-developed prototype board which is not fully spec compliant.
These references will all be removed over time.
Sponsored by: Intel
Contributions from: Joe Golio/EMC <joseph dot golio at emc dot com>
- Use callout(9) rather than timeout(9).
- Add a mutex as an I/O lock that protects the adapter and is used
for the I/O path.
- Add an sx lock as a configuration lock that protects the relationship
of configured volumes.
- Freeze the request queue when a DMA load is deferred with EINPROGRESS
and unfreeze the queue when the DMA callback is invoked.
- Explicitly poll the hardware while waiting to submit a command to
allow completed commands to free up slots in the command ring.
- Remove driver-wide 'initted' variable from mlx_*_fw_handshake() routines.
That state should be per-controller instead. Add it as an argument
since the first caller knows when it is the first caller.
- Remove explicit bus_space tag/handle and use bus_*() rather than
bus_space_*().
- Move duplicated PCI device ID probing into a mlx_pci_match() routine.
- Don't check for PCIM_CMD_MEMEN (the PCI bus will enable that when
allocating the resource) and use pci_enable_busmaster() rather than
manipulating the register directly.
Tested by: no one despite multiple requests (hope it works)
These are intended for software TX filtering support, where the NIC
decides there has been too many successive failues to a destination
and will filter it.
Although the filtering is done per-destination (via the keycache),
the state and queue is kept per-TID for now. It simplifies the overall
architecture design and locking.
Whilst here, add ATH_TID_UNLOCK_ASSERT().
* Don't treat high percentage failures as "sucessive failures" - high
MCS rates are very picky and will quite happily "fade" from low
to high failure % and back again within a few seconds. If they really
don't work, the aggregate will just plain fail.
* Only sample MCS rates +/- 3 from the current MCS. Sample will back off
quite quickly, so there's no need to sample _all_ MCS rates between
a high MCS rate and MCS0; there may be a lot of them.
* Modify the smoothing rate to be 75% rather than 95% - it's more adaptive
but it comes with a cost of being slightly less stable at times.
A per-node, hysterisis behaviour would be nicer.
(download microcode with offsets, save, and activate).
SATI translation layer was incorrectly using allocation length instead
of blocks, and was constructing the ATA command incorrectly.
Also change #define to specify that the 512 block size here is
specific for DOWNLOAD_MICROCODE, and does not relate to the device's
logical block size.
Submitted by: scottl (with small modifications)
MFC after: 3 days
trap checks (eg. printtrap()).
Generally this check is not needed anymore, as there is not a legitimate
case where curthread != NULL, after pcpu 0 area has been properly
initialized.
Reviewed by: bde, jhb
MFC after: 1 week
- Add constants for the rest of the fields in the PCI-express device
capability and control registers.
- Tweak some of the recently added PCI-e capability constants (always
use hex for offsets in config space, and include a shortened
version of the relevant register in the name of field constants).
MFC after: 1 week
I'm not sure where in the deep, distant past I found the AR_PHY_MODE
registers for half/quarter rate mode, but unfortunately that doesn't
seem to work "right" for non-AR9280 chips.
Specifically:
* don't touch AR_PHY_MODE
* set the PLL bits when configuring half/quarter rate
I've verified this on the AR9280 (5ghz fast clock) and the AR5416.
The AR9280 works in both half/quarter rate; the AR5416 unfortunately
only currently works at half rate. It fails to calibrate on quarter rate.
No, this isn't HT/5 and HT/10 support. This is the 11a half/quarter
rate support primarily used by the 4.9GHz and GSM band regulatory
domains.
This is definitely a work in progress.
TODO:
* everything in the last commit;
* lots more interoperability testing with the AR5212 half/quarter rate
support for the relevant chips;
* Do some interop testing on half/quarter rate support between _all_
the 11n chips - AR5416, AR9160, AR9280 (and AR9285/AR9287 when 2GHz
half/quarter rate support is coded up.)
used when running the chips in half/quarter rate.
This sets up some default parameters which are then overridden by the
driver (which manually configures things like slot timing at interface
start time.)
Although this is a copy-and-modify from the AR5212 HAL, I did peek
at the reference HAL and the ath9k driver to see what they did.
Ath9k in particular doesn't hard-code this - instead, their version
of ar5416InitUserSettings() does all of the relevant math.
TODO:
* do the math, not hard code things!
* fix the mac clock calculation for the AR9287; since it runs the
MAC clock at a higher rate, requiring all the duration calculations
to change;
* Do a whole lot more validation for half/quarter rates.
Obtained from: Qualcomm Atheros, Linux ath9k
Some of the math is a little wrong thanks to clocks in 11a mode running
at 44MHz when in fast clock mode (rather than 40MHz, which the chips
before AR9280 ran 11a in). That'll have to be addressed in a future commit.
This fixes the incorrect slot (and likely ACK/RTS timeout) values
which I see when enabling half/quarter rate support on the AR9280.
The resulting math matches the expected calculated default values.
of the DWC OTG is very simple in PIO mode, and we need to re-transmit
data when NAK is received among other things. We probably will need
to implement some kind of rate limitation on the NAK-ing.
drivers:
- Remove scsi_low_pisa.*, they were unused.
- Remove <compat/netbsd/physio_proc.h> and calls to the stubs in that
header. They were empty nops.
- Retire sl_xname and use device_get_nameunit() and device_printf() with
the underlying device_t instead.
- Remove unused {ct,ncv,nsp,stg}print() functions.
- Remove empty SOFT_INTR_REQUIRED() macro and the unused sl_irq member.
(1022) in HPET. But according to report they still haven't fixed problem
with level-triggered interrupts.
Make workaround used for earlier chipsets apply to this new ID also.
PR: amd64/171355
MFC after: 3 days
The DWC OTG host mode support should still be considered
experimental. Isochronous support for DWC OTG is not
fully implemented. Some code added derives from
Aleksandr Rybalko's dotg.c driver.
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.
This had the side effect of clearing HAL_TXDESC_CLRDMASK for a bunch of
frames, meaning they'd end up being potentially filtered if there were
an error. This is fine in the previous world as they'd just be
software retried but now that I'm working on filtered frames, these
descriptors would be endlessly retried until another valid frame would
come along that had CLRDMASK set.
do a BUS_DMASYNC_POSTWRITE over the DMA map. The way it currently is would
only do POSTREAD for read transactions.
Submitted by: Sascha Wildner
MFC after: 1 month
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
NetBSD/pc98 was never merged into the main NetBSD tree and is no longer
developed. Adding locking to these drivers would have made the compat
shims hard to impossible to maintain, so remove the shims to ease
future changes.
These changes were verified by md5. Some additional shims can be removed
that do affect the compiled results that I will probably do in another
round.
Approved by: nyan (tentatively)
have been chosen based on the bit names in the PCI Express Base
Specification 3.0, and to match the predominant style of the existing
bit definitions.
MFC after: 1 week
twe_start() to simulate the behavior on 4.x where the driver dropped
spl to allow interrupts to run to free up space in the command
queue. Be careful to only poll if we are going to make at least
one more attempt to queue the current command. Also, when polling,
be careful to not call twe_startio() to queue more commands to avoid
recursion.
- Move the buffer for formatting AEN messages into the softc instead of
using a single driver-wide static buffer.
Requested by: scottl (1)
Tested by: Mike Tancsa @ Sentex
generator, found on IvyBridge and supposedly later CPUs, accessible
with RDRAND instruction.
From the Intel whitepapers and articles about Bull Mountain, it seems
that we do not need to perform post-processing of RDRAND results, like
AES-encryption of the data with random IV and keys, which was done for
Padlock. Intel claims that sanitization is performed in hardware.
Make both Padlock and Bull Mountain random generators support code
covered by kernel config options, for the benefit of people who prefer
minimal kernels. Also add the tunables to disable hardware generator
even if detected.
Reviewed by: markm, secteam (simon)
Tested by: bapt, Michael Moll <kvedulv@kvedulv.de>
MFC after: 3 weeks
Fix the strong signal diversity capability setting - I had totally
messed up the indentation.
Set the default values to match what's in the .ini for now, rather than
what values I had previously gleaned from places. This seems to work
quite well for the early AR5212 NICs I have. Of course, later NICs
have different PHYs and the radar configuration is very card/board
dependent..
Tested:
* ath1: AR5212 mac 5.3 RF5111 phy 4.1
ath1: 2GHz radio: 0x0023; 5GHz radio: 0x0017
This detects 1, 5, 25, 50, 75, 100uS pulses reliably (with no interference.)
However, 10uS pulses don't detect reliably. That may be around the
transition between short and long pulses so some further tuning may
improve things.
up on (at least) the AR5413.
The 30 second summary - if a CRC error frame comes in during PHY error
processing, that CRC bit will be set for all subsequent frames until
a non-CRC error frame is processed.
So to allow for accurate PHY error processing (Radar, and ANI on the AR5212
HAL chips) just tag the frame as being both CRC and PHY - let the driver
decide what to do with it.
PR: kern/169362
this case, allocate a plain mbuf and copy the frame into it, then send the
copy up the stack, leaving the original mbuf+cluster in place in the
receive ring for immediate re-use. This saves a trip through 2 of the
3 zones of the compound mbuf allocator, a trip through busdma, and a trip
through the 1 of the 3 mbuf destructors. For our load at Netflix, this can
lower CPU consumption by as much as 20%. The copy algorithm is based on
investigative work from Luigi Rizzo earlier in the year.
Reviewed by: jfv
Obtained from: Netflix
the interface is brought up. Without this, the boot time interrupt
round-robin assignment does not think the allocated interrupt resources
are active and leaves them assigned to CPU 0.
While here, add descriptive tags to each interrupt handler when MSI-X
is used.
Reviewed by: np
MFC after: 1 week
some HAL definitions rather than local definitions.
The original source (ath9k) pulled this stuff from the QCA driver and
removed the HAL_* prefix. I'm just restoring the correct order of things.
Obtained from: Qualcomm Atheros
This routine is intended only for commands such as INQUIRY where
the controller may fill out a smaller amount of data than allocated
by the host.
The end result of this bug was that isci(4) would report non-zero
resid for successful SCSI_UNMAP commands.
Sponsored by: Intel
MFC after: 3 days
In fact, bus_dmamem_alloc() happily NULLs the dmat pointer passed in,
before replacing it with its own.
This fixes a MIPS crash when kldload'ing if_ath/if_ath_pci -
bus_dmamap_destroy() was passed in a NULL dmat pointer and was doing
all kinds of very bad things.
Reviewed by: scottl
This is a re-implementation based on the reference carrier code
for the AR5413.
Tested:
* Pulse detection for AR5212 and AR5413, to ensure the
correct behaviour for both chips
PR: kern/170904
Obtained from: Qualcomm Atheros
The comparison assumes maxFirstepLevel is a count, rather than a maximum
value. The array is 3 entries in size however 'maxFirstepLevel' is 2.
This bug also exists in the AR5212 HAL.
MPSAFE.
- Preallocate a full set of QCBs during attach rather than allocating new
ones on demand to avoid allocations in the I/O path.
- Remove the explicit bus space tag/handle and use bus_*() on the
relevant 'struct resource' instead.
- Defer logical drive probing to an intrhook.
- Fix ida_detach() to detach and delete child devices (logical drives).
- Update the DMA handling to support EINPROGRESS by moving the work to
submit a mapped request into the bus_dma callback routine as well as
add support for freezing the queue when EINPROGRESS is encountered.
Tested by: Marco Steinbach coco executive-computing de
Terasic DE-4 board. Allow LED configuration to be set using loader
tunables, not just from userspace, and preconfigure LED 8 as a kernel
heartbeat. For now, this is a Nexus-attached, BERI-only driver, but it
could be used with other hard and soft cores on Altera FPGAs as well, in
principle.
Sponsored by: DARPA, AFRL
used with Terasic's DE-4 and other similar FPGA boards. This display
is 800x480 and includes a capacitive touch screen, multi-touch
gesture recognition, etc. This device driver depends on a Cambridge-
provided IP core that allows the MTL device to be hooked up to the
Altera Avalon SoC bus, and also provides a VGA-like text frame buffer.
Although it is compiled as a single device driver, it actually
implements a number of different device nodes exporting various
aspects of this multi-function device to userspace:
- Simple memory-mapped driver for the MTL 24-bit pixel frame buffer.
- Simple memory-mapped driver for the MTL control register set.
- Simple memory-mapped driver for the MTL text frame buffer.
- syscons attachment for the MTL text frame buffer.
This driver attaches directly to Nexus as is common for SoC device
drivers, and for the time being is considered BERI-specific, although
in principle it might be used with other hard and soft cores on
Altera FPGAs.
Control registers, including touchscreen input, are simply memory
mapped; in the future it would be desirable to hook up a more
conventional device node that can stream events, support kqueue(2)/
poll(2)/select(2), etc.
This is the first use of syscons on MIPS, as far as I can tell, and
there are some loose ends, such as an inability to use the hardware
cursor. More fundamentally, it appears that syscons(4) assumes that
either a host is PC-like (i386, amd64) *or* it must be using a
graphical frame buffer. While the MTL supports a graphical frame
buffer, using the text frame buffer is preferable for console use.
Fixing this issue in syscons(4) requires non-trivial changes, as the
text frame buffer support assumes that direct memory access can be
done to the text frame buffer without using bus accessor methods,
which is not the case on MIPS. As a workaround for this, we instead
double-buffer and pretend to be a graphical frame buffer exposing
text accessor methods, leading to some quirks in syscons behaviour.
Sponsored by: DARPA, AFRL
The driver attempts to support all documented parts, but has only been
tested with the 512Mbit part on the Terasic DE4 FPGA board. It should be
trivial to adapt the driver's attach routine to other embedded boards
using with any parts in the family.
Also import isfctl(8) which can be used to erase sections of the flash.
Sponsored by: DARPA, AFRL
which presents a UART-like interface over the Avalon bus that can be
addressed over JTAG. This IP core proves extremely useful, allowing us to
connect trivially to the FreeBSD console over JTAG for FPGA-embedded hard
and soft cores. As interrupts are optionally configured for this soft
core, we support both interrupt-driven and polled modes of operation,
which must be selected using device.hints. UART instances appear in /dev
as ttyu0, ttyu1, etc.
However, it also contains a number of quirks, which make it difficult to
tell when JTAG is connected, and some buffering issues. We work around
these as best we can, using various heuristics.
While the majority of this device driver is not only not BERI-specific,
but also not MIPS-specific, for now add its defines in the BERI files
list, as the console-level parts are aware of where the first JTAG UART
is mapped on Avalon, and contain MIPS-specific address translation, to
use before Newbus and device.hints are available.
Sponsored by: DARPA, AFRL
which can be synthesised in Altera FPGAs. An altera_sdcardc device
probes during the boot, and /dev/altera_sdcard devices come and go as
inserted and removed. The device driver attaches directly to the
Nexus, as is common for system-on-chip device drivers.
This IP core suffers a number of significant limitations, including a
lack of interrupt-driven I/O -- we must implement timer-driven polling,
only CSD 0 cards (up to 2G) are supported, there are serious memory
access issues that require the driver to verify writes to memory-mapped
buffers, undocumented alignment requirements, and erroneous error
returns. The driver must therefore work quite hard, despite a fairly
simple hardware-software interface. The IP core also supports at most
one outstanding I/O at a time, so is not a speed demon.
However, with the above workarounds, and subject to performance
problems, it works quite reliably in practice, and we can use it for
read-write mounts of root file systems, etc.
Sponsored by: DARPA, AFRL
CPU cores on Altera FPGAs. The device driver allows memory-mapped devices
on Altera's Avalon SoC bus to be exported to userspace via device nodes.
device.hints directories dictate device name, permissible access methods,
physical address and length, and I/O alignment. Devices can be accessed
using read(2)/write(2), but also memory mapped in userspace using mmap(2).
Devices attach directly to the Nexus, as is common for embedded device
drivers; in the future something more mature might be desirable. There is
currently no facility to support directing device-originated interrupts to
userspace.
In the future, this device driver may be renamed to socgen(4), as it can
in principle also be used with other system-on-chip (SoC) busses, such as
Axi on ASICs and FPGAs. However, we have only tested it on Avalon busses
with memory-mapped ROMs, frame buffers, etc.
Sponsored by: DARPA, AFRL
on PowerPC support. This was clearly not something syscons was
designed to do (very specific assumptions about the nature of VGA
consoles on PCs), but fortunately others have long since blazed
the way on making it work regardless of that.
Sponsored by: DARPA, AFRL
relative priority of the gxemul console in line with its role as a
"seiral console". This allows it to override video console drivers
that might otherwise take precdence, subject to that boot flag.
Sponsored by: DARPA, AFRL
This ensures that any ccbs which immediately start during the call to
xpt_release_devq see an accurate picture of the frozen_lun_mask.
Sponsored by: Intel
MFC after: 3 days
DFS parameters fetched from the HAL.
Check whether the specific chipset supports RADAR reporting before
enabling DFS; or some of the (unset) DFS methods may fail.
Tested:
* AR5210 (correctly didn't enable radar PHY reporting)
* AR5212 (correctly enabled radar PHY reporting w/ the correct default
parameters.)
TODO:
* Now that I have this capability check in place, I could remove the
(empty) DFS methods from AR5210/AR5211.
* Test on AR5416, AR9160, AR9280.
PR: kern/170904
* mfp support;
* 4.9ghz support in the HAL;
* device type - specifically, the bus type and whether it's a HB63
NIC (which requires some subtle chainmask handling differences
in the AR5416 HAL.)
Obtained from: Qualcomm Atheros
Report errors indicated by the transport. If this is too chatty, I'll
throw it behind a debug write.
Remove commented out debugs that are no longer useful.
This addresses kernel panic observed when sending SCSI UNMAP
commands to SATA disks attached to isci(4).
1) Flesh out callback routines to allocate/free buffers needed for
translating SCSI UNMAP data to ATA DSM data.
2) Add controller-level pool for storing buffers previously allocated
for UNMAP translation, to lessen chance of no buffer available
under memory pressure.
3) Ensure driver properly handles case where buffer pool is empty
and contigmalloc returns NULL.
Sponsored by: Intel
Reported by: Maksim Yevmenkin <max at netflix dot com>
Discussed with: scottl
MFC after: 3 days
evicted from the syncache but a later syncache_expand succeeds because
of syncookies. The TOE driver has to resort to more direct means to
install its hooks in the socket in this case.
