If an SRP target is no longer reachable and srp_reset_host() fails to
reconnect then ib_srp will invoke scsi_remove_host(). That function
will invoke __scsi_remove_device() for each LUN. And that last
function will change the device state from SDEV_TRANSPORT_OFFLINE into
SDEV_CANCEL. Certain user space software, e.g. older versions of
multipathd, continue queueing I/O to SCSI devices that are in the
SDEV_CANCEL state.
If these I/O requests are submitted as SG_IO that means that the
REQ_PREEMPT flag will be set and hence that these requests will be
passed to srp_queuecommand(). These requests will time out. If new
requests are queued fast enough from user space these active requests
will prevent __scsi_remove_device() to finish.
Avoid this by failing I/O requests in the SDEV_CANCEL state if the
transport is offline. Introduce a new variable to keep track of the
transport state instead of failing requests if (!target->connected ||
target->qp_in_error), so that the SCSI error handler has a chance to
retry commands after a transport layer failure occurred.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Cc: <stable@vger.kernel.org> # 3.8
Signed-off-by: Roland Dreier <roland@purestorage.com>
If a SCSI command times out it is passed to the SCSI error
handler. The SCSI error handler will try to abort the commands that
timed out. If aborting fails, a device reset will be attempted. If
the device reset also fails a host reset will be attempted. If the
host reset also fails the whole procedure will be repeated.
srp_abort() and srp_reset_device() fail for a QP in the error state.
srp_reset_host() fails after host removal has started. Hence if the
SCSI error handler gets invoked after host removal has started and
with the QP in the error state an endless loop will be triggered.
Modify the SCSI error handling functions in ib_srp as follows:
- Abort SCSI commands properly even if the QP is in the error state.
- Make srp_reset_host() reset SCSI requests even after host removal
has already started or if reconnecting fails.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dave@thedillows.org>
Cc: <stable@vger.kernel.org> # 3.8
Signed-off-by: Roland Dreier <roland@purestorage.com>
Do not send a task management function if sending will fail anyway
because either there is no RDMA/RC connection or the QP is in the
error state.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dave@thedillows.org>
Cc: <stable@vger.kernel.org> # 3.8
Signed-off-by: Roland Dreier <roland@purestorage.com>
Remove an assignment that incorrectly overwrites the connection state
update by srp_connect_target().
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dave@thedillows.org>
Cc: <stable@vger.kernel.org> # 3.8
Signed-off-by: Roland Dreier <roland@purestorage.com>
Commit 1fb9fed6d4 ("IB/qib: Fix QP RCU sparse warning") broke QP
hash list deletion in qp_remove() badly.
This patch restores the former for loop behavior, while still fixing
the sparse warnings.
Cc: <stable@vger.kernel.org>
Reviewed-by: Gary Leshner <gary.s.leshner@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
After commit b13912bbb4 ("IPoIB: Call skb_dst_drop() once skb is
enqueued for sending"), using connected mode and running multithreaded
iperf for long time, ie
iperf -c <IP> -P 16 -t 3600
results in a crash.
After the above-mentioned patch, the driver is calling skb_orphan() and
skb_dst_drop() after calling post_send() in ipoib_cm.c::ipoib_cm_send()
(also in ipoib_ib.c::ipoib_send())
The problem with this is, as is written in a comment in both routines,
"it's entirely possible that the completion handler will run before we
execute anything after the post_send()." This leads to running the
skb cleanup routines simultaneously in two different contexts.
The solution is to always perform the skb_orphan() and skb_dst_drop()
before queueing the send work request. If an error occurs, then it
will be no different than the regular case where dev_free_skb_any() in
the completion path, which is assumed to be after these two routines.
Signed-off-by: Shlomo Pongratz <shlomop@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
CONFIG_HOTPLUG is going away as an option. As a result, the __dev*
markings need to be removed.
This change removes the use of __devinit, __devexit_p, __devinitdata,
and __devexit from these drivers.
Based on patches originally written by Bill Pemberton, but redone by me
in order to handle some of the coding style issues better, by hand.
Cc: Bill Pemberton <wfp5p@virginia.edu>
Cc: Tom Tucker <tom@opengridcomputing.com>
Cc: Steve Wise <swise@opengridcomputing.com>
Cc: Roland Dreier <roland@kernel.org>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Cc: Christoph Raisch <raisch@de.ibm.com>
Cc: Mike Marciniszyn <infinipath@intel.com>
Cc: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
- cxgb4 changes to fix lookup engine hash collisions
- mlx4 changes to make flow steering usable
- fix to IPoIB to avoid pinning dst reference for too long
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)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=lDXt
-----END PGP SIGNATURE-----
Merge tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband
Pull more infiniband changes from Roland Dreier:
"Second batch of InfiniBand/RDMA changes for 3.8:
- cxgb4 changes to fix lookup engine hash collisions
- mlx4 changes to make flow steering usable
- fix to IPoIB to avoid pinning dst reference for too long"
* tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
RDMA/cxgb4: Fix bug for active and passive LE hash collision path
RDMA/cxgb4: Fix LE hash collision bug for passive open connection
RDMA/cxgb4: Fix LE hash collision bug for active open connection
mlx4_core: Allow choosing flow steering mode
mlx4_core: Adjustments to Flow Steering activation logic for SR-IOV
mlx4_core: Fix error flow in the flow steering wrapper
mlx4_core: Add QPN enforcement for flow steering rules set by VFs
cxgb4: Add LE hash collision bug fix path in LLD driver
cxgb4: Add T4 filter support
IPoIB: Call skb_dst_drop() once skb is enqueued for sending
Retries active opens for INUSE errors.
Logs any active ofld_connect_wr error replies.
Sends ofld_connect_wr on same ctrlq. It needs to go on the same control txq as
regular CPL active/passive messages.
Retries on active open replies with EADDRINUSE.
Uses active open fw wr only if active filter region is set.
Adds stat for ofld_connect_wr failures.
This patch also adds debugfs file to show endpoints.
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
It establishes passive open connection through firmware work request. Passive
open connection will go through this path as now instead of listening server we
create a server filter which will redirect the incoming SYN packet to the
offload queue. After this driver tries to establish the connection using
firmware work request.
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
It enables establishing active open connection using fw_ofld_connection work
request when cpl_act_open_rpl says TCAM full error which may be because
of LE hash collision. Current support is only for IPv4 active open connections.
Sets ntuple bits in active open requests. For T4 firmware greater than 1.4.10.0
ntuple bits are required to be set.
Adds nocong and enable_ecn module parameter options.
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
[ Move all FW return values to t4fw_api.h. - Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
Currently, IPoIB delays collecting send completions for TX packets in
order to batch work more efficiently. It does skb_orphan() right after
queuing the packets so that destructors run early, to avoid problems
like holding socket send buffers for too long (since we might not
collect a send completion until a long time after the packet is
actually sent).
However, IPoIB clears IFF_XMIT_DST_RELEASE because it actually looks
at skb_dst() to update the PMTU when it gets a too-long packet. This
means that the packets sitting in the TX ring with uncollected send
completions are holding a reference on the dst. We've seen this lead
to pathological behavior with respect to route and neighbour GC. The
easy fix for this is to call skb_dst_drop() when we call skb_orphan().
Also, give packets sent via connected mode (CM) the same skb_orphan()
/ skb_dst_drop() treatment that packets sent via datagram mode get.
Signed-off-by: Roland Dreier <roland@purestorage.com>
Pull powerpc update from Benjamin Herrenschmidt:
"The main highlight is probably some base POWER8 support. There's more
to come such as transactional memory support but that will wait for
the next one.
Overall it's pretty quiet, or rather I've been pretty poor at picking
things up from patchwork and reviewing them this time around and Kumar
no better on the FSL side it seems..."
* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc: (73 commits)
powerpc+of: Rename and fix OF reconfig notifier error inject module
powerpc: mpc5200: Add a3m071 board support
powerpc/512x: don't compile any platform DIU code if the DIU is not enabled
powerpc/mpc52xx: use module_platform_driver macro
powerpc+of: Export of_reconfig_notifier_[register,unregister]
powerpc/dma/raidengine: add raidengine device
powerpc/iommu/fsl: Add PAMU bypass enable register to ccsr_guts struct
powerpc/mpc85xx: Change spin table to cached memory
powerpc/fsl-pci: Add PCI controller ATMU PM support
powerpc/86xx: fsl_pcibios_fixup_bus requires CONFIG_PCI
drivers/virt: the Freescale hypervisor driver doesn't need to check MSR[GS]
powerpc/85xx: p1022ds: Use NULL instead of 0 for pointers
powerpc: Disable relocation on exceptions when kexecing
powerpc: Enable relocation on during exceptions at boot
powerpc: Move get_longbusy_msecs into hvcall.h and remove duplicate function
powerpc: Add wrappers to enable/disable relocation on exceptions
powerpc: Add set_mode hcall
powerpc: Setup relocation on exceptions for bare metal systems
powerpc: Move initial mfspr LPCR out of __init_LPCR
powerpc: Add relocation on exception vector handlers
...
Pull target updates from Nicholas Bellinger:
"It has been a very busy development cycle this time around in target
land, with the highlights including:
- Kill struct se_subsystem_dev, in favor of direct se_device usage
(hch)
- Simplify reservations code by combining SPC-3 + SCSI-2 support for
virtual backends only (hch)
- Simplify ALUA code for virtual only backends, and remove left over
abstractions (hch)
- Pass sense_reason_t as return value for I/O submission path (hch)
- Refactor MODE_SENSE emulation to allow for easier addition of new
mode pages. (roland)
- Add emulation of MODE_SELECT (roland)
- Fix bug in handling of ExpStatSN wrap-around (steve)
- Fix bug in TMR ABORT_TASK lookup in qla2xxx target (steve)
- Add WRITE_SAME w/ UNMAP=0 support for IBLOCK backends (nab)
- Convert ib_srpt to use modern target_submit_cmd caller + drop
legacy ioctx->kref usage (nab)
- Convert ib_srpt to use modern target_submit_tmr caller (nab)
- Add link_magic for fabric allow_link destination target_items for
symlinks within target_core_fabric_configfs.c code (nab)
- Allocate pointers in instead of full structs for
config_group->default_groups (sebastian)
- Fix 32-bit highmem breakage for FILEIO (sebastian)
All told, hch was able to shave off another ~1K LOC by killing the
se_subsystem_dev abstraction, along with a number of PR + ALUA
simplifications. Also, a nice patch by Roland is the refactoring of
MODE_SENSE handling, along with the addition of initial MODE_SELECT
emulation support for virtual backends.
Sebastian found a long-standing issue wrt to allocation of full
config_group instead of pointers for config_group->default_group[]
setup in a number of areas, which ends up saving memory with big
configurations. He also managed to fix another long-standing BUG wrt
to broken 32-bit highmem support within the FILEIO backend driver.
Thank you again to everyone who contributed this round!"
* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending: (50 commits)
target/iscsi_target: Add NodeACL tags for initiator group support
target/tcm_fc: fix the lockdep warning due to inconsistent lock state
sbp-target: fix error path in sbp_make_tpg()
sbp-target: use simple assignment in tgt_agent_rw_agent_state()
iscsi-target: use kstrdup() for iscsi_param
target/file: merge fd_do_readv() and fd_do_writev()
target/file: Fix 32-bit highmem breakage for SGL -> iovec mapping
target: Add link_magic for fabric allow_link destination target_items
ib_srpt: Convert TMR path to target_submit_tmr
ib_srpt: Convert I/O path to target_submit_cmd + drop legacy ioctx->kref
target: Make spc_get_write_same_sectors return sector_t
target/configfs: use kmalloc() instead of kzalloc() for default groups
target/configfs: allocate only 6 slots for dev_cg->default_groups
target/configfs: allocate pointers instead of full struct for default_groups
target: update error handling for sbc_setup_write_same()
iscsit: use GFP_ATOMIC under spin lock
iscsi_target: Remove redundant null check before kfree
target/iblock: Forward declare bio helpers
target: Clean up flow in transport_check_aborted_status()
target: Clean up logic in transport_put_cmd()
...
The terminate timer needs to be initialized just once.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
To avoid nes tcp_timer crash for SMP architectures, add_timer is
replaced with mod_timer.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Make it possible to disconnect the IB RC connection used by the SRP
protocol to communicate with a target.
Have the SRP transport layer create a sysfs "delete" attribute for
initiator drivers that support this functionality.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Robert Jennings <rcj@linux.vnet.ibm.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Now that SRP recreates the CM ID, QP, and CQ for each connection,
there is no need to wait for the timewait state to complete.
Signed-off-by: Vu Pham <vu@mellanox.com>
Signed-off-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Roland Dreier <roland@purestorage.com>
HW QP FATAL errors persist over a reset operation, but we can recover
from that by recreating the QP and associated CQs for each connection.
Creating a new QP/CQ also completely forecloses any possibility of
getting stale completions or packets on the new connection.
Signed-off-by: Ishai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
[ updated to current code from OFED, cleaned up commit message ]
Signed-off-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Only queue removal work after having changed the target state
into SRP_TARGET_REMOVED and not if that state was already equal
to SRP_TARGET_REMOVED. That allows us to remove the state
SRP_TARGET_DEAD. Add a call to srp_disconnect_target() in
srp_remove_target() -- due to previous changes it is now safe to
invoke that function even if the IB connection has already
been disconnected. This change allows us to replace the target
removal code in srp_remove_one() by an (indirect) call to
srp_remove_target(). Rename srp_target_port.work into
srp_target_port.remove_work to reflect its usage.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Keep track of the connection state. Only report QP errors while
connected. Only invoke ib_send_cm_dreq() when connected so that
invoking srp_disconnect_target() after having received a DREQ does not
cause an error message to be printed.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
If the RDMA RC connection is closed, tell the SCSI mid-layer to
terminate all pending commands instead of only the first.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Introduce the function srp_handle_qp_err(), change the type of
qp_in_error from int into bool and move the initialization of that
variable from srp_reconnect_target() to srp_connect_target().
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Since scsi_remove_host() has been modified so that SCSI error handling
functions will no longer be invoked after scsi_remove_host() returns,
the test at the start of srp_send_tsk_mgmt() is now superfluous.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Some SCSI upper layer drivers, e.g. sd, issue SCSI commands from
inside scsi_remove_host() (see the sd_shutdown() call in sd_remove()).
Make sure that these commands have a chance to reach the SCSI device.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Block the SCSI host while reconnecting instead of representing the
reconnection activity as a distinct SRP target state. This allows us
to eliminate the target state SRP_TARGET_CONNECTING.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Increase the block layer timeout for disks so that it is above the
InfiniBand transport layer timeout.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: David Dillow <dillowda@ornl.gov>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Problem reported by Dan Carpenter <dan.carpenter@oracle.com>:
The patch 3c86aa70bf: "RDMA/cm: Add RDMA CM support for IBoE
devices" from Oct 13, 2010, leads to the following warning:
net/sunrpc/xprtrdma/svc_rdma_transport.c:722 svc_rdma_create()
error: passing non neg 1 to ERR_PTR
This bug would result in a NULL dereference. svc_rdma_create() is
supposed to return ERR_PTRs or valid pointers, but instead it returns
ERR_PTRs, valid pointers and 1.
The call tree is:
svc_rdma_create()
=> rdma_bind_addr()
=> cma_acquire_dev()
=> find_gid_port()
rdma_bind_addr() should return a valid errno. Fix this by having
find_gid_port() also return a valid errno. If we can't find the
specified GID on a given port, return -EADDRNOTAVAIL, rather than
-EAGAIN, to better indicate the error. We also drop using the
special return value of '1' and instead pass through the error
returned by the underlying verbs call. On such errors, rather
than aborting the search, we simply continue to check the next
device/port.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
lockdep warns about taking a hard-irq-unsafe lock (sriov->id_map_lock)
inside a hard-irq-safe lock (sriov->going_down_lock).
Since id_map_lock is never taken in the interrupt context, we can
simply reverse the order of taking the two spinlocks, thus avoiding
the warning and the depencency.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This patch converts the TMR path in srpt_handle_tsk_mgmt() to use
target_submit_tmr() with TARGET_SCF_ACK_KREF flag usage.
v2: Drop ununused res in target_submit_tmr (Fengguang.Wu)
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Roland Dreier <roland@kernel.org>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
This patch converts the main srpt_handle_cmd() I/O path to use modern
target_submit_cmd() with TARGET_SCF_ACK_KREF flag usage. This includes
dropping the original internal ioctx->kref + srpt_put_send_ioctx() usage
in favor of target_put_sess_cmd() w/ se_cmd_t->cmd_kref within ib_srpt
response callbacks.
It also updates srpt_abort_cmd() to call target_put_sess_cmd() for
completion of aborted commands, and adds target_wait_for_sess_cmds() into
srpt_release_channel_work() to allow outstanding I/O to complete during
session shutdown.
Also, go ahead and update srpt_handle_tsk_mgmt() to make the remaining
transport_init_se_cmd() to setup the ioctx->cmd with se_tmr_req.
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Roland Dreier <roland@kernel.org>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
Use WARN rather than printk followed by WARN_ON(1), for conciseness.
A simplified version of the semantic patch that makes this transformation
is as follows: (http://coccinelle.lip6.fr/)
// <smpl>
@@
expression list es;
@@
-printk(
+WARN(1,
es);
-WARN_ON(1);
// </smpl>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Use WARN rather than printk followed by WARN_ON(1), for conciseness.
A simplified version of the semantic patch that makes this transformation
is as follows: (http://coccinelle.lip6.fr/)
// <smpl>
@@
expression list es;
@@
-printk(
+WARN(1,
es);
-WARN_ON(1);
// </smpl>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
ConnectX-3 devices can use either 64- or 32-byte completion queue
entries (CQEs) and event queue entries (EQEs). Using 64-byte
EQEs/CQEs performs better because each entry is aligned to a complete
cacheline. This patch queries the HCA's capabilities, and if it
supports 64-byte CQEs and EQES the driver will configure the HW to
work in 64-byte mode.
The 32-byte vs 64-byte mode is global per HCA and not per CQ or EQ.
Since this mode is global, userspace (libmlx4) must be updated to work
with the configured CQE size, and guests using SR-IOV virtual
functions need to know both EQE and CQE size.
In case one of the 64-byte CQE/EQE capabilities is activated, the
patch makes sure that older guest drivers that use the QUERY_DEV_FUNC
command (e.g as done in mlx4_core of Linux 3.3..3.6) will notice that
they need an update to be able to work with the PPF. This is done by
changing the returned pf_context_behaviour not to be zero any more. In
case none of these capabilities is activated that value remains zero
and older guest drivers can run OK.
The SRIOV related flow is as follows
1. the PPF does the detection of the new capabilities using
QUERY_DEV_CAP command.
2. the PPF activates the new capabilities using INIT_HCA.
3. the VF detects if the PPF activated the capabilities using
QUERY_HCA, and if this is the case activates them for itself too.
Note that the VF detects that it must be aware to the new PF behaviour
using QUERY_FUNC_CAP. Steps 1 and 2 apply also for native mode.
User space notification is done through a new field introduced in
struct mlx4_ib_ucontext which holds device capabilities for which user
space must take action. This changes the binary interface so the ABI
towards libmlx4 exposed through uverbs is bumped from 3 to 4 but only
when **needed** i.e. only when the driver does use 64-byte CQEs or
future device capabilities which must be in sync by user space. This
practice allows to work with unmodified libmlx4 on older devices (e.g
A0, B0) which don't support 64-byte CQEs.
In order to keep existing systems functional when they update to a
newer kernel that contains these changes in VF and userspace ABI, a
module parameter enable_64b_cqe_eqe must be set to enable 64-byte
mode; the default is currently false.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Merge my own merge branch to get various fixes from there
and upstream, especially the hvc console tty refcouting fixes
which which testing is quite a bit harder...
Signed-off-by: Alan Cox <alan@linux.intel.com>
Acked-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Use WARN() rather than printk() followed by WARN_ON(1), for conciseness.
A simplified version of the semantic patch that makes this transformation
is as follows: (http://coccinelle.lip6.fr/)
// <smpl>
@@
expression list es;
@@
-printk(
+WARN(1,
es);
-WARN_ON(1);
// </smpl>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
[ Remove extra KERN_ERR from WARN() format. - Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
Locking fix to prevent race conditions. Fpdus (per qp) need to be
forwarded to hardware in the order of their sequence numbers.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Donald Wood <Donald.E.Wood@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Fix for incorrect ip header address when forwarding fpdus to hardware.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
