The ib_cm can incorrectly detect a stale connection (a new connection
request for a QPN that is already connected) as a duplicate connection
request. Separate the handling of potential duplicate REQs from stale
connections.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
commit 518b1646 ("IPoIB/cm: Fix SRQ WR leak") introduced a severe
performance regression on Mellanox cards, because keeping a QP in the
error state for extended periods of time moves hardware to the slow
path (until the QP is destroyed). For example, MPI latency goes from
~3 usecs to ~7 usecs.
Fix this by posting a send WR on one of the QPs that are being
flushed, instead of using a separate drain QP that is kept in the
error state.
This fixes bug <https://bugs.openfabrics.org/show_bug.cgi?id=636>,
reported and bisected by Scott Weitzenkamp at Cisco and debugged by
Sasha Mikheev at Voltaire.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
mthca_free_err_wqe() currently treats both send and receive CQEs
identically if a QP is using an SRQ. But for Tavor hardware, send
CQEs with error can be chained together even if the RQ is part of SRQ,
so we may miss some CQEs.
Fix by following the WQE chain for all send CQEs even for non-SRQ QPs.
This fixes crashes in IPoIB CM:
<https://bugs.openfabrics.org//show_bug.cgi?id=604>
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Since NAPI polling is disabled while ipoib_cm_dev_stop() is running,
ipoib_cm_dev_stop() must poll the CQ itself in order to see the
packets draining.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
time_after() was used backwards, so the timeout occurred immediately.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Due to a typo, the driver was reporting the wrong number of "actual send
WRs" after ehca_create_qp().
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
We need to initialize the owner bit of send queue WQEs to hardware
ownership whenever the QP is modified from reset to init, not just
when the QP is first allocated. This avoids having the hardware
process stale WQEs when the QP is moved to reset but not destroyed and
then modified to init again.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
If a QP is attached to a shared receive queue (SRQ), then it doesn't
have a receive queue (RQ). So don't allocate an RQ doorbell (or map a
doorbell from userspace for userspace QPs) for that QP.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
* 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband:
IB/cm: Improve local id allocation
IPoIB/cm: Fix SRQ WR leak
IB/ipoib: Fix typos in error messages
IB/mlx4: Check if SRQ is full when posting receive
IB/mlx4: Pass send queue sizes from userspace to kernel
IB/mlx4: Fix check of opcode in mlx4_ib_post_send()
mlx4_core: Fix array overrun in dump_dev_cap_flags()
IB/mlx4: Fix RESET to RESET and RESET to ERROR transitions
IB/mthca: Fix RESET to ERROR transition
IB/mlx4: Set GRH:HopLimit when sending globally routed MADs
IB/mthca: Set GRH:HopLimit when building MLX headers
IB/mlx4: Fix check of max_qp_dest_rdma in modify QP
IB/mthca: Fix use-after-free on device restart
IB/ehca: Return proper error code if register_mr fails
IPoIB: Handle P_Key table reordering
IB/core: Use start_port() and end_port()
IB/core: Add helpers for uncached GID and P_Key searches
IB/ipath: Fix potential deadlock with multicast spinlocks
IB/core: Free umem when mm is already gone
The IB CM uses an idr for local id allocations, with a running counter
as start_id. This fails to generate distinct ids if
1. An id is constantly created and destroyed
2. A chunk of ids just beyond the current next_id value is occupied
This in turn leads to an increased chance of connection request being
mis-detected as a duplicate, sometimes for several retries, until
next_id gets past the block of allocated ids. This has been observed
in practice.
As a fix, remember the last id allocated and start immediately above it.
This also fixes a problem with the old code, where next_id might
overflow and become negative.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Acked-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
SRQ WR leakage has been observed with IPoIB/CM: e.g. flipping ports on
and off will, with time, leak out all WRs and then all connections
will start getting RNR NAKs. Fix this in the way suggested by spec:
move the QP being destroyed to the error state, wait for "Last WQE
Reached" event and then post WR on a "drain QP" connected to the same
CQ. Once we observe a completion on the drain QP, it's safe to call
ib_destroy_qp.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
First thing mm.h does is including sched.h solely for can_do_mlock() inline
function which has "current" dereference inside. By dealing with can_do_mlock()
mm.h can be detached from sched.h which is good. See below, why.
This patch
a) removes unconditional inclusion of sched.h from mm.h
b) makes can_do_mlock() normal function in mm/mlock.c
c) exports can_do_mlock() to not break compilation
d) adds sched.h inclusions back to files that were getting it indirectly.
e) adds less bloated headers to some files (asm/signal.h, jiffies.h) that were
getting them indirectly
Net result is:
a) mm.h users would get less code to open, read, preprocess, parse, ... if
they don't need sched.h
b) sched.h stops being dependency for significant number of files:
on x86_64 allmodconfig touching sched.h results in recompile of 4083 files,
after patch it's only 3744 (-8.3%).
Cross-compile tested on
all arm defconfigs, all mips defconfigs, all powerpc defconfigs,
alpha alpha-up
arm
i386 i386-up i386-defconfig i386-allnoconfig
ia64 ia64-up
m68k
mips
parisc parisc-up
powerpc powerpc-up
s390 s390-up
sparc sparc-up
sparc64 sparc64-up
um-x86_64
x86_64 x86_64-up x86_64-defconfig x86_64-allnoconfig
as well as my two usual configs.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pass the number of WQEs for the send queue and their size from userspace
to the kernel to avoid having to keep the QP size calculations in sync
between the kernel driver and libmlx4. This fixes a bug seen with the
current mlx4_ib driver and current libmlx4 caused by a difference in the
calculated sizes for SQ WQEs. Also, this gives more flexibility for
userspace to experiment with using multiple WQE BBs for a single SQ WQE.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
wr->opcode is invalid if it's >= ARRAY_SIZE(mlx4_ib_opcode), not just
strictly >.
This was spotted by the Coverity checker (CID 1643).
Signed-off-by: Roland Dreier <rolandd@cisco.com>
According to the IB spec, a QP can be moved from RESET back to RESET
or to the ERROR state, but mlx4 firmware does not support this and
returns an error if we try. Fix the RESET to RESET transition by
just returning 0 without doing anything, and fix RESET to ERROR by
moving the QP from RESET to INIT with dummy parameters and then
transitioning from INIT to ERROR.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
According to the IB spec, a QP can be moved from RESET to the ERROR
state, but mthca firmware does not support this and returns an error if
we try. Work around this FW limitation by moving the QP from RESET to
INIT with dummy parameters and then transitioning from INIT to ERROR.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Global CM packets used by rmda_cm were being sent with a GRH:hopLimit
of zero, causing them to be dropped by the router. The problem is a
missing initialization of the hop_limit field in mthca_read_ah(),
which was called by build_mlx_header() when sending a MAD on QP1.
Signed-off-by: Rolf Manderscheid <rvm@obsidianresearch.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
max_qp_dest_rdma is already in natural units - no need to shift. This
was discovered by a test that deliberately requests more outstanding
atomic operation than the device supports.
Found by Sagi Rotem at Mellanox.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Set the return code of ehca_register_mr() to ENOMEM if the corresponding
firmware call fails due to out of resources. Some other error codes
were explicitly mapped to EINVAL -- just remove those cases so they
get mapped to the default case, which already returns EINVAL anyway.
Signed-off-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
SM reconfiguration or failover possibly causes a shuffling of the values
in the P_Key table. Right now, IPoIB only queries for the P_Key index
once when it creates the device QP, and hence there are problems if the
index of a P_Key value changes. Fix this by using the PKEY_CHANGE event
to trigger a recheck of the P_Key index.
Signed-off-by: Yosef Etigin <yosefe@voltaire.com>
Acked-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Clean up ib_query_port() and ib_modify_port() slightly by using the
just-added start_port() and end_port() helpers.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Add ib_find_gid() and ib_find_pkey() functions that use uncached device
queries. The calls might block but the returns are always up-to-date.
Cache P_Key and GID table lengths in core to avoid extra port info queries.
Signed-off-by: Yosef Etigin <yosefe@voltaire.com>
Acked-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Lockdep found the following potential deadlock between mcast_lock and
n_mcast_grps_lock: mcast_lock is taken from both interrupt context and
process context, so spin_lock_irqsave() must be used to take it.
n_mcast_grps_lock is only taken from process context, so at first it
seems safe to take it with plain spin_lock(); however, it also nests
inside mcast_lock, and hence we could deadlock:
cpu A cpu B
ipath_mcast_add():
spin_lock_irq(&mcast_lock);
ipath_mcast_detach():
spin_lock(&n_mcast_grps_lock);
<enter interrupt>
ipath_mcast_find():
spin_lock_irqsave(&mcast_lock);
spin_lock(&n_mcast_grps_lock);
Fix this by using spin_lock_irq() to take n_mcast_grps_lock.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Free umem when task's mm is already destroyed by the time
ib_umem_release gets called.
Found by Dotan Barak at Mellanox.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
* 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband:
IPoIB/cm: Optimize stale connection detection
IB/mthca: Set cleaned CQEs back to HW ownership when cleaning CQ
IB/mthca: Fix posting >255 recv WRs for Tavor
RDMA/cma: Add check to validate that cm_id is bound to a device
RDMA/cma: Fix synchronization with device removal in cma_iw_handler
RDMA/cma: Simplify device removal handling code
IB/ehca: Disable scaling code by default, bump version number
IB/ehca: Beautify sysfs attribute code and fix compiler warnings
IB/ehca: Remove _irqsave, move #ifdef
IB/ehca: Fix AQP0/1 QP number
IB/ehca: Correctly set GRH mask bit in ehca_modify_qp()
IB/ehca: Serialize hypervisor calls in ehca_register_mr()
IB/ipath: Shadow the gpio_mask register
IB/mlx4: Fix uninitialized spinlock for 32-bit archs
mlx4_core: Remove unused doorbell_lock
net: Trivial MLX4_DEBUG dependency fix.
In the presence of some running RX connections, we repeat
queue_delayed_work calls each 4 RX WRs, which is a waste. It's enough
to start stale task when a first passive connection is added, and
rerun it every IPOIB_CM_RX_DELAY as long as there are outstanding
passive connections.
This removes some code from RX data path.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
mthca_cq_clean() updates the CQ consumer index without moving CQEs
back to HW ownership. As a result, the same WRID might get reported
twice, resulting in a use-after-free. This was observed in IPoIB CM.
Fix by moving all freed CQEs to HW ownership.
This fixes <https://bugs.openfabrics.org/show_bug.cgi?id=617>
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Fix posting lists of > 255 receive WRs for Tavor: rq.next_ind must
be updated each doorbell, otherwise the next doorbell will use an
incorrect index.
Found by Ronni Zimmermann at Mellanox.
Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Several checks in the rdma_cm check against the state of the
cm_id, but only to validate that the cm_id is bound to an underlying
transport specific CM and an RDMA device. Make the check explicit
in what we're trying to check for, since we're not synchronizing
against the cm_id state.
This will allow a user to disconnect a cm_id or reject a connection
after receiving a device removal event.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The cma_iw_handler needs to validate the state of the rdma_cm_id before
processing a new connection request to ensure that a device removal is
not already being processed for the same rdma_cm_id. Without the state
check, the user can receive simultaneous callbacks for the same cm_id, or
a callback after they've destroyed the cm_id.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Add a new routine and rename another to encapsulate common code for
synchronizing with device removal.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
- Scaling code is still considered experimental, so disable it by default
- Increase version to SVNEHCA_0023
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
eHCA's sysfs attributes are now being created via sysfs_create_group(),
making the process neatly table-driven. The return value is checked, thus
fixing a few compiler warnings.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
- In ehca_process_eq(), we're IRQ safe throughout the whole function, so we
don't need another _irqsave in the middle of flight.
- take_over_work() is only called by comp_pool_callback(), so it can move
into the same #ifdef block.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
AQP0/1 should report qp_num={0|1} and the actual QP# should be stored
in struct ehca_qp, not the other way round.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The driver needs to always supply the "GRH present" flag to the
hypervisor, whether it's true or false. Not supplying it (i.e. not
setting the corresponding mask bit) amounts to a "perhaps", which we
don't want.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Some pSeries hypervisor versions show a race condition in the allocate
MR hCall. Serialize this call per adapter to circumvent this problem.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Once upon a time, GPIO interrupts were rare. But then a chip bug in
the waldo series forced the use of a GPIO interrupt to signal packet
reception. This greatly increased the frequency of GPIO interrupts
which have the gpio_mask bits set on the waldo chips. Other bits in
the gpio_status register are used for I2C clock and data lines, these
bits are usually on. An "unlikely" annotation leftover from the old
days was improperly applied to these bits, and an unnecessary chip
mmio read was being accessed in the interrupt fast path on waldo.
Remove the stagnant unlikely annotation in the interrupt handler and
keep a shadow copy of the gpio_mask register to avoid the slow mmio
read when testing for interruptable GPIO bits.
Signed-off-by: Arthur Jones <arthur.jones@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
uar_lock spinlock was used in mlx4_ib_cq_arm without being initialized
(this only affects 32-bit archs, because uar_lock is not used on
64-bit archs and MLX4_INIT_DOORBELL_LOCK() is a NOP).
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Add "depends on HAS_IOMEM" to a number of menus to make them
disappear for s390 which does not have I/O memory.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
* 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband:
IB/mlx4: Add a driver Mellanox ConnectX InfiniBand adapters
IB: Put rlimit accounting struct in struct ib_umem
IB/uverbs: Export ib_umem_get()/ib_umem_release() to modules
Since nonboot CPUs are now disabled after tasks and devices have been
frozen and the CPU hotplug infrastructure is used for this purpose, we need
special CPU hotplug notifications that will help the CPU-hotplug-aware
subsystems distinguish normal CPU hotplug events from CPU hotplug events
related to a system-wide suspend or resume operation in progress. This
patch introduces such notifications and causes them to be used during
suspend and resume transitions. It also changes all of the
CPU-hotplug-aware subsystems to take these notifications into consideration
(for now they are handled in the same way as the corresponding "normal"
ones).
[oleg@tv-sign.ru: cleanups]
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add an InfiniBand driver for Mellanox ConnectX adapters. Because
these adapters can also be used as ethernet NICs and Fibre Channel
HBAs, the driver is split into two modules:
mlx4_core: Handles low-level things like device initialization and
processing firmware commands. Also controls resource allocation
so that the InfiniBand, ethernet and FC functions can share a
device without stepping on each other.
mlx4_ib: Handles InfiniBand-specific things; plugs into the
InfiniBand midlayer.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
When memory pinned with ib_umem_get() is released, ib_umem_release()
needs to subtract the amount of memory being unpinned from
mm->locked_vm. However, ib_umem_release() may be called with
mm->mmap_sem already held for writing if the memory is being released
as part of an munmap() call, so it is sometimes necessary to defer
this accounting into a workqueue.
However, the work struct used to defer this accounting is dynamically
allocated before it is queued, so there is the possibility of failing
that allocation. If the allocation fails, then ib_umem_release has no
choice except to bail out and leave the process with a permanently
elevated locked_vm.
Fix this by allocating the structure to defer accounting as part of
the original struct ib_umem, so there's no possibility of failing a
later allocation if creating the struct ib_umem and pinning memory
succeeds.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Export ib_umem_get()/ib_umem_release() and put low-level drivers in
control of when to call ib_umem_get() to pin and DMA map userspace,
rather than always calling it in ib_uverbs_reg_mr() before calling the
low-level driver's reg_user_mr method.
Also move these functions to be in the ib_core module instead of
ib_uverbs, so that driver modules using them do not depend on
ib_uverbs.
This has a number of advantages:
- It is better design from the standpoint of making generic code a
library that can be used or overridden by device-specific code as
the details of specific devices dictate.
- Drivers that do not need to pin userspace memory regions do not
need to take the performance hit of calling ib_mem_get(). For
example, although I have not tried to implement it in this patch,
the ipath driver should be able to avoid pinning memory and just
use copy_{to,from}_user() to access userspace memory regions.
- Buffers that need special mapping treatment can be identified by
the low-level driver. For example, it may be possible to solve
some Altix-specific memory ordering issues with mthca CQs in
userspace by mapping CQ buffers with extra flags.
- Drivers that need to pin and DMA map userspace memory for things
other than memory regions can use ib_umem_get() directly, instead
of hacks using extra parameters to their reg_phys_mr method. For
example, the mlx4 driver that is pending being merged needs to pin
and DMA map QP and CQ buffers, but it does not need to create a
memory key for these buffers. So the cleanest solution is for mlx4
to call ib_umem_get() in the create_qp and create_cq methods.
Signed-off-by: Roland Dreier <rolandd@cisco.com>
