IB/mlx5: Fire the CQ completion handler from tasklet
Previously, mlx5_ib_cq_comp was executed from interrupt context.
Under heavy load, this could cause the CPU core to be in an interrupt
context too long.
Instead of executing the handler from the interrupt context we
execute it from a much friendly tasklet context.
net/mlx5_core: Use tasklet for user-space CQ completion events
Previously, we've fired all our completion callbacks straight from
our ISR.
Some of those callbacks were lightweight (for example, mlx5 Ethernet
napi callbacks), but some of them did more work (for example,
the user-space RDMA stack uverbs' completion handler). Besides that,
doing more than the minimal work in ISR is generally considered wrong,
it could even lead to a hard lockup of the system. Since when a lot
of completion events are generated by the hardware, the loop over
those events could be so long, that we'll get into a hard lockup by
the system watchdog.
In order to avoid that, add a new way of invoking completion events
callbacks. In the interrupt itself, we add the CQs which receive
completion event to a per-EQ list and schedule a tasklet. In the
tasklet context we loop over all the CQs in the list and invoke the
user callback.
IB/core: Do not require CAP_NET_ADMIN for packet sniffing
In the Ethernet/TCP world, CAP_NET_RAW is sufficient to allow a program
to listen to all incoming packets on a specific interface, and the
higher CAP_NET_ADMIN is required to set the interface into promiscuous
mode. We want to emulate that same basic division of privilege in the
RDMA stack, so when dealing with Raw Ethernet QPs, allow apps with
CAP_NET_RAW to listen to all incoming flows (and direct them as they see
fit in their own listen stream). Do not require CAP_NET_ADMIN just to
listen to traffic already incoming. Reserve CAP_NET_ADMIN if we attempt
to set promiscuous mode.
Signed-off-by: Christoph Lameter <cl@linux.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/mlx4: Fix unaligned access in send_reply_to_slave
The problem is that the function 'send_reply_to_slave' gets the
'req_sa_mad' as a pointer whose address is only aliged to 4 bytes
but is 8 bytes in size. This can result in unaligned access faults
on certain architectures.
Sowmini Varadhan pointed to this reply from Dave Miller that say
that memcpy should not be used to solve alignment issues:
https://lkml.org/lkml/2015/10/21/352
Optimization of memcpy to 'ldx' instruction can only happen if the
compiler knows that the size of the data we are copying is 8 bytes
and it assumes it is aligned to 8 bytes. If the compiler know the
type is not aligned to 8 it must not optimize the 8 byte copy.
Defining the data type as aligned to 4 forces the compiler to treat
all accesses as though they aren't aligned and avoids the 'ldx'
optimization.
Full credit for the idea goes to Jason Gunthorpe
<jgunthorpe@obsidianresearch.com>.
Colin Ian King [Mon, 28 Mar 2016 11:14:46 +0000 (12:14 +0100)]
i40iw: pass hw_stats by reference rather than by value
passing hw_stats by value requires a 280 byte copy so instead
pass it by reference is much more efficient.
Signed-off-by: Colin Ian King <colin.king@canonical.com> Acked-by: Chien Tin Tung <chien.tin.tung@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
i40iw: Remove unnecessary synchronize_irq() before free_irq()
Calling synchronize_irq() right before free_irq() is quite useless. On one
hand the IRQ can easily fire again before free_irq() is entered, on the
other hand free_irq() itself calls synchronize_irq() internally (in a race
condition free way), before any state associated with the IRQ is freed.
Patch was generated using the following semantic patch:
// <smpl>
@@
expression irq;
@@
-synchronize_irq(irq);
free_irq(irq, ...);
// </smpl>
Guy Levi [Wed, 27 Apr 2016 13:49:50 +0000 (16:49 +0300)]
IB/mlx5: Add UARs write-combining and non-cached mapping
By this patch, the user space library will be able to improve
performance using appropriate ringing DoorBell method according to the
memory type it asked for.
Currently only one mapping command is allowed for UARs:
MLX5_IB_MMAP_REGULAR_PAGE. Using this mapping, the kernel maps the
UARs to write-combining (WC) if the system supports it.
If the system is not supporting WC the UARs are mapped to
non-cached(NC). In this case the user space library can't tell which
mapping is applied.
This patch adds 2 new mapping commands: MLX5_IB_MMAP_WC_PAGE and
MLX5_IB_MMAP_NC_PAGE. For these commands the kernel maps exactly as
requested and fails if it can't.
Since there is no generic way to check if the requested memory region
can be mapped as WC, driver enables conclusive WC mapping only for
x86, PowerPC and ARM which support WC for the device's memory region.
Signed-off-by: Guy Levy <guyle@mellanox.com> Signed-off-by: Moshe Lazer <moshel@mellanox.com> Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/mlx5: Allow mapping the free running counter on PROT_EXEC
The current mlx5 code disallows mapping the free running counter of
mlx5 based hardwares when PROT_EXEC is set.
Although this behaviour is correct, Linux does add an implicit VM_EXEC
to the vm_flags if the READ_IMPLIES_EXEC bit is set in the process
personality. This happens for example if the process stack is
executable.
This causes libmlx5 to output a warning and prevents the user from
reading the free running clock.
Executing the init segment of the hardware isn't a security risk
(at least no more than executing a process own stack), so we just
prevent writes to there.
Fixes: d69e3bcf7976 ('IB/mlx5: Mmap the HCA's core clock register to
user-space') Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Mark Bloch [Fri, 6 May 2016 19:45:27 +0000 (22:45 +0300)]
IB/SA: Use correct free function
Fixes a direct call to kfree_skb when nlmsg_free should be used.
Fixes: 2ca546b92a02 ('IB/sa: Route SA pathrecord query through netlink') Signed-off-by: Mark Bloch <markb@mellanox.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Reviewed-by: Ira Weiny <ira.weiny@intel.com> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Mark Bloch [Fri, 6 May 2016 19:45:26 +0000 (22:45 +0300)]
IB/core: Fix a potential array overrun in CMA and SA agent
Fix array overrun when going over callback table.
In declaration of callback table, the max size isn't provided and
in registration phase, it is provided.
There is potential scenario where a new operation is added
and it is not supported by current client. The acceptance of
such operation by ib_netlink will cause to array overrun.
Fixes: 809d5fc9bf65 ("infiniband: pass rdma_cm module to netlink_dump_start") Fixes: b493d91d333e ("iwcm: common code for port mapper") Fixes: 2ca546b92a02 ("IB/sa: Route SA pathrecord query through netlink") Signed-off-by: Mark Bloch <markb@mellanox.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
Mark Bloch [Fri, 6 May 2016 19:45:24 +0000 (22:45 +0300)]
IB/IWPM: Fix a potential skb leak
In case ibnl_put_msg fails in send_nlmsg_done,
the function returns with -ENOMEM without freeing.
This patch fixes this behavior.
Fixes: 30dc5e63d6a5 ("RDMA/core: Add support for iWARP Port Mapper user space service") Signed-off-by: Mark Bloch <markb@mellanox.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Reviewed-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Andy Shevchenko [Fri, 6 May 2016 17:19:42 +0000 (20:19 +0300)]
RDMA/nes: replace custom print_hex_dump()
There is no need to duplicate a lot of code that is in the kernel library for ages.
Replace duplicating code by calling to print_hex_dump() directly.
Note that output is slightly changed:
- hex and ascii parts have just two spaces delimeter
- there is no delimeter for ascii portions
- file and line removed from prefix (they were redundant anyway since previous
output shows same closer enough)
Colin Ian King [Mon, 25 Apr 2016 19:26:50 +0000 (20:26 +0100)]
IB/mlx4: trivial fix of spelling mistake on "argument"
fix spelling mistake, argumant -> argument
Signed-off-by: Colin Ian King <colin.king@canonical.com> Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Tatyana Nikolova [Tue, 29 Mar 2016 17:58:37 +0000 (12:58 -0500)]
RDMA/nes: Adding queue drain functions
Adding sq and rq drain functions, which block until all
previously posted wr-s in the specified queue have completed.
A completion object is signaled to unblock the thread,
when the last cqe for the corresponding queue is processed.
IPoIB collects statistics of traffic including number of packets
sent/received, number of bytes transferred, and certain errors. This
patch makes these statistics available to be queried by ethtool.
The last user of pkey_mutex was removed in db84f8803759 ("IB/ipoib: Use
P_Key change event instead of P_Key polling mechanism") but the lock
remained.
This patch removes it.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Colin Ian King [Mon, 28 Mar 2016 11:14:46 +0000 (12:14 +0100)]
i40iw: pass hw_stats by reference rather than by value
passing hw_stats by value requires a 280 byte copy so instead
pass it by reference is much more efficient.
Signed-off-by: Colin Ian King <colin.king@canonical.com> Acked-by: Chien Tin Tung <chien.tin.tung@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
i40iw: Remove unnecessary synchronize_irq() before free_irq()
Calling synchronize_irq() right before free_irq() is quite useless. On one
hand the IRQ can easily fire again before free_irq() is entered, on the
other hand free_irq() itself calls synchronize_irq() internally (in a race
condition free way), before any state associated with the IRQ is freed.
Patch was generated using the following semantic patch:
// <smpl>
@@
expression irq;
@@
-synchronize_irq(irq);
free_irq(irq, ...);
// </smpl>
__force casts should be avoided if there is a better alternative.
Hence modify the comparison of s_addr with INADDR_ANY such that the
__force cast is no longer necessary.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Vipul Pandya <vipul@chelsio.com> Acked-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Hariprasad S [Fri, 6 May 2016 16:48:09 +0000 (22:18 +0530)]
RDMA/iw_cxgb4: Always wake up waiter in c4iw_peer_abort_intr()
Currently c4iw_peer_abort_intr() does not wake up the waiter if the
endpoint state indicates we're using MPAv2 and we're currently trying to
connect. This was introduced with commit 7c0a33d61187a ("RDMA/cxgb4:
Don't wakeup threads for MPAv2")
However, this original fix is flawed because it introduces a race that
can cause a deadlock of the iwarp stack. Here is the race:
->local side sets up an active offload connection.
->local side sends MPA_START request.
->peer sends MPA_START response.
->local side ingress cpl thread begins processing the MPA_START response,
but before it changes the state from MPA_REQ_SENT to FPDU_MODE:
->peer sends a RST which results in a ABORT_REQ_RSS. This triggers
peer_abort_intr() which sees the state in MPA_REQ_SENT and since mpa_rev
is 2, it will avoid waking up the endpoint with -ECONNRESET, assuming the
stack will re-attempt the connection using MPAv1.
->Meanwhile, the cpl thread moves the state to FPDU_MODE and calls
c4iw_modify_rc_qp() which calls rdma_init() which sends a RI_WR/INIT WR
to firmware. But since HW sent an abort, FW correctly drops the RI_WR/INIT
WR.
->So the cpl thread is stuck waiting for a reply and cannot process the
ABORT_REQ_RSS cpl sitting in its input queue. Thus everything comes to a
halt because no more ingress cpls are processed by the stack...
The correct fix for the issue is to always do the wake up in
c4iw_abort_intr() but reinitialize the wait object in c4iw_reconnect().
Fixes: 7c0a33d61187a ("RDMA/cxgb4: Don't wakeup threads for MPAv2") Signed-off-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
Hariprasad S [Fri, 6 May 2016 16:48:07 +0000 (22:18 +0530)]
RDMA/iw_cxgb4: atomic find and reference for listening endpoints
Add get_ep_from_stid() which will atomically find and reference the
endpoint struct if found. This avoids touch-after-free races between
threads destroying listening endpoints and the CPL processing thread
processing an incoming PASS_ACCEPT_REQ CPL.
Hariprasad S [Fri, 6 May 2016 16:48:05 +0000 (22:18 +0530)]
RDMA/iw_cxgb4: Release ep for for FPDU_MODE and MPA_REQ_RCVD in process_timeout
ARP failure may also happen when ep in FPDU_MODE and these failures need
to be handled by process_timeout(). process_timeout() also has to handle
case MPA_REQ_RCVD, setting abort to 1, leading to ep resource release.
Hariprasad S [Fri, 6 May 2016 16:48:03 +0000 (22:18 +0530)]
RDMA/iw_cxgb4: atomically lookup ep and get a reference
There is a race between ULP threads calling c4iw_ep_disconnect() via
c4iw_modify_rc_qp() and the ingress CPL thread where the ULP thread
can free the endpoint just after the ingress CPL thread finds the ep
pointer in the tid table. To avoid this, we now use the hwtid_idr table
for lookups instead of the LLD tid table so we can lock around insert,
remove, and lookup+get_ep to avoid the race. The CPL handlers now will
either find the ep ptr and have a ref on it, or not find it and they
can discard the CPL. Callers of get_ep_from_tid() will have a ref
on the ep if found, and thus must deref when they are done.
Negative advice in peer_abort_intr() need to dereference the ep.
therefore peer_abort() is scheduled to dereference the ep later.
Hariprasad S [Fri, 6 May 2016 16:48:02 +0000 (22:18 +0530)]
RDMA/iw_cxgb4: Handle return value of c4iw_ofld_send() in abort_arp_failure()
In abort_arp_failure(), the return value from c4iw_ofld_send() is
ignored and thus if the CPL isn't sent, the endpoint is stuck and never
gets aborted. Failure of c4iw_ofld_send() is treated as fatal error, and
the ep resources are released in a safer context through process_work().
Hariprasad S [Fri, 6 May 2016 16:48:01 +0000 (22:18 +0530)]
RDMA/iw_cxgb4: in process_timeout() don't move ep state to ABORTING
Moving the state to ABORTING causes the ep to get stuck because
c4iw_ep_timeout() thinks the ABORT has already been done. So leave the
state alone and let c4iw_ep_disconnect() do the right thing given the
ep state.
Hariprasad S [Fri, 6 May 2016 16:48:00 +0000 (22:18 +0530)]
RDMA/iw_cxgb4: handle return value of c4iw_l2t_send() and send_mpa_req()
->In act_open_rpl(), CPL_ERR_TCAM_FULL error handling branch, there is
no handling of the return value of send_fw_act_open_req().
->In send_fw_act_open_req(), there is no handling of return value of
c4iw_l2t_send(), which may cause a ep leak and won't notify upper layers
on connection establish failure.
->send_mpa_req() should act on the return from c4iw_l2t_send() and
return the error to the caller.
->In case of c4iw_l2t_send() failure in send_mpa_req(), returns without
starting the timer and not changing the ep state, which is further
handled by act_establish()
-> In act_establish()?if send_mpa_request's get_skb returns an error,
may cause an ep leak. So handle return value of send_mpa_req()
Hariprasad S [Fri, 6 May 2016 16:47:59 +0000 (22:17 +0530)]
RDMA/iw_cxgb4: stop_ep_timer() after MPA negotiation
->Stop the ep timer after MPA negotiation so that the arp failures
during send_mpa_reply/reject will be handled by process_timeout() after
the ep timer expires.
->Added case MPA_REP_SENT in process_timeout().
->For MPA reject, c4iw_ep_disconnect tries to start an already started
timer, which leads to warning message "timer already started".
-> In case of mpa reject stop the timer and call send_mpa_reject().
-> Added new ep flag STOP_MPA_TIMER to tell fw4_ack() to stop the timer
only for send_mpa_reply(), which is set in c4iw_accept_cr().
Hariprasad S [Fri, 6 May 2016 16:47:57 +0000 (22:17 +0530)]
RDMA/iw_cxgb4: parent_ep has to be dereferenced in case of passive accept failure
-> On passive side of connection parent_ep referenced during connection
request has to be dereferenced during the passive accept failure.
-> As passive accept failure error handlinglogic runs in atomic context,
the parent ep is dereferenced by scheduling work request.
Hariprasad S [Fri, 6 May 2016 16:47:56 +0000 (22:17 +0530)]
RDMA/iw_cxgb4: set the correct FID value in DSGL commands
The FID value in a ULP_MEMIO command needs to be set to an IQ ID of
a queue configured for our PF. The FID/IQ id is used to index into the
PCIE FID table, to find out on which function the DMA needs to be
issued. Essentially, every DMA needs to have the ingress queue. The exact
ingress queue doesn't matter, but it needs to be an ingress queue
associated with the function you want to see the DMA on.
Hariprasad S [Fri, 6 May 2016 16:47:54 +0000 (22:17 +0530)]
RDMA/iw_cxgb4: Add few history bits for ep
- add EP_DISC_FAIL history bit
- add QP_REFED/DEREFED history bits
- Add functions to ref/deref the cm_id and add history bit for the same
- add CLOSE_CON_RPL history
Colin Ian King [Mon, 28 Mar 2016 11:14:46 +0000 (12:14 +0100)]
i40iw: pass hw_stats by reference rather than by value
passing hw_stats by value requires a 280 byte copy so instead
pass it by reference is much more efficient.
Signed-off-by: Colin Ian King <colin.king@canonical.com> Acked-by: Chien Tin Tung <chien.tin.tung@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
i40iw: Remove unnecessary synchronize_irq() before free_irq()
Calling synchronize_irq() right before free_irq() is quite useless. On one
hand the IRQ can easily fire again before free_irq() is entered, on the
other hand free_irq() itself calls synchronize_irq() internally (in a race
condition free way), before any state associated with the IRQ is freed.
Patch was generated using the following semantic patch:
// <smpl>
@@
expression irq;
@@
-synchronize_irq(irq);
free_irq(irq, ...);
// </smpl>
Bart Van Assche [Thu, 12 May 2016 17:50:35 +0000 (10:50 -0700)]
IB/srp: Prevent mapping failures
If both max_sectors and the queue_depth are high enough it can
happen that the MR pool is depleted temporarily. This causes
the SRP initiator to report mapping failures. Although the SRP
initiator recovers from such mapping failures, prevent that
this can happen by allocating more memory regions.
Additionally, only enable memory registration if at least two
pages can be registered per memory region.
Bart Van Assche [Thu, 12 May 2016 17:49:15 +0000 (10:49 -0700)]
IB/core: Enhance ib_map_mr_sg()
The SRP initiator allows to set max_sectors to a value that exceeds
the largest amount of data that can be mapped at once with an mlx4
HCA using fast registration and a page size of 4 KB. Hence modify
ib_map_mr_sg() such that it can map partial sg-elements. If an
sg-element has been mapped partially, let the caller know
which fraction has been mapped by adjusting *sg_offset.
Bart Van Assche [Thu, 12 May 2016 17:48:13 +0000 (10:48 -0700)]
IB/srp: Fix a memory descriptor leak in an error path
If an error occurs after srp_fr_pool_get() succeeded and before the
descriptor is stored in srp_map_state (*state->fr.next++ = desc)
then srp_unmap_data() won't free the newly allocated memory
descriptor. Hence free the descriptor explicitly.
Fixes: f7f7aab1a5c0 ("IB/srp: Convert to new registration API") Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Tested-by: Laurence Oberman <loberman@redhat.com> Cc: Sagi Grimberg <sai@grimberg.me> Cc: Christoph Hellwig <hch@lst.de> Cc: <stable@vger.kernel.org> # v4.4+ Signed-off-by: Doug Ledford <dledford@redhat.com>
Bart Van Assche [Thu, 12 May 2016 17:47:38 +0000 (10:47 -0700)]
IB/srp: Print "ib_srp: " prefix once
pr_debug() already prints prefix PFX. Avoid that PFX is printed
twice if the debug statement in srp_add_target() is enabled.
Fixes: 34aa654ecb8e ("IB/srp: Avoid that I/O hangs due to a cable pull during LUN scanning") Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Tested-by: Laurence Oberman <loberman@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/isert: convert to the generic RDMA READ/WRITE API
Replace the homegrown RDMA READ/WRITE code in isert with the generic API,
which also adds iWarp support to the I/O path as a side effect. Note
that full iWarp operation will need a few additional patches from Steve.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/srpt: convert to the generic RDMA READ/WRITE API
Replace the homegrown RDMA READ/WRITE code in srpt with the generic API.
The only real twist here is that we need to allocate one Linux scatterlist
per direct buffer in the SRP command, and chain them before handing them
off to the target core.
As a side-effect of the conversion the driver will also chain the SEND
of the SRP response to the RDMA WRITE WRs for a DATA OUT command, and
properly account for RDMA WRITE WRs instead of just for RDMA READ WRs
like the driver previously did.
We now allocate half of the SQ size to RDMA READ/WRITE contexts, assuming
by default one RDMA READ or WRITE operation per command. If a command
has multiple operations it will eat into the budget but will still succeed,
possible after waiting for WQEs to be available.
Also ensure the QPs request the maximum allowed SGEs so that RDMA R/W API
works correctly.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
target: enhance and export target_alloc_sgl/target_free_sgl
The SRP target driver will need to allocate and chain it's own SGLs soon.
For this export target_alloc_sgl, and add a new argument to it so that it
can allocate an additional chain entry that doesn't point to a page. Also
export transport_free_sgl after renaming it to target_free_sgl to free
these SGLs again.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
This supports both manual mapping of lots of SGEs, as well as using MRs
from the QP's MR pool, for iWarp or other cases where it's more optimal.
For now, MRs are only used for iWARP transports. The user of the RDMA-RW
API must allocate the QP MR pool as well as size the SQ accordingly.
Thanks to Steve Wise for testing, fixing and rewriting the iWarp support,
and to Sagi Grimberg for ideas, reviews and fixes.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
Adding the needed mlx5_ifc hardware bits and structs
for the following features:
* Add vport to steering commands for SRIOV ACL support
* Add mlcr, pcmr and mcia registers for dump module EEPROM
* Add support for FCS, beacon led and disable_link bits to
hca caps
* Add CQE period mode bit in CQ context for CQE based CQ
moderation support
* Add umr SQ bit for fragmented memory registration
* Add needed bits and caps for Striding RQ support
In-order to avoid possible future conflicts between rdma and
net-next we added all expected updates to this file for this release.
If more changes will be submitted, we plan to do it only through
one of the subsystems, probably net-next.
All updated bits in this patch will be later used in
the up-coming submissions to net-next and rdma trees.
Since all srp_map_finish_fr() callers pass a non-zero value as
the fourth argument (sg_nents), the sg_nents == 0 check in that
function can be removed. Add a count == 0 check in the caller
of that function.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Sagi Grimberg <sagi@grimberg.me> Cc: Laurence Oberman <loberman@redhat.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
IB/srp: Avoid that mapping failure triggers an infinite loop
The srp_queuecommand() function translates ENOMEM into QUEUE_FULL
which causes the SCSI mid-layer to retry the command. All other
error codes are translated into DID_ERROR which causes the SCSI
command to fail. Return E2BIG if mapping will always fail to
prevent that the SCSI mid-layer keeps resubmitting a command
forever.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Sagi Grimberg <sagi@grimberg.me> Cc: Laurence Oberman <loberman@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
Ensure that req->nmdesc is set correctly in srp_map_sg() if mapping
fails. Avoid that mapping failure causes a memory descriptor leak.
Report srp_map_sg() failure to the caller.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Sagi Grimberg <sagi@grimberg.me> Cc: Laurence Oberman <loberman@redhat.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Doug Ledford <dledford@redhat.com>
Hariprasad S [Wed, 4 May 2016 19:57:36 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: move QP -> ERROR on fatal disconnect errors
In c4iw_ep_disconnect(), if we fail to initiate a close operation, then
move the qp to ERROR to disassociate the ep from the qp. Failure to do
this will leak the ep resources.
Hariprasad S [Wed, 4 May 2016 19:57:35 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: don't use abort_connection in process_mpa_request()
Instead return whether the caller needs to disconnect. This is part of
getting rid of abort_connection() altogether so we properly clean up on
send_abort() failures.
Hariprasad S [Wed, 4 May 2016 19:57:32 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: remove connection abort from process_mpa_reply
Instead, have the caller, rx_data() handle the close/abort like
it does for process_mpa_request(). This is part of getting rid of
abort_connection() altogether so we properly clean up on send_abort()
failures.
Hariprasad S [Wed, 4 May 2016 19:57:31 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: ensure eps don't get freed while the mutex is held
In rx_data(), with the ep in FPDU_MODE, refcnt=2, if we get unexpected
streaming data, we call c4iw_modify_rc_qp() and move the qp from
RTS -> TERMINATE. In c4iw_modify_rc_qp(), if rdma_fini() returns
an error, the ep will be dereferenced (refcnt=1). Then rx_data()
calls c4iw_ep_disconnect() which starts the close operation.
But if send_halfclose() fails in c4iw_ep_disconnect(), we will call
release_ep_resources() derefing the ep which reduces the refcnt to 0 and
and frees the ep. However we still has the ep mutex at that point, so we
have a touch-after-free bug. There is a similar issue where
peer_close() calls c4iw_ep_disconnect().
The solution is to add a reference to the ep in c4iw_ep_disconnect()
after acquiring the mutex, and release it after releasing the mutex.
Hariprasad S [Wed, 4 May 2016 19:57:30 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: stop ep timer on close failure
In c4iw_ep_disconnect(), if we start the ep timer to begin a close,
but send_halfclose() fails, we need to stop the timer and send a CLOSE
event up to the IWCM before releasing the resources. Otherwise, we can
crash when the ep timer fires if the ep is referencing a previous instance
of the device. This can happen as part of adapter reset/recovery, for
instance.
Hariprasad S [Wed, 4 May 2016 19:57:29 +0000 (01:27 +0530)]
RDMA/iw_cxgb4: release ep resources on accept arp failure
If ARP fails before the CPL_PASS_ACCEPT_RPL is seen by hardware, the tid
will be stuck in SYN_PEND and never released. So create an arp failure
handler specifically for this message to release the endpoint resources.
In pass_accept_rpl_arp_failure(), put the parent endpoint so it will
be freed when destroyed. Also we don't need to call release_tid() here
because _c4iw_free_ep() calls cxgb4_remove_tid() which releases the
hwtid.
If we get an ABORT_REQ_RSS instead of a PASS_ESTABLISH (because the
peer's ACK to our SYN is never received), then put the parent as well
in peer_abort().
Treat accept_cr() failures just like arp failures: put the parent ep
and release the ep resources destroying the tid
The ARP failure handlers are called in an atomic context, so we need to
schedule some of the processing which might block. Namely _c4iw_free_ep()
which needs a mutex. So create a "special" CPL opcode and handler and
schedule it via sched() to be run by process_work() in a blockable context.
Also rework the active open arp failure handler to make use of
release_ep_resources(). This allows both the active and passive arp
failure handlers to use the same deferred cleanup function.
iSER currently has a couple places that set max_sectors in either the host
template or SCSI host, and all of them get it wrong.
This patch instead uses a single assignment that (hopefully) gets it right:
the max_sectors value must be derived from the number of segments in the
FR or FMR structure, but actually be one lower than the page size multiplied
by the number of sectors, as it has to handle the case of non-aligned I/O.
Without this I get trivial to reproduce hangs when running xfstests
(on XFS) over iSER to Linux targets.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Max Gurtovoy <maxg@mellanox.com> Acked-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Doug Ledford <dledford@redhat.com>
RDMA/i40iw: Fix for checking if the QP is destroyed
Fix for checking if the QP associated with a completion
has been destroyed while processing CQ elements.
If that is the case, move the CQ head to the next element
and continue completion processing.
projects / linux-beck.git / log