- 26 Feb, 2013 1 commit
-
-
Roland Dreier authored
-
- 25 Feb, 2013 13 commits
-
-
Shani Michaeli authored
Indicate memory windows support through device capabilities, kernel verb entries and the relevant uverbs command mask entries. Signed-off-by:
Haggai Eran <haggaie@mellanox.com> Signed-off-by:
Shani Michaeli <shanim@mellanox.com> Signed-off-by:
Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by:
Roland Dreier <roland@purestorage.com>
-
Shani Michaeli authored
* Implement memory windows binding in mlx4_ib_post_send. * Implement mlx4_ib_bind_mw by deferring to mlx4_ib_post_send. * Rename MLX4_WQE_FMR_PERM_* flags to MLX4_WQE_FMR_AND_BIND_PERM_*, indicating that they are used both for fast registration work requests, and for memory window bind work requests. Signed-off-by:
-
Shani Michaeli authored
Implement MW allocation and deallocation in mlx4_core and mlx4_ib. Pass down the enable bind flag when registering memory regions. Signed-off-by:
-
Shani Michaeli authored
Add memory windows-related code to INIT_HCA and QUERY_HCA. Signed-off-by:
-
Shani Michaeli authored
Do not enable memory windows allocation for virtual functions. In addition, add a few safety checks, such as: * Verifying the PD of a new MPT matches the VF. * Making sure binding memory window isn't enabled for FMRs, and that new memory windows are not FMR themselves. Signed-off-by:
-
Roland Dreier authored
If IPoIB fails to look up a path record (eg if it tries during an SM failover when one SM is dead but the new one hasn't taken over yet), the driver ends up with a neighbour structure but no address handle (AH). There's no mechanism to recover from this: any further packets sent to this destination will be silently dumped in ipoib_start_xmit(). Fix this by freeing the neighbour structures when a path rec query fails, so that the next packet queued to be sent will trigger a new path record query. Signed-off-by: -
Bart Van Assche authored
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:
-
Bart Van Assche authored
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:
David Dillow <dave@thedillows.org> Cc: <stable@vger.kernel.org> # 3.8 Signed-off-by:
-
Bart Van Assche authored
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 authored
Remove an assignment that incorrectly overwrites the connection state update by srp_connect_target(). Signed-off-by:
-
Syam Sidhardhan authored
kfree on NULL pointer is a no-op. Signed-off-by:
Syam Sidhardhan <s.syam@samsung.com> Signed-off-by:
-
Paul Bolle authored
Building qp.o triggers this gcc warning: drivers/infiniband/hw/mlx4/qp.c: In function ‘mlx4_ib_post_send’: drivers/infiniband/hw/mlx4/qp.c:1862:62: warning: ‘vlan’ may be used uninitialized in this function [-Wmaybe-uninitialized] drivers/infiniband/hw/mlx4/qp.c:1752:6: note: ‘vlan’ was declared here Looking at the code it is clear 'vlan' is only set and used if 'is_eth' is non-zero. But by initializing 'vlan' to 0xffff, on gcc (Ubuntu 4.7.2-22ubuntu1) 4.7.2 on x86-64 at least, we fix the warning, and the compiler was already setting 'vlan' to 0 in the generated code, so there's no real downside. Signed-off-by:Paul Bolle <pebolle@tiscali.nl> [ Get rid of unnecessary move of 'is_vlan' initialization. - Roland ] Signed-off-by:
-
Roland Dreier authored
Matches the way they're used, and actually lets at least x86-64 generate better code: add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-38 (-38) function old new delta mlx4_ib_post_send 4416 4378 -38 Signed-off-by:
-
- 22 Feb, 2013 3 commits
-
-
Or Gerlitz authored
Reuse the "SG unaligned for FMR" driver flow to make the initiator functional when running over driver instance which doesn't support FMRs, such as a mlx4 virtual function. Signed-off-by:
Alex Tabachnik <alext@mellanox.com> Signed-off-by:
-
Or Gerlitz authored
Under IO/CPU stress its possible that the FMR pool might not have a free FMR mapping element for iSER to use because of incomplete background unmapping processing. In that case we get -EAGAIN and the IO is pushed back to the SCSI layer which soon retries it. No need to be so verbose about that. Signed-off-by:
-
Or Gerlitz authored
ISER_DEF_CMD_PER_LUN was meant to be ISCSI_DEF_XMIT_CMDS_MAX, not plain 128 Signed-off-by:
Roi Dayan <roid@mellanox.com> Signed-off-by:
-
- 21 Feb, 2013 5 commits
-
-
Shani Michaeli authored
The existing user/kernel uverbs API has IB_USER_VERBS_CMD_ALLOC/DEALLOC_MW. Implement these calls, along with destroying user memory windows during process cleanup. Signed-off-by:
-
Shani Michaeli authored
This patch enhances the IB core support for Memory Windows (MWs). MWs allow an application to have better/flexible control over remote access to memory. Two types of MWs are supported, with the second type having two flavors: Type 1 - associated with PD only Type 2A - associated with QPN only Type 2B - associated with PD and QPN Applications can allocate a MW once, and then repeatedly bind the MW to different ranges in MRs that are associated to the same PD. Type 1 windows are bound through a verb, while type 2 windows are bound by posting a work request. The 32-bit memory key is composed of a 24-bit index and an 8-bit key. The key is changed with each bind, thus allowing more control over the peer's use of the memory key. The changes introduced are the following: * add memory window type enum and a corresponding parameter to ib_alloc_mw. * type 2 memory window bind work request support. * create a struct that contains the common part of the bind verb struct ibv_mw_bind and the bind work request into a single struct. * add the ib_inc_rkey helper function to advance the tag part of an rkey. Consumer interface details: * new device capability flags IB_DEVICE_MEM_WINDOW_TYPE_2A and IB_DEVICE_MEM_WINDOW_TYPE_2B are added to indicate device support for these features. Devices can set either IB_DEVICE_MEM_WINDOW_TYPE_2A or IB_DEVICE_MEM_WINDOW_TYPE_2B if it supports type 2A or type 2B memory windows. It can set neither to indicate it doesn't support type 2 windows at all. * modify existing provides and consumers code to the new param of ib_alloc_mw and the ib_mw_bind_info structure Signed-off-by: -
Shani Michaeli authored
MR deregistration fails when memory windows are bound to the MR. Handle such failures by propagating them to the caller ULP. Signed-off-by:
-
Shani Michaeli authored
The MPT - Memory Protection Table - is used by both memory windows and memory regions. Hence, all MPT references are relevant for both types of memory objects. Rename the relevant functions to start with mpt_ instead of the current mr_ prefix. Signed-off-by:
-
Shani Michaeli authored
Remove unused fields from the local invalidate WQE segment structure. Signed-off-by:
-
- 19 Feb, 2013 3 commits
-
-
Itai Garbi authored
If the ipoib client info isn't found on the _remove_one callback, we must not attempt to scan the returned null list. Found by Coverity. Signed-off-by:
Itai Garbi <igarbi@mellanox.com> Signed-off-by:
-
Yan Burman authored
Implement version info as well as report firmware version and bus info of the underlying IB HW device. Signed-off-by:
Yan Burman <yanb@mellanox.com> Signed-off-by:
-
Shlomo Pongratz authored
The hash function introduced in commit b63b70d8 ("IPoIB: Use a private hash table for path lookup in xmit path") was designd to use the 3 octets of the IPoIB HW address that holds the remote QPN. However, this currently isn't the case on little-endian machines, because the the code there uses the flags part (octet[0]) and not the last octet of the QPN (octet[3]). Fix this. The fix caused a checkpatch warning on line over 80 characters, to solve that changed the name of the temp variable that holds the daddr. Signed-off-by:
Shlomo Pongratz <shlomop@mellanox.com> Signed-off-by:
-
- 15 Feb, 2013 5 commits
-
-
Julia Lawall authored
Delete successive tests to the same location. The code tested the result of a previous allocation, that itself was already tested. It is changed to test the result of the most recent allocation. A simplified version of the semantic match that finds this problem is as follows: (http://coccinelle.lip6.fr/) // <smpl> @s exists@ local idexpression y; expression x,e; @@ *if ( \(x == NULL\|IS_ERR(x)\|y != 0\) ) { ... when forall return ...; } ... when != \(y = e\|y += e\|y -= e\|y |= e\|y &= e\|y++\|y--\|&y\) when != \(XT_GETPAGE(...,y)\|WMI_CMD_BUF(...)\) *if ( \(x == NULL\|IS_ERR(x)\|y != 0\) ) { ... when forall return ...; } // </smpl> Signed-off-by:
Julia Lawall <Julia.Lawall@lip6.fr> Signed-off-by:
-
Dan Carpenter authored
We have to decrement "i" before calling mlx4_ib_free_demux_ctx() or we free something that wasn't allocated. That's fine for free_pv_object() but it would lead to a NULL dereference calling mlx4_ib_free_demux_ctx(). The null dereference is because ->tun is NULL when we check: if (!ctx->tun[i]) Also we didn't free ->sriov.demux[0] so it was a small leak. Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
-
Wei Yongjun authored
Use the module_pci_driver() macro to make the code simpler by eliminating module_init and module_exit calls. dpatch engine is used to auto generate this patch. (https://github.com/weiyj/dpatch) Signed-off-by:
Wei Yongjun <yongjun_wei@trendmicro.com.cn> Reviewed-by:
Steve WIse <swise@opengridcomputing.com> Signed-off-by:
-
Stefan Hasko authored
Fix compile warning about cast to pointer from integer of different size. Signed-off-by:
Stefan Hasko <hasko.stevo@gmail.com> Acked-by:
-
Mike Marciniszyn authored
remove_qp() can execute concurrently with a qib_lookup_qpn() on another CPU, which in of itself, is ok, given the RCU locking. The issue is that remove_qp() NULLs out the qp->next field so that a qib_lookup_qpn() might fail to find a qp if it occurs after the one that is being deleted. This is a momentary issue and subsequent qib_lookup_qpn() calls would find the qp's since the search restarts from the bucket head. At scale, the issue might causes dropped packets and unnecessary retransmissions. The fix just deletes the qp->next NULL assignment to prevent the remove_qp() from hiding qp's from qib_lookup_qpn(). Reviewed-by:
Dean Luick <dean.luick@intel.com> Signed-off-by:
Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by:
-
- 14 Feb, 2013 10 commits
-
-
Paul Bolle authored
Work requests are passed between the host and the firmware with a "cookie". This cookie is swapped to big-endian when passed to the firmware and back to host endianness on return. This swapping seems to be implemented incorrectly. Moreover, the byte swapping triggers GCC warnings on 32 bit: drivers/infiniband/hw/cxgb4/cm.c: In function ‘passive_ofld_conn_reply’: drivers/infiniband/hw/cxgb4/cm.c:2803:12: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast] drivers/infiniband/hw/cxgb4/cm.c: In function ‘send_fw_pass_open_req’: drivers/infiniband/hw/cxgb4/cm.c:2941:16: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast] [...] But byte swapping isn't needed as the firmware doesn't actually touch the cookie. Dropping byte swapping makes the warnings go away too. Signed-off-by: -
Vipul Pandya authored
Fixe the following types of sparse warnings - cast to pointer from integer of different size - cast from pointer to integer of different size - incorrect type in assignment (different base types) - incorrect type in argument 1 (different base types) - cast from restricted __be64 - cast from restricted __be32 Signed-off-by:
Vipul Pandya <vipul@chelsio.com> Signed-off-by:
-
Vipul Pandya authored
CPL_ABORT_REQ_RSS can come before TCP connection is established. In such case peer_abort was trying to remove the hwtid, which was not inserted. To avoid this we insert the hwtid when we are sure that we are surely going to send passive accept request. Signed-off-by:
-
Vipul Pandya authored
Don't wakeup threads blocked in rdma_init/rdma_fini if we are on MPAv2, and want to retry connection with MPAv1. Stop ep-timer on getting MPA version mismatch, before doing the abort_connection - in process_mpa_request. Take care to stop ep-timer in error paths for process_mpa_request. Signed-off-by:
-
Vipul Pandya authored
Only reconnect if the endpoint wasn't freed. peer_abort() should only attempt to reconnect if the endpoint wasn't freed. Also remove hwtid from the debugfs idr. Add missing check for peer2peer in MPAv2 code Use correct mpa version on reject. Signed-off-by:
-
Vipul Pandya authored
The endpoint timeout logic had a race that could cause an endpoint object to be freed while it was still on the timedout list. This can happen if the timer is stopped after it had fired, but before the timedout thread processed the endpoint timeout. Signed-off-by:
-
Vipul Pandya authored
With newer firmware, we can get streaming data due to connection errors before the driver moves the QP out of RTS. Signed-off-by:
-
Vipul Pandya authored
Log AEs even if the QP isn't in RTS. It is useful information. Signed-off-by:
-
Vipul Pandya authored
The driver is currently releasing the last ref on the QP too early. This can cause bus errors due to HW still fetching WRs from the HW queue. The fix is to keep a qp ref until we release the HW TID. Signed-off-by:
-
Vipul Pandya authored
With later firmware, the chances of getting streaming mode data after we exit RTS is likely, so we don't need to warn for it. The only real case where we don't expect it is when the QP is in RTS. Move QP to ERROR when streaming mode data received. Signed-off-by:
-
