- Mar 02, 2010
-
-
Roland Dreier authored
-
Roland Dreier authored
-
Roland Dreier authored
-
Roland Dreier authored
-
Roland Dreier authored
-
Roland Dreier authored
-
Roland Dreier authored
-
- Feb 26, 2010
-
-
Chien Tung authored
Add support for KR device id 0x0110. While at it, cleanup nes_init_phy() by splitting it into nes_init_1g_phy() and nes_init_2025_phy(). Remove support for NES_PHY_TYPE_IRIS, which was used on an XFP board that was only manufactured in small quantities and given out for evals in even smaller quantities. Signed-off-by:
Chien Tung <chien.tin.tung@intel.com> Signed-off-by:
Roland Dreier <rolandd@cisco.com>
-
- Feb 25, 2010
-
-
Steve Wise authored
If cxgb3 calls the iw_cxgb3 t3cclient remove function due to a device removal event, then the iwch device must be marked with CXIO_ERROR_FATAL since the device below us is going away. Otherwise, we can get stuck in a deadlock as RDMA ULPs try and deallocate objects (like MRs, QPs, etc). So always mark the device with CXIO_ERROR_FATAL when removing. Signed-off-by:
Steve Wise <swise@opengridcomputing.com> Signed-off-by:
-
Steve Wise authored
Only kernel mode CQs need the SW queue memory allocated. The SW queue for user mode CQs is allocated in userspace by libcxgb3. Signed-off-by:
-
Steve Wise authored
Signed-off-by:
-
Steve Wise authored
T3 hardware doorbell FIFO overflows can cause application stalls due to lost doorbell ring events. This has been seen when running large NP IMB alltoall MPI jobs. The T3 hardware supports an xon/xoff-type flow control mechanism to help avoid overflowing the HW doorbell FIFO. This patch uses these interrupts to disable RDMA QP doorbell rings when we near an overflow condition, and then turn them back on (and ring all the active QP doorbells) when when the doorbell FIFO empties out. In addition if an doorbell ring is dropped by the hardware, the code will now recover. Design: cxgb3: - enable these DB interrupts - in the interrupt handler, schedule work tasks to call the ULPs event handlers with the new events. - ring all the qset txqs when an overflow is detected. iw_cxgb3: - disable db ringing on all active qps when we get the DB_FULL event - enable db ringing on all active qps and ring all active dbs when we get the DB_EMPTY event - On DB_DROP event: - disable db rings in the event handler - delay-schedule a work task which rings and enables the dbs on all active qps. - in post_send and post_recv logic, don't ring the db if it's disabled. Signed-off-by: -
Alexander Chiang authored
A small change to reduce the size of ib_device to 1112 bytes (from 1128). Signed-off-by:
Alex Chiang <achiang@hp.com> Signed-off-by:
-
Alexander Chiang authored
As shown when 'let c_space_errors=1' is set in vim. Signed-off-by:
-
Alexander Chiang authored
Some large systems may support more than IB_UCM_MAX_DEVICES (currently 32). This change allows us to support more devices in a backwards-compatible manner. the first IB_UCM_MAX_DEVICES keep the same major/minor device numbers they've always had. If there are more than IB_UCM_MAX_DEVICES, then we dynamically request a new major device number (new minors start at 0). Signed-off-by:
-
Alexander Chiang authored
This change is not useful by itself, but sets us up for a future change that allows us to support more than IB_UCM_MAX_DEVICES. Signed-off-by:
-
Alexander Chiang authored
This change is not useful by itself, but sets us up for a future change that allows us to dynamically allocate device numbers in case we have more than IB_UCM_MAX_DEVICES in the system. Signed-off-by:
-
Alexander Chiang authored
Clean errors as shown when 'let c_space_errors=1' is set in vim. Signed-off-by:
-
Alexander Chiang authored
Some large systems may support more than IB_UMAD_MAX_PORTS (currently 64). This change allows us to support more ports in a backwards-compatible manner. The first IB_UMAD_MAX_PORTS keep the same major/minor device numbers they've always had. If there are more than IB_UMAD_MAX_PORTS, we then dynamically request a new major device number (new minors start at 0). Signed-off-by:
-
Alexander Chiang authored
This change is not useful by itself, but sets us up for a future change that allows us to support more than IB_UMAD_MAX_PORTS in a system. Signed-off-by:
-
Alexander Chiang authored
This change is not useful by itself, but sets us up for a future change that allows us to dynamically allocate device numbers in case we have more than IB_UMAD_MAX_PORTS in the system. Signed-off-by:
-
Alexander Chiang authored
We no longer need this data structure, as it was used to associate an inode back to a struct ib_umad_port during ->open(). But now that we're embedding a struct cdev in struct ib_umad_port, we can use the container_of() macro to go from the inode back to the device instead. Signed-off-by:
-
Alexander Chiang authored
Instead of storing pointers to cdev and sm_cdev, embed the full structures instead. This change allows us to use the container_of() macro in ib_umad_open() and ib_umad_sm_open() in a future patch. This change increases the size of struct ib_umad_port to 320 bytes from 128. Signed-off-by:
-
Alexander Chiang authored
Clean up the errors as shown when 'let c_space_errors=1' is set in vim. Signed-off-by:
-
Alexander Chiang authored
Eliminate some padding in the structure by rearranging the members. sizeof(struct ib_uverbs_event_file) is now 72 bytes (from 80) and more members now fit in the first cacheline. Signed-off-by:
-
Alexander Chiang authored
Some large systems may support more than IB_UVERBS_MAX_DEVICES (currently 32). This change allows us to support more devices in a backwards-compatible manner. The first IB_UVERBS_MAX_DEVICES keep the same major/minor device numbers that they've always had. If there are more than IB_UVERBS_MAX_DEVICES, we then dynamically request a new major device number (new minors start at 0). This change increases the maximum number of HCAs to 64 (from 32). Signed-off-by:
-
Alexander Chiang authored
This change is not useful by itself, but sets us up for a future change that allows us to support more than IB_UVERBS_MAX_DEVICES in a system. Signed-off-by:
-
Alexander Chiang authored
This change is not useful by itself, but it sets us up for a future change that allows us to dynamically allocate device numbers in case we have more than IB_UVERBS_MAX_DEVICES in the system. Signed-off-by:
-
Alexander Chiang authored
dev_table's raison d'etre was to associate an inode back to a struct ib_uverbs_device. However, now that we've converted ib_uverbs_device to contain an embedded cdev (instead of a *cdev), we can use the container_of() macro and cast back to the containing device. There's no longer any need for dev_table, so get rid of it. Signed-off-by:
-
Alexander Chiang authored
Instead of storing a pointer to a cdev, embed the entire struct cdev. This change allows us to use the container_of() macro in ib_uverbs_open() in a future patch. This change increases the size of struct ib_uverbs_device to 168 bytes across 3 cachelines from 80 bytes in 2 cachelines. However, we rearrange the members so that everything fits into the first cacheline except for the struct cdev. Finally, we don't touch the cdev in any fastpaths, so this change shouldn't negatively affect performance. Signed-off-by:
-
Or Gerlitz authored
Currently the iSER receive completion flow takes the session lock twice. Optimize it to avoid the first one by letting iser_task_rdma_finalize() be called only from the cleanup_task callback invoked by iscsi_free_task, thus reducing the contention on the session lock between the scsi command submission to the scsi command completion flows. Signed-off-by:
Or Gerlitz <ogerlitz@voltaire.com> Reviewed-by:
Mike Christie <michaelc@cs.wisc.edu> Signed-off-by:
-
Or Gerlitz authored
libiscsi passthrough mode invokes the transport xmit calls directly without first going through an internal queue, unlike the other mode, which uses a queue and a xmitworker thread. Now that the "cant_sleep" prerequisite of iscsi_host_alloc is met, move to use it. Handling xmit errors is now done by the passthrough flow of libiscsi. Since the queue/worker aren't used in this mode, the code that schedules the xmitworker is removed. Signed-off-by:
-
Or Gerlitz authored
Remove unnecessary checks for the IB connection state and for QP overflow, as conn state changes are reported by iSER to libiscsi and handled there. QP overflow is theoretically possible only when unsolicited data-outs are used; anyway it's being checked and handled by HW drivers. Signed-off-by:
-
Or Gerlitz authored
Two minor flows in iSER's data path still use allocations; move them to be atomic as a preperation step towards moving to use libiscsi passthrough mode. Signed-off-by:
-
Or Gerlitz authored
Simplify and shrink the logic/code used for the send descriptors. Changes include removing struct iser_dto (an unnecessary abstraction), using struct iser_regd_buf only for handling SCSI commands, using dma_sync instead of dma_map/unmap, etc. Signed-off-by:
-
Or Gerlitz authored
Use a different CQ for send completions, where send completions are polled by the interrupt-driven receive completion handler. Therefore, interrupts aren't used for the send CQ. Signed-off-by:
-
Or Gerlitz authored
Now that both the posting and reaping of receive buffers is done in the completion path, the counter of outstanding buffers not be atomic. Signed-off-by:
-
Or Gerlitz authored
Currently, the recv buffer posting logic is based on the transactional nature of iSER which allows for posting a buffer before sending a PDU. Change this to post only when the number of outstanding recv buffers is below a water mark and in a batched manner, thus simplifying and optimizing the data path. Use a pre-allocated ring of recv buffers instead of allocating from kmem cache. A special treatment is given to the login response buffer whose size must be 8K unlike the size of buffers used for any other purpose which is 128 bytes. Signed-off-by:
-
Or Gerlitz authored
We will make a major change in the recv buffer posting logic, after which the problem commit bba7ebba "avoid recv buffer exhaustion caused by unexpected PDUs" comes to solve doesn't exist any more, so revert it. Signed-off-by:
-
- Feb 20, 2010
-
-
Or Gerlitz authored
Change the nes driver to return -ENOMEM on SQ/RQ overflow to match the return code of other RDMA HW drivers (e.g cxgb3, ehca, mlx4, mthca). Signed-off-by:
-
