- 19 Dec, 2018 12 commits
-
-
Ewan D. Milne authored
Certain older adapters such as the OneConnect OCe10100 may not have a valid wqpcnt value. In this case, do not set queue->page_count to 0 in lpfc_sli4_queue_alloc() as this will prevent the driver from initializing. Fixes: 895427bd ("scsi: lpfc: NVME Initiator: Base modifications") Cc: stable@vger.kernel.org # 4.11+ Signed-off-by:
Ewan D. Milne <emilne@redhat.com> Reviewed-by:
Laurence Oberman <loberman@redhat.com> Tested-by:
Martin K. Petersen <martin.petersen@oracle.com>
-
Christoph Hellwig authored
The same effects can be achieved by setting the dma_boundary to PAGE_SIZE - 1 and the max_segment_size to PAGE_SIZE, so shift those settings into the drivers. Note that in many cases the setting might be bogus, but this keeps the status quo. [mkp: fix myrs and myrb] Signed-off-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
-
Christoph Hellwig authored
mac53c94 has no limitations on crossing pages for segments. Just make the 65535 byte segment size limit explicit, even if it matches the current block layer limit. Signed-off-by:
Paul Mackerras <paulus@ozlabs.org> Signed-off-by:
-
Christoph Hellwig authored
mesh has no limitations on crossing pages for segments. Just make the 65535 byte segment size limit explicit, even if it matches the current block layer limit. Signed-off-by:
-
Christoph Hellwig authored
There is no such limitation in the protocol or implementation, so remove it. Signed-off-by:
Juergen Gross <jgross@suse.com> Signed-off-by:
-
Christoph Hellwig authored
This driver already sets the dma_boundary to PAGE_SIZE - 1, which has the same result. Signed-off-by:
-
Christoph Hellwig authored
This driver already sets the dma_boundary to PAGE_SIZE - 1, which has the same result. Signed-off-by:
-
Christoph Hellwig authored
This allows the host driver to indicate the maximum supported segment size in a nice an easy way, so that the driver doesn't have to worry about DMA-layer imposed limitations. Signed-off-by:
Bart Van Assche <bvanassche@acm.org> Signed-off-by:
-
Christoph Hellwig authored
Most SCSI drivers want to enable "clustering", that is merging of segments so that they might span more than a single page. Remove the ENABLE_CLUSTERING define, and require drivers to explicitly set DISABLE_CLUSTERING to disable this feature. Signed-off-by:
-
Christoph Hellwig authored
A few drivers were not setting the use_clustering flag at all and thus default to disable. Fix them up to explicitly set this field in preparation for additional cleanups. Signed-off-by:
-
Wei Yongjun authored
memory allocated by kmem_cache_alloc() should be freed using kmem_cache_free(), not kfree(). Fixes: ad669505 ("scsi: target/core: Make sure that target_wait_for_sess_cmds() waits long enough") Signed-off-by:
Wei Yongjun <weiyongjun1@huawei.com> Reviewed-by:
-
Bart Van Assche authored
Commit b5b6e8c8 ("scsi: virtio_scsi: fix IO hang caused by automatic irq vector affinity") removed all virtio_scsi hostdata users. Since the SCSI host data is no longer used, also remove the host data itself. Cc: Paolo Bonzini <pbonzini@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Ming Lei <ming.lei@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Signed-off-by:
Ming Lei <ming.lei@redhat.com> Signed-off-by:
-
- 13 Dec, 2018 16 commits
-
-
YueHaibing authored
Remove .owner field if calls are used which set it automatically Generated by: scripts/coccinelle/api/platform_no_drv_owner.cocci Signed-off-by:
YueHaibing <yuehaibing@huawei.com> Signed-off-by:
-
Suganath Prabu authored
Update driver version from 27.100.00.00 to 27.101.00.00. Signed-off-by:
Suganath Prabu <suganath-prabu.subramani@broadcom.com> Signed-off-by:
-
Suganath Prabu authored
Use ioc->base_readl to restrict the readl retries to only Aero controllers. Signed-off-by:
-
Suganath Prabu authored
Sometimes Aero controllers appears to be returning bad data (0) for doorbell register read and if retries are performed immediately after the bad read, they return good data. Workaround is added to retry read from doorbell registers for maximum three times if driver get the zero. Added functions base_readl_aero for Aero IOC and base_readl for gen35 and other controllers. Signed-off-by:
-
Suganath Prabu authored
Adding flag "is_aero_ioc" to differentiate aero based controllers from other gen35 controllers. Signed-off-by:
-
Xiang Chen authored
Sht->sg_tablesize is set in the driver, and it will be assigned to shost->sg_tablesize in SCSI mid-layer. So it is not necessary to assign shost->sg_table one more time in the driver. In addition to the change, change each scsi_host_template.sg_tablesize to HISI_SAS_SGE_PAGE_CNT instead of SG_ALL. Signed-off-by:
Xiang Chen <chenxiang66@hisilicon.com> Signed-off-by:
John Garry <john.garry@huawei.com> Signed-off-by:
-
Xiang Chen authored
Relocate the codes related to dma_map/unmap in hisi_sas_task_prep() to reduce complexity, with a view to add DIF/DIX support. Signed-off-by:
-
John Garry authored
This patchset fixes some warnings detected by the sparse tool, like these: drivers/scsi/hisi_sas/hisi_sas_main.c:1469:52: warning: incorrect type in assignment (different base types) drivers/scsi/hisi_sas/hisi_sas_main.c:1469:52: expected unsigned short [unsigned] [assigned] [usertype] tag_of_task_to_be_managed drivers/scsi/hisi_sas/hisi_sas_main.c:1469:52: got restricted __le16 [usertype] <noident> drivers/scsi/hisi_sas/hisi_sas_main.c:1723:52: warning: incorrect type in assignment (different base types) drivers/scsi/hisi_sas/hisi_sas_main.c:1723:52: expected unsigned short [unsigned] [assigned] [usertype] tag_of_task_to_be_managed drivers/scsi/hisi_sas/hisi_sas_main.c:1723:52: got restricted __le16 [usertype] <noident> Signed-off-by:
-
Himanshu Madhani authored
Signed-off-by:
Himanshu Madhani <hmadhani@marvell.com> Signed-off-by:
-
Giridhar Malavali authored
This patch fixes NVMe discovery by setting SKIP_PRLI flag, so that PRLI is driven by driver and is retried when the NPIV port is detected to have NVMe capability. Signed-off-by:
Giridhar Malavali <gmalavali@marvell.com> Signed-off-by:
-
Himanshu Madhani authored
This patch fixes issues with NPIV port with FC-NVMe. Clean up code for remoteport delete and also call nvme_delete when deleting VPs. Signed-off-by:
-
Anil Gurumurthy authored
Signed-off-by:
Anil Gurumurthy <agurumurthy@marvell.com> Signed-off-by:
-
Arnd Bergmann authored
The event_wait semaphore has completion semantics, so we can change it over to the completion interface for clarity without changing the behavior. Signed-off-by:
Arnd Bergmann <arnd@arndb.de> Reviewed-by:
Johannes Thumshirn <jthumshirn@suse.de> Reviewed-by:
Dave Carroll <david.carroll@microsemi.com> Signed-off-by:
-
Arnd Bergmann authored
The wait_sem member is used like a completion, so we should use the respective API. The behavior is unchanged. Signed-off-by:
-
James Smart authored
Addition of support for if_type=6 missed several checks for interface type, resulting in the failure of several key management features such as firmware dump and loopback testing. Correct the checks on the if_type so that both SLI4 IF_TYPE's 2 and 6 are supported. Signed-off-by:
Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by:
James Smart <jsmart2021@gmail.com> Reviewed-by:
-
Martin K. Petersen authored
This reverts commit 287aba25. We killed the bad firmware and this mod is no longer necessary. Signed-off-by:
James Smart <james.smart@broadcom.com> Signed-off-by:
-
- 08 Dec, 2018 12 commits
-
-
James Smart authored
Update the driver version to 12.0.0.9 Signed-off-by:
-
James Smart authored
When dif and first burst is used in a write command wqe, the driver was not properly setting fields in the io command request. This resulted in no dif bytes being sent and invalid xfer_rdy's, resulting in the io being aborted by the hardware. Correct the wqe initializaton when both dif and first burst are used. Signed-off-by:
-
James Smart authored
On driver termination, after the driver stops fw logging by writing a register on the chip, the driver immediately unmaps and frees the logging buffer, without confirming in any way that the chip has received the write and terminated the logging. As termination on the chip is not immediate, the chip may issue a dma request to the now unmapped dma buffer, resulting in a iommu fault. Change the driver to receive a confirmation that logging ahs been terminated. As the driver always issues an SLI reset with the device as part of shutdown, and as part of that is receiving confirmation that the reset is complete - the driver was modified to perform the write to disable fw logging prior to the SLI reset and only free the fw log buffer after the SLI reset is complete. That guarantees use of the fw log buffer is fully terminated when it is unmapped. Signed-off-by:
-
James Smart authored
Driver missed classifying the chip type for G7 when reporting supported topologies. This resulted in loop being shown as supported on FC links that are not supported per the standard. Add the chip classifications to the topology checks in the driver. Signed-off-by:
-
James Smart authored
Driver is setting bits in word 10 of the SLI4 ABORT WQE (the wqid). The field was a carry over from a prior SLI revision. The field does not exist in SLI4, and the action may result in an overlap with future definition of the WQE. Remove the setting of WQID in the ABORT WQE. Also cleaned up WQE field settings - initialize to zero, don't bother to set fields to zero. Signed-off-by:
-
James Smart authored
The current discovery state machine the driver treated FLOGI oddly. When point to point, an FLOGI is to be exchanged by the two ports, with the port with the most significant WWN then proceeding with PLOGI. The implementation in the driver was keyed to closely with "what have I sent", not with what has happened between the two endpoints. Thus, it blatantly would ACC an FLOGI, but reject PLOGI's until it had its FLOGI ACC'd. The problem is - the sending of FLOGI may be delayed for some reason, or the response to FLOGI held off by the other side. In the failing situation the other side sent an FLOGI, which was ACC'd, then sent PLOGIs which were then rjt'd until the retry count for the PLOGIs were exceeded and the port gave up. The FLOGI may have been very late in transmit, or the response held off until the PLOGIs failed. Given the other port had the higher WWN, no PLOGIs would occur and communication stopped. Correct the situation by changing the FLOGI handling. Defer any response to an FLOGI until the driver has sent its FLOGI as well. Then, upon either completion of the sent FLOGI, or upon sending an ACC to a received FLOGI (which may be received before or just after FLOGI was sent). the driver will act on who has the higher WWN. if the other port does, the driver will noop any handling of an FLOGI response (if outstanding) and wait for PLOGI. If the local port does, the driver will transition to sending PLOGI and will noop any action on responding to an FLOGI (if not yet received). Fortunately, to implement this, it only took another state flag and deferring any FLOGI response if the FLOGI has yet to be transmit. All subsequent actions were already in place. Signed-off-by:
-
James Smart authored
In some link initialization sequences, the fw generates an erroneous FLOGI payload to the driver without an intervening link bounce. The driver, when it sees a 2nd FLOGI without an intervening link bounce, automatically performs a link bounce. In this, the link bounce causes the situate to repeat and in a nasty loop of link bounces. Resolve the issue by validating the FLOGI payload. The erroneous FLOGI will contain VVL signatures that are not normal. When the driver sees these, it will simply reject the flogi rather than bouncing the link. The reject is consumed within the firmware. Signed-off-by:
-
James Smart authored
Two initiator ports were cable swapped and after swap both went down. The driver internally swaps the nlp nodes based on matching node wwn's but not the same nport id as before. After detecting a change in the nodes RPI, the driver sends an UNREG_RPI command and clears the NLP_RPI_REGISTERED flag, then swaps the node information with the other node. But the other node's NLP_RPI_REGISTERED flag is also cleared, but it is done so without an UNREG_RPI being sent, which causes the later REG_RPI for that other node to fail as the hardware believes its still registered. Additionally, if the node swap occurred while the two nodes had PLOGI's in flight, the fc4_types weren't properly getting swapped such that when the PLOGIs commpleted and PRLI's were then sent, the PRLI's acted on bad protocol types so the PRLI was for the wrong protocol. NVME devices saw SCSI FCP PRLIs and vice versa. Clean up the node swap so that the NLP_RPI_REGISTERED flag is handled properly. Fix the handling of the fc4_types when the nodes are swapped as well Signed-off-by:
-
James Smart authored
Depending on the chipset, the number of NPIV vports may vary and be in excess of what most switches support (256). To avoid confusion with the users, limit the reported NPIV vports to 256. Additionally correct the 16G adapter which is reporting a bogus NPIV vport number if the link is down. Signed-off-by:
-
James Smart authored
Driver is hitting null pring pointers in lpfc_do_work(). Pointer assignment occurs based on SLI-revision. If recovering after an error, its possible the sli revision for the port was cleared, making the lpfc_phba_elsring() not return a ring pointer, thus the null pointer. Add SLI revision checking to lpfc_phba_elsring() and status checking to all callers. Signed-off-by:
-
James Smart authored
Renumber one of the 0711 log messages so there isn't a duplication. Signed-off-by:
-
James Smart authored
The driver is getting hit with 100s of RSCNs during remote port address changes. Each of those RSCN's ends up generating UNREG_RPI and REG_PRI mailbox commands. The discovery engine within the driver doesn't wait for the mailbox command completions. Instead it sets state flags and moves forward. At some point, there's a massive backlog of mailbox commands which take time for the adapter to process. Additionally, it appears there were duplicate events from the switch so the driver generated duplicate mailbox commands for the same remote port. During this window, failures on PLOGI and PRLI ELS's are see as the adapter is rejecting them as they are for remote ports that still have pending mailbox commands. Streamline the discovery engine so that PLOGI log checks for outstanding UNREG_RPIs and defer the processing until the commands complete. This better synchronizes the ELS transmission vs the RPI registrations. Filter out multiple UNREG_RPIs being queued up for the same remote port. Beef up log messages in this area. Signed-off-by:
-
