- 01 Jun, 2015 1 commit
-
-
Webb Scales authored
Synchronize completion the reset with completion of outstanding commands Extending the newly-added synchronous abort functionality, now also synchronize resets with the completion of outstanding commands. Rename the wait queue to reflect the fact that it's being used for both types of waits. Also, don't complete commands which are terminated due to a reset operation. fix for controller lockup during reset Reviewed-by:
Scott Teel <scott.teel@pmcs.com> Reviewed-by:
Kevin Barnett <kevin.barnett@pmcs.com> Reviewed-by:
Tomas Henzl <thenzl@redhat.com> Reviewed-by:
Hannes Reinecke <hare@Suse.de> Signed-off-by:
Webb Scales <webbnh@hp.com> Signed-off-by:
Don Brace <don.brace@pmcs.com> Reviewed-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
James Bottomley <JBottomley@Odin.com>
-
- 31 May, 2015 39 commits
-
-
Robert Elliott authored
If hpsa_wait_for_board_state fails, hpsa_kdump_soft_reset should propagate its return value (e.g., -ENODEV) rather than just returning -1. Reviewed-by:
Robert Elliott <elliott@hp.com> Signed-off-by:
-
Robert Elliott authored
Rather than numbering the hpsa controllers with an incrementing 0..n value (e.g., that shows up in /proc/interrupts), use the scsi midlayer host_no (e.g. matching /sys/class/scsi_host/hostNN). Reviewed-by:
-
Webb Scales authored
Rework slave allocation: - separate the tagging support setup from the hostdata setup - make the hostdata setup act consistently when the lookup fails - make the hostdata setup act consistently when the device is not added - set up the queue depth consistently across these scenarios - if the block layer mq support is not available, explicitly enable and activate the SCSI layer tcq support (and do this at allocation-time so that the tags will be available for INQUIRY commands) Tweak slave configuration so that devices which are masked are also not attached. Reviewed-by: -
Robert Elliott authored
Add the interrupt number to the interrupt names that appear in /proc/interrupts, so they are unique Also, delete the IRQ and DAC prints. Other parts of the kernel already print the IRQ assignments, and dual-address-cycle support has not been interesting since the parallel PCI bus went from 32 to 64 bits wide. Reviewed-by:
-
Robert Elliott authored
Don't create the resubmit workqueue in hpsa_init_one until everything else is ready to use, so everything can be freed in reverse order of when they were allocated without risking freeing things while workqueue items are still active. Destroy the workqueue in the right order in hpsa_undo_allocations_after_kdump_soft_reset too. Reviewed-by:
-
Robert Elliott authored
If registering the special interrupt handlers in hpsa_init_one before a soft reset fails, the error exit needs to deallocate everything that was allocated before. Reviewed-by:
-
Robert Elliott authored
In hpsa_undo_allocations_after_kdump_soft_reset, the things allocated in hpsa_init_one step 2 - h->resubmit_wq and h->lockup_detected need to be freed, in the right order. Reviewed-by:
-
Robert Elliott authored
If try_soft_reset fails to re-allocate irqs, the error exit starts with free_irq calls, which generate kernel WARN messages since they were already freed a few lines earlier. Jump to the next exit label to skip the free_irq calls. Reviewed-by:
-
Robert Elliott authored
Despite the fact that PCI devices are enabled in this order: 1. pci_enable_device 2. pci_request_regions Documentation/PCI/pci.txt specifies that they be undone in this order 1. pci_disable_device 2. pci_release_regions Tested by injecting error in the call to pci_enable_device in hpsa_init_one -> hpsa_pci_init: [ 9.095001] hpsa 0000:04:00.0: failed to enable PCI device [ 9.095005] hpsa: probe of 0000:04:00.0 failed with error -22 (-22 is -EINVAL) and then in the call pci_request_regions: [ 9.178623] hpsa 0000:04:00.0: failed to obtain PCI resources [ 9.178671] hpsa: probe of 0000:04:00.0 failed with error -16 (-16 is -EBUSY) and then by adding reset_devices to the kernel command line and inject errors into the two calls to pci_enable_device and the call to pci_request_regions in hpsa_init_one -> hpsa_init_reset_devices. (inject on 6th call, 1st to hpsa2) [ 62.413750] hpsa 0000:04:00.0: Failed to enable PCI device (inject on 7th call, 2nd to hpsa2) [ 62.807571] hpsa 0000:04:00.0: failed to enable device. (inject on 8th call, 3rd to hpsa2) [ 62.697198] hpsa 0000:04:00.0: failed to obtain PCI resources [ 62.697234] hpsa: probe of 0000:04:00.0 failed with error -16 The reset_devices path calls return -ENODEV on failure rather than passing the result, which apparently doesn't cause the pci driver to print anything. Reviewed-by: -
Webb Scales authored
Divide the loop in hpsa_scatter_gather() into two, one for the initial SG list and a second one for the chained list, if any. This allows the conditional check which resets the indicies for the chained list to be performed outside the loop instead of being done on every iteration inside the loop. Reviewed-by:
-
Webb Scales authored
Factor out the code which sends the TEST_UNIT_READY from wait_for_device_to_become_ready() into its own function. Move the code which waits for the TEST_UNIT_READY from wait_for_device_to_become_ready() into its own function. If a logical drive has failed, resetting it will ensure outstanding commands are completed, but polling it with TURs after the reset will not work because the TURs will never report good status. So successful TUR should not be a condition of success for the device reset error handler. Reviewed-by:
-
Webb Scales authored
Don't return from the abort request until the target command is complete. Mark outstanding commands which have a pending abort, and do not send them to the host if we can avoid it. If the current command has been aborted, do not call the SCSI command completion routine from the I/O path: when the abort returns successfully, the SCSI mid-layer will handle the completion implicitly. The following race was possible in theory. 1. LLD is requested to abort a scsi command 2. scsi command completes 3. The struct CommandList associated with 2 is made available. 4. new io request to LLD to another LUN re-uses struct CommandList 5. abort handler follows scsi_cmnd->host_scribble and finds struct CommandList and tries to aborts it. Now we have aborted the wrong command. Fix by resetting the scsi_cmd field of struct CommandList upon completion and making the abort handler check that the scsi_cmd pointer in the CommadList struct matches the scsi_cmnd that it has been asked to abort. Reviewed-by:
-
Webb Scales authored
cleanup command completions Reviewed-by:
-
Stephen Cameron authored
add support for tmf when in ioaccel2 mode Reviewed-by:
Joe Handzik <joseph.t.handzik@hp.com> Signed-off-by:
-
Robert Elliott authored
The SCSI midlayer already prints more detail about completions, and has logging level options to filter them if not wanted. These just slow down the system if a lot of errors occur, stressing error handling even more. Reviewed-by:
-
Robert Elliott authored
report more useful information on aborts Reviewed-by:
-
Robert Elliott authored
Improve initialization error handling in hpsa_init_one Clean up style and indent issues Rename functions for consistency Improve error messaging on allocations Fix return status from hpsa_put_ctlr_into_performant_mode Correct free order in hpsa_init_one using new function hpsa_free_performant_mode Prevent inadvertent use of null pointers by nulling out the parent structures and zeroing out associated size variables. Reviewed-by:
-
Robert Elliott authored
correct return codes for error conditions Reviewed-by:
-
Robert Elliott authored
cmd_alloc can no longer return NULL, so don't check for NULL any more (which is unreachable code). Reviewed-by:
-
Joe Handzik authored
improve ioaccel2 error handling, including better handling of underrun statuses Reviewed-by:
-
Webb Scales authored
Increase the request size for ioaccel2 path. The error, if any, returned by hpsa_allocate_ioaccel2_sg_chain_blocks to hpsa_alloc_ioaccel2_cmd_and_bft should be returned upstream rather than assumed to be -ENOMEM. This differs slightly from hpsa_alloc_ioaccel1_cmd_and_bft, which does not call another hpsa_allocate function and only has -ENOMEM to return from some kmalloc calls. Reviewed-by:
-
Robert Elliott authored
refactor freeing of resources into more logical functions Reviewed-by:
-
Robert Elliott authored
refactor error cleanup and shutdown disable interrupts and pci_disable_device on critical failures add hpsa_free_cfgtables function Reviewed-by:
-
Robert Elliott authored
replace calls to hpsa_free_irqs_and_disable_msix with hpsa_free_irqs and hpsa_disable_interrupt_mode Reviewed-by:
-
Joe Handzik authored
get drive queue depth to help avoid task set full conditions. Reviewed-by:
-
Joe Handzik authored
use ioaccel2 path to submit I/O to physical drives in HBA mode Reviewed-by:
-
Robert Elliott authored
offload_enabled changes are deferred until after the added/updated prints occur, so the values are incorrect. defer printing SSD Smart Path Enabled status information until the information is correct Reviewed-by:
-
Webb Scales authored
clean up command submission Reviewed-by:
-
Stephen Cameron authored
allow the controller firmware to queue up commands when the ioaccel device queue is full. Reviewed-by:
-
Stephen Cameron authored
add error handling for failure when registering with SCSI subsystem. Reviewed-by:
-
Stephen Cameron authored
Factor out hpsa_cmd_init from cmd_alloc(). We also need this for resubmitting commands down the default RAID path when they have returned from the ioaccel paths with errors. In particular, reinitialize the cmd_type and busaddr fields as these will not be correct for submitting down the RAID stack path after ioaccel command completion. This saves time when submitting commands. Reviewed-by:
-
Robert Elliott authored
make function names more consistent and meaningful Reviewed-by:
-
Stephen Cameron authored
expose a detected lockup via sysfs Reviewed-by:
-
Stephen Cameron authored
In hba mode, we could get sense data in descriptor format so we need to handle that. It's possible for CommandStatus to have value 0x0D "TMF Function Status", which we should handle. We will get this from a P1224 when aborting a non-existent tag, for example. The "ScsiStatus" field of the errinfo field will contain the TMF function status value. Reviewed-by:
-
Stephen Cameron authored
make tracking of outstanding commands more robust Reviewed-by:
-
Stephen Cameron authored
Do not send aborts to logical devices that do not support aborts Instead of relying on what the Smart Array claims for supporting logical drives, simply try an abort and see how it responds at device discovery time. This way devices that do support aborts (e.g. MSA2000) can work and we do not waste time trying to send aborts to logical drives that do not support them (important for high IOPS devices.) While rescanning devices only test whether devices support aborts the first time we encounter a device rather than every time. Some Smart Arrays required aborts to be sent with tags in the wrong endian byte order. To avoid having to know about this, we would send two aborts with tags with each endian order. On high IOPS devices, this turns out to be not such a hot idea. So we now have a list of the devices that got the tag backwards, and we only send it one way. If all available commands are outstanding and the abort handler is invoked, the abort handler may not be able to allocate a command and may busy-wait excessivly. Reserve a small number of commands for the abort handler and limit the number of concurrent abort requests to the number of reserved commands. Reviewed-by:
-
Webb Scales authored
Allow driver initiated commands to have a timeout. It does not yet try to do anything with timeouts on such commands. We are sending a reset in order to get rid of a command we want to abort. If we make it return on the same reply queue as the command we want to abort, the completion of the aborted command will not race with the completion of the reset command. Rename hpsa_scsi_do_simple_cmd_core() to hpsa_scsi_do_simple_cmd(), since this function is the interface for issuing commands to the controller and not the "core" of that implementation. Add a parameter to it which allows the caller to specify the reply queue to be used. Modify existing callers to specify the default reply queue. Rename __hpsa_scsi_do_simple_cmd_core() to hpsa_scsi_do_simple_cmd_core(), since this routine is the "core" implementation of the "do simple command" function and there is no longer any other function with a similar name. Modify the existing callers of this routine (other than hpsa_scsi_do_simple_cmd()) to instead call hpsa_scsi_do_simple_cmd(), since it will now accept the reply_queue paramenter, and it provides a controller lock-up check. (Also, tweak two related message strings to make them distinct from each other.) Submitting a command to a locked up controller always results in a timeout, so check for controller lock-up before submitting. This is to enable fixing a race between command completions and abort completions on different reply queues in a subsequent patch. We want to be able to specify which reply queue an abort completion should occur on so that it cannot race the completion of the command it is trying to abort. The following race was possible in theory: 1. Abort command is sent to hardware. 2. Command to be aborted simultaneously completes on another reply queue. 3. Hardware receives abort command, decides command has already completed and indicates this to the driver via another different reply queue. 4. driver processes abort completion finds that the hardware does not know about the command, concludes that therefore the command cannot complete, returns SUCCESS indicating to the mid-layer that the scsi_cmnd may be re-used. 5. Command from step 2 is processed and completed back to scsi mid layer (after we already promised that would never happen.) Fix by forcing aborts to complete on the same reply queue as the command they are aborting. Piggybacking device rescanning functionality onto the lockup detection thread is not a good idea because if the controller locks up during device rescanning, then the thread could get stuck, then the lockup isn't detected. Use separate work queues for device rescanning and lockup detection. Detect controller lockup in abort handler. After a lockup is detected, return DO_NO_CONNECT which results in immediate termination of commands rather than DID_ERR which results in retries. Modify detect_controller_lockup() to return the result, to remove the need for a separate check. Reviewed-by: -
Webb Scales authored
We had a mix of formats used for specifying controller, bus, target, and lun address of devices. change to the format used by the scsi midlayer and upper layer (2:3:0:0) so you can easily follow the information from hpsa to scsi midlayer to sd upper layer. Also add this information: - product ID - vendor ID - RAID level - SSD Smath Path capable and enabled - exposure level (sg-only) Example: hpsa 0000:04:00.0: added scsi 2:0:0:0: Direct-Access HP LOGICAL VOLUME RAID-0 SSDSmartPathCap+ En+ Exp=4 scsi 2:0:0:0: Direct-Access HP LOGICAL VOLUME 10.0 PQ: 0 ANSI: 5 sd 2:0:0:0: [sdr] 12501713072 512-byte logical blocks: (6.40 TB/5.82 TiB) sd 2:0:0:0: [sdr] 4096-byte physical blocks sd 2:0:0:0: [sdr] Attached SCSI disk sd 2:0:0:0: Attached scsi generic sg20 type 0 Reviewed-by:
-
Stephen Cameron authored
Cache the ioaccel handle so that when we need to abort commands sent down the ioaccel2 path, we can look up the LUN ID in h->dev[] instead of having to do I/O to the controller. Add a field to elements in h->dev[] to keep track of how the device is exposed to the SCSI mid layer: Not at all, without an upper level driver (no_uld_attach) or normally exposed. Since masked physical devices are now present in h->dev[] array it would be perfectly possible to do echo scsi add-single-device 2 2 0 0 > /proc/scsi/scsi and bring them online. This was previously not allowed for masked physical devices. Ensure that the mapping of physical disks to logical drives gets updated in a consistent way when a RAID migration occurs and is not touched until updates to it are complete. now instead of doing CISS_REPORT_PHYSICAL to get the LUNID for the physical disk in hpsa_get_pdisk_of_ioaccel2(), just get it out of h->dev[] where we already have it cached. do not touch phys_disk[] for ioaccel enabled logical drives during rescan Reviewed-by:
-
