Commit 0e9973ed33 ("scsi: aacraid: Add periodic checks to see IOP reset
status") changed the way driver checks if a reset succeeded. Now, after an
IOP reset, aacraid immediately start polling a register to verify the reset
is complete.
This behavior cause regressions on the reset path in PowerPC (at least).
Since the delay after the IOP reset was removed by the aforementioned patch,
the fact driver just starts to read a register instantly after the reset
was issued (by writing in another register) "corrupts" the reset procedure,
which ends up failing all the time.
The issue highly impacted kdump on PowerPC, since on kdump path we
proactively issue a reset in adapter (through the reset_devices kernel
parameter).
This patch (re-)adds a delay right after IOP reset is issued. Empirically
we measured that 3 seconds is enough, but for safety reasons we delay
for 5s (and since it was 30s before, 5s is still a small amount).
For reference, without this patch we observe the following messages
on kdump kernel boot process:
[ 76.294] aacraid 0003:01:00.0: IOP reset failed
[ 76.294] aacraid 0003:01:00.0: ARC Reset attempt failed
[ 86.524] aacraid 0003:01:00.0: adapter kernel panic'd ff.
[ 86.524] aacraid 0003:01:00.0: Controller reset type is 3
[ 86.524] aacraid 0003:01:00.0: Issuing IOP reset
[146.534] aacraid 0003:01:00.0: IOP reset failed
[146.534] aacraid 0003:01:00.0: ARC Reset attempt failed
Fixes: 0e9973ed33 ("scsi: aacraid: Add periodic checks to see IOP reset status")
Cc: stable@vger.kernel.org # v4.13+
Signed-off-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com>
Acked-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Fix possible indexing array of bound for &aac->hba_map[bus][cid], where
bus and cid boundary check happens later.
Fixes: 0d643ff3c3 ("scsi: aacraid: use aac_tmf_callback for reset fib")
Signed-off-by: Nikola Pajkovsky <npajkovsky@suse.cz>
Reviewed-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The logic for supporting large drives was previously tied to 4Kn support
for SmartIOC-2000. As SmartIOC-2000 does not support volumes using 4Kn
drives, use the intended option flag AAC_OPT_NEW_COMM_64 to determine
support for volumes greater than 2T.
Cc: <stable@vger.kernel.org>
Signed-off-by: Dave Carroll <david.carroll@microsemi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
aac_convert_sgraw2() kmalloc memory and return -1 on error, which should
be -ENOMEM. However, nobody is checking return value, so with this
change, -ENOMEM is propagated to upper layer.
Signed-off-by: Nikola Pajkovsky <npajkovsky@suse.cz>
Reviewed-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
unsigned long byte_count = 0;
nseg = scsi_dma_map(scsicmd);
if (nseg < 0)
return nseg;
if (nseg) {
...
}
return byte_count;
is equal to
unsigned long byte_count = 0;
nseg = scsi_dma_map(scsicmd);
if (nseg <= 0)
return nseg;
...
return byte_count;
No other code has changed.
[mkp: fix checkpatch complaints]
Signed-off-by: Nikola Pajkovsky <npajkovsky@suse.cz>
Reviewed-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This fixes a potential race condition observed on Power systems.
Several places throughout the aacraid driver call aac_fib_send or
similar to send a command to the aacraid adapter, then check the return
code to determine if the command was actually sent to the adapter, then
update the phase field in the scsi command scratch pad area to track
that the firmware now owns this command. However, there is nothing that
ensures that by the time the aac_fib_send function returns and we go to
write to the scsi command, that the command hasn't already completed and
the scsi command has been freed. This was causing random crashes in the
TCP stack which was tracked down to be caused by memory that had been a
struct request + scsi_cmnd being now used for an skbuff. Memory
poisoning was enabled in the kernel to debug this which showed that the
last owner of the memory that had been freed was aacraid and that it was
a struct request. The memory that was corrupted was the exact data
pattern of AAC_OWNER_FIRMWARE and it was at the same offset that aacraid
writes, which is scsicmd->SCp.phase. The patch below resolves this
issue.
Cc: <stable@vger.kernel.org>
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Tested-by: Wen Xiong <wenxiong@linux.vnet.ibm.com>
Reviewed-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
aac_hba_send() will return FAILED for any non-SCSI command requests,
failing any TMFs. This patch updates the check to allow TMFs.
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When sending a reset fib we shouldn't rely on the scsi command,
but rather set the TMF status in the map_info->reset_state variable.
That allows us to send a TMF independent on a scsi command.
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
"qd.id" comes directly from the copy_from_user() on the line before so
we should verify that it's within bounds.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Both aac_send_raw_srb() and aac_get_hba_info() may copy stack allocated
structs to userspace without initializing all members of these
structs. Clear out this memory to prevent information leaks.
Fixes: 423400e64d ("scsi: aacraid: Include HBA direct interface")
Fixes: c799d519bf ("scsi: aacraid: Retrieve HBA host information ioctl")
Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The fields sense_data_size and sense_data are unitialized garbage from
the stack and are being copied back to userspace. Fix this leak of
stack information by ensuring they are zero'd.
Detected by CoverityScan, CID#1435473 ("Uninitialized scalar variable")
Fixes: 423400e64d ("scsi: aacraid: Include HBA direct interface")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The command thread checks the ctrl health periodically before sending
updates to the controller. The function that it uses is aac_check_health
which does more than get the health status.
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: David Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Now the driver issues a soft reset and waits for the controller to be up
and running by periodically checking on the status of the controller
health registers. Also prevents ARC adapters from issuing soft reset if
IOP resets failed.
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: David Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The default queue depth for non NATIVE RAW disks is calculated from the
number of fibs and number of disks or a max of 256. This causes poor disk
IO performance.
The fix is to set default qd based on the type of disks
(SATA -32 and SAS -64)
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: David Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The qd for ARC Native disks is calculated by dividing the max IO 1024
by the number of disks or 256 which ever is lower. This causes poor
disk IO performance.
The fix is set the qd based on the type of disk (SAS - 64 and SATA -
32).
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: David Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The driver changed the DMA consistent map after consistent memory was
allocated, this invalidated the IOMMU identity mapping. The fix was to
make sure that we set the DMA consistent mask setting once depending on
the controller card.
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The raw srb commands do not requires memory that in the ZONE_DMA memory
space. For 32bit srb commands use GFP_DMA32 to limit the memory to 32bit
memory range (4GB).
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
There were pci_alloc_consistent() failures on ARM64 platform. Use
dma_alloc_coherent() with GFP_KERNEL flag DMA memory allocations.
Signed-off-by: Mahesh Rajashekhara <mahesh.rajashekhara@microsemi.com>
[hch: tweaked indentation, removed memsets]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
During a PCI error recovery, if aac_check_health() is not aware that a
PCI error happened and we have an offline PCI channel, it might trigger
some errors (like NULL pointer dereference) and inhibit the error
recovery process to complete.
This patch makes the health check procedure aware of PCI channel issues,
and in case of error recovery process, the function
aac_adapter_check_health() returns -1 and let the recovery process to
complete successfully. This patch was tested on upstream kernel
v4.11-rc5 in PowerPC ppc64le architecture with adapter 9005:028d
(VID:DID) - the error recovery procedure was able to recover fine.
Fixes: 5c63f7f710 ("aacraid: Added EEH support")
Cc: stable@vger.kernel.org # v4.6+
Signed-off-by: Guilherme G. Piccoli <gpiccoli@linux.vnet.ibm.com>
Reviewed-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently, command threads fails to return ioctls commands for older
controller versions, since it returns when all the fibs have been
allocated. Another issue is even all the fibs have not been allocated,
the correct allocated fibs is not updated nor freed.
Fixes: 113156bcea (scsi: aacraid: Reworked aac_command_thread)
Reported-by: Tomas Henzl <thenzl@redhat.com>
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Pull more SCSI updates from James Bottomley:
"This is the set of stuff that didn't quite make the initial pull and a
set of fixes for stuff which did.
The new stuff is basically lpfc (nvme), qedi and aacraid. The fixes
cover a lot of previously submitted stuff, the most important of which
probably covers some of the failing irq vectors allocation and other
fallout from having the SCSI command allocated as part of the block
allocation functions"
* tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (59 commits)
scsi: qedi: Fix memory leak in tmf response processing.
scsi: aacraid: remove redundant zero check on ret
scsi: lpfc: use proper format string for dma_addr_t
scsi: lpfc: use div_u64 for 64-bit division
scsi: mac_scsi: Fix MAC_SCSI=m option when SCSI=m
scsi: cciss: correct check map error.
scsi: qla2xxx: fix spelling mistake: "seperator" -> "separator"
scsi: aacraid: Fixed expander hotplug for SMART family
scsi: mpt3sas: switch to pci_alloc_irq_vectors
scsi: qedf: fixup compilation warning about atomic_t usage
scsi: remove scsi_execute_req_flags
scsi: merge __scsi_execute into scsi_execute
scsi: simplify scsi_execute_req_flags
scsi: make the sense header argument to scsi_test_unit_ready mandatory
scsi: sd: improve TUR handling in sd_check_events
scsi: always zero sshdr in scsi_normalize_sense
scsi: scsi_dh_emc: return success in clariion_std_inquiry()
scsi: fix memory leak of sdpk on when gd fails to allocate
scsi: sd: make sd_devt_release() static
scsi: qedf: Add QLogic FastLinQ offload FCoE driver framework.
...
The check for ret being zero is redundant as a few statements earlier we
break out of the while loop if ret is non-zero. Thus we can remove the
zero check and also the dead-code non-zero case too.
Detected by CoverityScan, CID#1411632 ("Logically Dead Code")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Current driver Hotplug processing code skips over Enclosure channel,
therefore any addition/removal of expander enclosure is not processed.
Additionally device addition code relies on older device type, which
prevents the hotplug of adapter expanders.
Fixed by removing code that skips over Enclosure channels and using the
latest device type for addition or removal or enclosure expanders.
Fixes: 6223a39fe6 (scsi: aacraid: Added support for hotplug)
Signed-off-by: Raghava Aditya Renukunta <RaghavaAditya.Renukunta@microsemi.com>
Reviewed-by: Dave Carroll <david.carroll@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
