Re: [BUG][bisected 270065e] linux-next fails to boot on powerpc

[Brian King]

On 08/17/2017 10:52 AM, Bart Van Assche wrote:
> On Wed, 2017-08-16 at 18:18 -0500, Brian King wrote:
>> On 08/16/2017 12:21 PM, Bart Van Assche wrote:
>>> On Wed, 2017-08-16 at 22:30 +0530, Abdul Haleem wrote:
>>>> As of next-20170809, linux-next on powerpc boot hung with below trace
>>>> message.
>>>>
>>>> [ ... ]
>>>>
>>>> A bisection resulted in first bad commit (270065e92 - scsi: scsi-mq:
>>>> Always unprepare ...) in the merge branch 'scsi/for-next'
>>>>
>>>> System booted fine when the below commit is reverted: 
>>>>
>>>> commit 270065e92c317845d69095ec8e3d18616b5b39d5
>>>> Author: Bart Van Assche <bart.vanassche@xxxxxxx>
>>>> Date:   Thu Aug 3 14:40:14 2017 -0700
>>>>
>>>>     scsi: scsi-mq: Always unprepare before requeuing a request
>>>
>>> Hello Brian and Michael,
>>>
>>> Do you agree that this probably indicates a bug in the PowerPC block driver
>>> that is used to access the boot disk? Anyway, since a solution is not yet
>>> available, I will submit a revert for this patch.
>>
>> I've been looking at this a bit, and can recreate the issue, but haven't
>> got to root cause of the issue as of yet. If I do a sysrq-w while the system is hung
>> during boot I see this:
>>
>> [   25.561523] Workqueue: events_unbound async_run_entry_fn
>> [   25.561527] Call Trace:
>> [   25.561529] [c0000001697873f0] [c000000169701600] 0xc000000169701600 (unreliable)
>> [   25.561534] [c0000001697875c0] [c00000000001ab78] __switch_to+0x2e8/0x430
>> [   25.561539] [c000000169787620] [c00000000091ccb0] __schedule+0x310/0xa00
>> [   25.561543] [c0000001697876f0] [c00000000091d3e0] schedule+0x40/0xb0
>> [   25.561548] [c000000169787720] [c000000000921e40] schedule_timeout+0x200/0x430
>> [   25.561553] [c000000169787810] [c00000000091db10] io_schedule_timeout+0x30/0x70
>> [   25.561558] [c000000169787840] [c00000000091e978] wait_for_common_io.constprop.3+0x178/0x280
>> [   25.561563] [c0000001697878c0] [c00000000047f7ec] blk_execute_rq+0x7c/0xd0
>> [   25.561567] [c000000169787910] [c000000000614cd0] scsi_execute+0x100/0x230
>> [   25.561572] [c000000169787990] [c00000000060d29c] scsi_report_opcode+0xbc/0x170
>> [   25.561577] [c000000169787a50] [d000000004fe6404] sd_revalidate_disk+0xe04/0x1620 [sd_mod]
>> [   25.561583] [c000000169787b80] [d000000004fe6d84] sd_probe_async+0xb4/0x230 [sd_mod]
>> [   25.561588] [c000000169787c00] [c00000000010fc44] async_run_entry_fn+0x74/0x210
>> [   25.561593] [c000000169787c90] [c000000000102f48] process_one_work+0x198/0x480
>> [   25.561598] [c000000169787d30] [c0000000001032b8] worker_thread+0x88/0x510
>> [   25.561603] [c000000169787dc0] [c00000000010b030] kthread+0x160/0x1a0
>> [   25.561608] [c000000169787e30] [c00000000000b3a4] ret_from_kernel_thread+0x5c/0xb8
>>
>> I was noticing that we are commonly in scsi_report_opcode. Since ipr RAID arrays don't support
>> the MAINTENANCE_IN / MI_REPORT_SUPPORTED_OPERATION_CODES, I tried setting sdev->no_report_opcodes = 1
>> in ipr's slave configure. This seems to eliminate the boot hang for me, but is only working around
>> the issue. Since this command is not supported by ipr, it should return with an illegal request.
>> When I'm hung at this point, there is nothing outstanding to the adapter / driver. I'll continue
>> debugging...
> 
> (+linux-scsi)
> 
> Hello Brian,
> 
> Is kernel debugging enabled on your test system? Is lockdep enabled?
> Anyway, stack traces like the above usually mean that a request got stuck in
> a block or scsi driver (ipr in this case). Information about pending requests,
> including the SCSI CDB, is available under /sys/kernel/debug/block (see also
> commit 0eebd005dd07 ("scsi: Implement blk_mq_ops.show_rq()")).

I think I have an understanding what is going on and why Bart's patch is causing problems for ipr.
I can work around the boot hang in ipr, but ultimately I think we need to figure out a fix
in scsi / block. I added some tracing and confirmed its not a matter of commands getting stuck
in ipr. The issue is we are retrying failed commands until we finally run out of time. This is
what I see:

1. sd_revalidate_disk calls scsi_report_opcode
2. ipr RAID arrays don't support MAINTENANCE_IN / MI_REPORT_SUPPORTED_OPERATION_CODES
3. ipr returns the command with DID_ERROR
4. scsi_decide_disposition goes to maybe_retry, increments scmd->retries, and returns NEEDS_RETRY
5. scsi_softirq_done calls scsi_queue_insert to requeue the command, which calls scsi_mq_requeue_cmd
6. With Bart's change, we then clear RQF_DONTPREP in this path, while prior we did not
7. This results in the command getting scmd->retries zeroed out when it gets re-queued,
   since we go through prep again and we lose our retry counter, resulting in lots and lots of retries.
8. Since the default command timeout for an ipr RAID array is 120 seconds, these retries go on for
   quite a long time...
9. Finally, the command has been retried so long we trip over the overall retry timer
   in scsi_softirq_done and we timeout the command.

I'll follow up with a patch to ipr to workaround the hang, but I think we need to somehow preserve
the retry counter in the scsi command, as this will likely cause issues with other drivers. 

-Brian

-- 
Brian King
Power Linux I/O
IBM Linux Technology Center