On Tue, Jul 31, 2018 at 01:51:22PM -0400, Peter Geis wrote:

On 07/31/2018 12:25 PM, Ming Lei wrote:

quoted

Hi Peter,

On Tue, Jul 31, 2018 at 08:47:45AM -0400, Peter Geis wrote:

quoted

Good Morning,

On 07/30/2018 09:38 PM, Ming Lei wrote:

quoted

Hi Peter,

Thanks for collecting the log.

On Mon, Jul 30, 2018 at 02:55:42PM -0400, Peter Geis wrote:

quoted

On 07/28/2018 09:37 AM, Ming Lei wrote:

...

quoted

[   10.887209] systemd--112     0.n.1 2411122us : blk_mq_make_request: make
rq -1
[   10.890274] kworker/-98      0...1 2411506us : blk_mq_free_request:
complete: rq -1
[   10.893313] systemd--107     0...1 2412025us : blk_mq_make_request: make
rq -1
[   10.896354] systemd--107     0.... 2412323us : mmc_mq_queue_rq: queue rq
-1, 0
[   10.899388] systemd--107     0.... 2412327us :
blk_mq_try_issue_list_directly: issue direct: rq -1, ret 0
[   10.902463] (direxec-111     1...1 2413829us : blk_mq_make_request: make
rq -1
[   10.905513] systemd--114     1...1 2415159us : blk_mq_make_request: make
rq -1

Above is the most interesting part in the log. MMC sets hw queue depth
as 1, and you are using none scheduler, that means the max number of
in-flight requests should be one, but the above log shows that there may
be 3 in-flight requests.

That's odd, I have CFQ set as the default, is something changing this during
boot?
CONFIG_CFQ_GROUP_IOSCHED=y
CONFIG_DEFAULT_IOSCHED="cfq"

No, now mmc has been converted to blk-mq, and the default mq io sched
should have been mq-deadline for mmc, but not sure why it is none in
your case.

Thanks, you just answered the question.
It would be helpful if CONFIG_MQ_IOSCHED_DEADLINE=y is set.

Would you like to continue tracking down why if that is unset or the module
is unavailable the whole thing breaks?

Sure, I already sent you one simple debug patch in last email, which might show
where the issue is. And up to now, the issue is highly related with
mmc's internal sync.

Could you apply the attached debug patch and collect the trace log again?


Thanks,
Ming