Il giorno 7 mag 2021, alle ore 10:30, Yi Zhang [off-list ref] ha scritto:

On Mon, May 3, 2021 at 5:35 PM Paolo Valente [off-list ref] wrote:

quoted

quoted

Il giorno 20 apr 2021, alle ore 09:33, Paolo Valente [off-list ref] ha scritto:

quoted

Il giorno 20 apr 2021, alle ore 04:00, Yi Zhang [off-list ref] ha scritto:


On Wed, Apr 7, 2021 at 11:15 PM Yi Zhang [off-list ref] wrote:


On 4/2/21 9:39 PM, Paolo Valente wrote:

quoted

quoted

Il giorno 1 apr 2021, alle ore 03:27, Yi Zhang [off-list ref] ha scritto:

Hi

Hi

quoted

We reproduced this bfq regression[3] on ppc64le with blktests[2] on the latest linux-block/for-next branch, seems it was introduced with [1] from my bisecting, pls help check it. Let me know if you need any testing for it, thanks.

Thanks for reporting this bug and finding the candidate offending commit. Could you try this test with my dev kernel, which might provide more information? The kernel is here:
https://github.com/Algodev-github/bfq-mq

Alternatively, I could try to provide you with patches to instrument your kernel.

HI Paolo
I tried your dev kernel, but with no luck to reproduce it, could you
provide the debug patch based on latest linux-block/for-next?

Hi Paolo
This issue has been consistently reproduced with LTP/fstests/blktests on recent linux-block/for-next, do you have a chance to check it?

Hi Yi, all,
I've been working hard to port my code-instrumentation layer to the kernel in for-next. I seem I finished the porting yesterday. I tested it but the system crashed. I'm going to analyze the oops. Maybe this freeze is caused by mistakes in this layer, maybe the instrumentation is already detecting a bug. In the first case, I'll fix the mistakes and try the tests suggested in this thread.

Hi Yi, all,
I seem to have made it.  I've attached a patch series, which applies
on top of for-next, as it was when you reported this failure (i.e., on
top of 816e1d1c2f7d Merge branch 'for-5.13/io_uring' into for-next).
If patches are to be applied on top of a different HEAD, and they
don't apply cleanly, I'll take care of rebasing them.

Of course I've tried your test myself, but with no failure at all.

Looking forward to your feedback,
Paolo

Hi Paolo

With the patch series, blktests nvme-tcp nvme/011 passed on
linux-block/for-next, thanks.

Great!

That series contained only one fix.  I do hope that it has been that
fix to eliminate the failure.

So I'm about to propose such a fix as an individual patch for
mainline.  Please test it if you can.

Thanks,
Paolo

Thanks
Yi