Re: [PATCH] fs/aio: obey min_nr when doing wakeups

[Kent Overstreet <kent.overstreet@linux.dev>]

On Fri, Mar 29, 2024 at 12:09:10PM -0500, Mitchell Augustin wrote:
> I was able to reproduce this panic with the following ktest:
> 
> ~/ktest/tests/stress_ng.ktest:
> 
> #!/usr/bin/env bash
> 
> . $(dirname $(readlink -e "${BASH_SOURCE[0]}"))/test-libs.sh
> 
> test_stressng()
> {
>     apt install stress-ng
>     count=15
>     for i in $(seq $count); do
>         echo "Starting Stress #${i}/${count} for $(uname -r)"
>         mkdir /tmp/kerneltest
>         stress-ng --aggressive --verify --timeout 240 --temp-path
> //tmp/kerneltest --hdd-opts dsync --readahead-bytes 16M -k --aiol 0
>         rm -rf /tmp/kerneltest
>         echo "Completed Stress #${i}/${count} for $(uname -r)"
>     done
> }
> 
> main "$@"
> 
> by running ~/ktest/build-test-kernel run -I ~/ktest/tests/stress_ng.ktest
> 
> Note that the panic may not necessarily happen on the first run of
> that stress-ng command, so you might have to wait several iterations.
> 
> Panic:
> Running test stress_ng.ktest on gunyolk at /home/ubuntu/upstream/linux
> building kernel... done
> Kernel version: 6.9.0-rc1-ktest-00061-g8d025e2092e2

Thanks for the quick reproducer.

I got it to pop - a few times, actually - but now that I've added debug
code (cookie values to check for some sort of a stray write), it's now
decided not to pop anymore, after 24 hours of testing, with and without
my debug changes.

Hmm.

We may just have to revert this for now, unless anyone else has any
bright ideas.