On Fri, Nov 12, 2021 at 1:55 AM Nikolay Borisov [off-list ref] wrote:

On 11.11.21 г. 22:57, Chris Murphy wrote:

quoted

On Thu, Nov 11, 2021 at 3:24 PM Chris Murphy [off-list ref] wrote:

quoted

Soon after logging in and launching some apps, I get a hang. Although
there's lots of btrfs stuff in the call traces, I think we're stuck in
writeback so everything else just piles up and it all hangs
indefinitely.

Happening since at least
5.16.0-0.rc0.20211109gitd2f38a3c6507.9.fc36.x86_64 and is still happening with
5.16.0-0.rc0.20211111gitdebe436e77c7.11.fc36.x86_64

Full dmesg including sysrq+w when the journal becomes unresponsive and
then a bunch of block tasks  > 120s roll in on their own.

https://bugzilla-attachments.redhat.com/attachment.cgi?id=1841283

The btrfs traces in this one doesn't look interesting, what's
interesting is you have a bunch of tasks, including btrfs transaction
commit which are stuck waiting to get a tag from the underlying block
device - blk_mq_get_tag function. This indicates something's going on
with the underlying block device.

Well the hang doesn't ever happen with 5.14.x or 5.15.x kernels, only
the misc-next (Fedora rc0) kernels. And also I just discovered that
it's not happening (or not as quickly) with IO scheduler none. I've
been using kyber and when I switch back to it, the hang happens almost
immediately.




-- 
Chris Murphy