Write I/O queue hangup at random on recent Linus' kernels

[Jens Axboe]

On 11/23/21 2:05 PM, Kenneth R. Crudup wrote:
> 
> (Please forgive the SPAMmy nature of the To: list; I'm not exactly sure whose
> subsystem this issue belongs to, so please trim as appropriate).
> 
> I've got a Kioxia NVMe SSD on my Dell XPS-7390 2-in-1 running an i7-1065G7 CPU
> with 32GB RAM.  If you need more info (and I suspect so), please let me know.
> 
> I'm sorry I don't have a better description of the problem, but I run Linus'
> master branch (and sometimes I weed out problems like this). I'm current as of
> his commit 1360572566 (the 5.16-rc2 tag).
> 
> For about two weeks now every now and then my block/NVMe/...? subsystem comes to
> a total halt on writes, and I get a system that can no longer issue writes
> (reads/pageins still seem to work) until I reboot. SysRq-S/U/B still leaves a
> dirty ext4 filesystem requring recovery on reboot.
> 
> It happens at random- twice today as a matter of fact- and there doesn't seem to
> be any particular action that causes it:

It looks like some missed accounting. You can just disable wbt for now, would
be a useful data point to see if that fixes it. Just do:

echo 0 > /sys/block/nvme0n1/queue/wbt_lat_usec

and that will disable writeback throttling on that device.

I'll take a look at this, but most likely not until start next week...

-- 
Jens Axboe