On 05/04/2024 18:13, Christoph Hellwig wrote:
On Fri, Apr 05, 2024 at 05:43:24PM +0100, John Garry wrote:
This actually looks like a kernel issue - that being that the mutex API is
not annotated for lock checking.
Oh. Yeah, that would explain the weird behavior.
I would need to investigate further for any progress in adding that lock
checking to the mutex API, but it did not look promising from that
patchset. For now I suppose you can either:
a. remove current annotation.
I can send a patch for that.
A reminder on this one.
I can send a patch if you like.
b. change to a spinlock - I don't think that anything requiring scheduling
is happening when updating the limits, but would need to audit to be sure.
With SCSI we'll hold it over the new ->device_configure, which must
be able to sleep.
