On Thu, 03 Oct 2024, Keith Busch wrote:
Oh, that's pretty neat. Thanks for testing that! I think that reference counting patch is pretty safe too. I can rebase the series with that one included next time.
Well the thing is it is crashing on me, as reported. I was able to reproduce the deadlock with rescan_remove_lock on a switched CXL topology (via sysfs remove/rescan parent/child). This is why I tested your full pci-bus-locking-2024-09-09 branch hoping the last patch would fix it, but still cannot confirm because of that nil ptr deref.
