On Thu, Aug 15, 2024 at 04:21:26PM +0100, Jonathan Cameron wrote: > > The global rescan_remove lock has deadlocks during concurrent removals > > because it is used within interrupt handlers. Use a bus specific lock > > instead. > > > > Signed-off-by: Keith Busch <kbusch@xxxxxxxxxx> > > Looks reasonable to me. I'm not particularly confident on this one > so more eyes (plus testing) would be good. Thanks for the reviews! For the hardware I have, it is limited to x86 and I tested concurrent pcie native hotplugs and errors with PCIe switches. I could get that to reliably deadlock or crash before, and those are fixed with this patch set. But indeed, there are many more paths using this common pci code than what I could test. I am confident about the earlier patches from this series, but I also feel this last patch ought to be tried on a more diversified set of platforms.
