On Wed, Jul 16, 2025 at 07:47:04AM -0600, Keith Busch wrote: > On Wed, Jul 16, 2025 at 09:32:01AM -0300, Jason Gunthorpe wrote: > > > > You should be making a matching change to iommufd too.. > > Yeah, okay. My colleauge I've been working with on this hasn't yet > reported CPU stalls using iommufd though, so I'm not sure which path > iommufd takes to know where to place cond_resched(). Blindly stumbly > through it, my best guess right now is arriving at the loop in > iopt_fill_domains_pages(), called through iommufd_vfio_map_dma(). Am I > close? Yeah, I would guess maybe put it in pfn_reader_next() But iommufd won't be faulting the MMIO memory from VFIO so you can't hit the same exact condition you are describing.. Jason
