On Tue, 2012-10-09 at 17:00 -0600, Alex Williamson wrote: > On Tue, 2012-10-09 at 08:21 -0600, Alex Williamson wrote: > > On Tue, 2012-10-09 at 21:12 +0900, Linus Torvalds wrote: > > > On Tue, Oct 9, 2012 at 7:02 PM, Eric Dumazet <eric.dumazet@xxxxxxxxx> wrote: > > > > > > > > It seems drivers/vfio/pci/vfio_pci.c uses VM_RESERVED > > > > > > Yeah, I just pushed out what I think is the right (trivial) fix. > > > > Thank you, looks correct to me as well. > > Well, that might still be correct, but it's actually b3b9c293 (mm, x86, > pat: rework linear pfn-mmap tracking) that breaks vfio. As soon as I > add that commit our use of mmap'd device areas stops working, both > mapping them through the iommu and through kvm. kvm hits the BUG_ON > from !kvm_is_mmio_pfn in virt/kvm/kvm_main.c:hva_to_pfn. Thanks, It looks like vfio was for some reason relying on the cow special case in remap_pfn_range() to set vma->vm_pgoff. Since we're not doing a is_cow_mapping, vm_pgoff is no longer updated in that function after b3b9c293. I'll add a vfio patch to fix this. Thanks, Alex -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>