Unbinding virtio_pci_modern does not release BAR4 in Linux 6.5.0-rc4
Hi,
After running "driverctl --nosave set-override 0000:01:00.0 vfio-pci" on a virtio-blk-pci device, /proc/iomem shows that BAR4 is still owned by virtio_pci_modern even though the vfio-pci driver is now bound to the PCI device.
This regression was introduced after 6.4.7 but I don't see the culprit in the git logs.
Unfortunately I don't have time to investigate further right now but I've included instructions on how to reproduce this below.
Can anyone else reproduce this and can we still fix it for the upcoming Linux 6.5?
Thanks,
Stefan
---
$ qemu-system-x86_64 \
-M q35,accel=kvm,kernel-irqchip=split \
-cpu host \
-m 1G \
-device intel-iommu,intremap=on,device-iotlb=on \
--blockdev file,filename=test.img,cache.direct=on,node-name=drive0 \
--device virtio-blk-pci,drive=drive0 \
-blockdev file,filename=test2.img,cache.direct=on,node-name=drive2 \
--device ioh3420,id=pcie.1,chassis=1 \
--device virtio-blk-pci,disable-legacy=on,disable-modern=off,drive=drive2,iommu_platform=on,ats=on,bus=pcie.1(guest)# driverctl --nosave set-override 0000:01:00.0 vfio-pci (guest)# cat /proc/iomem
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Quoted 7 lines · Stefan Hajnoczi
Hi,
After running "driverctl --nosave set-override 0000:01:00.0 vfio-pci" on a virtio-blk-pci device,/proc/iomemshows that BAR4 is still owned byvirtio_pci_moderneven though the vfio-pci driver is now bound to the PCI device.
Quoted 8 lines
This regression was introduced after 6.4.7 but I don't see the culprit in the git logs.
Unfortunately I don't have time to investigate further right now but I've included instructions on how to reproduce this below.
Can anyone else reproduce this and can we still fix it for the upcoming Linux 6.5?
This seems to be fixed by:
https://lore.kernel.org/lkml/[email protected]/T/
Thanks
Quoted 15 lines
Thanks,
Stefan
---$ qemu-system-x86_64 \ -M q35,accel=kvm,kernel-irqchip=split \ -cpu host \ -m 1G \ -device intel-iommu,intremap=on,device-iotlb=on \ --blockdev file,filename=test.img,cache.direct=on,node-name=drive0 \ --device virtio-blk-pci,drive=drive0 \ -blockdev file,filename=test2.img,cache.direct=on,node-name=drive2 \ --device ioh3420,id=pcie.1,chassis=1 \ --device virtio-blk-pci,disable-legacy=on,disable-modern=off,drive=drive2,iommu_platform=on,ats=on,bus=pcie.1
Quoted 2 lines
(guest)# driverctl --nosave set-override 0000:01:00.0 vfio-pci (guest)# cat
/proc/iomem
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Quoted 14 lines · Jason Wang
Quoted 13 lines · Stefan Hajnoczi
Hi,
After running "driverctl --nosave set-override 0000:01:00.0 vfio-pci" on a virtio-blk-pci device,/proc/iomemshows that BAR4 is still owned byvirtio_pci_moderneven though the vfio-pci driver is now bound to the PCI device.This regression was introduced after 6.4.7 but I don't see the culprit in the git logs.
Unfortunately I don't have time to investigate further right now but I've included instructions on how to reproduce this below.
Quoted 6 lines
Quoted 2 lines
Can anyone else reproduce this and can we still fix it for the upcoming Linux 6.5?
This seems to be fixed by:
Awesome, thanks for letting me know!
Stefan
Quoted 21 lines
Thanks
Quoted 17 lines
Thanks,
Stefan
---$ qemu-system-x86_64 \ -M q35,accel=kvm,kernel-irqchip=split \ -cpu host \ -m 1G \ -device intel-iommu,intremap=on,device-iotlb=on \ --blockdev file,filename=test.img,cache.direct=on,node-name=drive0 \ --device virtio-blk-pci,drive=drive0 \ -blockdev file,filename=test2.img,cache.direct=on,node-name=drive2 \ --device ioh3420,id=pcie.1,chassis=1 \ --device virtio-blk-pci,disable-legacy=on,disable-modern=off,drive=drive2,iommu_platform=on,ats=on,bus=pcie.1(guest)# driverctl --nosave set-override 0000:01:00.0 vfio-pci (guest)# cat
/proc/iomem
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization