pcie, virtio, nvme: preserve and quiesce emulated PCIe devices across guest reset#3915
Open
bitranox wants to merge 1 commit into
Open
pcie, virtio, nvme: preserve and quiesce emulated PCIe devices across guest reset#3915bitranox wants to merge 1 commit into
bitranox wants to merge 1 commit into
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR hardens device reset behavior so emulated PCIe and storage devices remain present and do not corrupt guest memory across guest reboot/reset cycles.
Changes:
- Virtio: introduce
abort_queue()and switch reset/disable teardown to use it so in-flight IO can be awaited/canceled without publishing completions into torn-down queue memory. - Virtio-blk: implement
abort_queue()to await io_uring-backed IOs to completion while discarding used-ring completions on reset/disable. - PCIe & NVMe: preserve PCIe presence-detect across port reset, and clear NVMe shadow-doorbell mappings on controller reset; add regression tests for PCIe presence.
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| vm/devices/virtio/virtio/src/transport/task.rs | Switch reset/disable teardown to abort_queue() for all queues. |
| vm/devices/virtio/virtio/src/device.rs | Add abort_queue() to the virtio device traits (typed + dyn). |
| vm/devices/virtio/virtio_blk/src/lib.rs | Implement virtio-blk reset/disable abort path by draining in-flight IO and discarding used-ring completions. |
| vm/devices/storage/nvme/src/workers/coordinator.rs | Reset NVMe doorbell backing store on controller reset to prevent stale shadow mappings. |
| vm/devices/storage/nvme/src/queue.rs | Add DoorbellMemory::reset() helper to restore power-on host-backed doorbells. |
| vm/devices/pci/pcie/src/switch.rs | Reset downstream ports via port reset to preserve presence detect; add regression test behind switch. |
| vm/devices/pci/pcie/src/root.rs | Reset root ports via port reset to preserve presence detect. |
| vm/devices/pci/pcie/src/port.rs | Add PcieDownstreamPort::reset() that re-asserts presence when a link is still connected; add regression test. |
bitranox
force-pushed
the
fix-pcie-device-reset
branch
from
July 10, 2026 11:24
e4e37fd to
92c61c5
Compare
bitranox
force-pushed
the
fix-pcie-device-reset
branch
from
July 10, 2026 11:33
92c61c5 to
b3e04ed
Compare
bitranox
force-pushed
the
fix-pcie-device-reset
branch
from
July 10, 2026 11:40
b3e04ed to
7245cf7
Compare
… guest reset On a guest reboot, emulated PCIe devices either vanished from the guest or corrupted the rebooted guest's memory. Three fixes: pcie: a VM reset resets device config space, which clears the downstream-port slot-status Presence Detect bit. The connected device (the port's link) survives the reset, so the slot is still occupied, but the guest's pciehp driver reads the port after reboot, sees no presence, and removes the still-attached device: an emulated disk or NIC disappears on every reboot until a manual PCI rescan. Re-derive presence from the link on reset, for both root-complex root ports and switch downstream ports. virtio: the reset/disable teardown drained each queue with stop_queue, which posts completions to the used ring. A disk read submitted before the reset completes asynchronously and writes into guest memory the rebooted guest has already reused. Add VirtioDevice::abort_queue, used by the teardown path: it awaits the in-flight IO to completion, so it finishes while the guest is stopped, then discards the completion without posting. The default falls back to stop_queue; virtio-blk overrides it. The suspend/resume path still posts. nvme: a shadow-doorbell mapping installed by Doorbell Buffer Config survived controller reset, so after a reboot the controller wrote shadow event_idx into the stale guest address the rebooted guest had reused. Reset the doorbell store to its power-on host-backed state on controller reset. Tested by rebooting a Linux guest with virtio-blk, nvme, and virtio-net attached in a loop: before, guest memory was corrupted on the first reboot (random anon_vma / slab / rbtree oopses); after, it survives repeated reboots with the devices intact.
bitranox
force-pushed
the
fix-pcie-device-reset
branch
from
July 10, 2026 11:45
7245cf7 to
2fa5c2e
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
On a guest reboot, emulated PCIe devices either vanished from the guest or corrupted the rebooted guest's memory. Three independent fixes here:
PCIe presence detect. A VM reset resets device config space, which clears the downstream-port slot-status Presence Detect bit. The connected device (the port's link) survives the reset, so the slot is still occupied, but the guest's pciehp driver reads the port after reboot, sees no presence, and removes the still-attached device: an emulated disk or NIC disappears on every reboot until a manual PCI rescan. Re-derive presence from the link on reset, for both root-complex root ports and switch downstream ports.
virtio-blk in-flight I/O. The reset/disable teardown drained the queue and posted completions to the used ring. A disk read submitted before the reset completes asynchronously and writes into guest memory the rebooted guest has already reused. Await the in-flight I/O to completion (so it finishes while the guest is stopped) but discard the completion rather than posting it. The suspend/resume path is unchanged and still posts.
nvme shadow doorbell. A shadow-doorbell mapping installed by Doorbell Buffer Config survived controller reset, so after a reboot the controller wrote shadow event_idx into the stale guest address the rebooted guest had reused. Reset the doorbell store to its power-on host-backed state on controller reset.
Tested by rebooting a Linux guest with virtio-blk, nvme, and virtio-net attached in a loop: before, guest memory was corrupted on the first reboot (random anon_vma / slab / rbtree oopses); after, it survives repeated reboots with the devices intact.