pci_pm_reset() actually depends on the device starting in the D0 power state, therefore any time we're using a flavor of pci_reset_function() we should make sure the device is fully powered-up in case the PM reset method is used. It's not uncommon that shutdown of a VM will put the device into a D3 state such that vfio_pci_disable() is managing a device in this low power state. The reset state of a device is the D0 power state, so it's also reasonable to put the device into this state prior to reset. Signed-off-by: Alex Williamson <alex.williamson@xxxxxxxxxx> --- drivers/vfio/pci/vfio_pci.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c index a4f44ea52fa3..e3faa1eb1a8c 100644 --- a/drivers/vfio/pci/vfio_pci.c +++ b/drivers/vfio/pci/vfio_pci.c @@ -454,6 +454,15 @@ static void vfio_pci_disable(struct vfio_pci_device *vdev) vdev->needs_reset = true; + /* + * Userspace may have left the device in a low power state which + * affects our ability to trigger a PM reset, restore to D0 and + * toss any saved state from the previous session. + */ + pci_set_power_state(pdev, PCI_D0); + kfree(vdev->pm_save); + vdev->pm_save = NULL; + /* * If we have saved state, restore it. If we can reset the device, * even better. Resetting with current state seems better than @@ -1013,6 +1022,9 @@ static long vfio_pci_ioctl(struct vfio_device *core_vdev, if (!vdev->reset_works) return -EINVAL; + /* PM reset depends on the device not already being in D3 */ + vfio_pci_set_power_state(vdev, PCI_D0); + vfio_pci_zap_and_down_write_memory_lock(vdev); ret = pci_try_reset_function(vdev->pdev); up_write(&vdev->memory_lock);