The current logic can allow libvirt to issue a secondary bus reset
despite the host having devices on the same bus as a device that's being
assigned to a guest. The following patches refactor the code a little
and then update the secondary bus reset logic to properly deal with
active devices. This has been lightly tested, testing is still incomplete,
hence the RFC to get any style comments out of the way once testing
verifies this is working as expected.
The following scenarios have been successfully tested:
- Do not issue a secondary bus reset:
- start guest w/ domain conf including hostdev that shares bus w/
devices owned by host
- use attach-device to dynamically add hostdev to guest, hostdev
shares bus w/ devices owned by host
- attempt to node-reset a device assigned to guest
- Issue a secondary bus reset:
- start guest w/ domain conf including all devices on bus[1]
- use attach-device to dynamically add hostdev to guest, hostdev
does not share bus with other devices
Not yet tested:
- Issue a secondary bus reset:
- when shutting down domain
- when detaching device from domain
[1] guest did not succesfully start w/ these devices, unclear if this
patch series is implicated.
Chris Wright (4):
qemuGetPciHostDeviceList take hostdev list directly
Add helpers qemuPrepareHostdevPCIDevices and qemuDomainReAttachHostdevDevices
qemudDomainAttachHostPciDevice refactor to use new helpers
pciResetDevice: use inactive devices to determine safe reset
src/qemu/qemu_driver.c | 87 ++++++++++++++++++++----------------------------
src/util/pci.c | 20 ++++++-----
src/util/pci.h | 3 +-
src/xen/xen_driver.c | 2 +-
4 files changed, 50 insertions(+), 62 deletions(-)
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list