Re: [libvirt] [PATCH 0/2] Adjust order of calls during qemuDomainRemoveHostDevice
On 03/28/2016 05:17 PM, Laine Stump wrote: On 03/28/2016 07:52 AM, John Ferlan wrote: While reviewing patch: http://www.redhat.com/archives/libvir-list/2016-March/msg00194.html I noted that the order of operations during the code path was slightly different and could cause issues if the reattachment code was called first. It would seem that resetting the hostdev label and cgroup prior to reattaching the device to the host would be a better corollary to the attachment code path which detaches the devices from the host, then sets the cgroup, then sets the hostdev label. I've noticed similar issues at other places in the code (teardown not being the exact mirror image of setup) but didn't touch it because I didn't want to accidentally introduce regressions. I wanted to test these patches to verify that the sequence change doesn't uncover some odd regression, but now I find that hotplug of type='hostdev' network devices is completely broken (even before applying your patches), so I'll have to figure out the cause of that first :-/ I was finally able to try this out (extensively, because I've been testing the fixes for the other problems I found on top of your patches). Everything checks out and the logic makes sense, so ACK. -- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list
Re: [libvirt] [PATCH 0/2] Adjust order of calls during qemuDomainRemoveHostDevice
On 03/28/2016 07:52 AM, John Ferlan wrote: While reviewing patch: http://www.redhat.com/archives/libvir-list/2016-March/msg00194.html I noted that the order of operations during the code path was slightly different and could cause issues if the reattachment code was called first. It would seem that resetting the hostdev label and cgroup prior to reattaching the device to the host would be a better corollary to the attachment code path which detaches the devices from the host, then sets the cgroup, then sets the hostdev label. I've noticed similar issues at other places in the code (teardown not being the exact mirror image of setup) but didn't touch it because I didn't want to accidentally introduce regressions. I wanted to test these patches to verify that the sequence change doesn't uncover some odd regression, but now I find that hotplug of type='hostdev' network devices is completely broken (even before applying your patches), so I'll have to figure out the cause of that first :-/ -- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list
[libvirt] [PATCH 0/2] Adjust order of calls during qemuDomainRemoveHostDevice
While reviewing patch: http://www.redhat.com/archives/libvir-list/2016-March/msg00194.html I noted that the order of operations during the code path was slightly different and could cause issues if the reattachment code was called first. It would seem that resetting the hostdev label and cgroup prior to reattaching the device to the host would be a better corollary to the attachment code path which detaches the devices from the host, then sets the cgroup, then sets the hostdev label. John Ferlan (2): qemu: Restore label before reattach device to host qemu: Tear down the cgroup before reattach device to host src/qemu/qemu_hotplug.c | 24 +++- 1 file changed, 11 insertions(+), 13 deletions(-) -- 2.5.5 -- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list
