Message ID | 20210421124340.301916-1-seth.forshee@canonical.com |
---|---|
Headers | show |
Series | Fix disk hot removal regression in s390x under kvm | expand |
On Wed, Apr 21, 2021 at 9:44 AM Seth Forshee <seth.forshee@canonical.com> wrote: > > BugLink: https://bugs.launchpad.net/bugs/1925211 > > [Impact] > > Hot removal of disks under kvm on s390 does not result in the kernel > removing the block device, which can lead to hung tasks and other > issues. > > [Fix] > > A bisect identified commit 8cc0dcfdc1c0 "s390/cio: remove pm support > from ccw bus driver" as introducing the regression. Reverting this patch > fixed the problem in testing. > > [Test Plan] > > See steps to reproduce the bug in the original description below. To > test, execute these steps and confirm that the block device gets removed > as expected. > > [Where problems could occur] > > The fix is a revert of the changes which introduced this regression. The > original commit was a removal of supposedly unused code, but it seems a > mistake was made in the logic around unregistering of disks. Reverting > the changes could have potential to introduce bugs related to other virt > devices, especially if it interacts badly with subsequent driver > changes. However, the patch reverted cleanly, and reverting restores the > code to the state which has been working well in previous kernels and > seems like the lowest risk option until a proper fix is available > upstream. > Thanks Seth, LGTM: Acked-by: Guilherme G. Piccoli <gpiccoli@canonical.com>
On 21/04/2021 13:43, Seth Forshee wrote: > BugLink: https://bugs.launchpad.net/bugs/1925211 > > [Impact] > > Hot removal of disks under kvm on s390 does not result in the kernel > removing the block device, which can lead to hung tasks and other > issues. > > [Fix] > > A bisect identified commit 8cc0dcfdc1c0 "s390/cio: remove pm support > from ccw bus driver" as introducing the regression. Reverting this patch > fixed the problem in testing. > > [Test Plan] > > See steps to reproduce the bug in the original description below. To > test, execute these steps and confirm that the block device gets removed > as expected. > > [Where problems could occur] > > The fix is a revert of the changes which introduced this regression. The > original commit was a removal of supposedly unused code, but it seems a > mistake was made in the logic around unregistering of disks. Reverting > the changes could have potential to introduce bugs related to other virt > devices, especially if it interacts badly with subsequent driver > changes. However, the patch reverted cleanly, and reverting restores the > code to the state which has been working well in previous kernels and > seems like the lowest risk option until a proper fix is available > upstream. > > --- > > Seth Forshee (1): > UBUNTU: SAUCE: Revert "s390/cio: remove pm support from ccw bus > driver" > > arch/s390/include/asm/ccwdev.h | 10 ++ > drivers/s390/cio/cmf.c | 5 + > drivers/s390/cio/device.c | 247 ++++++++++++++++++++++++++++++++- > drivers/s390/cio/device.h | 1 + > drivers/s390/cio/device_fsm.c | 6 + > drivers/s390/cio/io_sch.h | 1 + > 6 files changed, 268 insertions(+), 2 deletions(-) > > Yep, seems good to me. Acked-by: Colin Ian King <colin.king@canonical.com>
On 21.04.21 14:43, Seth Forshee wrote: > BugLink: https://bugs.launchpad.net/bugs/1925211 > > [Impact] > > Hot removal of disks under kvm on s390 does not result in the kernel > removing the block device, which can lead to hung tasks and other > issues. > > [Fix] > > A bisect identified commit 8cc0dcfdc1c0 "s390/cio: remove pm support > from ccw bus driver" as introducing the regression. Reverting this patch > fixed the problem in testing. > > [Test Plan] > > See steps to reproduce the bug in the original description below. To > test, execute these steps and confirm that the block device gets removed > as expected. > > [Where problems could occur] > > The fix is a revert of the changes which introduced this regression. The > original commit was a removal of supposedly unused code, but it seems a > mistake was made in the logic around unregistering of disks. Reverting > the changes could have potential to introduce bugs related to other virt > devices, especially if it interacts badly with subsequent driver > changes. However, the patch reverted cleanly, and reverting restores the > code to the state which has been working well in previous kernels and > seems like the lowest risk option until a proper fix is available > upstream. > > --- > > Seth Forshee (1): > UBUNTU: SAUCE: Revert "s390/cio: remove pm support from ccw bus > driver" > > arch/s390/include/asm/ccwdev.h | 10 ++ > drivers/s390/cio/cmf.c | 5 + > drivers/s390/cio/device.c | 247 ++++++++++++++++++++++++++++++++- > drivers/s390/cio/device.h | 1 + > drivers/s390/cio/device_fsm.c | 6 + > drivers/s390/cio/io_sch.h | 1 + > 6 files changed, 268 insertions(+), 2 deletions(-) > > Applied to hirsute:linux/master-next. Thanks. -Stefan