On Wed, Aug 31, 2022 at 11:22:04AM +0100, Richard W.M. Jones wrote:> > After running those tests for an hour or two I got a nasty kernel > memory corruption issue in the test VM, see attached.>From the dmesg log, the direct issue is from zram.The userspace ublksrv shouldn't trigger this kind of issue, and the only related code is to discard IO pages if they are idle for enough time via madvise(MADV_DONTNEED). In the ublk driver side, we just pin and unpin pages, and it shouldn't be hard to verify if they work correctly . I will trace in that area. Except for pin & unpin pages, ublk driver doesn't play any other mm game. thanks, Ming
Richard W.M. Jones
2022-Aug-31 11:06 UTC
[Libguestfs] [PATCH libnbd] ublk: Add new nbdublk program
On Wed, Aug 31, 2022 at 06:33:25PM +0800, Ming Lei wrote:> On Wed, Aug 31, 2022 at 11:22:04AM +0100, Richard W.M. Jones wrote: > > > > After running those tests for an hour or two I got a nasty kernel > > memory corruption issue in the test VM, see attached. > > >From the dmesg log, the direct issue is from zram. > > The userspace ublksrv shouldn't trigger this kind of issue, and the only > related code is to discard IO pages if they are idle for enough time via > madvise(MADV_DONTNEED). > > In the ublk driver side, we just pin and unpin pages, and it shouldn't > be hard to verify if they work correctly . I will trace in that area. > Except for pin & unpin pages, ublk driver doesn't play any other mm > game.In case it's an unrelated bug in the very new RC2 kernel I've started another test which just uses a local (virtio-scsi) disk, to see if I observe the same thing or not. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com Fedora Windows cross-compiler. Compile Windows programs, test, and build Windows installers. Over 100 libraries supported. http://fedoraproject.org/wiki/MinGW
On Wed, Aug 31, 2022 at 12:06:37PM +0100, Richard W.M. Jones wrote:> On Wed, Aug 31, 2022 at 06:33:25PM +0800, Ming Lei wrote: > > On Wed, Aug 31, 2022 at 11:22:04AM +0100, Richard W.M. Jones wrote: > > > > > > After running those tests for an hour or two I got a nasty kernel > > > memory corruption issue in the test VM, see attached. > > > > >From the dmesg log, the direct issue is from zram. > > > > The userspace ublksrv shouldn't trigger this kind of issue, and the only > > related code is to discard IO pages if they are idle for enough time via > > madvise(MADV_DONTNEED). > > > > In the ublk driver side, we just pin and unpin pages, and it shouldn't > > be hard to verify if they work correctly . I will trace in that area. > > Except for pin & unpin pages, ublk driver doesn't play any other mm > > game. > > In case it's an unrelated bug in the very new RC2 kernel I've started > another test which just uses a local (virtio-scsi) disk, to see if I > observe the same thing or not.Maybe you should use one nbd disk, which has the closest code path with nbdublk. Thanks, Ming