x86 VM Boot hang with latest linux-next

x86 VM Boot hang with latest linux-next

[Alexander Duyck]

So I have been seeing an issue with an intermittent boot hang on my
x86 KVM VM with the latest linux-next and have bisected it down to the
following commit:
1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7 is the first bad commit
commit 1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7
Author: Mike Snitzer <snitzer@redhat.com>
Date:   Fri Feb 22 11:23:01 2019 -0500

    dm: must allocate dm_noclone for stacked noclone devices

    Otherwise various lvm2 testsuite tests fail because the lower layers of
    the stacked noclone device aren't updated to allocate a new 'struct
    dm_clone' that reflects the upper layer bio that was issued to it.

    Fixes: 97a89458020b38 ("dm: improve noclone bio support")
    Reported-by: Mikulas Patocka <mpatocka@redhat.com>
    Signed-off-by: Mike Snitzer <snitzer@redhat.com>

What I am seeing is in about 3 out of 4 boots the startup just hangs
at the filesystem check stage with the following message:
[  OK  ] Reached target Local File Systems (Pre).
         Starting File System Check on /dev/…127-ad57-426f-bb45-363950544c0c...
[    **] (1 of 2) A start job is running for…n on device 252:2 (19s / no limit)

I did some googling and it looks like a similar issue has been
reported for s390. Based on the request for data there I have the
following info:
[root@localhost ~]# dmsetup ls --tree
fedora-swap (253:1)
 └─ (252:2)
fedora-root (253:0)
 └─ (252:2)

[root@localhost ~]# dmsetup table
fedora-swap: 0 4194304 linear 252:2 2048
fedora-root: 0 31457280 linear 252:2 4196352

Re: x86 VM Boot hang with latest linux-next

[Mike Snitzer]

On Sat, Mar 02 2019 at  6:34pm -0500,
Alexander Duyck <alexander.duyck@gmail.com> wrote:

> So I have been seeing an issue with an intermittent boot hang on my
> x86 KVM VM with the latest linux-next and have bisected it down to the
> following commit:
> 1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7 is the first bad commit
> commit 1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7
> Author: Mike Snitzer <snitzer@redhat.com>
> Date:   Fri Feb 22 11:23:01 2019 -0500
> 
>     dm: must allocate dm_noclone for stacked noclone devices
> 
>     Otherwise various lvm2 testsuite tests fail because the lower layers of
>     the stacked noclone device aren't updated to allocate a new 'struct
>     dm_clone' that reflects the upper layer bio that was issued to it.
> 
>     Fixes: 97a89458020b38 ("dm: improve noclone bio support")
>     Reported-by: Mikulas Patocka <mpatocka@redhat.com>
>     Signed-off-by: Mike Snitzer <snitzer@redhat.com>
> 
> What I am seeing is in about 3 out of 4 boots the startup just hangs
> at the filesystem check stage with the following message:
> [  OK  ] Reached target Local File Systems (Pre).
>          Starting File System Check on /dev/…127-ad57-426f-bb45-363950544c0c...
> [    **] (1 of 2) A start job is running for…n on device 252:2 (19s / no limit)
> 
> I did some googling and it looks like a similar issue has been
> reported for s390. Based on the request for data there I have the
> following info:
> [root@localhost ~]# dmsetup ls --tree
> fedora-swap (253:1)
>  └─ (252:2)
> fedora-root (253:0)
>  └─ (252:2)
> 
> [root@localhost ~]# dmsetup table
> fedora-swap: 0 4194304 linear 252:2 2048
> fedora-root: 0 31457280 linear 252:2 4196352

Thanks, which version of Fedora are you running?

Your case is more straightforward in that you're clearly using bio-based
DM linear (which was updated to leverage "noclone" support); whereas the
s390 case is using request-based DM which isn't impacted by the commit
in question at all.

I'll attempt to reproduce first thing Monday.

Mike

Re: x86 VM Boot hang with latest linux-next

[Alexander Duyck]

On Sat, Mar 2, 2019 at 7:48 PM Mike Snitzer <snitzer@redhat.com> wrote:
>
> On Sat, Mar 02 2019 at  6:34pm -0500,
> Alexander Duyck <alexander.duyck@gmail.com> wrote:
>
> > So I have been seeing an issue with an intermittent boot hang on my
> > x86 KVM VM with the latest linux-next and have bisected it down to the
> > following commit:
> > 1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7 is the first bad commit
> > commit 1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7
> > Author: Mike Snitzer <snitzer@redhat.com>
> > Date:   Fri Feb 22 11:23:01 2019 -0500
> >
> >     dm: must allocate dm_noclone for stacked noclone devices
> >
> >     Otherwise various lvm2 testsuite tests fail because the lower layers of
> >     the stacked noclone device aren't updated to allocate a new 'struct
> >     dm_clone' that reflects the upper layer bio that was issued to it.
> >
> >     Fixes: 97a89458020b38 ("dm: improve noclone bio support")
> >     Reported-by: Mikulas Patocka <mpatocka@redhat.com>
> >     Signed-off-by: Mike Snitzer <snitzer@redhat.com>
> >
> > What I am seeing is in about 3 out of 4 boots the startup just hangs
> > at the filesystem check stage with the following message:
> > [  OK  ] Reached target Local File Systems (Pre).
> >          Starting File System Check on /dev/…127-ad57-426f-bb45-363950544c0c...
> > [    **] (1 of 2) A start job is running for…n on device 252:2 (19s / no limit)
> >
> > I did some googling and it looks like a similar issue has been
> > reported for s390. Based on the request for data there I have the
> > following info:
> > [root@localhost ~]# dmsetup ls --tree
> > fedora-swap (253:1)
> >  └─ (252:2)
> > fedora-root (253:0)
> >  └─ (252:2)
> >
> > [root@localhost ~]# dmsetup table
> > fedora-swap: 0 4194304 linear 252:2 2048
> > fedora-root: 0 31457280 linear 252:2 4196352
>
> Thanks, which version of Fedora are you running?

The VM is running Fedora 27 with a kernel built off of latest
linux-next as of March 1st.

> Your case is more straightforward in that you're clearly using bio-based
> DM linear (which was updated to leverage "noclone" support); whereas the
> s390 case is using request-based DM which isn't impacted by the commit
> in question at all.
>
> I'll attempt to reproduce first thing Monday.
>
> Mike

Thanks. The behavior of it has me wondering if we are looking at
something like an uninitialized data issue or something like that
since as I mentioned I don't see this occur on every boot, just on
most of them. So every now and then I can boot up the VM without any
issues, but most of the time it will boot and then get stuck waiting
on jobs that take forever.

- Alex

Re: x86 VM Boot hang with latest linux-next

[Mike Snitzer]

On Sun, Mar 03 2019 at 12:06pm -0500,
Alexander Duyck <alexander.duyck@gmail.com> wrote:

> Thanks. The behavior of it has me wondering if we are looking at
> something like an uninitialized data issue or something like that
> since as I mentioned I don't see this occur on every boot, just on
> most of them. So every now and then I can boot up the VM without any
> issues, but most of the time it will boot and then get stuck waiting
> on jobs that take forever.

The commit in question (1efa3bb79d3 "dm: must allocate dm_noclone for
stacked noclone devices") conditionally allocates the 'struct
dm_noclone' and I can only infer that the change is causing a negative
side-effect in virtualized environments (both x86_64 and s390).

I haven't been able to reproduce on x86_64 kvm with virtio-scsi on
fedora 25 though.  Will try fedora 29 shortly, and also try virtio-blk.

Could you please provide your guest's .config (even if just off-list)?
Also, what kernel are you running on the host?

Thanks,
Mike

Re: x86 VM Boot hang with latest linux-next

[Mike Snitzer]

On Sun, Mar 03 2019 at 12:06pm -0500,
Alexander Duyck <alexander.duyck@gmail.com> wrote:

> On Sat, Mar 2, 2019 at 7:48 PM Mike Snitzer <snitzer@redhat.com> wrote:
> >
> > On Sat, Mar 02 2019 at  6:34pm -0500,
> > Alexander Duyck <alexander.duyck@gmail.com> wrote:
> >
> > > So I have been seeing an issue with an intermittent boot hang on my
> > > x86 KVM VM with the latest linux-next and have bisected it down to the
> > > following commit:
> > > 1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7 is the first bad commit
> > > commit 1efa3bb79d3de8ca1b7f6770313a1fc0bebe25c7
> > > Author: Mike Snitzer <snitzer@redhat.com>
> > > Date:   Fri Feb 22 11:23:01 2019 -0500
> > >
> > >     dm: must allocate dm_noclone for stacked noclone devices
> > >
> > >     Otherwise various lvm2 testsuite tests fail because the lower layers of
> > >     the stacked noclone device aren't updated to allocate a new 'struct
> > >     dm_clone' that reflects the upper layer bio that was issued to it.
> > >
> > >     Fixes: 97a89458020b38 ("dm: improve noclone bio support")
> > >     Reported-by: Mikulas Patocka <mpatocka@redhat.com>
> > >     Signed-off-by: Mike Snitzer <snitzer@redhat.com>
> > >
> > > What I am seeing is in about 3 out of 4 boots the startup just hangs
> > > at the filesystem check stage with the following message:
> > > [  OK  ] Reached target Local File Systems (Pre).
> > >          Starting File System Check on /dev/…127-ad57-426f-bb45-363950544c0c...
> > > [    **] (1 of 2) A start job is running for…n on device 252:2 (19s / no limit)
> > >
> > > I did some googling and it looks like a similar issue has been
> > > reported for s390. Based on the request for data there I have the
> > > following info:
> > > [root@localhost ~]# dmsetup ls --tree
> > > fedora-swap (253:1)
> > >  └─ (252:2)
> > > fedora-root (253:0)
> > >  └─ (252:2)
> > >
> > > [root@localhost ~]# dmsetup table
> > > fedora-swap: 0 4194304 linear 252:2 2048
> > > fedora-root: 0 31457280 linear 252:2 4196352
> >
> > Thanks, which version of Fedora are you running?
> 
> The VM is running Fedora 27 with a kernel built off of latest
> linux-next as of March 1st.
> 
> > Your case is more straightforward in that you're clearly using bio-based
> > DM linear (which was updated to leverage "noclone" support); whereas the
> > s390 case is using request-based DM which isn't impacted by the commit
> > in question at all.
> >
> > I'll attempt to reproduce first thing Monday.
> >
> > Mike
> 
> Thanks. The behavior of it has me wondering if we are looking at
> something like an uninitialized data issue or something like that
> since as I mentioned I don't see this occur on every boot, just on
> most of them. So every now and then I can boot up the VM without any
> issues, but most of the time it will boot and then get stuck waiting
> on jobs that take forever.

I just copied you on another related thread, but for the benefit of
anyone on LKML, please see the following for a fix that works for me:

https://www.redhat.com/archives/dm-devel/2019-March/msg00027.html