Re: WARNING in vcpu_enter_guest

[Sean Christopherson <sean.j.christopherson@intel.com>]

On Thu, Mar 19, 2020 at 10:35:49AM -0700, Sean Christopherson wrote:
> On Thu, Mar 19, 2020 at 04:14:55PM +0100, Paolo Bonzini wrote:
> > On 19/03/20 15:49, Sean Christopherson wrote:
> > > On Thu, Mar 19, 2020 at 03:35:16AM -0700, syzbot wrote:
> > >> syzbot has found a reproducer for the following crash on:
> > >>
> > >> HEAD commit:    5076190d mm: slub: be more careful about the double cmpxch..
> > >> git tree:       upstream
> > >> console output: https://syzkaller.appspot.com/x/log.txt?x=143ca61de00000
> > >> kernel config:  https://syzkaller.appspot.com/x/.config?x=9f894bd92023de02
> > >> dashboard link: https://syzkaller.appspot.com/bug?extid=00be5da1d75f1cc95f6b
> > >> compiler:       gcc (GCC) 9.0.0 20181231 (experimental)
> > >> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=10bb4023e00000
> > >>
> > >> IMPORTANT: if you fix the bug, please add the following tag to the commit:
> > >> Reported-by: syzbot+00be5da1d75f1cc95f6b@syzkaller.appspotmail.com
> > > Reproduced with a little tweaking of the reproducer, debug in progress.
> > > 
> > 
> > I think the WARN_ON at x86.c:2447 is just bogus.  You can always get it
> > to trigger if garbage is passed to KVM_SET_CLOCK.
> 
> Yep.  I worked through logic/math, mostly to gain a wee bit of knowledge
> about the clock stuff, and it's sound.  The KVM_SET_CLOCK from syzkaller
> is simply making time go backwards.

Actually, would it make sense to return -EINVAL for KVM_SET_CLOCK if the
user tries to make kvmclock_offset go backwards?