Re: [RFC][PATCH] qemu-kvm: Drop vga dirty logging workarounds

[Avi Kivity <avi@redhat.com>]

On 11/15/2010 03:26 PM, Jan Kiszka wrote:
> Am 15.11.2010 14:16, Avi Kivity wrote:
> >  On 11/15/2010 12:32 PM, Jan Kiszka wrote:
> >>  These diffs to upstream should all date back to the days qemu-kvm
> >>  supported vga dirty logging with restricted/broken kvm kernel modules.
> >>  We no longer do, so there is no need for those workarounds. Even worse
> >>  they can trigger internal bug checks these days:
> >>
> >>  BUG: kvm_dirty_pages_log_change: invalid parameters 00000000000a8000-00000000000affff
> >
> >  I'd like to apply this.  What kind of testing has this seen?  autotest
> >  likely isn't a good enough test.
>
> No systematic testing.
>
> It's based on the fact that I'm not aware of any VGA issues in upstream
> when KVM is enabled,

I don't think upstream+kvm sees a lot of testing.

>   on the fact that explicit log enable/disable became
> obsolete when switching to upstream's logging backend, and the annoying
> BUG messages only issued under qemu-kvm. There just remains a slight
> uncertainty about what mark_dirty and s[123] were once doing - or may
> still be useful for.

I remember re-adding some kvm-specific dirty logging code after 
regressions were uncovered, so I'm a little worried.  I guess we can 
risk it, we can always revert it (or fix) if it turns out badly.

-- 
error compiling committee.c: too many arguments to function