On Thu, Sep 16, 2021 at 1:49 AM Andrew Jones <drjones@xxxxxxxxxx> wrote:
>
> On Wed, Sep 15, 2021 at 02:55:03PM -0700, Ben Gardon wrote:
> > On Wed, Sep 15, 2021 at 2:30 PM David Matlack <dmatlack@xxxxxxxxxx> wrote:
> > >
> > > The calculation to get the per-slot dirty bitmap was incorrect leading
> > > to a buffer overrun. Fix it by dividing the number of pages by
> > > BITS_PER_LONG, since each element of the bitmap is a long and there is
> > > one bit per page.
> > >
> > > Fixes: 609e6202ea5f ("KVM: selftests: Support multiple slots in dirty_log_perf_test")
> > > Signed-off-by: David Matlack <dmatlack@xxxxxxxxxx>
> >
> > I was a little confused initially because we're allocating only one
> > dirty bitmap in userspace even when we have multiple slots, but that's
> > not a problem.
>
> It's also confusing to me. Wouldn't it be better to create a bitmap per
> slot? I think the new constraint that host mem must be a multiple of 64
> is unfortunate.
I don't think think the multiple-of-64 (256K) constraint will matter
much in practice. But it wouldn't be very complicated to switch to a
bitmap per slot, and would prevent further confusion.
I'll switch it over in v2.
>
> Thanks,
> drew
>
