Re: [PATCH 2/3] kvm: x86/mmu: Ensure TDP MMU roots are freed after yield

[Ben Gardon]

On Wed, Jan 6, 2021 at 9:37 AM Maciej S. Szmigiero
<maciej.szmigiero@xxxxxxxxxx> wrote:
>
> On 06.01.2021 18:28, Ben Gardon wrote:
> > On Wed, Jan 6, 2021 at 1:26 AM Maciej S. Szmigiero
> > <maciej.szmigiero@xxxxxxxxxx> wrote:
> >>
> >> Thanks for looking at it Ben.
> >>
> >> On 06.01.2021 00:38, Ben Gardon wrote:
> >> (..)
> >>>
> >>> +Sean Christopherson, for whom I used a stale email address.
> >>> .
> >>> I tested this series by running kvm-unit-tests on an Intel Skylake
> >>> machine. It did not introduce any new failures. I also ran the
> >>> set_memory_region_test
> >>
> >> It's "memslot_move_test" that is crashing the kernel - a memslot
> >> move test based on "set_memory_region_test".
> >
> > I apologize if I'm being very dense, but I can't find this test
> > anywhere.
>
> No problem, the reproducer is available here:
> https://gist.github.com/maciejsszmigiero/890218151c242d99f63ea0825334c6c0
> as I stated in my original report.

Ah, that makes sense now. I didn't realize there were more files below
the .config. I added your test and can now reproduce the issue!

>
> > Is this something you have in-house but haven't upstreamed
> > or just the test_move_memory_region(); testcase from
> > set_memory_region_test? I have a similar memslot-moving-stress-test in
> > the pipeline I need to send out, but I didn't think such a test
> > existed yet and my test hadn't caught this issue.
>
> The reproducer at that GitHub link is taken from my KVM memslot
> test mini-set, itself based on set_memory_region_test.c from
> KVM selftests.
> The full mini-set will be posted as soon as I finish it :)

Awesome, thank you for writing this test!

>
> Thanks,
> Maciej