Re: [PATCH v2 2/2] KVM: selftests: access_tracking_perf_test: add option to skip the sanity check

[James Houghton]

On Wed, Mar 26, 2025 at 11:41 AM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
>
> On Tue, Mar 25, 2025, James Houghton wrote:
> > On Mon, Mar 24, 2025 at 6:57 PM Maxim Levitsky <mlevitsk@xxxxxxxxxx> wrote:
> > >
> > > Add an option to skip sanity check of number of still idle pages,
> > > and set it by default to skip, in case hypervisor or NUMA balancing
> > > is detected.
> > >
> > > Signed-off-by: Maxim Levitsky <mlevitsk@xxxxxxxxxx>
> >
> > Thanks Maxim! I'm still working on a respin of this test with MGLRU
> > integration, like [1]. Sorry it's taking me so long. I'll apply my
> > changes on top of yours.
> >
> > [1]: https://lore.kernel.org/kvm/20241105184333.2305744-12-jthoughton@xxxxxxxxxx/
> >
> > > ---
> > >  .../selftests/kvm/access_tracking_perf_test.c | 33 ++++++++++++++++---
> > >  .../testing/selftests/kvm/include/test_util.h |  1 +
> > >  tools/testing/selftests/kvm/lib/test_util.c   |  7 ++++
> > >  3 files changed, 37 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/tools/testing/selftests/kvm/access_tracking_perf_test.c b/tools/testing/selftests/kvm/access_tracking_perf_test.c
> > > index 3c7defd34f56..6d50c829f00c 100644
> > > --- a/tools/testing/selftests/kvm/access_tracking_perf_test.c
> > > +++ b/tools/testing/selftests/kvm/access_tracking_perf_test.c
> > > @@ -65,6 +65,8 @@ static int vcpu_last_completed_iteration[KVM_MAX_VCPUS];
> > >  /* Whether to overlap the regions of memory vCPUs access. */
> > >  static bool overlap_memory_access;
> > >
> > > +static int warn_on_too_many_idle_pages = -1;
> > > +
> > >  struct test_params {
> > >         /* The backing source for the region of memory. */
> > >         enum vm_mem_backing_src_type backing_src;
> > > @@ -184,11 +186,10 @@ static void mark_vcpu_memory_idle(struct kvm_vm *vm,
> > >          * are cached and the guest won't see the "idle" bit cleared.
> > >          */
> > >         if (still_idle >= pages / 10) {
> > > -#ifdef __x86_64__
> > > -               TEST_ASSERT(this_cpu_has(X86_FEATURE_HYPERVISOR),
> > > +               TEST_ASSERT(warn_on_too_many_idle_pages,
> >
> > I think this assertion is flipped (or how warn_on_too_many_idle_pages
> > is being set is flipped, see below).
> >
> > >                             "vCPU%d: Too many pages still idle (%lu out of %lu)",
> > >                             vcpu_idx, still_idle, pages);
> > > -#endif
> > > +
> > >                 printf("WARNING: vCPU%d: Too many pages still idle (%lu out of %lu), "
> > >                        "this will affect performance results.\n",
> > >                        vcpu_idx, still_idle, pages);
> > > @@ -342,6 +343,8 @@ static void help(char *name)
> > >         printf(" -v: specify the number of vCPUs to run.\n");
> > >         printf(" -o: Overlap guest memory accesses instead of partitioning\n"
> > >                "     them into a separate region of memory for each vCPU.\n");
> > > +       printf(" -w: Skip or force enable the check that after dirtying the guest memory, most (90%%) of \n"
> > > +              "it is reported as dirty again (0/1)");
> > >         backing_src_help("-s");
> > >         puts("");
> > >         exit(0);
> > > @@ -359,7 +362,7 @@ int main(int argc, char *argv[])
> > >
> > >         guest_modes_append_default();
> > >
> > > -       while ((opt = getopt(argc, argv, "hm:b:v:os:")) != -1) {
> > > +       while ((opt = getopt(argc, argv, "hm:b:v:os:w:")) != -1) {
> > >                 switch (opt) {
> > >                 case 'm':
> > >                         guest_modes_cmdline(optarg);
> > > @@ -376,6 +379,11 @@ int main(int argc, char *argv[])
> > >                 case 's':
> > >                         params.backing_src = parse_backing_src_type(optarg);
> > >                         break;
> > > +               case 'w':
> > > +                       warn_on_too_many_idle_pages =
> > > +                               atoi_non_negative("1 - enable warning, 0 - disable",
> > > +                                                 optarg);
> >
> > We still get a "warning" either way, right? Maybe this should be
> > called "fail_on_too_many_idle_pages" (in which case the above
> > assertion is indeed flipped). Or "warn_on_too_many_idle_pages" should
> > mean *only* warn, i.e., *don't* fail, in which case, below we need to
> > flip how we set it below.
>
>
> Agreed.  I like the "warn" terminology,  Maybe this?
>
>         printf(" -w: Control whether the test warns or fails if more than 10%\n"
>                "     of pages are still seen as idle/old after accessing guest\n"
>                "     memory.  >0 == warn only, 0 == fail, <0 == auto.  For auto\n"
>                "     mode, the test fails by default, but switches to warn only\n"
>                "     if NUMA balancing is enabled or the test detects it's running\n"
>                "     in a VM.");

LGTM.

>
> And let the user explicitly select auto:
>
>                 case 'w':
>                         warn_only = atoi_paranoid(optarg);
>                         break;
>
> Then the auto resolving works as below, and as James pointed out, the assert
> becomes
>
>                 TEST_ASSERT(!warn_only, ....);

I think the auto-resolving below needs to be flipped, and the
TEST_ASSERT should be for `warn_only`, not `!warn_only`.

If warn_only == 1, the assert should pass.

>
> >
> > > +                       break;
> > >                 case 'h':
> > >                 default:
> > >                         help(argv[0]);
> > > @@ -386,6 +394,23 @@ int main(int argc, char *argv[])
> > >         page_idle_fd = open("/sys/kernel/mm/page_idle/bitmap", O_RDWR);
> > >         __TEST_REQUIRE(page_idle_fd >= 0,
> > >                        "CONFIG_IDLE_PAGE_TRACKING is not enabled");
> > > +       if (warn_on_too_many_idle_pages == -1) {
> > > +#ifdef __x86_64__
> > > +               if (this_cpu_has(X86_FEATURE_HYPERVISOR)) {
> > > +                       printf("Skipping idle page count sanity check, because the test is run nested\n");
> > > +                       warn_on_too_many_idle_pages = 0;
> > > +               } else
> > > +#endif
> > > +               if (is_numa_balancing_enabled()) {
> > > +                       printf("Skipping idle page count sanity check, because NUMA balance is enabled\n");
> > > +                       warn_on_too_many_idle_pages = 0;
> > > +               } else {
> > > +                       warn_on_too_many_idle_pages = 1;
> > > +               }
> > > +       } else if (!warn_on_too_many_idle_pages) {
> > > +               printf("Skipping idle page count sanity check, because this was requested by the user\n");
>
> Eh, I vote to omit this.  The sanity check is still there, it's just degraded to
> a warn.  I'm not totally against it, just seems superfluous and potentially confusing.

I agree, it's not adding much.

Separately: I've finished the MGLRU version of this test. It uses
MGLRU if it is available, and marking pages as idle is much faster
when using it. If MGLRU is enabled but otherwise not usable, the test
fails, as the idle page bitmap is no longer usable for this test.

I'm happy to post a new version of Maxim's patch with the MGLRU
patches too, Maxim, if you're okay with that.