KVM: selftests: touch all pages of args on each memstress iteration

Access the same memory addresses on each iteration of the memstress
guest code.  This ensures that the state of KVM's page tables
is the same after every iteration, including the pages that host the
guest page tables for args and vcpu_args.

This difference is visible when running the proposed
dirty_log_page_splitting_test[*] on AMD, or on Intel with pml=0 and
eptad=0.  The tests fail due to different semantics of dirty bits for
page-table pages on AMD (and eptad=0) and Intel.  Both AMD and Intel with
eptad=0 treat page-table accesses as writes, therefore more pages are
dropped before the repopulation phase when dirty logging is disabled.

The "missing" page had been included in the population phase because it
hosts the page tables for vcpu_args, but repopulation does not need it."

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Vipin Sharma <vipinsh@google.com>
Link: https://lore.kernel.org/r/20230412200913.1570873-1-pbonzini@redhat.com
[sean: add additional details in changelog]
Signed-off-by: Sean Christopherson <seanjc@google.com>

diff --git a/tools/testing/selftests/kvm/lib/memstress.c b/tools/testing/selftests/kvm/lib/memstress.c
index 5f1d3173c238cb7d570b3c391945276a316b8336..7d2f812e7c9ae03ff26d0714af84fa600c87110d 100644 (file)
--- a/tools/testing/selftests/kvm/lib/memstress.c
+++ b/tools/testing/selftests/kvm/lib/memstress.c
@@ -64,6 +64,9 @@ void memstress_guest_code(uint32_t vcpu_idx)
        GUEST_ASSERT(vcpu_args->vcpu_idx == vcpu_idx);
 
        while (true) {
+               for (i = 0; i < sizeof(memstress_args); i += args->guest_page_size)
+                       (void) *((volatile char *)args + i);
+
                for (i = 0; i < pages; i++) {
                        if (args->random_access)
                                page = guest_random_u32(&rand_state) % pages;