[PATCH v7 7/8] KVM: selftests: Only warn for idle-page checks on powerpc pseries LPAR

[Ritesh Harjani (IBM)]

Same nested TLB/access bit issue like x86. PowerPC nested KVM (pSeries
LPAR) can hide accessed/idle, since KVM (L1) doesn't explicitly flush
the TLB for access bit update, so L0 keeps translations cached. That
also means that the L2 accesses do not refault and L1 thinks that the
pages for L2 memory are still idle.

Use kvm_is_ppc_pseries_lpar() check to detect when running nested on
powerpc pseries LPAR and only warn on the idle-page count.

Reported-by: Anushree Mathur <anushree.mathur@xxxxxxxxxxxxx>
Tested-by: Anushree Mathur <anushree.mathur@xxxxxxxxxxxxx>
Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@xxxxxxxxx>
---
 tools/testing/selftests/kvm/Makefile.kvm               |  1 +
 .../testing/selftests/kvm/access_tracking_perf_test.c  | 10 ++++++++++
 2 files changed, 11 insertions(+)

diff --git a/tools/testing/selftests/kvm/Makefile.kvm b/tools/testing/selftests/kvm/Makefile.kvm
index 4b7f6ab0ee54..692bbae44a65 100644
--- a/tools/testing/selftests/kvm/Makefile.kvm
+++ b/tools/testing/selftests/kvm/Makefile.kvm
@@ -257,6 +257,7 @@ TEST_GEN_PROGS_loongarch += set_memory_region_test
 TEST_GEN_PROGS_loongarch += steal_time
 
 TEST_GEN_PROGS_powerpc = $(TEST_GEN_PROGS_COMMON)
+TEST_GEN_PROGS_powerpc += access_tracking_perf_test
 
 SPLIT_TESTS += arch_timer
 SPLIT_TESTS += get-reg-list
diff --git a/tools/testing/selftests/kvm/access_tracking_perf_test.c b/tools/testing/selftests/kvm/access_tracking_perf_test.c
index 4415c94b2866..eb6939065e79 100644
--- a/tools/testing/selftests/kvm/access_tracking_perf_test.c
+++ b/tools/testing/selftests/kvm/access_tracking_perf_test.c
@@ -471,6 +471,16 @@ static int access_tracking_unreliable(void)
 		return 1;
 	}
 #endif
+	/*
+	 * PowerPC: Nested KVM (pSeries LPAR) can hide accessed/idle, since KVM
+	 * doesn't explicitely flush the TLB for access bit update. L0 keeps
+	 * translations cached so the guest won't see the idle bit cleared.
+	 */
+	if (kvm_is_ppc_pseries_lpar()) {
+		puts("Skipping idle page count sanity check, because the test is run nested");
+		return 1;
+	}
+
 	/*
 	 * When NUMA balancing is enabled, guest memory will be unmapped to get
 	 * NUMA faults, dropping the Accessed bits.
-- 
2.39.5