Hi,
This patch series aims to improve the PMU event filter settings with a cleaner
and more organized structure and adds several test cases related to PMU event
filters.
The first patch of this series introduces a custom "__kvm_pmu_event_filter"
structure that simplifies the event filter setup and improves overall code
readability and maintainability.
The second patch adds test cases to check that unsupported input values in the
PMU event filters are rejected, covering unsupported "action" values,
unsupported "flags" values, and unsupported "nevents" values, as well as the
setting of non-existent fixed counters in the fixed bitmap.
The third patch includes tests for the PMU event filter's behavior when applied
to fixed performance counters, ensuring the correct operation in cases where no
fixed counters exist (e.g., Intel guest PMU version=1 or AMD guest).
Finally, the fourth patch adds a test to verify that setting both generic and
fixed performance event filters does not impact the consistency of the fixed
performance filter behavior.
These changes help to ensure that KVM's PMU event filter functions as expected
in all supported use cases. These patches have been tested and verified to
function properly.
Any feedback or suggestions are greatly appreciated.
Please note that following patches should be applied before this patch series:
https://lore.kernel.org/kvm/[email protected]
https://lore.kernel.org/kvm/[email protected]
This will ensure that macro definitions such as X86_INTEL_MAX_FIXED_CTR_NUM,
INTEL_PMC_IDX_FIXED, etc. can be used.
Sincerely,
Jinrong Liang
Changes log:
v3:
- Rebased to 31b4fc3bc64a(tag: kvm-x86-next-2023.06.02).
- Dropped the patch "KVM: selftests: Replace int with uint32_t for nevents". (Sean)
- Dropped the patch "KVM: selftests: Test pmu event filter with incompatible
kvm_pmu_event_filter". (Sean)
- Introduce __kvm_pmu_event_filter to replace the original method of creating
PMU event filters. (Sean)
- Use the macro definition of kvm_cpu_property to find the number of supported
fixed counters instead of calculating it via the vcpu's cpuid. (Sean)
- Remove the wrappers that are single line passthroughs. (Sean)
- Optimize function names and variable names. (Sean)
- Optimize comments to make them more rigorous. (Sean)
v2:
- Wrap the code from the documentation in a block of code. (Bagas Sanjaya)
v1:
https://lore.kernel.org/kvm/[email protected]
Jinrong Liang (4):
KVM: selftests: Introduce __kvm_pmu_event_filter to improved event
filter settings
KVM: selftests: Test unavailable event filters are rejected
KVM: selftests: Check if event filter meets expectations on fixed
counters
KVM: selftests: Test gp event filters don't affect fixed event filters
.../kvm/x86_64/pmu_event_filter_test.c | 341 +++++++++++++-----
1 file changed, 246 insertions(+), 95 deletions(-)
base-commit: 31b4fc3bc64aadd660c5bfa5178c86a7ba61e0f7
prerequisite-patch-id: 909d42f185f596d6e5c5b48b33231c89fa5236e4
prerequisite-patch-id: ba0dd0f97d8db0fb6cdf2c7f1e3a60c206fc9784
--
2.31.1
From: Jinrong Liang <[email protected]>
Add a test to ensure that setting both generic and fixed performance
event filters does not affect the consistency of the fixed performance
filter behavior in KVM. This test helps to ensure that the fixed
performance filter works as expected even when generic performance
event filters are also set.
Signed-off-by: Jinrong Liang <[email protected]>
---
.../kvm/x86_64/pmu_event_filter_test.c | 32 +++++++++++++++++++
1 file changed, 32 insertions(+)
diff --git a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
index 72f7fdb821e5..79bfdb64820b 100644
--- a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
+++ b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
@@ -853,6 +853,19 @@ static uint64_t test_with_fixed_counter_filter(struct kvm_vcpu *vcpu,
return run_vcpu_to_sync(vcpu);
}
+static uint64_t test_set_gp_and_fixed_event_filter(struct kvm_vcpu *vcpu,
+ uint32_t action,
+ uint32_t bitmap)
+{
+ struct __kvm_pmu_event_filter f = base_event_filter;
+
+ f.action = action;
+ f.fixed_counter_bitmap = bitmap;
+ do_vcpu_set_pmu_event_filter(vcpu, &f);
+
+ return run_vcpu_to_sync(vcpu);
+}
+
static void __test_fixed_counter_bitmap(struct kvm_vcpu *vcpu, uint8_t idx,
uint8_t nr_fixed_counters)
{
@@ -881,6 +894,25 @@ static void __test_fixed_counter_bitmap(struct kvm_vcpu *vcpu, uint8_t idx,
bitmap);
TEST_ASSERT(!!count == !(bitmap & BIT(idx)),
"Fixed event filter does not work as expected.");
+
+ /*
+ * Check that setting both events[] and fixed_counter_bitmap
+ * does not affect the consistency of the fixed ctrs' behaviour.
+ *
+ * Note, the fixed_counter_bitmap has higher priority than the
+ * events list.
+ */
+ count = test_set_gp_and_fixed_event_filter(vcpu,
+ KVM_PMU_EVENT_ALLOW,
+ bitmap);
+ TEST_ASSERT(!!count == !!(bitmap & BIT(idx)),
+ "Fixed event filter does not work as expected.");
+
+ count = test_set_gp_and_fixed_event_filter(vcpu,
+ KVM_PMU_EVENT_DENY,
+ bitmap);
+ TEST_ASSERT(!!count == !(bitmap & BIT(idx)),
+ "Fixed event filter does not work as expected.");
}
}
--
2.31.1
From: Jinrong Liang <[email protected]>
Add custom "__kvm_pmu_event_filter" structure to improve pmu event
filter settings. Simplifies event filter setup by organizing event
filter parameters in a cleaner, more organized way. Improves overall
code readability and maintainability.
Signed-off-by: Jinrong Liang <[email protected]>
---
.../kvm/x86_64/pmu_event_filter_test.c | 180 +++++++++---------
1 file changed, 88 insertions(+), 92 deletions(-)
diff --git a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
index 40507ed9fe8a..26f674c32cde 100644
--- a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
+++ b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
@@ -28,6 +28,10 @@
#define NUM_BRANCHES 42
+/* Matches KVM_PMU_EVENT_FILTER_MAX_EVENTS in pmu.c */
+#define MAX_FILTER_EVENTS 300
+#define MAX_TEST_EVENTS 10
+
/*
* This is how the event selector and unit mask are stored in an AMD
* core performance event-select register. Intel's format is similar,
@@ -69,21 +73,33 @@
#define INST_RETIRED EVENT(0xc0, 0)
+struct __kvm_pmu_event_filter {
+ __u32 action;
+ __u32 nevents;
+ __u32 fixed_counter_bitmap;
+ __u32 flags;
+ __u32 pad[4];
+ __u64 events[MAX_FILTER_EVENTS];
+};
+
/*
* This event list comprises Intel's eight architectural events plus
* AMD's "retired branch instructions" for Zen[123] (and possibly
* other AMD CPUs).
*/
-static const uint64_t event_list[] = {
- EVENT(0x3c, 0),
- INST_RETIRED,
- EVENT(0x3c, 1),
- EVENT(0x2e, 0x4f),
- EVENT(0x2e, 0x41),
- EVENT(0xc4, 0),
- EVENT(0xc5, 0),
- EVENT(0xa4, 1),
- AMD_ZEN_BR_RETIRED,
+static const struct __kvm_pmu_event_filter base_event_filter = {
+ .nevents = ARRAY_SIZE(base_event_filter.events),
+ .events = {
+ EVENT(0x3c, 0),
+ INST_RETIRED,
+ EVENT(0x3c, 1),
+ EVENT(0x2e, 0x4f),
+ EVENT(0x2e, 0x41),
+ EVENT(0xc4, 0),
+ EVENT(0xc5, 0),
+ EVENT(0xa4, 1),
+ AMD_ZEN_BR_RETIRED,
+ },
};
struct {
@@ -225,51 +241,16 @@ static bool sanity_check_pmu(struct kvm_vcpu *vcpu)
return !r;
}
-static struct kvm_pmu_event_filter *alloc_pmu_event_filter(uint32_t nevents)
-{
- struct kvm_pmu_event_filter *f;
- int size = sizeof(*f) + nevents * sizeof(f->events[0]);
-
- f = malloc(size);
- TEST_ASSERT(f, "Out of memory");
- memset(f, 0, size);
- f->nevents = nevents;
- return f;
-}
-
-
-static struct kvm_pmu_event_filter *
-create_pmu_event_filter(const uint64_t event_list[], int nevents,
- uint32_t action, uint32_t flags)
-{
- struct kvm_pmu_event_filter *f;
- int i;
-
- f = alloc_pmu_event_filter(nevents);
- f->action = action;
- f->flags = flags;
- for (i = 0; i < nevents; i++)
- f->events[i] = event_list[i];
-
- return f;
-}
-
-static struct kvm_pmu_event_filter *event_filter(uint32_t action)
-{
- return create_pmu_event_filter(event_list,
- ARRAY_SIZE(event_list),
- action, 0);
-}
-
/*
* Remove the first occurrence of 'event' (if any) from the filter's
* event list.
*/
-static struct kvm_pmu_event_filter *remove_event(struct kvm_pmu_event_filter *f,
+static struct kvm_pmu_event_filter *remove_event(struct __kvm_pmu_event_filter *__f,
uint64_t event)
{
bool found = false;
int i;
+ struct kvm_pmu_event_filter *f = (void *)__f;
for (i = 0; i < f->nevents; i++) {
if (found)
@@ -315,66 +296,70 @@ static void test_without_filter(struct kvm_vcpu *vcpu)
}
static void test_with_filter(struct kvm_vcpu *vcpu,
- struct kvm_pmu_event_filter *f)
+ struct __kvm_pmu_event_filter *__f)
{
+ struct kvm_pmu_event_filter *f = (void *)__f;
+
vm_ioctl(vcpu->vm, KVM_SET_PMU_EVENT_FILTER, f);
run_vcpu_and_sync_pmc_results(vcpu);
}
static void test_amd_deny_list(struct kvm_vcpu *vcpu)
{
- uint64_t event = EVENT(0x1C2, 0);
- struct kvm_pmu_event_filter *f;
+ struct __kvm_pmu_event_filter f = base_event_filter;
- f = create_pmu_event_filter(&event, 1, KVM_PMU_EVENT_DENY, 0);
- test_with_filter(vcpu, f);
- free(f);
+ f.action = KVM_PMU_EVENT_DENY;
+ f.nevents = 1;
+ f.events[0] = EVENT(0x1C2, 0);
+ test_with_filter(vcpu, &f);
ASSERT_PMC_COUNTING_INSTRUCTIONS();
}
static void test_member_deny_list(struct kvm_vcpu *vcpu)
{
- struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_DENY);
+ struct __kvm_pmu_event_filter f = base_event_filter;
- test_with_filter(vcpu, f);
- free(f);
+ f.action = KVM_PMU_EVENT_DENY;
+ test_with_filter(vcpu, &f);
ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS();
}
static void test_member_allow_list(struct kvm_vcpu *vcpu)
{
- struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_ALLOW);
+ struct __kvm_pmu_event_filter f = base_event_filter;
- test_with_filter(vcpu, f);
- free(f);
+ f.action = KVM_PMU_EVENT_ALLOW;
+ test_with_filter(vcpu, &f);
ASSERT_PMC_COUNTING_INSTRUCTIONS();
}
static void test_not_member_deny_list(struct kvm_vcpu *vcpu)
{
- struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_DENY);
+ struct __kvm_pmu_event_filter f = base_event_filter;
+
+ f.action = KVM_PMU_EVENT_DENY;
- remove_event(f, INST_RETIRED);
- remove_event(f, INTEL_BR_RETIRED);
- remove_event(f, AMD_ZEN_BR_RETIRED);
- test_with_filter(vcpu, f);
- free(f);
+ remove_event(&f, INST_RETIRED);
+ remove_event(&f, INTEL_BR_RETIRED);
+ remove_event(&f, AMD_ZEN_BR_RETIRED);
+ test_with_filter(vcpu, &f);
ASSERT_PMC_COUNTING_INSTRUCTIONS();
}
static void test_not_member_allow_list(struct kvm_vcpu *vcpu)
{
- struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_ALLOW);
+ struct __kvm_pmu_event_filter f = base_event_filter;
- remove_event(f, INST_RETIRED);
- remove_event(f, INTEL_BR_RETIRED);
- remove_event(f, AMD_ZEN_BR_RETIRED);
- test_with_filter(vcpu, f);
- free(f);
+ f.action = KVM_PMU_EVENT_ALLOW;
+
+ remove_event(&f, INST_RETIRED);
+ remove_event(&f, INTEL_BR_RETIRED);
+ remove_event(&f, AMD_ZEN_BR_RETIRED);
+ test_with_filter(vcpu, &f);
ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS();
}
@@ -569,19 +554,16 @@ static void run_masked_events_test(struct kvm_vcpu *vcpu,
const uint64_t masked_events[],
const int nmasked_events)
{
- struct kvm_pmu_event_filter *f;
+ struct __kvm_pmu_event_filter f = {
+ .nevents = nmasked_events,
+ .action = KVM_PMU_EVENT_ALLOW,
+ .flags = KVM_PMU_EVENT_FLAG_MASKED_EVENTS,
+ };
- f = create_pmu_event_filter(masked_events, nmasked_events,
- KVM_PMU_EVENT_ALLOW,
- KVM_PMU_EVENT_FLAG_MASKED_EVENTS);
- test_with_filter(vcpu, f);
- free(f);
+ memcpy(f.events, masked_events, sizeof(uint64_t) * nmasked_events);
+ test_with_filter(vcpu, &f);
}
-/* Matches KVM_PMU_EVENT_FILTER_MAX_EVENTS in pmu.c */
-#define MAX_FILTER_EVENTS 300
-#define MAX_TEST_EVENTS 10
-
#define ALLOW_LOADS BIT(0)
#define ALLOW_STORES BIT(1)
#define ALLOW_LOADS_STORES BIT(2)
@@ -753,17 +735,27 @@ static void test_masked_events(struct kvm_vcpu *vcpu)
run_masked_events_tests(vcpu, events, nevents);
}
-static int run_filter_test(struct kvm_vcpu *vcpu, const uint64_t *events,
- int nevents, uint32_t flags)
+static int do_vcpu_set_pmu_event_filter(struct kvm_vcpu *vcpu,
+ struct __kvm_pmu_event_filter *__f)
{
- struct kvm_pmu_event_filter *f;
- int r;
+ struct kvm_pmu_event_filter *f = (void *)__f;
+
+ return __vm_ioctl(vcpu->vm, KVM_SET_PMU_EVENT_FILTER, f);
+}
- f = create_pmu_event_filter(events, nevents, KVM_PMU_EVENT_ALLOW, flags);
- r = __vm_ioctl(vcpu->vm, KVM_SET_PMU_EVENT_FILTER, f);
- free(f);
+static int set_pmu_single_event_filter(struct kvm_vcpu *vcpu, uint64_t event,
+ uint32_t flags, uint32_t action)
+{
+ struct __kvm_pmu_event_filter f = {
+ .nevents = 1,
+ .flags = flags,
+ .action = action,
+ .events = {
+ event,
+ },
+ };
- return r;
+ return do_vcpu_set_pmu_event_filter(vcpu, &f);
}
static void test_filter_ioctl(struct kvm_vcpu *vcpu)
@@ -775,14 +767,18 @@ static void test_filter_ioctl(struct kvm_vcpu *vcpu)
* Unfortunately having invalid bits set in event data is expected to
* pass when flags == 0 (bits other than eventsel+umask).
*/
- r = run_filter_test(vcpu, &e, 1, 0);
+ r = set_pmu_single_event_filter(vcpu, e, 0, KVM_PMU_EVENT_ALLOW);
TEST_ASSERT(r == 0, "Valid PMU Event Filter is failing");
- r = run_filter_test(vcpu, &e, 1, KVM_PMU_EVENT_FLAG_MASKED_EVENTS);
+ r = set_pmu_single_event_filter(vcpu, e,
+ KVM_PMU_EVENT_FLAG_MASKED_EVENTS,
+ KVM_PMU_EVENT_ALLOW);
TEST_ASSERT(r != 0, "Invalid PMU Event Filter is expected to fail");
e = KVM_PMU_ENCODE_MASKED_ENTRY(0xff, 0xff, 0xff, 0xf);
- r = run_filter_test(vcpu, &e, 1, KVM_PMU_EVENT_FLAG_MASKED_EVENTS);
+ r = set_pmu_single_event_filter(vcpu, e,
+ KVM_PMU_EVENT_FLAG_MASKED_EVENTS,
+ KVM_PMU_EVENT_ALLOW);
TEST_ASSERT(r == 0, "Valid PMU Event Filter is failing");
}
--
2.31.1
On Wed, Jun 07, 2023, Jinrong Liang wrote:
> -static struct kvm_pmu_event_filter *remove_event(struct kvm_pmu_event_filter *f,
> +static struct kvm_pmu_event_filter *remove_event(struct __kvm_pmu_event_filter *__f,
> uint64_t event)
Can you tack on a patch to drop the return? None of the callers consume it, and
it incorrectly implies that the incoming filter isn't modified.
> {
> bool found = false;
> int i;
> + struct kvm_pmu_event_filter *f = (void *)__f;
Nit, reverse xmas tree is preferred:
struct kvm_pmu_event_filter *f = (void *)__f;
bool found = false;
int i;
Hoever, I don't think this one needs to cast, the cast is only necessary when
invoking a KVM ioctl(), e.g. I believe this should work:
static void remove_event(struct __kvm_pmu_event_filter *f, uint64_t event)
{
bool found = false;
int i;
for (i = 0; i < f->nevents; i++) {
if (found)
f->events[i - 1] = f->events[i];
else
found = f->events[i] == event;
}
if (found)
f->nevents--;
}
> @@ -569,19 +554,16 @@ static void run_masked_events_test(struct kvm_vcpu *vcpu,
> const uint64_t masked_events[],
> const int nmasked_events)
> {
> - struct kvm_pmu_event_filter *f;
> + struct __kvm_pmu_event_filter f = {
> + .nevents = nmasked_events,
> + .action = KVM_PMU_EVENT_ALLOW,
> + .flags = KVM_PMU_EVENT_FLAG_MASKED_EVENTS,
Tabs, not spaces please.
> +static int set_pmu_single_event_filter(struct kvm_vcpu *vcpu, uint64_t event,
> + uint32_t flags, uint32_t action)
> +{
> + struct __kvm_pmu_event_filter f = {
> + .nevents = 1,
> + .flags = flags,
> + .action = action,
> + .events = {
> + event,
Tabs.
On Wed, Jun 07, 2023, Jinrong Liang wrote:
> static void __test_fixed_counter_bitmap(struct kvm_vcpu *vcpu, uint8_t idx,
> uint8_t nr_fixed_counters)
> {
> @@ -881,6 +894,25 @@ static void __test_fixed_counter_bitmap(struct kvm_vcpu *vcpu, uint8_t idx,
> bitmap);
> TEST_ASSERT(!!count == !(bitmap & BIT(idx)),
> "Fixed event filter does not work as expected.");
> +
> + /*
> + * Check that setting both events[] and fixed_counter_bitmap
> + * does not affect the consistency of the fixed ctrs' behaviour.
Spell out "counters". And this isn't about "consistency", the fixed_counter_bitmap
should straight up win, i.e.
/*
* Check that fixed_counter_bitmap has higher priority than
* events[] when both are set.
*/
Sean Christopherson <[email protected]> 于2023年6月29日周四 05:19写道：
>
> On Wed, Jun 07, 2023, Jinrong Liang wrote:
> > -static struct kvm_pmu_event_filter *remove_event(struct kvm_pmu_event_filter *f,
> > +static struct kvm_pmu_event_filter *remove_event(struct __kvm_pmu_event_filter *__f,
> > uint64_t event)
>
> Can you tack on a patch to drop the return? None of the callers consume it, and
> it incorrectly implies that the incoming filter isn't modified.
Thank you very much for your suggestion! I'm more than happy to follow
your advice and modify the code accordingly.
>
> > {
> > bool found = false;
> > int i;
> > + struct kvm_pmu_event_filter *f = (void *)__f;
>
> Nit, reverse xmas tree is preferred:
>
> struct kvm_pmu_event_filter *f = (void *)__f;
> bool found = false;
> int i;
>
> Hoever, I don't think this one needs to cast, the cast is only necessary when
> invoking a KVM ioctl(), e.g. I believe this should work:
>
> static void remove_event(struct __kvm_pmu_event_filter *f, uint64_t event)
> {
> bool found = false;
> int i;
>
> for (i = 0; i < f->nevents; i++) {
> if (found)
> f->events[i - 1] = f->events[i];
> else
> found = f->events[i] == event;
> }
> if (found)
> f->nevents--;
> }
> > @@ -569,19 +554,16 @@ static void run_masked_events_test(struct kvm_vcpu *vcpu,
> > const uint64_t masked_events[],
> > const int nmasked_events)
> > {
> > - struct kvm_pmu_event_filter *f;
> > + struct __kvm_pmu_event_filter f = {
> > + .nevents = nmasked_events,
> > + .action = KVM_PMU_EVENT_ALLOW,
> > + .flags = KVM_PMU_EVENT_FLAG_MASKED_EVENTS,
>
> Tabs, not spaces please.
>
> > +static int set_pmu_single_event_filter(struct kvm_vcpu *vcpu, uint64_t event,
> > + uint32_t flags, uint32_t action)
> > +{
> > + struct __kvm_pmu_event_filter f = {
> > + .nevents = 1,
> > + .flags = flags,
> > + .action = action,
> > + .events = {
> > + event,
>
> Tabs.
I will include these change in the new patch version and ensure that
any related code is adjusted accordingly.
Once again, I truly appreciate your guidance!