Hi Alexandru,
On 4/21/23 13:13, Alexandru Elisei wrote:
> Hi,
>
> On Wed, Mar 15, 2023 at 12:07:24PM +0100, Eric Auger wrote:
>> Add a new basic test that runs MEM_ACCESS loop over
>> 100 iterations and make sure the number of measured
>> MEM_ACCESS never overflows the margin. Some other
>> pmu tests rely on this pattern and if the MEM_ACCESS
>> measurement is not reliable, it is better to report
>> it beforehand and not confuse the user any further.
>>
>> Without the subsequent patch, this typically fails on
>> ThunderXv2 with the following logs:
>>
>> INFO: pmu: pmu-memaccess-reliability: 32-bit overflows:
>> overflow=1 min=21 max=41 COUNT=20 MARGIN=15
>> FAIL: pmu: pmu-memaccess-reliability: 32-bit overflows:
>> memaccess is reliable
>>
>> Signed-off-by: Eric Auger <eric.auger@xxxxxxxxxx>
>> ---
>> arm/pmu.c | 52 +++++++++++++++++++++++++++++++++++++++++++++++
>> arm/unittests.cfg | 6 ++++++
>> 2 files changed, 58 insertions(+)
>>
>> diff --git a/arm/pmu.c b/arm/pmu.c
>> index af679667..c3d2a428 100644
>> --- a/arm/pmu.c
>> +++ b/arm/pmu.c
>> @@ -56,6 +56,7 @@
>> #define EXT_COMMON_EVENTS_HIGH 0x403F
>>
>> #define ALL_SET_32 0x00000000FFFFFFFFULL
>> +#define ALL_SET_64 0xFFFFFFFFFFFFFFFFULL
>> #define ALL_CLEAR 0x0000000000000000ULL
>> #define PRE_OVERFLOW_32 0x00000000FFFFFFF0ULL
>> #define PRE_OVERFLOW_64 0xFFFFFFFFFFFFFFF0ULL
>> @@ -67,6 +68,10 @@
>> * for some observed variability we take into account a given @MARGIN
>> */
>> #define PRE_OVERFLOW2_32 (ALL_SET_32 - COUNT - MARGIN)
>> +#define PRE_OVERFLOW2_64 (ALL_SET_64 - COUNT - MARGIN)
>> +
>> +#define PRE_OVERFLOW2(__overflow_at_64bits) \
>> + (__overflow_at_64bits ? PRE_OVERFLOW2_64 : PRE_OVERFLOW2_32)
>>
>> #define PRE_OVERFLOW(__overflow_at_64bits) \
>> (__overflow_at_64bits ? PRE_OVERFLOW_64 : PRE_OVERFLOW_32)
>> @@ -746,6 +751,50 @@ static void disable_chain_counter(int even)
>> isb();
>> }
>>
>> +static void test_memaccess_reliability(bool overflow_at_64bits)
>> +{
>> + uint32_t events[] = {MEM_ACCESS};
>> + void *addr = malloc(PAGE_SIZE);
>> + uint64_t count, max = 0, min = pmevcntr_mask();
>> + uint64_t pre_overflow2 = PRE_OVERFLOW2(overflow_at_64bits);
>> + uint64_t pmcr_lp = overflow_at_64bits ? PMU_PMCR_LP : 0;
>> + bool overflow = false;
>> +
>> + if (!satisfy_prerequisites(events, ARRAY_SIZE(events)) ||
>> + !check_overflow_prerequisites(overflow_at_64bits))
>> + return;
>> +
>> + pmu_reset();
>> + write_regn_el0(pmevtyper, 0, MEM_ACCESS | PMEVTYPER_EXCLUDE_EL0);
>> + for (int i = 0; i < 100; i++) {
>> + pmu_reset();
>> + write_regn_el0(pmevcntr, 0, pre_overflow2);
>> + write_sysreg_s(0x1, PMCNTENSET_EL0);
>> + isb();
>> + mem_access_loop(addr, COUNT, pmu.pmcr_ro | PMU_PMCR_E | pmcr_lp);
>> + count = read_regn_el0(pmevcntr, 0);
>> + if (count < pre_overflow2) {
>> + count += COUNT + MARGIN;
>> + if (count > max)
>> + max = count;
>> + if (count < min)
>> + min = count;
>> + overflow = true;
>> + report_info("iter=%d count=%ld min=%ld max=%ld overflow!!!",
>> + i, count, min, max);
>> + continue;
>> + }
>> + count -= pre_overflow2;
>> + if (count > max)
>> + max = count;
>> + if (count < min)
>> + min = count;
> I'm having difficulties following the above maze of conditions. That's not going
> to be easy to maintain.
>
> If I understand the commit message correctly, the point of this test is to check
> that PRE_OVERFLOW2 + COUNT doesn't overflow, but PRE_OVERFLOW2 + 2 * COUNT does.
> How about this simpler approach instead:
>
> for (int i = 0; i < 100; i++) {
> pmu_reset();
> write_regn_el0(pmevcntr, 0, pre_overflow2);
> write_sysreg_s(0x1, PMCNTENSET_EL0);
> isb();
>
> mem_access_loop(addr, COUNT, pmu.pmcr_ro | PMU_PMCR_E | pmcr_lp);
> count = read_regn_el0(pmevcntr, 0);
> /* Counter overflowed when it shouldn't. */
> if (count < pre_overflow2) {
> report_fail("reliable memaccess loop");
> return;
> }
>
> mem_access_loop(addr, COUNT, pmu.pmcr_ro | PMU_PMCR_E | pmcr_lp);
> count = read_regn_el0(pmevcntr, 0);
> /* Counter didn't overflow when it should. */
> if (count >= pre_overflow2) {
> report_fail("reliable memaccess loop");
> return;
> }
> }
>
> report_success("reliable memaccess loop");
The test only checks that loop with PRE_OVERFLOW2 init value does not
overflow (in which case the mem access count is considered as 'reliable'
for subsequent tests using the same kind of loop). Besides doing that
check, the test also records the min/max mem access count values over
100 iterations to have an idea about how much the counting is [un]reliable.
I rearranged the logic and added comments. Hopefully this will be easier
to read.
Thank you for the review!
Eric
>
> Thanks,
> Alex
>
>> static void test_chain_promotion(bool unused)
>> {
>> uint32_t events[] = {MEM_ACCESS, CHAIN};
>> @@ -1203,6 +1252,9 @@ int main(int argc, char *argv[])
>> } else if (strcmp(argv[1], "pmu-basic-event-count") == 0) {
>> run_event_test(argv[1], test_basic_event_count, false);
>> run_event_test(argv[1], test_basic_event_count, true);
>> + } else if (strcmp(argv[1], "pmu-memaccess-reliability") == 0) {
>> + run_event_test(argv[1], test_memaccess_reliability, false);
>> + run_event_test(argv[1], test_memaccess_reliability, true);
>> } else if (strcmp(argv[1], "pmu-mem-access") == 0) {
>> run_event_test(argv[1], test_mem_access, false);
>> run_event_test(argv[1], test_mem_access, true);
>> diff --git a/arm/unittests.cfg b/arm/unittests.cfg
>> index 5e67b558..301261aa 100644
>> --- a/arm/unittests.cfg
>> +++ b/arm/unittests.cfg
>> @@ -90,6 +90,12 @@ groups = pmu
>> arch = arm64
>> extra_params = -append 'pmu-mem-access'
>>
>> +[pmu-memaccess-reliability]
>> +file = pmu.flat
>> +groups = pmu
>> +arch = arm64
>> +extra_params = -append 'pmu-memaccess-reliability'
>> +
>> [pmu-sw-incr]
>> file = pmu.flat
>> groups = pmu
>> --
>> 2.38.1
>>
