On 09/01/2020 16:54, Auger Eric wrote:
Hi Eric,
> On 1/3/20 7:12 PM, Andre Przywara wrote:
>> On Mon, 16 Dec 2019 21:47:51 +0100
>> Eric Auger <eric.auger@xxxxxxxxxx> wrote:
>>
>> Hi Eric,
>>
>>> If event counters are implemented check the common events
>>> required by the PMUv3 are implemented.
>>>
>>> Some are unconditionally required (SW_INCR, CPU_CYCLES,
>>> either INST_RETIRED or INST_SPEC). Some others only are
>>> required if the implementation implements some other features.
>>>
>>> Check those wich are unconditionally required.
>>>
>>> This test currently fails on TCG as neither INST_RETIRED
>>> or INST_SPEC are supported.
>>>
>>> Signed-off-by: Eric Auger <eric.auger@xxxxxxxxxx>
>>>
>>> ---
>>>
>>> v1 ->v2:
>>> - add a comment to explain the PMCEID0/1 splits
>>> ---
>>> arm/pmu.c | 71 +++++++++++++++++++++++++++++++++++++++++++++++
>>> arm/unittests.cfg | 6 ++++
>>> 2 files changed, 77 insertions(+)
>>>
>>> diff --git a/arm/pmu.c b/arm/pmu.c
>>> index d24857e..d88ef22 100644
>>> --- a/arm/pmu.c
>>> +++ b/arm/pmu.c
>>> @@ -101,6 +101,10 @@ static inline void precise_instrs_loop(int loop, uint32_t pmcr)
>>> : [pmcr] "r" (pmcr), [z] "r" (0)
>>> : "cc");
>>> }
>>> +
>>> +/* event counter tests only implemented for aarch64 */
>>> +static void test_event_introspection(void) {}
>>> +
>>> #elif defined(__aarch64__)
>>> #define ID_AA64DFR0_PERFMON_SHIFT 8
>>> #define ID_AA64DFR0_PERFMON_MASK 0xf
>>> @@ -139,6 +143,70 @@ static inline void precise_instrs_loop(int loop, uint32_t pmcr)
>>> : [pmcr] "r" (pmcr)
>>> : "cc");
>>> }
>>> +
>>> +#define PMCEID1_EL0 sys_reg(11, 3, 9, 12, 7)
>>> +
>>> +static bool is_event_supported(uint32_t n, bool warn)
>>> +{
>>> + uint64_t pmceid0 = read_sysreg(pmceid0_el0);
>>> + uint64_t pmceid1 = read_sysreg_s(PMCEID1_EL0);
>>> + bool supported;
>>> + uint32_t reg;
>>> +
>>> + /*
>>> + * The low 32-bits of PMCEID0/1 respectly describe
>>> + * event support for events 0-31/32-63. Their High
>>> + * 32-bits describe support for extended events
>>> + * starting at 0x4000, using the same split.
>>> + */
>>> + if (n >= 0x0 && n <= 0x1F)
>>> + reg = pmceid0 & 0xFFFFFFFF;
>>> + else if (n >= 0x4000 && n <= 0x401F)
>>> + reg = pmceid0 >> 32;
>>> + else if (n >= 0x20 && n <= 0x3F)
>>> + reg = pmceid1 & 0xFFFFFFFF;
>>> + else if (n >= 0x4020 && n <= 0x403F)
>>> + reg = pmceid1 >> 32;
>>> + else
>>> + abort();
>>> +
>>> + supported = reg & (1 << n);
>>
>> Don't we need to mask off everything but the lowest 5 bits of "n"? Probably also using "1U" is better.
> I added an assert to check n is less or equal than 0x3F
But "n" will definitely be bigger than that in case of an extended
event, won't it? So you adjust "reg" accordingly, but miss to do
something similar to "n"?
>>
>>> + if (!supported && warn)
>>> + report_info("event %d is not supported", n);
>>> + return supported;
>>> +}
>>> +
>>> +static void test_event_introspection(void)
>>
>> "introspection" sounds quite sophisticated. Are you planning to extend this? If not, could we maybe rename it to "test_available_events"?
> Yes this test is a placeholder for looking at the PMU characteristics
> and we may add some other queries there.
>>
>>> +{
>>> + bool required_events;
>>> +
>>> + if (!pmu.nb_implemented_counters) {
>>> + report_skip("No event counter, skip ...");
>>> + return;
>>> + }
>>> +
>>> + /* PMUv3 requires an implementation includes some common events */
>>> + required_events = is_event_supported(0x0, true) /* SW_INCR */ &&
>>> + is_event_supported(0x11, true) /* CPU_CYCLES */ &&
>>> + (is_event_supported(0x8, true) /* INST_RETIRED */ ||
>>> + is_event_supported(0x1B, true) /* INST_PREC */);
>>> +
>>> + if (pmu.version == 0x4) {
>>> + /* ARMv8.1 PMU: STALL_FRONTEND and STALL_BACKEND are required */
>>> + required_events = required_events ||
>>> + is_event_supported(0x23, true) ||
>>
>> Shouldn't those two operators be '&&' instead?
> yes definitively
>>
>>> + is_event_supported(0x24, true);
>>> + }
>>> +
>>> + /*
>>> + * L1D_CACHE_REFILL(0x3) and L1D_CACHE(0x4) are only required if
>>> + * L1 data / unified cache. BR_MIS_PRED(0x10), BR_PRED(0x12) are only
>>> + * required if program-flow prediction is implemented.
>>> + */
>>
>> Is this a TODO?
> yes. Added TODO. I do not know how to check whether the conditions are
> satisfied? Do you have any idea?
Well, AFAICS KVM doesn't filter PMCEIDn, right? So some basic checks are
surely fine, but I wouldn't go crazy about checking every possible
aspect of it. After all you would just check the hardware, as we pass
this register on.
Cheers,
Andre.
> Thank you for the review!
>
> Eric
>>
>> Cheers,
>> Andre
>>
>>
>>> +
>>> + report(required_events, "Check required events are implemented");
>>> +}
>>> +
>>> #endif
>>>
>>> /*
>>> @@ -326,6 +394,9 @@ int main(int argc, char *argv[])
>>> "Monotonically increasing cycle count");
>>> report(check_cpi(cpi), "Cycle/instruction ratio");
>>> pmccntr64_test();
>>> + } else if (strcmp(argv[1], "event-introspection") == 0) {
>>> + report_prefix_push(argv[1]);
>>> + test_event_introspection();
>>> } else {
>>> report_abort("Unknown sub-test '%s'", argv[1]);
>>> }
>>> diff --git a/arm/unittests.cfg b/arm/unittests.cfg
>>> index 79f0d7a..4433ef3 100644
>>> --- a/arm/unittests.cfg
>>> +++ b/arm/unittests.cfg
>>> @@ -66,6 +66,12 @@ file = pmu.flat
>>> groups = pmu
>>> extra_params = -append 'cycle-counter 0'
>>>
>>> +[pmu-event-introspection]
>>> +file = pmu.flat
>>> +groups = pmu
>>> +arch = arm64
>>> +extra_params = -append 'event-introspection'
>>> +
>>> # Test PMU support (TCG) with -icount IPC=1
>>> #[pmu-tcg-icount-1]
>>> #file = pmu.flat
>>
>
