Re: [kvm-unit-tests PATCH v2 6/9] arm: pmu: Test chained counter

2020-03-05 Thread Andrew Jones
On Thu, Jan 30, 2020 at 12:25:07PM +0100, Eric Auger wrote:
> Add 2 tests exercising chained counters. The first one uses
> CPU_CYCLES and the second one uses SW_INCR.
> 
> Signed-off-by: Eric Auger 
> ---
>  arm/pmu.c | 128 ++
>  arm/unittests.cfg |  12 +
>  2 files changed, 140 insertions(+)
> 
> diff --git a/arm/pmu.c b/arm/pmu.c
> index 1b0101f..538fbeb 100644
> --- a/arm/pmu.c
> +++ b/arm/pmu.c
> @@ -113,6 +113,8 @@ static void test_event_introspection(void) {}
>  static void test_event_counter_config(void) {}
>  static void test_basic_event_count(void) {}
>  static void test_mem_access(void) {}
> +static void test_chained_counters(void) {}
> +static void test_chained_sw_incr(void) {}
>  
>  #elif defined(__aarch64__)
>  #define ID_AA64DFR0_PERFMON_SHIFT 8
> @@ -458,6 +460,126 @@ static void test_mem_access(void)
>   read_sysreg(pmovsclr_el0));
>  }
>  
> +static void test_chained_counters(void)
> +{
> + uint32_t events[] = { 0x11 /* CPU_CYCLES */, 0x1E /* CHAIN */};
> +
> + if (!satisfy_prerequisites(events, ARRAY_SIZE(events)))
> + return;
> +
> + pmu_reset();
> +
> + write_regn(pmevtyper, 0, events[0] | PMEVTYPER_EXCLUDE_EL0);
> + write_regn(pmevtyper, 1, events[1] | PMEVTYPER_EXCLUDE_EL0);
> + /* enable counters #0 and #1 */
> + write_sysreg_s(0x3, PMCNTENSET_EL0);
> + /* preset counter #0 at 0xFFF0 */
> + write_regn(pmevcntr, 0, 0xFFF0);
> +
> + precise_instrs_loop(22, pmu.pmcr_ro | PMU_PMCR_E);
> +
> + report(read_regn(pmevcntr, 1) == 1, "CHAIN counter #1 incremented");
> + report(!read_sysreg(pmovsclr_el0), "check no overflow is recorded");
> +
> + /* test 64b overflow */
> +
> + pmu_reset();
> + write_sysreg_s(0x3, PMCNTENSET_EL0);
> +
> + write_regn(pmevcntr, 0, 0xFFF0);
> + write_regn(pmevcntr, 1, 0x1);
> + precise_instrs_loop(22, pmu.pmcr_ro | PMU_PMCR_E);
> + report_info("overflow reg = 0x%lx", read_sysreg(pmovsclr_el0));
> + report(read_regn(pmevcntr, 1) == 2, "CHAIN counter #1 incremented");
> + report(!read_sysreg(pmovsclr_el0), "check no overflow is recorded");
> +
> + write_regn(pmevcntr, 0, 0xFFF0);
> + write_regn(pmevcntr, 1, 0x);
> +
> + precise_instrs_loop(22, pmu.pmcr_ro | PMU_PMCR_E);
> + report_info("overflow reg = 0x%lx", read_sysreg(pmovsclr_el0));
> + report(!read_regn(pmevcntr, 1), "CHAIN counter #1 wrapped");
> + report(read_sysreg(pmovsclr_el0) == 0x2,
> + "check no overflow is recorded");
> +}
> +
> +static void test_chained_sw_incr(void)
> +{
> + uint32_t events[] = { 0x0 /* SW_INCR */, 0x0 /* SW_INCR */};
> + int i;
> +
> + if (!satisfy_prerequisites(events, ARRAY_SIZE(events)))
> + return;
> +
> + pmu_reset();
> +
> + write_regn(pmevtyper, 0, events[0] | PMEVTYPER_EXCLUDE_EL0);
> + write_regn(pmevtyper, 1, events[1] | PMEVTYPER_EXCLUDE_EL0);
> + /* enable counters #0 and #1 */
> + write_sysreg_s(0x3, PMCNTENSET_EL0);
> +
> + /* preset counter #0 at 0xFFF0 */
> + write_regn(pmevcntr, 0, 0xFFF0);
> +
> + for (i = 0; i < 100; i++)
> + write_sysreg(0x1, pmswinc_el0);
> +
> + report_info("SW_INCR counter #0 has value %ld", read_regn(pmevcntr, 0));
> + report(read_regn(pmevcntr, 0) == 0xFFF0,
> + "PWSYNC does not increment if PMCR.E is unset");
> +
> + pmu_reset();
> +
> + write_regn(pmevcntr, 0, 0xFFF0);
> + write_sysreg_s(0x3, PMCNTENSET_EL0);
> + set_pmcr(pmu.pmcr_ro | PMU_PMCR_E);
> +
> + for (i = 0; i < 100; i++)
> + write_sysreg(0x3, pmswinc_el0);
> +
> + report(read_regn(pmevcntr, 0)  == 84, "counter #1 after + 100 SW_INCR");
> + report(read_regn(pmevcntr, 1)  == 100,
> + "counter #0 after + 100 SW_INCR");
> + report_info("counter values after 100 SW_INCR #0=%ld #1=%ld",
> + read_regn(pmevcntr, 0), read_regn(pmevcntr, 1));
> + report(read_sysreg(pmovsclr_el0) == 0x1,
> + "overflow reg after 100 SW_INCR");
> +
> + /* 64b SW_INCR */
> + pmu_reset();
> +
> + events[1] = 0x1E /* CHAIN */;
> + write_regn(pmevtyper, 1, events[1] | PMEVTYPER_EXCLUDE_EL0);
> + write_regn(pmevcntr, 0, 0xFFF0);
> + write_sysreg_s(0x3, PMCNTENSET_EL0);
> + set_pmcr(pmu.pmcr_ro | PMU_PMCR_E);
> + for (i = 0; i < 100; i++)
> + write_sysreg(0x3, pmswinc_el0);
> +
> + report(!read_sysreg(pmovsclr_el0) && (read_regn(pmevcntr, 1) == 1),
> + "overflow reg after 100 SW_INCR/CHAIN");
> + report_info("overflow=0x%lx, #0=%ld #1=%ld", read_sysreg(pmovsclr_el0),
> + read_regn(pmevcntr, 0), read_regn(pmevcntr, 1));
> +
> + /* 64b SW_INCR and overflow on CHAIN counter*/
> + pmu_reset();
> +
> + write_regn(pmevtyper, 1, events[1] | PMEVTYPER_EXCLUDE_EL0);
> + write_regn(pmevcntr, 0, 0xFFF0);

Re: [kvm-unit-tests PATCH v2 6/9] arm: pmu: Test chained counter

2020-02-11 Thread Auger Eric
Hi Peter,

On 2/11/20 5:24 PM, Peter Maydell wrote:
> On Thu, 30 Jan 2020 at 11:26, Eric Auger  wrote:
>>
>> Add 2 tests exercising chained counters. The first one uses
>> CPU_CYCLES and the second one uses SW_INCR.
>>
>> Signed-off-by: Eric Auger 
>> +static void test_chained_sw_incr(void)
>> +{
>> +   uint32_t events[] = { 0x0 /* SW_INCR */, 0x0 /* SW_INCR */};
> 
> Cut-n-paste error? This test relies on the CHAIN event but it
> isn't present in this list of events to pass to satisfy_prerequisites(),
> so I suspect the second element should be "0x1e /* CHAIN */" ?
No that's not a cut-n-paste error. I may rename the test into test_sw_incr.

It starts by testing unchained SW_INCR.

chained SW_INCR testing start with
/* 64b SW_INCR */


> 
> (This makes the test fail on QEMU TCG, because we don't implement
> CHAIN.)
OK

Thanks

Eric
> 
> thanks
> -- PMM
> 

___
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm


Re: [kvm-unit-tests PATCH v2 6/9] arm: pmu: Test chained counter

2020-02-11 Thread Peter Maydell
On Thu, 30 Jan 2020 at 11:26, Eric Auger  wrote:
>
> Add 2 tests exercising chained counters. The first one uses
> CPU_CYCLES and the second one uses SW_INCR.
>
> Signed-off-by: Eric Auger 
> +static void test_chained_sw_incr(void)
> +{
> +   uint32_t events[] = { 0x0 /* SW_INCR */, 0x0 /* SW_INCR */};

Cut-n-paste error? This test relies on the CHAIN event but it
isn't present in this list of events to pass to satisfy_prerequisites(),
so I suspect the second element should be "0x1e /* CHAIN */" ?

(This makes the test fail on QEMU TCG, because we don't implement
CHAIN.)

thanks
-- PMM
___
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm