[PATCH v2] perf: RISC-V: Check standard event availability
Alexandre Ghiti
alex at ghiti.fr
Tue Jun 18 00:48:39 PDT 2024
Hi Samuel, Atish,
On 24/05/2024 22:00, Alexandre Ghiti wrote:
> Hi Samuel,
>
> On 26/04/2024 17:38, Atish Patra wrote:
>> On 4/17/24 18:46, Samuel Holland wrote:
>>> The RISC-V SBI PMU specification defines several standard hardware and
>>> cache events. Currently, all of these events are exposed to userspace,
>>> even when not actually implemented. They appear in the `perf list`
>>> output, and commands like `perf stat` try to use them.
>>>
>>> This is more than just a cosmetic issue, because the PMU driver's .add
>>> function fails for these events, which causes pmu_groups_sched_in() to
>>> prematurely stop scheduling in other (possibly valid) hardware events.
>>>
>>> Add logic to check which events are supported by the hardware (i.e. can
>>> be mapped to some counter), so only usable events are reported to
>>> userspace. Since the kernel does not know the mapping between events
>>> and
>>> possible counters, this check must happen during boot, when no counters
>>> are in use. Make the check asynchronous to minimize impact on boot
>>> time.
>>>
>>> Signed-off-by: Samuel Holland <samuel.holland at sifive.com>
>>> ---
>>> Before this patch:
>>> $ perf list hw
>>>
>>> List of pre-defined events (to be used in -e or -M):
>>>
>>> branch-instructions OR branches [Hardware event]
>>> branch-misses [Hardware event]
>>> bus-cycles [Hardware event]
>>> cache-misses [Hardware event]
>>> cache-references [Hardware event]
>>> cpu-cycles OR cycles [Hardware event]
>>> instructions [Hardware event]
>>> ref-cycles [Hardware event]
>>> stalled-cycles-backend OR idle-cycles-backend [Hardware event]
>>> stalled-cycles-frontend OR idle-cycles-frontend [Hardware event]
>>>
>>> $ perf stat -ddd true
>>>
>>> Performance counter stats for 'true':
>>>
>>> 4.36 msec task-clock # 0.744
>>> CPUs utilized
>>> 1 context-switches # 229.325
>>> /sec
>>> 0 cpu-migrations # 0.000 /sec
>>> 38 page-faults # 8.714 K/sec
>>> 4,375,694 cycles # 1.003
>>> GHz (60.64%)
>>> 728,945 instructions # 0.17
>>> insn per cycle
>>> 79,199 branches # 18.162
>>> M/sec
>>> 17,709 branch-misses # 22.36%
>>> of all branches
>>> 181,734 L1-dcache-loads # 41.676
>>> M/sec
>>> 5,547 L1-dcache-load-misses # 3.05% of
>>> all L1-dcache accesses
>>> <not counted> LLC-loads (0.00%)
>>> <not counted> LLC-load-misses (0.00%)
>>> <not counted> L1-icache-loads (0.00%)
>>> <not counted> L1-icache-load-misses (0.00%)
>>> <not counted> dTLB-loads (0.00%)
>>> <not counted> dTLB-load-misses (0.00%)
>>> <not counted> iTLB-loads (0.00%)
>>> <not counted> iTLB-load-misses (0.00%)
>>> <not counted> L1-dcache-prefetches (0.00%)
>>> <not counted> L1-dcache-prefetch-misses (0.00%)
>>>
>>> 0.005860375 seconds time elapsed
>>>
>>> 0.000000000 seconds user
>>> 0.010383000 seconds sys
>>>
>>> After this patch:
>>> $ perf list hw
>>>
>>> List of pre-defined events (to be used in -e or -M):
>>>
>>> branch-instructions OR branches [Hardware event]
>>> branch-misses [Hardware event]
>>> cache-misses [Hardware event]
>>> cache-references [Hardware event]
>>> cpu-cycles OR cycles [Hardware event]
>>> instructions [Hardware event]
>>>
>>> $ perf stat -ddd true
>>>
>>> Performance counter stats for 'true':
>>>
>>> 5.16 msec task-clock # 0.848
>>> CPUs utilized
>>> 1 context-switches # 193.817
>>> /sec
>>> 0 cpu-migrations # 0.000 /sec
>>> 37 page-faults # 7.171 K/sec
>>> 5,183,625 cycles # 1.005 GHz
>>> 961,696 instructions # 0.19
>>> insn per cycle
>>> 85,853 branches # 16.640
>>> M/sec
>>> 20,462 branch-misses # 23.83%
>>> of all branches
>>> 243,545 L1-dcache-loads # 47.203
>>> M/sec
>>> 5,974 L1-dcache-load-misses # 2.45% of
>>> all L1-dcache accesses
>>> <not supported> LLC-loads
>>> <not supported> LLC-load-misses
>>> <not supported> L1-icache-loads
>>> <not supported> L1-icache-load-misses
>>> <not supported> dTLB-loads
>>> 19,619 dTLB-load-misses
>>> <not supported> iTLB-loads
>>> 6,831 iTLB-load-misses
>>> <not supported> L1-dcache-prefetches
>>> <not supported> L1-dcache-prefetch-misses
>>>
>>> 0.006085625 seconds time elapsed
>>>
>>> 0.000000000 seconds user
>>> 0.013022000 seconds sys
>>>
>>>
>>> Changes in v2:
>>> - Move the event checking to a workqueue to make it asynchronous
>>> - Add more details to the commit message based on the v1 discussion
>>>
>>> drivers/perf/riscv_pmu_sbi.c | 45
>>> +++++++++++++++++++++++++++++++++---
>>> 1 file changed, 42 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/perf/riscv_pmu_sbi.c
>>> b/drivers/perf/riscv_pmu_sbi.c
>>> index 8cbe6e5f9c39..c326954af066 100644
>>> --- a/drivers/perf/riscv_pmu_sbi.c
>>> +++ b/drivers/perf/riscv_pmu_sbi.c
>>> @@ -20,6 +20,7 @@
>>> #include <linux/cpu_pm.h>
>>> #include <linux/sched/clock.h>
>>> #include <linux/soc/andes/irq.h>
>>> +#include <linux/workqueue.h>
>>> #include <asm/errata_list.h>
>>> #include <asm/sbi.h>
>>> @@ -109,7 +110,7 @@ struct sbi_pmu_event_data {
>>> };
>>> };
>>> -static const struct sbi_pmu_event_data pmu_hw_event_map[] = {
>>> +static struct sbi_pmu_event_data pmu_hw_event_map[] = {
>>> [PERF_COUNT_HW_CPU_CYCLES] = {.hw_gen_event = {
>>> SBI_PMU_HW_CPU_CYCLES,
>>> SBI_PMU_EVENT_TYPE_HW, 0}},
>>> @@ -143,7 +144,7 @@ static const struct sbi_pmu_event_data
>>> pmu_hw_event_map[] = {
>>> };
>>> #define C(x) PERF_COUNT_HW_CACHE_##x
>>> -static const struct sbi_pmu_event_data
>>> pmu_cache_event_map[PERF_COUNT_HW_CACHE_MAX]
>>> +static struct sbi_pmu_event_data
>>> pmu_cache_event_map[PERF_COUNT_HW_CACHE_MAX]
>>> [PERF_COUNT_HW_CACHE_OP_MAX]
>>> [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
>>> [C(L1D)] = {
>>> @@ -288,6 +289,34 @@ static const struct sbi_pmu_event_data
>>> pmu_cache_event_map[PERF_COUNT_HW_CACHE_M
>>> },
>>> };
>>> +static void pmu_sbi_check_event(struct sbi_pmu_event_data *edata)
>>> +{
>>> + struct sbiret ret;
>>> +
>>> + ret = sbi_ecall(SBI_EXT_PMU, SBI_EXT_PMU_COUNTER_CFG_MATCH,
>>> + 0, cmask, 0, edata->event_idx, 0, 0);
>>> + if (!ret.error) {
>>> + sbi_ecall(SBI_EXT_PMU, SBI_EXT_PMU_COUNTER_STOP,
>>> + ret.value, 0x1, SBI_PMU_STOP_FLAG_RESET, 0, 0, 0);
>>> + } else if (ret.error == SBI_ERR_NOT_SUPPORTED) {
>>> + /* This event cannot be monitored by any counter */
>>> + edata->event_idx = -EINVAL;
>>> + }
>>> +}
>>> +
>>> +static void pmu_sbi_check_std_events(struct work_struct *work)
>>> +{
>>> + for (int i = 0; i < ARRAY_SIZE(pmu_hw_event_map); i++)
>>> + pmu_sbi_check_event(&pmu_hw_event_map[i]);
>>> +
>>> + for (int i = 0; i < ARRAY_SIZE(pmu_cache_event_map); i++)
>>> + for (int j = 0; j < ARRAY_SIZE(pmu_cache_event_map[i]); j++)
>>> + for (int k = 0; k <
>>> ARRAY_SIZE(pmu_cache_event_map[i][j]); k++)
>>> + pmu_sbi_check_event(&pmu_cache_event_map[i][j][k]);
>>> +}
>>> +
>>> +static DECLARE_WORK(check_std_events_work, pmu_sbi_check_std_events);
>>> +
>>> static int pmu_sbi_ctr_get_width(int idx)
>>> {
>>> return pmu_ctr_list[idx].width;
>>> @@ -473,6 +502,12 @@ static int pmu_sbi_event_map(struct perf_event
>>> *event, u64 *econfig)
>>> u64 raw_config_val;
>>> int ret;
>>> + /*
>>> + * Ensure we are finished checking standard hardware events for
>>> + * validity before allowing userspace to configure any events.
>>> + */
>>> + flush_work(&check_std_events_work);
>>> +
>>> switch (type) {
>>> case PERF_TYPE_HARDWARE:
>>> if (config >= PERF_COUNT_HW_MAX)
>>> @@ -634,7 +669,8 @@ static inline void pmu_sbi_stop_all(struct
>>> riscv_pmu *pmu)
>>> * which may include counters that are not enabled yet.
>>> */
>>> sbi_ecall(SBI_EXT_PMU, SBI_EXT_PMU_COUNTER_STOP,
>>> - 0, pmu->cmask, 0, 0, 0, 0);
>>> + 0, pmu->cmask, SBI_PMU_STOP_FLAG_RESET, 0, 0, 0);
>>> +
>>
>> Why is this required for this patch?
>>
>> If the intention is a generic improvement to reset mhpmevent at boot
>> time, it should be separate patch.
>>
>>> }
>>> static inline void pmu_sbi_stop_hw_ctrs(struct riscv_pmu *pmu)
>>> @@ -1108,6 +1144,9 @@ static int pmu_sbi_device_probe(struct
>>> platform_device *pdev)
>>> register_sysctl("kernel", sbi_pmu_sysctl_table);
>>> + /* Asynchronously check which standard events are available */
>>> + schedule_work(&check_std_events_work);
>>> +
>>> return 0;
>>> out_unregister:
>>
>> Until we have the new SBI interface defined to optimize these SBI
>> calls, we can use this individual call approach to fix the current
>> issue.
>>
>> lgtm otherwise.
>>
>> Reviewed-by: Atish Patra <atishp at rivosinc.com>
>> Tested-by: Atish Patra <atishp at rivosinc.com>
>
>
> Can you add a Fixes tag for this so that we can merge in 6.10-rcX?
>
> Thanks,
>
> Alex
So I'd choose the following commit as a Fixes tag: e9991434596f
("RISC-V: Add perf platform driver based on SBI PMU extension")
In addition, this patch does not apply cleanly on top of 6.10, the
resolution I came up with is there
https://github.com/linux-riscv/linux-riscv/pull/1043/commits/0a440d8b9e90117261f02f4d019359847c64cb38
Can you guys check it's correct?
Thanks,
Alex
>
>
>>
>>
>> _______________________________________________
>> linux-riscv mailing list
>> linux-riscv at lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-riscv
>
> _______________________________________________
> linux-riscv mailing list
> linux-riscv at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv
More information about the linux-arm-kernel
mailing list