[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