diff options
| author | Yang Jihong <yangjihong1@huawei.com> | 2021-04-30 09:26:59 +0800 |
|---|---|---|
| committer | Will Deacon <will@kernel.org> | 2021-06-01 14:17:01 +0100 |
| commit | fdbef8c4e68ad423416aa6cc93d1616d6f8ac5b3 (patch) | |
| tree | b113b95475251efafd97def7fdc24f7e7f5c943c /arch/arm/kernel/perf_event_v7.c | |
| parent | eb2b22f024c3615d576cead56f2a7d2c90355716 (diff) | |
| download | lwn-fdbef8c4e68ad423416aa6cc93d1616d6f8ac5b3.tar.gz lwn-fdbef8c4e68ad423416aa6cc93d1616d6f8ac5b3.zip | |
arm_pmu: Fix write counter incorrect in ARMv7 big-endian mode
Commit 3a95200d3f89 ("arm_pmu: Change API to support 64bit counter values")
changes the input "value" type from 32-bit to 64-bit, which introduces the
following problem: ARMv7 PMU counters is 32-bit width, in big-endian mode,
write counter uses high 32-bit, which writes an incorrect value.
Before:
Performance counter stats for 'ls':
2.22 msec task-clock # 0.675 CPUs utilized
0 context-switches # 0.000 K/sec
0 cpu-migrations # 0.000 K/sec
49 page-faults # 0.022 M/sec
2150476593 cycles # 966.663 GHz
2148588788 instructions # 1.00 insn per cycle
2147745484 branches # 965435.074 M/sec
2147508540 branch-misses # 99.99% of all branches
None of the above hw event counters are correct.
Solution:
"value" forcibly converted to 32-bit type before being written to PMU register.
After:
Performance counter stats for 'ls':
2.09 msec task-clock # 0.681 CPUs utilized
0 context-switches # 0.000 K/sec
0 cpu-migrations # 0.000 K/sec
46 page-faults # 0.022 M/sec
2807301 cycles # 1.344 GHz
1060159 instructions # 0.38 insn per cycle
250496 branches # 119.914 M/sec
23192 branch-misses # 9.26% of all branches
Fixes: 3a95200d3f89 ("arm_pmu: Change API to support 64bit counter values")
Cc: <stable@vger.kernel.org>
Signed-off-by: Yang Jihong <yangjihong1@huawei.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Link: https://lore.kernel.org/r/20210430012659.232110-1-yangjihong1@huawei.com
Signed-off-by: Will Deacon <will@kernel.org>
Diffstat (limited to 'arch/arm/kernel/perf_event_v7.c')
| -rw-r--r-- | arch/arm/kernel/perf_event_v7.c | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c index 2924d7910b10..eb2190477da1 100644 --- a/arch/arm/kernel/perf_event_v7.c +++ b/arch/arm/kernel/perf_event_v7.c @@ -773,10 +773,10 @@ static inline void armv7pmu_write_counter(struct perf_event *event, u64 value) pr_err("CPU%u writing wrong counter %d\n", smp_processor_id(), idx); } else if (idx == ARMV7_IDX_CYCLE_COUNTER) { - asm volatile("mcr p15, 0, %0, c9, c13, 0" : : "r" (value)); + asm volatile("mcr p15, 0, %0, c9, c13, 0" : : "r" ((u32)value)); } else { armv7_pmnc_select_counter(idx); - asm volatile("mcr p15, 0, %0, c9, c13, 2" : : "r" (value)); + asm volatile("mcr p15, 0, %0, c9, c13, 2" : : "r" ((u32)value)); } } |