Re: [PATCH V2] powerpc/perf: Enable PMU counters post partition migration if PMU is active
flat view
From: Nathan Lynch <hidden>
Date: 2021-10-21 17:18:33
Athira Rajeev [off-list ref] writes:
During Live Partition Migration (LPM), it is observed that perf
counter values reports zero post migration completion. However
'perf stat' with workload continues to show counts post migration
since PMU gets disabled/enabled during sched switches. But incase
of system/cpu wide monitoring, zero counts were reported with 'perf
stat' after migration completion.
Example:
./perf stat -e r1001e -I 1000
time counts unit events
1.001010437 22,137,414 r1001e
2.002495447 15,455,821 r1001e
<<>> As seen in next below logs, the counter values shows zero
after migration is completed.
<<>>
86.142535370 129,392,333,440 r1001e
87.144714617 0 r1001e
88.146526636 0 r1001e
89.148085029 0 r1001e
Confirmed in my environment:
51.099987985 300,338 cache-misses
52.101839374 296,586 cache-misses
53.116089796 263,150 cache-misses
54.117949249 232,290 cache-misses
55.602029375 68,700,421,711 cache-misses
56.610073969 0 cache-misses
57.614732000 0 cache-misses
I wonder what it means that there is a very unlikely huge value before
the counter stops working -- I believe your example has this phenomenon
too.
quoted hunk ↗ jump to hunk
diff --git a/arch/powerpc/platforms/pseries/mobility.c b/arch/powerpc/platforms/pseries/mobility.c index e83e089..ff7a77c 100644 --- a/arch/powerpc/platforms/pseries/mobility.c +++ b/arch/powerpc/platforms/pseries/mobility.c@@ -476,6 +476,8 @@ static int do_join(void *arg) retry: /* Must ensure MSR.EE off for H_JOIN. */ hard_irq_disable(); + /* Disable PMU before suspend */ + mobility_pmu_disable(); hvrc = plpar_hcall_norets(H_JOIN); switch (hvrc) {@@ -530,6 +532,8 @@ static int do_join(void *arg) * reset the watchdog. */ touch_nmi_watchdog(); + /* Enable PMU after resuming */ + mobility_pmu_enable(); return ret; }
We should minimize calls into other subsystems from this context (the callback function we've passed to stop_machine); it's fairly sensitive. Can this be moved out to pseries_migrate_partition() or similar?