x86, perf, p4: Counter corruption when using lots of perf groups
Subsystems (2)
PATCH 1/1 x86, perf, p4: Counter corruption when using lots of perf groups
On a P4 box stressing perf with
./perf record -o perf.data ./perf stat -v ./perf bench all
it was noticed that a slew of unknown NMIs would pop out rather quickly.
Painfully debugging this ancient platform, led me to notice cross cpu counter corruption.
The P4 machine is special in that it has 18 counters, half are used for cpu0 and the other half is for cpu1 (or all 18 if hyperthreading is disabled). But the splitting of the counters has to be actively managed by the software.
In this particular bug, one of the cpu0 specific counters was being used by cpu1 and caused all sorts of random unknown nmis.
I am not entirely sure on the corruption path, but what happens is:
o perf schedules a group with p4_pmu_schedule_events()
o inside p4_pmu_schedule_events(), it notices an hwc pointer is being reused but for a different cpu, so it 'swaps' the config bits and returns the updated 'assign' array with a _new_ index.
o perf schedules another group with p4_pmu_schedule_events()
o inside p4_pmu_schedule_events(), it notices an hwc pointer is being reused (the same one as above) but for the _same_ cpu [BUG!!], so it updates the 'assign' array to use the _old_ (wrong cpu) index because the _new_ index is in an earlier part of the 'assign' array (and hasn't been committed yet).
o perf commits the transaction using the wrong index and corrupts the other cpu
The [BUG!!] is because the 'hwc->config' is updated but not the 'hwc->idx'. So the check for 'p4_should_swap_ts()' is correct the first time around but incorrect the second time around (because hwc->config was updated in between).
I think the spirit of perf was to not modify anything until all the transactions had a chance to 'test' if they would succeed, and if so, commit atomically. However, P4 breaks this spirit by touching the hwc->config element.
So my fix is to continue the un-perf like breakage, by assigning hwc->idx to -1 on swap to tell follow up group scheduling to find a new index.
Of course if the transaction fails rolling this back will be difficult, but that is not different than how the current code works. :-) And I wasn't sure how much effort to cleanup the code I should do for a platform that is almost 10 years old by now.
Hence the lazy fix.
--- arch/x86/kernel/cpu/perf_event_p4.c | 19 ++++++++++++++++++- 1 files changed, 18 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +18 -1
@@ -1257,7 +1257,24 @@ again: pass++; goto again; } - +/* +* Perf does test runs to see if a whole group can be assigned +* together succesfully. There can be multiple rounds of this. +* Unfortunately, p4_pmu_swap_config_ts touches the hwc->config +* bits, such that the next round of group assignments will +* cause the above p4_should_swap_ts to pass instead of fail. +* This leads to counters exclusive to thread0 being used by +* thread1. +* +* Solve this with a cheap hack, reset the idx back to -1 to +* force a new lookup (p4_next_cntr) to get the right counter +* for the right thread. +* +* This probably doesn't comply with the general spirit of how +* perf wants to work, but P4 is special. :-( +*/ +if (p4_should_swap_ts (hwc->config, cpu)) + hwc->idx =-1 ;p4_pmu_swap_config_ts (hwc, cpu);if (assign) assign[i] = cntr_idx;
Signature · 1 line
1.7.1
Quoted 18 lines · Don Zickus
On a P4 box stressing perf with
./perf record -o perf.data ./perf stat -v ./perf bench all
it was noticed that a slew of unknown NMIs would pop out rather quickly.
Painfully debugging this ancient platform, led me to notice cross cpu counter corruption.
The P4 machine is special in that it has 18 counters, half are used for cpu0 and the other half is for cpu1 (or all 18 if hyperthreading is disabled). But the splitting of the counters has to be actively managed by the software.
In this particular bug, one of the cpu0 specific counters was being used by cpu1 and caused all sorts of random unknown nmis.
I am not entirely sure on the corruption path, but what happens is:
Quoted 10 lines
o perf schedules a group with
p4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused but for a different cpu, so it 'swaps' the config bits and returns the updated 'assign' array with a _new_ index.
o perf schedules another group withp4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused (the same one as above) but for the _same_ cpu [BUG!!], so it updates the 'assign' array to use the _old_ (wrong cpu) index because the _new_ index is in an earlier part of the 'assign' array (and hasn't been committed yet).
o perf commits the transaction using the wrong index and corrupts the other cpu
Thanks for the fix Don! I fear I won't be able to look precisely tonight, so could it wait until tomorrow? (If it's critical sure such fix should do the trick).
Quoted 30 lines · Cyrill Gorcunov
Quoted 18 lines · Don Zickus
On a P4 box stressing perf with
./perf record -o perf.data ./perf stat -v ./perf bench all
it was noticed that a slew of unknown NMIs would pop out rather quickly.
Painfully debugging this ancient platform, led me to notice cross cpu counter corruption.
The P4 machine is special in that it has 18 counters, half are used for cpu0 and the other half is for cpu1 (or all 18 if hyperthreading is disabled). But the splitting of the counters has to be actively managed by the software.
In this particular bug, one of the cpu0 specific counters was being used by cpu1 and caused all sorts of random unknown nmis.
I am not entirely sure on the corruption path, but what happens is:
Quoted 10 lines
o perf schedules a group with
p4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused but for a different cpu, so it 'swaps' the config bits and returns the updated 'assign' array with a _new_ index.
o perf schedules another group withp4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused (the same one as above) but for the _same_ cpu [BUG!!], so it updates the 'assign' array to use the _old_ (wrong cpu) index because the _new_ index is in an earlier part of the 'assign' array (and hasn't been committed yet).
o perf commits the transaction using the wrong index and corrupts the other cpu
Quoted 3 lines
Thanks for the fix Don! I fear I won't be able to look precisely tonight, so could it wait until tomorrow? (If it's critical sure such fix should do the trick).
There is no rush. Early next week is fine too. :-)
Cheers,
Don
Cyrill Gorcunov
Acked-byQuoted 13 lines · Don Zickus
Quoted 13 lines
Quoted 12 lines
I am not entirely sure on the corruption path, but what happens is:
o perf schedules a group with
p4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused but for a different cpu, so it 'swaps' the config bits and returns the updated 'assign' array with a _new_ index.
o perf schedules another group withp4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused (the same one as above) but for the _same_ cpu [BUG!!], so it updates the 'assign' array to use the _old_ (wrong cpu) index because the _new_ index is in an earlier part of the 'assign' array (and hasn't been committed yet).
o perf commits the transaction using the wrong index and corrupts the other cpu
Quoted 5 lines
Quoted 3 lines
Thanks for the fix Don! I fear I won't be able to look precisely tonight, so could it wait until tomorrow? (If it's critical sure such fix should do the trick).
There is no rush. Early next week is fine too. :-)
Hi Don, sorry for delay. I thought maybe extending match_prev_assignment() would be better (ie to figure out if previous event can run without reprogramming the counter) but this makes code only harder (and what is worse -- having no physical accees to p4 machine leaves no chance to test changes). So eventually I think your patch does the same thing as I had in mind but in different way. Thus
thanks a lot!
Quoted 27 lines · Cyrill Gorcunov
Quoted 13 lines · Don Zickus
Quoted 13 lines
Quoted 12 lines
I am not entirely sure on the corruption path, but what happens is:
o perf schedules a group with
p4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused but for a different cpu, so it 'swaps' the config bits and returns the updated 'assign' array with a _new_ index.
o perf schedules another group withp4_pmu_schedule_events()
o insidep4_pmu_schedule_events(), it notices an hwc pointer is being reused (the same one as above) but for the _same_ cpu [BUG!!], so it updates the 'assign' array to use the _old_ (wrong cpu) index because the _new_ index is in an earlier part of the 'assign' array (and hasn't been committed yet).
o perf commits the transaction using the wrong index and corrupts the other cpuQuoted 5 lines
Quoted 3 lines
Thanks for the fix Don! I fear I won't be able to look precisely tonight, so could it wait until tomorrow? (If it's critical sure such fix should do the trick).
There is no rush. Early next week is fine too. :-)
Hi Don, sorry for delay. I thought maybe extending
match_prev_assignment()would be better (ie to figure out if previous event can run without reprogramming the counter) but this makes code only harder (and what is worse -- having no physical accees to p4 machine leaves no chance to test changes). So eventually I think your patch does the same thing as I had in mind but in different way. Thus
Quoted 3 lines
Acked-by Cyrill Gorcunov <[email protected]>thanks a lot!
thanks!
Cheers,
Don
tip-bot for Don Zickus
merge bot[tip:perf/core] perf/x86/p4: Fix counter corruption when using lots of perf groups
In reply to Don Zickusperf/x86/p4: Fix counter corruption when using lots of perf groups
Commit-ID: 13beacee817d27a40ffc6f065ea0042685611dd5 Gitweb: http://git.kernel.org/tip/13beacee817d27a40ffc6f065ea0042685611dd5 Author: Don Zickus <[email protected]> AuthorDate: Wed, 29 Jan 2014 14:37:50 -0500 Committer: Ingo Molnar <[email protected]> CommitDate: Sun, 9 Feb 2014 13:17:23 +0100
perf/x86/p4: Fix counter corruption when using lots of perf groups
On a P4 box stressing perf with:
./perf record -o perf.data ./perf stat -v ./perf bench all
it was noticed that a slew of unknown NMIs would pop out rather quickly.
Painfully debugging this ancient platform, led me to notice cross cpu counter corruption.
The P4 machine is special in that it has 18 counters, half are used for cpu0 and the other half is for cpu1 (or all 18 if hyperthreading is disabled). But the splitting of the counters has to be actively managed by the software.
In this particular bug, one of the cpu0 specific counters was being used by cpu1 and caused all sorts of random unknown nmis.
I am not entirely sure on the corruption path, but what happens is:
o perf schedules a group with p4_pmu_schedule_events()
o inside p4_pmu_schedule_events(), it notices an hwc pointer is being reused but for a different cpu, so it 'swaps' the config bits and returns the updated 'assign' array with a _new_ index.
o perf schedules another group with p4_pmu_schedule_events()
o inside p4_pmu_schedule_events(), it notices an hwc pointer is being reused (the same one as above) but for the _same_ cpu [BUG!!], so it updates the 'assign' array to use the _old_ (wrong cpu) index because the _new_ index is in an earlier part of the 'assign' array (and hasn't been committed yet).
o perf commits the transaction using the wrong index and corrupts the other cpu
The [BUG!!] is because the 'hwc->config' is updated but not the 'hwc->idx'. So the check for 'p4_should_swap_ts()' is correct the first time around but incorrect the second time around (because hwc->config was updated in between).
I think the spirit of perf was to not modify anything until all the transactions had a chance to 'test' if they would succeed, and if so, commit atomically. However, P4 breaks this spirit by touching the hwc->config element.
So my fix is to continue the un-perf like breakage, by assigning hwc->idx to -1 on swap to tell follow up group scheduling to find a new index.
Of course if the transaction fails rolling this back will be difficult, but that is not different than how the current code works. :-) And I wasn't sure how much effort to cleanup the code I should do for a platform that is almost 10 years old by now.
Hence the lazy fix.
--- arch/x86/kernel/cpu/perf_event_p4.c | 19 ++++++++++++++++++- 1 file changed, 18 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/perf_event_p4.c +18 -1
@@ -1257,7 +1257,24 @@ again: pass++; goto again; } - +/* +* Perf does test runs to see if a whole group can be assigned +* together succesfully. There can be multiple rounds of this. +* Unfortunately, p4_pmu_swap_config_ts touches the hwc->config +* bits, such that the next round of group assignments will +* cause the above p4_should_swap_ts to pass instead of fail. +* This leads to counters exclusive to thread0 being used by +* thread1. +* +* Solve this with a cheap hack, reset the idx back to -1 to +* force a new lookup (p4_next_cntr) to get the right counter +* for the right thread. +* +* This probably doesn't comply with the general spirit of how +* perf wants to work, but P4 is special. :-( +*/ +if (p4_should_swap_ts (hwc->config, cpu)) + hwc->idx =-1 ;p4_pmu_swap_config_ts (hwc, cpu);if (assign) assign[i] = cntr_idx;