posix-cpu-timers: don't account cpu timer after stopped thread runtime accounting
authorKOSAKI Motohiro <[email protected]>
Sun, 26 May 2013 21:35:41 +0000 (17:35 -0400)
committerFrederic Weisbecker <[email protected]>
Thu, 4 Jul 2013 16:02:30 +0000 (18:02 +0200)
commitfa18f7bde3ad4568d1d343b60d963bfbd8dc3991
treeaa51270ce1fcc53f2547571b615a53e4e1bce464
parenta0b2062b0904ef07944c4a6e4d0f88ee44f1e9f2
posix-cpu-timers: don't account cpu timer after stopped thread runtime accounting

When tsk->signal->cputimer->running is 1, signal->cputimer (i.e. per process
timer account) and tsk->sum_sched_runtime (i.e. per thread timer account)
increase at the same pace because update_curr() increases both accounting.

However, there is one exception. When thread exiting, __exit_signal() turns
over task's sum_shced_runtime to sig->sum_sched_runtime, but it doesn't stop
signal->cputimer accounting.

This inconsistency makes POSIX timer wake up too early. This patch fixes it.

Original-patch-by: Olivier Langlois <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Acked-by: Peter Zijlstra <[email protected]>
Signed-off-by: Olivier Langlois <[email protected]>
Signed-off-by: KOSAKI Motohiro <[email protected]>
Signed-off-by: Frederic Weisbecker <[email protected]>
kernel/sched/stats.h