perf: Fix forgotten preempt_enable by nested writers
authorFrederic Weisbecker <fweisbec@gmail.com>
Thu, 20 May 2010 19:28:34 +0000 (21:28 +0200)
committerFrederic Weisbecker <fweisbec@gmail.com>
Thu, 20 May 2010 19:28:34 +0000 (21:28 +0200)
commitacd35a463cb2a8d2b28e094d718cf6e653ad7191
tree65f9392ef5b670eafa819138602a8c49458040ab
parentdfacc4d6c98b89609250269f518c1f54c30454ef
perf: Fix forgotten preempt_enable by nested writers

A writer that gets a reference to the buffer handle disables
preemption. When we put that reference, we check if we are
the outer most writer and if not, we simply return and defer
the head update to the outer most writer. The problem here
is that preemption is only reenabled by the outer most, that
produces preemption count imbalance for every nested writer
that exit.

So just don't forget to always re-enable preemption when we
put the buffer reference, whoever we are.

Fixes lots of sleeping in atomic warnings, visible with lock
events recording.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Robert Richter <robert.richter@amd.com>
kernel/perf_event.c