arm64: Don't trace __switch_to if function graph tracer is enabled
authorJoel Fernandes <joelaf@google.com>
Wed, 21 Dec 2016 22:44:46 +0000 (14:44 -0800)
committerWill Deacon <will.deacon@arm.com>
Tue, 10 Jan 2017 11:05:08 +0000 (11:05 +0000)
commit8f4b326d663b92e1c2a53a6857bef42e91aea5a6
treef193952a90f4cd415a80a66e36069c1708245270
parenta121103c922847ba5010819a3f250f1f7fc84ab8
arm64: Don't trace __switch_to if function graph tracer is enabled

Function graph tracer shows negative time (wrap around) when tracing
__switch_to if the nosleep-time trace option is enabled.

Time compensation for nosleep-time is done by an ftrace probe on
sched_switch. This doesn't work well for the following events (with
letters representing timestamps):
A - sched switch probe called for task T switch out
B - __switch_to calltime is recorded
C - sched_switch probe called for task T switch in
D - __switch_to rettime is recorded

If C - A > D - B, then we end up over compensating for the time spent in
__switch_to giving rise to negative times in the trace output.

On x86, __switch_to is not traced if function graph tracer is enabled.
Do the same for arm64 as well.

Cc: Todd Kjos <tkjos@google.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Joel Fernandes <joelaf@google.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
arch/arm64/kernel/process.c