diff options
author | Mel Gorman <mgorman@techsingularity.net> | 2020-04-09 11:40:34 +0100 |
---|---|---|
committer | Steven Rostedt (VMware) <rostedt@goodmis.org> | 2020-06-05 20:08:26 -0400 |
commit | 388d8bdb87e01bcea6d0b2bf797b5f6d7b2401fb (patch) | |
tree | 089b5a342c116668658ef3ea8c5f109840300b2a /kernel/trace/trace.c | |
parent | daceabf1b494c9e40a93fcc323b1258f557f65a1 (diff) | |
download | linux-stable-388d8bdb87e01bcea6d0b2bf797b5f6d7b2401fb.tar.gz linux-stable-388d8bdb87e01bcea6d0b2bf797b5f6d7b2401fb.tar.bz2 linux-stable-388d8bdb87e01bcea6d0b2bf797b5f6d7b2401fb.zip |
tracing: Remove obsolete PREEMPTIRQ_EVENTS kconfig option
The PREEMPTIRQ_EVENTS option is unused after commit c3bc8fd637a9 ("tracing:
Centralize preemptirq tracepoints and unify their usage"). Remove it.
Note that this option is hazardous as it stands. It enables TRACE_IRQFLAGS
event on non-preempt configurations without the irqsoff tracer enabled.
TRACE_IRQFLAGS as it stands incurs significant overhead on each IRQ
entry/exit. This is because trace_hardirqs_[on|off] does all the per-cpu
manipulations and NMI checks even if tracing is completely disabled for
some insane reason. For example, netperf running UDP_STREAM on localhost
incurs a 4-6% performance penalty without any tracing if IRQFLAGS is
set. It can be put behind a static brach but even the function entry/exit
costs a little bit.
Link: https://lkml.kernel.org/r/20200409104034.GJ3818@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Diffstat (limited to 'kernel/trace/trace.c')
0 files changed, 0 insertions, 0 deletions