ring-buffer: Micro-optimize with some strategic inlining

By using inline and noinline, we are able to make the fast path of
recording an event 4% faster.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
This commit is contained in:
Steven Rostedt 2010-10-19 13:17:08 -04:00 committed by Steven Rostedt
parent 140ff89127
commit d9abde2138

View File

@ -2078,7 +2078,7 @@ static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
local_inc(&cpu_buffer->commits);
}
static void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
{
unsigned long commits;
@ -2193,13 +2193,9 @@ rb_reserve_next_event(struct ring_buffer *buffer,
#define TRACE_RECURSIVE_DEPTH 16
static int trace_recursive_lock(void)
/* Keep this code out of the fast path cache */
static noinline void trace_recursive_fail(void)
{
current->trace_recursion++;
if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH))
return 0;
/* Disable all tracing before we do anything else */
tracing_off_permanent();
@ -2211,10 +2207,21 @@ static int trace_recursive_lock(void)
in_nmi());
WARN_ON_ONCE(1);
}
static inline int trace_recursive_lock(void)
{
current->trace_recursion++;
if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH))
return 0;
trace_recursive_fail();
return -1;
}
static void trace_recursive_unlock(void)
static inline void trace_recursive_unlock(void)
{
WARN_ON_ONCE(!current->trace_recursion);