summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorbde <bde@FreeBSD.org>2007-11-29 02:01:21 +0000
committerbde <bde@FreeBSD.org>2007-11-29 02:01:21 +0000
commit723157380283ac617fbbbb23fb4ae6120293007b (patch)
treedafc46e7794ee0a35610452328fb817732ffbd66
parent35b85a2fdb998527546900f82b6aca6f012c6561 (diff)
downloadFreeBSD-src-723157380283ac617fbbbb23fb4ae6120293007b.zip
FreeBSD-src-723157380283ac617fbbbb23fb4ae6120293007b.tar.gz
Don't use plain "ret" instructions at targets of jump instructions,
since the branch caches on at least Athlon XP through Athlon 64 CPU's don't understand such instructions and guarantee a cache miss taking at least 10 cycles. Use the documented workaround "ret $0" instead ("nop; ret" also works, but "ret $0" is probably faster on old CPUs). Normal code (even asm code) doesn't branch to "ret", since there is usually some cleanup to do, but the __mcount, .mcount and .mexitcount entry points were optimized too well to have the minimum number of instructions (3 instructions each if profiling is not enabled) and they did this. I didn't see a significant number of cache misses for .mexitcount, but for the shared "ret" for __mcount and .mcount I observed cache misses costing 26 cycles each. For a send(2) syscall that makes about 70 function calls, the cost of these cache misses alone increased the syscall time from about 4000 cycles to about 7000 cycles. 4000 is for a profiling (GUPROF) kernel with profiling disabled; after this fix, configuring profiling only costs about 600 cycles in the 4000, which is consistent with almost perfect branch prediction in the mcounting calls.
-rw-r--r--sys/amd64/amd64/prof_machdep.c4
-rw-r--r--sys/i386/isa/prof_machdep.c4
2 files changed, 4 insertions, 4 deletions
diff --git a/sys/amd64/amd64/prof_machdep.c b/sys/amd64/amd64/prof_machdep.c
index cdccc92..f49eb11 100644
--- a/sys/amd64/amd64/prof_machdep.c
+++ b/sys/amd64/amd64/prof_machdep.c
@@ -135,7 +135,7 @@ __mcount: \n\
popq %rdx \n\
popq %rax \n\
.mcount_exit: \n\
- ret \n\
+ ret $0 \n\
");
#else /* !__GNUCLIKE_ASM */
#error "this file needs to be ported to your compiler"
@@ -187,7 +187,7 @@ GMON_PROF_HIRES = 4 \n\
popq %rdx \n\
popq %rax \n\
.mexitcount_exit: \n\
- ret \n\
+ ret $0 \n\
");
#endif /* __GNUCLIKE_ASM */
diff --git a/sys/i386/isa/prof_machdep.c b/sys/i386/isa/prof_machdep.c
index 8fd0fc0..7c7edf4 100644
--- a/sys/i386/isa/prof_machdep.c
+++ b/sys/i386/isa/prof_machdep.c
@@ -113,7 +113,7 @@ __mcount: \n\
addl $8,%esp \n\
popfl \n\
.mcount_exit: \n\
- ret \n\
+ ret $0 \n\
");
#else /* !__GNUCLIKE_ASM */
#error "this file needs to be ported to your compiler"
@@ -157,7 +157,7 @@ GMON_PROF_HIRES = 4 \n\
popl %eax \n\
popl %edx \n\
.mexitcount_exit: \n\
- ret \n\
+ ret $0 \n\
");
#endif /* __GNUCLIKE_ASM */
OpenPOWER on IntegriCloud