Date: Wed, 8 Aug 2001 11:18:44 -0700
From: Larry McVoy <lm@bitmover.com>
Someobdy really ought to take the time to make a cache miss counter program
that works like /bin/time. So I could do
$ cachemiss lat_ctx 2
10123 instruction, 22345 data, 50432 TLB flushes
Has anyone done that? If so, then what would be cool is if each of these
wonderful new features that people propose come with cachemiss results for
the related part of LMbench or some other benchmark.
On some platforms, such an app can basically be written already.
The kernel support is there for sparc64 wrt. the cache
miss stuff. It uses the cpu performance counter stuff.
Have a look at linux/include/asm-sparc64/perfctr.h to see
what I mean.
The performance counters are fancy enough that you can
ask them to do stuff like:
1) tell me D-cache misses in user and/or kernel mode
2) tell me D-cache misses that hit the E-cache
in user and/or kernel mode
3) tell me I-cache misses, but only those which actually
ended up stalling the pipeline
4) tell me E-cache misses, where the chip was not able
to get granted to memory bus immediately
5) Same as #4, but how many total bus cycles were spent
waiting for bus grant for the E-cache miss
Later,
David S. Miller
davem@redhat.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Aug 15 2001 - 21:00:19 EST