ARM: perf: don't pretend to support counting of L1I writes

ARM has a harvard cache architecture and cannot write directly to the
I-side.

This patch removes the L1I write events from the cache map (which
previously returned *read* events in many cases).

Reported-by: Mike Williams <michael.williams@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
This commit is contained in:
Will Deacon 2013-01-16 12:01:59 +00:00
parent 1764c591df
commit 40c390c768
3 changed files with 12 additions and 12 deletions

View file

@ -106,7 +106,7 @@ static const unsigned armv6_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
},
[C(OP_WRITE)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = ARMV6_PERFCTR_ICACHE_MISS,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
[C(OP_PREFETCH)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
@ -259,7 +259,7 @@ static const unsigned armv6mpcore_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
},
[C(OP_WRITE)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = ARMV6MPCORE_PERFCTR_ICACHE_MISS,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
[C(OP_PREFETCH)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,