]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
perf/x86: Don't mark DataLA addresses as store
authorAndi Kleen <ak@linux.intel.com>
Mon, 11 Aug 2014 19:27:11 +0000 (21:27 +0200)
committerIngo Molnar <mingo@kernel.org>
Wed, 13 Aug 2014 05:51:14 +0000 (07:51 +0200)
Haswell supports reporting the data address for a range
of PEBS events, including:

UOPS_RETIRED.ALL
MEM_UOPS_RETIRED.STLB_MISS_LOADS
MEM_UOPS_RETIRED.STLB_MISS_STORES
MEM_UOPS_RETIRED.LOCK_LOADS
MEM_UOPS_RETIRED.SPLIT_LOADS
MEM_UOPS_RETIRED.SPLIT_STORES
MEM_UOPS_RETIRED.ALL_LOADS
MEM_UOPS_RETIRED.ALL_STORES
MEM_LOAD_UOPS_RETIRED.L1_HIT
MEM_LOAD_UOPS_RETIRED.L2_HIT
MEM_LOAD_UOPS_RETIRED.L3_HIT
MEM_LOAD_UOPS_RETIRED.L1_MISS
MEM_LOAD_UOPS_RETIRED.L2_MISS
MEM_LOAD_UOPS_RETIRED.L3_MISS
MEM_LOAD_UOPS_RETIRED.HIT_LFB
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_MISS
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_HIT
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_HITM
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_NONE
MEM_LOAD_UOPS_L3_MISS_RETIRED.LOCAL_DRAM

This facility was already enabled earlier with the original Haswell
perf changes.

However these addresses were always reports as stores by perf, which is wrong,
as they could be loads too.  The hardware does not distinguish loads and stores
for these instructions, so there's no (cheap) way for the profiler
to find out.

Change the type to PERF_MEM_OP_NA instead.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Link: http://lkml.kernel.org/r/1407785233-32193-3-git-send-email-eranian@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/kernel/cpu/perf_event_intel_ds.c

index aca77e99e6761edc821b51fa121c1c1a07c53bb5..a9b60f32064fa24a9e4b179da0ddb04f0b5f24c2 100644 (file)
@@ -114,7 +114,7 @@ static u64 precise_store_data_hsw(struct perf_event *event, u64 status)
        u64 cfg = event->hw.config & INTEL_ARCH_EVENT_MASK;
 
        dse.val = 0;
-       dse.mem_op = PERF_MEM_OP_STORE;
+       dse.mem_op = PERF_MEM_OP_NA;
        dse.mem_lvl = PERF_MEM_LVL_NA;
 
        /*