[PATCH 1/2] perf, x86: Don't mark DataLA addresses as store

mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed

* [PATCH 1/2] perf, x86: Don't mark DataLA addresses as store
@ 2014-04-18 21:31 Andi Kleen
  2014-04-18 21:31 ` [PATCH 2/2] x86, perf: Update Haswell PEBS event table Andi Kleen
  0 siblings, 1 reply; 2+ messages in thread
From: Andi Kleen @ 2014-04-18 21:31 UTC (permalink / raw)
  To: peterz; +Cc: linux-kernel, Andi Kleen

From: Andi Kleen <ak@linux.intel.com>

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>
---
 arch/x86/kernel/cpu/perf_event_intel_ds.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c
index ae96cfa..7e9546a 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
@@ -113,7 +113,7 @@ static u64 precise_store_data_hsw(u64 status)
 	union perf_mem_data_src dse;
 
 	dse.val = 0;
-	dse.mem_op = PERF_MEM_OP_STORE;
+	dse.mem_op = PERF_MEM_OP_NA;
 	dse.mem_lvl = PERF_MEM_LVL_NA;
 	if (status & 1)
 		dse.mem_lvl = PERF_MEM_LVL_L1;
-- 
1.9.0


^ permalink raw reply	[flat|nested] 2+ messages in thread

* [PATCH 2/2] x86, perf: Update Haswell PEBS event table
  2014-04-18 21:31 [PATCH 1/2] perf, x86: Don't mark DataLA addresses as store Andi Kleen
@ 2014-04-18 21:31 ` Andi Kleen
  0 siblings, 0 replies; 2+ messages in thread
From: Andi Kleen @ 2014-04-18 21:31 UTC (permalink / raw)
  To: peterz; +Cc: linux-kernel, Andi Kleen

From: Andi Kleen <ak@linux.intel.com>

- We were allowing some sub events in c4 (BR_INST_RETIRED.*)
that are not in the event list.  Tighten the check slightly.

- We were missing some valid subevents in d1-d2 (MEM_LOAD_UOPS_RETIRED.*,
MEM_LOAD_UOPS_RETIRED_L3_HIT_RETIRED.*) Allow all subevents.

The full event list supported by this table now:

INST_RETIRED.PREC_DIST cpu/event=0xC0,umask=0x01/pp
UOPS_RETIRED.ALL cpu/event=0xC2,umask=0x01/pp
UOPS_RETIRED.RETIRE_SLOTS cpu/event=0xC2,umask=0x02/pp
BR_INST_RETIRED.CONDITIONAL cpu/event=0xC4,umask=0x01/pp
BR_INST_RETIRED.NEAR_CALL cpu/event=0xC4,umask=0x02/pp
BR_INST_RETIRED.NEAR_RETURN cpu/event=0xC4,umask=0x08/pp
BR_INST_RETIRED.NEAR_TAKEN cpu/event=0xC4,umask=0x20/pp
BR_INST_RETIRED.ALL_BRANCHES_PEBS cpu/event=0xC4,umask=0x04/pp
BR_INST_RETIRED.NEAR_CALL_R3 cpu/event=0xC4,umask=0x02/pp
BR_MISP_RETIRED.CONDITIONAL cpu/event=0xC5,umask=0x01/pp
BR_MISP_RETIRED.ALL_BRANCHES_PEBS cpu/event=0xC5,umask=0x04/pp
HLE_RETIRED.ABORTED cpu/event=0xc8,umask=0x04/pp
RTM_RETIRED.ABORTED cpu/event=0xc9,umask=0x04/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_4 cpu/event=0xCD,umask=0x01,ldlat=0x4/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_8 cpu/event=0xCD,umask=0x01,ldlat=0x8/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_16 cpu/event=0xCD,umask=0x01,ldlat=0x10/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_32 cpu/event=0xCD,umask=0x01,ldlat=0x20/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_64 cpu/event=0xCD,umask=0x01,ldlat=0x40/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_128 cpu/event=0xCD,umask=0x01,ldlat=0x80/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_256 cpu/event=0xCD,umask=0x01,ldlat=0x100/pp
MEM_TRANS_RETIRED.LOAD_LATENCY_GT_512 cpu/event=0xCD,umask=0x01,ldlat=0x200/pp
MEM_UOPS_RETIRED.STLB_MISS_LOADS cpu/event=0xD0,umask=0x11/pp
MEM_UOPS_RETIRED.STLB_MISS_STORES cpu/event=0xD0,umask=0x12/pp
MEM_UOPS_RETIRED.LOCK_LOADS cpu/event=0xD0,umask=0x21/pp
MEM_UOPS_RETIRED.SPLIT_LOADS cpu/event=0xD0,umask=0x41/pp
MEM_UOPS_RETIRED.SPLIT_STORES cpu/event=0xD0,umask=0x42/pp
MEM_UOPS_RETIRED.ALL_LOADS cpu/event=0xD0,umask=0x81/pp
MEM_UOPS_RETIRED.ALL_STORES cpu/event=0xD0,umask=0x82/pp
MEM_LOAD_UOPS_RETIRED.L1_HIT cpu/event=0xD1,umask=0x01/pp
MEM_LOAD_UOPS_RETIRED.L2_HIT cpu/event=0xD1,umask=0x02/pp
MEM_LOAD_UOPS_RETIRED.L3_HIT cpu/event=0xD1,umask=0x04/pp
MEM_LOAD_UOPS_RETIRED.L1_MISS cpu/event=0xD1,umask=0x08/pp
MEM_LOAD_UOPS_RETIRED.L2_MISS cpu/event=0xD1,umask=0x10/pp
MEM_LOAD_UOPS_RETIRED.L3_MISS cpu/event=0xD1,umask=0x20/pp
MEM_LOAD_UOPS_RETIRED.HIT_LFB cpu/event=0xD1,umask=0x40/pp
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_MISS cpu/event=0xD2,umask=0x01/pp
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_HIT cpu/event=0xD2,umask=0x02/pp
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_HITM cpu/event=0xD2,umask=0x04/pp
MEM_LOAD_UOPS_L3_HIT_RETIRED.XSNP_NONE cpu/event=0xD2,umask=0x08/pp

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
 arch/x86/kernel/cpu/perf_event_intel_ds.c | 21 +++++++++------------
 1 file changed, 9 insertions(+), 12 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c
index 7e9546a..07feaa0 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
@@ -641,7 +641,11 @@ struct event_constraint intel_hsw_pebs_event_constraints[] = {
 	INTEL_UEVENT_CONSTRAINT(0x01c0, 0x2), /* INST_RETIRED.PRECDIST */
 	INTEL_PST_HSW_CONSTRAINT(0x01c2, 0xf), /* UOPS_RETIRED.ALL */
 	INTEL_UEVENT_CONSTRAINT(0x02c2, 0xf), /* UOPS_RETIRED.RETIRE_SLOTS */
-	INTEL_EVENT_CONSTRAINT(0xc4, 0xf),    /* BR_INST_RETIRED.* */
+	INTEL_UEVENT_CONSTRAINT(0x1c4, 0xf),	/* BR_INST_RETIRED.CONDITIONAL */
+	INTEL_UEVENT_CONSTRAINT(0x2c4, 0xf),	/* BR_INST_RETIRED.NEAR_CALL */
+	INTEL_UEVENT_CONSTRAINT(0x4c4, 0xf),	/* BR_INST_RETIRED.ALL_BRANCHES */
+	INTEL_UEVENT_CONSTRAINT(0x8c4, 0xf),	/* BR_INST_RETIRED.NEAR_RETURN */
+	INTEL_UEVENT_CONSTRAINT(0x20c4, 0xf),	/* BR_INST_RETIRED.NEAR_TAKEN */
 	INTEL_UEVENT_CONSTRAINT(0x01c5, 0xf), /* BR_MISP_RETIRED.CONDITIONAL */
 	INTEL_UEVENT_CONSTRAINT(0x04c5, 0xf), /* BR_MISP_RETIRED.ALL_BRANCHES */
 	INTEL_UEVENT_CONSTRAINT(0x20c5, 0xf), /* BR_MISP_RETIRED.NEAR_TAKEN */
@@ -656,17 +660,10 @@ struct event_constraint intel_hsw_pebs_event_constraints[] = {
 	INTEL_UEVENT_CONSTRAINT(0x42d0, 0xf),
 	INTEL_UEVENT_CONSTRAINT(0x81d0, 0xf), /* MEM_UOPS_RETIRED.ALL_LOADS */
 	INTEL_PST_HSW_CONSTRAINT(0x82d0, 0xf), /* MEM_UOPS_RETIRED.ALL_STORES */
-	INTEL_UEVENT_CONSTRAINT(0x01d1, 0xf), /* MEM_LOAD_UOPS_RETIRED.L1_HIT */
-	INTEL_UEVENT_CONSTRAINT(0x02d1, 0xf), /* MEM_LOAD_UOPS_RETIRED.L2_HIT */
-	INTEL_UEVENT_CONSTRAINT(0x04d1, 0xf), /* MEM_LOAD_UOPS_RETIRED.L3_HIT */
-	/* MEM_LOAD_UOPS_RETIRED.HIT_LFB */
-	INTEL_UEVENT_CONSTRAINT(0x40d1, 0xf),
-	/* MEM_LOAD_UOPS_LLC_HIT_RETIRED.XSNP_MISS */
-	INTEL_UEVENT_CONSTRAINT(0x01d2, 0xf),
-	/* MEM_LOAD_UOPS_LLC_HIT_RETIRED.XSNP_HIT */
-	INTEL_UEVENT_CONSTRAINT(0x02d2, 0xf),
-	/* MEM_LOAD_UOPS_LLC_MISS_RETIRED.LOCAL_DRAM */
-	INTEL_UEVENT_CONSTRAINT(0x01d3, 0xf),
+	INTEL_EVENT_CONSTRAINT(0xd1, 0xf),	/* MEM_LOAD_UOPS_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd2, 0xf),	/* MEM_LOAD_UOPS_L3_HIT_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd3, 0xf),	/* MEM_LOAD_UOPS_L3_MISS_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd3, 0xf),	/* MEM_LOAD_UOPS_L3_MISS_RETIRED.* */
 	INTEL_UEVENT_CONSTRAINT(0x04c8, 0xf), /* HLE_RETIRED.Abort */
 	INTEL_UEVENT_CONSTRAINT(0x04c9, 0xf), /* RTM_RETIRED.Abort */
 
-- 
1.9.0


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2014-04-18 21:47 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-04-18 21:31 [PATCH 1/2] perf, x86: Don't mark DataLA addresses as store Andi Kleen
2014-04-18 21:31 ` [PATCH 2/2] x86, perf: Update Haswell PEBS event table Andi Kleen

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®