Skip to contentSkip to conversation

Basic perf PMU support for Haswell v8

PATCH In review Andi Kleen 2 participants 9 messages Feb 18, 2013 – Feb 24, 2013

Subsystems (2)
lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

This is based on v7 of the full Haswell PMU support, but ported to the latest perf/core and stripped down to the bare bones

Only for very extremly basic usage.

Most interesting new features are not in this patchkit
(full version is git://git.kernel.org/pub/scm/linux/kernel/git/ak/linux-misc.git hsw/pmu5)

Contains support for:
- Basic Haswell PMU and PEBS support
- Late unmasking of the PMI
- Support for wide counters

v2: Addressed Stephane's feedback. See individual patches for details.
v3: now even more bite-sized. Qualifier constraints merged earlier.
v4: Rename some variables, add some comments and other minor changes. Add some Reviewed/Tested-bys.
v5: Address some minor review feedback. Port to latest perf/core
v6: Add just some variable names, add comments, edit descriptions, some more testing, rebased to latest perf/core
v7: Expand comment
v8: Rename structure field.

-Andi

author
#1 lore raw

PATCH 1/5 perf, x86: Add Haswell PEBS record support v4

From: Andi Kleen <[email protected]>

Add support for the Haswell extended (fmt2) PEBS format.

It has a superset of the nhm (fmt1) PEBS fields, but has a longer record so we need to adjust the code paths.

The main advantage is the new "EventingRip" support which directly gives the instruction, not off-by-one instruction. So with precise == 2 we use that directly and don't try to use LBRs and walking basic blocks. This lowers the overhead of using precise significantly.

Some other features are added in later patches.

v2: Rename various identifiers. Add more comments. Get rid of a cast.
v3: fmt2->hsw rename
v4: ip_of_the_event->real_ip rename

---
 arch/x86/kernel/cpu/perf_event.c          |    2 +-
 arch/x86/kernel/cpu/perf_event_intel_ds.c |  113 +++++++++++++++++++++++------
 2 files changed, 91 insertions(+), 24 deletions(-)
arch/x86/kernel/cpu/perf_event.c +1 -1
index bf0f01a..7d3b9bd 100644
@@ -397,7 +397,7 @@ int x86_pmu_hw_config(struct perf_event *event)
 		 * check that PEBS LBR correction does not conflict with
 		 * whatever the user is asking with attr->branch_sample_type
 		 */
-		if (event->attr.precise_ip > 1) {
+		if (event->attr.precise_ip > 1 && x86_pmu.intel_cap.pebs_format < 2) {
 			u64 *br_type = &event->attr.branch_sample_type;
 
 			if (has_branch_stack(event)) {
arch/x86/kernel/cpu/perf_event_intel_ds.c +90 -23

161 lines not on this page · raw message

Signature · 1 line
1.7.7.6
author
#2 lore raw

PATCH 2/5 perf, x86: Basic Haswell PMU support v5

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Add basic Haswell PMU support.

Similar to SandyBridge, but has a few new events and two new counter bits.

There are some new counter flags that need to be prevented from being set on fixed counters, and allowed to be set for generic counters.

Also we add support for the counter 2 constraint to handle all raw events.

Contains fixes from Stephane Eranian

v2: Folded TSX bits into standard FIXED_EVENT_CONSTRAINTS
v3: Use SNB LBR init code. Comment fix (Stephane Eranian)
v4: Add the counter2 constraints. Fix comment in the right place.
v5: Expand comment

---
 arch/x86/include/asm/perf_event.h      |    5 ++
 arch/x86/kernel/cpu/perf_event.h       |    5 ++-
 arch/x86/kernel/cpu/perf_event_intel.c |   73 ++++++++++++++++++++++++++++++++
 3 files changed, 82 insertions(+), 1 deletions(-)
arch/x86/include/asm/perf_event.h +5 -0
index 2234eaaec..80ab268 100644
@@ -29,8 +29,13 @@
 #define ARCH_PERFMON_EVENTSEL_INV			(1ULL << 23)
 #define ARCH_PERFMON_EVENTSEL_CMASK			0xFF000000ULL
 
+#define HSW_INTX					(1ULL << 32)
+#define HSW_INTX_CHECKPOINTED				(1ULL << 33)
+
 #define AMD64_EVENTSEL_GUESTONLY			(1ULL << 40)
 #define AMD64_EVENTSEL_HOSTONLY				(1ULL << 41)
+#define AMD_PERFMON_EVENTSEL_GUESTONLY			(1ULL << 40)
+#define AMD_PERFMON_EVENTSEL_HOSTONLY			(1ULL << 41)
 
 #define AMD64_EVENTSEL_EVENT	\
 	(ARCH_PERFMON_EVENTSEL_EVENT | (0x0FULL << 32))
arch/x86/kernel/cpu/perf_event.h +4 -1
index 7f5c75c..9845672 100644
@@ -219,11 +219,14 @@ struct cpu_hw_events {
  *  - inv
  *  - edge
  *  - cnt-mask
+ *  - intx
+ *  - intx_cp
  *  The other filters are supported by fixed counters.
  *  The any-thread option is supported starting with v3.
  */
+#define FIXED_EVENT_FLAGS (X86_RAW_EVENT_MASK|HSW_INTX|HSW_INTX_CHECKPOINTED)
 #define FIXED_EVENT_CONSTRAINT(c, n)	\
-	EVENT_CONSTRAINT(c, (1ULL << (32+n)), X86_RAW_EVENT_MASK)
+	EVENT_CONSTRAINT(c, (1ULL << (32+n)), FIXED_EVENT_FLAGS)
 
 /*
  * Constraint on the Event code + UMask
arch/x86/kernel/cpu/perf_event_intel.c +73 -0

97 lines not on this page · raw message

Signature · 1 line
1.7.7.6
author
#3 lore raw

PATCH 3/5 perf, x86: Basic Haswell PEBS support v4

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Add basic PEBS support for Haswell.
The constraints are similar to SandyBridge with a few new events.

v2: Readd missing pebs_aliases
v3: Readd missing hunk. Fix some constraints.
v4: Fix typo in PEBS event table (Stephane Eranian)

---
 arch/x86/kernel/cpu/perf_event.h          |    2 ++
 arch/x86/kernel/cpu/perf_event_intel.c    |    6 ++++--
 arch/x86/kernel/cpu/perf_event_intel_ds.c |   29 +++++++++++++++++++++++++++++
 3 files changed, 35 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/perf_event.h +2 -0
index 9845672..ded4667 100644
@@ -591,6 +591,8 @@ extern struct event_constraint intel_snb_pebs_event_constraints[];
 
 extern struct event_constraint intel_ivb_pebs_event_constraints[];
 
+extern struct event_constraint intel_hsw_pebs_event_constraints[];
+
 struct event_constraint *intel_pebs_constraints(struct perf_event *event);
 
 void intel_pmu_pebs_enable(struct perf_event *event);
arch/x86/kernel/cpu/perf_event_intel.c +4 -2
index bae54bb..74540ac 100644
@@ -827,7 +827,8 @@ static inline bool intel_pmu_needs_lbr_smpl(struct perf_event *event)
 		return true;
 
 	/* implicit branch sampling to correct PEBS skid */
-	if (x86_pmu.intel_cap.pebs_trap && event->attr.precise_ip > 1)
+	if (x86_pmu.intel_cap.pebs_trap && event->attr.precise_ip > 1 &&
+	    x86_pmu.intel_cap.pebs_format < 2)
 		return true;
 
 	return false;
@@ -2169,8 +2170,9 @@ __init int intel_pmu_init(void)
 		intel_pmu_lbr_init_snb();
 
 		x86_pmu.event_constraints = intel_hsw_event_constraints;
-
+		x86_pmu.pebs_constraints = intel_hsw_pebs_event_constraints;
 		x86_pmu.extra_regs = intel_snb_extra_regs;
+		x86_pmu.pebs_aliases = intel_pebs_aliases_snb;
 		/* all extra regs are per-cpu when HT is on */
 		x86_pmu.er_flags |= ERF_HAS_RSP_1;
 		x86_pmu.er_flags |= ERF_NO_HT_SHARING;
arch/x86/kernel/cpu/perf_event_intel_ds.c +29 -0
index 783b728..d1bc10e 100644
@@ -437,6 +437,35 @@ struct event_constraint intel_ivb_pebs_event_constraints[] = {
         EVENT_CONSTRAINT_END
 };
 
+struct event_constraint intel_hsw_pebs_event_constraints[] = {
+	INTEL_UEVENT_CONSTRAINT(0x01c0, 0x2), /* INST_RETIRED.PRECDIST */
+	INTEL_UEVENT_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(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 */
+	INTEL_EVENT_CONSTRAINT(0xcd, 0x8),    /* MEM_TRANS_RETIRED.* */
+	INTEL_UEVENT_CONSTRAINT(0x11d0, 0xf), /* MEM_UOPS_RETIRED.STLB_MISS_LOADS */
+	INTEL_UEVENT_CONSTRAINT(0x12d0, 0xf), /* MEM_UOPS_RETIRED.STLB_MISS_STORES */
+	INTEL_UEVENT_CONSTRAINT(0x21d0, 0xf), /* MEM_UOPS_RETIRED.LOCK_LOADS */
+	INTEL_UEVENT_CONSTRAINT(0x41d0, 0xf), /* MEM_UOPS_RETIRED.SPLIT_LOADS */
+	INTEL_UEVENT_CONSTRAINT(0x42d0, 0xf), /* MEM_UOPS_RETIRED.SPLIT_STORES */
+	INTEL_UEVENT_CONSTRAINT(0x81d0, 0xf), /* MEM_UOPS_RETIRED.ALL_LOADS */
+	INTEL_UEVENT_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 */
+	INTEL_UEVENT_CONSTRAINT(0x40d1, 0xf), /* MEM_LOAD_UOPS_RETIRED.HIT_LFB */
+	INTEL_UEVENT_CONSTRAINT(0x01d2, 0xf), /* MEM_LOAD_UOPS_LLC_HIT_RETIRED.XSNP_MISS */
+	INTEL_UEVENT_CONSTRAINT(0x02d2, 0xf), /* MEM_LOAD_UOPS_LLC_HIT_RETIRED.XSNP_HIT */
+	INTEL_UEVENT_CONSTRAINT(0x01d3, 0xf), /* MEM_LOAD_UOPS_LLC_MISS_RETIRED.LOCAL_DRAM */
+	INTEL_UEVENT_CONSTRAINT(0x04c8, 0xf), /* HLE_RETIRED.Abort */
+	INTEL_UEVENT_CONSTRAINT(0x04c9, 0xf), /* RTM_RETIRED.Abort */
+
+	EVENT_CONSTRAINT_END
+};
+
 struct event_constraint *intel_pebs_constraints(struct perf_event *event)
 {
 	struct event_constraint *c;
Signature · 1 line
1.7.7.6
author
#4 lore raw

PATCH 4/5 perf, x86: Support full width counting v3

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

Recent Intel CPUs like Haswell and IvyBridge have a new alternative MSR range for perfctrs that allows writing the full counter width. Enable this range if the hardware reports it using a new capability bit.

This lowers the overhead of perf stat slightly because it has to do less interrupts to accumulate the counter value. On Haswell it also avoids some problems with TSX aborting when the end of the counter range is reached.

v2: Print the feature at boot
v3: Rename field. Add comment.

---
 arch/x86/include/uapi/asm/msr-index.h  |    3 +++
 arch/x86/kernel/cpu/perf_event.h       |    5 +++++
 arch/x86/kernel/cpu/perf_event_intel.c |    7 +++++++
 3 files changed, 15 insertions(+), 0 deletions(-)
arch/x86/include/uapi/asm/msr-index.h +3 -0
index 433a59f..af41a77 100644
@@ -163,6 +163,9 @@
 #define MSR_KNC_EVNTSEL0               0x00000028
 #define MSR_KNC_EVNTSEL1               0x00000029
 
+/* Alternative perfctr range with full access. */
+#define MSR_IA32_PMC0			0x000004c1
+
 /* AMD64 MSRs. Not complete. See the architecture manual for a more
    complete list. */
 
arch/x86/kernel/cpu/perf_event.h +5 -0
index ded4667..ebac06d 100644
@@ -278,6 +278,11 @@ union perf_capabilities {
 		u64	pebs_arch_reg:1;
 		u64	pebs_format:4;
 		u64	smm_freeze:1;
+		/*
+		 * PMU supports separate counter range for writing
+		 * values > 32bit.
+		 */
+		u64	full_width_write:1;
 	};
 	u64	capabilities;
 };
arch/x86/kernel/cpu/perf_event_intel.c +7 -0
index 74540ac..f4d3377 100644
@@ -2230,5 +2230,12 @@ __init int intel_pmu_init(void)
 		}
 	}
 
+	/* Support full width counters using alternative MSR range */
+	if (x86_pmu.intel_cap.full_width_write) {
+		x86_pmu.max_period = x86_pmu.cntval_mask;
+		x86_pmu.perfctr = MSR_IA32_PMC0;
+		pr_cont("full-width counters, ");
+	}
+
 	return 0;
 }
Signature · 1 line
1.7.7.6
#5 lore raw

Re: [PATCH 4/5] perf, x86: Support full width counting v3

Quoted 11 lines · Andi Kleen

From: Andi Kleen <[email protected]>

Recent Intel CPUs like Haswell and IvyBridge have a new alternative MSR range for perfctrs that allows writing the full counter width. Enable this range if the hardware reports it using a new capability bit.

This lowers the overhead of perf stat slightly because it has to do less interrupts to accumulate the counter value. On Haswell it also avoids some problems with TSX aborting when the end of the counter range is reached.

The changelog does not adequately explain why this patch is critical for basic Haswell enablement. "Avoids some problems with TSX aborting" is not very helpful.

Thanks,

Ingo

author
#6 lore raw

Re: [PATCH 4/5] perf, x86: Support full width counting v3

Quoted 10 lines · Ingo Molnar
Quoted 7 lines · Andi Kleen

From: Andi Kleen <[email protected]>

Recent Intel CPUs like Haswell and IvyBridge have a new alternative MSR range for perfctrs that allows writing the full counter width. Enable this range if the hardware reports it using a new capability bit.

Quoted 8 lines
Quoted 4 lines

This lowers the overhead of perf stat slightly because it has to do less interrupts to accumulate the counter value. On Haswell it also avoids some problems with TSX aborting when the end of the counter range is reached.

The changelog does not adequately explain why this patch is critical for basic Haswell enablement. "Avoids some problems with TSX aborting" is not very helpful.

Updated description below. I don't want you from stop reviewing the rest of the patches, so I'm not posting a new series for now just for a new description. Please keep reading. You may make it through to the end.

---

Recent Intel CPUs like Haswell and IvyBridge have a new alternative MSR range for perfctrs that allows writing the full counter width. Enable this range if the hardware reports it using a new capability bit.

This lowers the overhead of perf stat slightly because it has to do less interrupts to accumulate the counter value. On Haswell it also avoids some problems with TSX aborting when the end of the counter range is reached. This can be observed when the checkpoint flag has been set, which has been enabled by the basic PMU patch. An overflow will abort the transaction and set the counter back. If the counter is near the overflow before the transaction this could happen continuously, forcing a transaction to continuously abort.
This is a partial fix, but it makes the overflows much less likely by using a larger counter, to lower the probability of the event. Additional counter measures are in the additional extended Haswell patchkit.

#7 lore raw

Re: [PATCH 4/5] perf, x86: Support full width counting v3

Quoted 17 lines · Andi Kleen

Recent Intel CPUs like Haswell and IvyBridge have a new alternative MSR range for perfctrs that allows writing the full counter width. Enable this range if the hardware reports it using a new capability bit.

This lowers the overhead of perf stat slightly because it has to do less interrupts to accumulate the counter value. On Haswell it also avoids some problems with TSX aborting when the end of the counter range is reached.

This can be observed when the checkpoint flag has been set, which has been enabled by the basic PMU patch. An overflow will abort the transaction and set the counter back. If the counter is near the overflow before the transaction this could happen continuously, forcing a transaction to continuously abort.

Quoted 4 lines

This is a partial fix, but it makes the overflows much less likely by using a larger counter, to lower the probability of the event. Additional counter measures are in the additional extended Haswell patchkit.

It would actually be _much_ more useful to first try to fix that condition - then extend the counter range. As you say it in the changelog it can happen anyway: and it's much more testable if the counter width is narrower initially.

Mind restructuring the basic patches thusly, putting the fix first and moving the counter extension to the later patches?

(If you don't have the time for that we can delay it all to v3.10, it's pretty late already even for v3.9.)

Thanks,

Ingo

author
#8 lore raw

PATCH 5/5 perf, x86: Move NMI clearing to end of PMI handler after the counter registers are reset

In reply to Andi Kleen

From: Andi Kleen <[email protected]>

This avoids some problems with spurious PMIs on Haswell. Haswell seems to behave more like P4 in this regard. Do the same thing as the P4 perf handler by unmasking the NMI only at the end. Shouldn't make any difference for earlier family 6 cores.

Tested on Haswell, IvyBridge, Westmere, Saltwell (Atom)

---
 arch/x86/kernel/cpu/perf_event_intel.c |   16 ++++++----------
 1 files changed, 6 insertions(+), 10 deletions(-)
arch/x86/kernel/cpu/perf_event_intel.c +6 -10
index f4d3377..23e1dcb 100644
@@ -1122,16 +1122,6 @@ static int intel_pmu_handle_irq(struct pt_regs *regs)
 
 	cpuc = &__get_cpu_var(cpu_hw_events);
 
-	/*
-	 * Some chipsets need to unmask the LVTPC in a particular spot
-	 * inside the nmi handler.  As a result, the unmasking was pushed
-	 * into all the nmi handlers.
-	 *
-	 * This handler doesn't seem to have any issues with the unmasking
-	 * so it was left at the top.
-	 */
-	apic_write(APIC_LVTPC, APIC_DM_NMI);
-
 	intel_pmu_disable_all();
 	handled = intel_pmu_drain_bts_buffer();
 	status = intel_pmu_get_status();
@@ -1191,6 +1181,12 @@ again:
 
 done:
 	intel_pmu_enable_all(0);
+	/*
+	 * Only unmask the NMI after the overflow counters
+	 * have been reset. This avoids spurious NMIs on
+	 * Haswell CPUs.
+	 */
+	apic_write(APIC_LVTPC, APIC_DM_NMI);
 	return handled;
 }
 
Signature · 1 line
1.7.7.6