Skip to contentSkip to conversation

perf vendor events arm64: Update N2 and V2 metrics and events using Arm telemetry repo

PATCH In review James Clark 4 participants 19 messages Aug 16, 2023 – Aug 30, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v6 cover perf vendor events arm64: Update N2 and V2 metrics and events using Arm telemetry repo

This is a completely new approach from V3 [1], although the metrics and event descriptions are autogenerated, the topdown metrics have been manually edited to use #no_stall_errata (now directly comparing on the CPUID in v5). This removes the need to duplicate the whole set of JSONs when only the topdown metrics are different between N2 and V2.

The CPU ID comparison function still needs to change so that the new literal can compare on versions, but now no change is needed to mapfile or the PMU event generation code because we still only support one set of JSONs per CPU.

[1]: https://lore.kernel.org/lkml/[email protected]/

------

Changes since v5:
* Split patch 5 into one to add the new expression builtin and one to update the Arm metric formulas
* Make _get_cpuid() still return an error if reading the file fails
* Improve the return value comment on strcmp_cpuid_str()
* Remove the "The events json file with the highest matching version is used." comment because that was only true in v2
* Drop patch 1 because it was applied (I kept patch 4 even though it was applied so that there is some context)

Changes since v4:
* Replace the #no_stall_errata literal with a more generic function for comparing CPU IDs. This will hopefully keep configuration out of the code and inside the JSONs

Changes since v3:
* Instead of duplicating all the metrics, add a new expression literal that can be used to share the same metrics between N2 and V2
* Move tests to arch/arm64/tests
* Remove changes from jevents.py and mapfile.csv

Changes since v2:
* version -> variant in second commit message
* Add a bit more detail about version matching in the second commit message
* Update the comments in pmu-events/arch/arm64/mapfile.csv to say that variant and revision fields are now used
* Increase the CC list

Changes since v1:
* Split last change into two so it doesn't hit the mailing list size limit

James Clark (6):
perf arm64: Allow version comparisons of CPU IDs
perf test: Add a test for the new Arm CPU ID comparison behavior perf vendor events arm64: Update scale units and descriptions of
common topdown metrics
perf jevents: Add a new expression builtin strcmp_cpuid_str() perf vendor events arm64: Update stall_slot workaround for N2 r0p3 perf vendor events arm64: Update N2 and V2 metrics and events using
Arm telemetry repo

 tools/perf/arch/arm64/include/arch-tests.h    |   3 +
 tools/perf/arch/arm64/tests/Build             |   1 +
 tools/perf/arch/arm64/tests/arch-tests.c      |   4 +
 tools/perf/arch/arm64/tests/cpuid-match.c     |  38 ++
 tools/perf/arch/arm64/util/header.c           |  67 +++-
 tools/perf/arch/arm64/util/pmu.c              |  18 +-
 .../arch/arm64/arm/neoverse-n2-v2/branch.json |   8 -
 .../arch/arm64/arm/neoverse-n2-v2/bus.json    |  18 +-
 .../arch/arm64/arm/neoverse-n2-v2/cache.json  | 155 --------
 .../arm64/arm/neoverse-n2-v2/exception.json   |  45 ++-
 .../arm/neoverse-n2-v2/fp_operation.json      |  22 ++
 .../arm64/arm/neoverse-n2-v2/general.json     |  10 +
 .../arm64/arm/neoverse-n2-v2/instruction.json | 143 -------
 .../arm64/arm/neoverse-n2-v2/l1d_cache.json   |  54 +++
 .../arm64/arm/neoverse-n2-v2/l1i_cache.json   |  14 +
 .../arm64/arm/neoverse-n2-v2/l2_cache.json    |  50 +++
 .../arm64/arm/neoverse-n2-v2/l3_cache.json    |  22 ++
 .../arm64/arm/neoverse-n2-v2/ll_cache.json    |  10 +
 .../arch/arm64/arm/neoverse-n2-v2/memory.json |  39 +-
 .../arm64/arm/neoverse-n2-v2/metrics.json     | 365 ++++++++++--------
 .../arm64/arm/neoverse-n2-v2/pipeline.json    |  23 --
 .../arm64/arm/neoverse-n2-v2/retired.json     |  30 ++
 .../arch/arm64/arm/neoverse-n2-v2/spe.json    |  12 +-
 .../arm/neoverse-n2-v2/spec_operation.json    | 110 ++++++
 .../arch/arm64/arm/neoverse-n2-v2/stall.json  |  30 ++
 .../arch/arm64/arm/neoverse-n2-v2/sve.json    |  50 +++
 .../arch/arm64/arm/neoverse-n2-v2/tlb.json    |  66 ++++
 .../arch/arm64/arm/neoverse-n2-v2/trace.json  |  27 +-
 tools/perf/pmu-events/arch/arm64/sbsa.json    |  24 +-
 tools/perf/pmu-events/metric.py               |  17 +-
 tools/perf/util/expr.c                        |  18 +
 tools/perf/util/expr.h                        |   1 +
 tools/perf/util/expr.l                        |   1 +
 tools/perf/util/expr.y                        |   8 +-
 tools/perf/util/pmu.c                         |  17 +
 tools/perf/util/pmu.h                         |   1 +
 36 files changed, 923 insertions(+), 598 deletions(-)
 create mode 100644 tools/perf/arch/arm64/tests/cpuid-match.c
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/branch.json
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/fp_operation.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/general.json
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/instruction.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l1d_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l1i_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l2_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l3_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/ll_cache.json
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/pipeline.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/retired.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/spec_operation.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/stall.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/sve.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/tlb.json
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#1 lore raw

PATCH v6 1/6 perf arm64: Allow version comparisons of CPU IDs

Currently variant and revision fields are masked out of the MIDR so it's not possible to compare different versions of the same CPU. In a later commit a workaround will be removed just for N2 r0p3, so enable comparisons on version.

This has the side effect of changing the MIDR stored in the header of the perf.data file to no longer have masked version fields. It also affects the lookups in mapfile.csv, but as that currently only has zeroed version fields, it has no actual effect. The mapfile.csv documentation also states to zero the version fields, so unless this isn't done it will continue to have no effect.

There is an existing weak default strcmp_cpuid_str() function, and an x86 version. This adds another version for arm64.

---
 tools/perf/arch/arm64/util/header.c | 67 ++++++++++++++++++++++-------
 1 file changed, 52 insertions(+), 15 deletions(-)
tools/perf/arch/arm64/util/header.c +52 -15
index 80b9f6287fe2..a2eef9ec5491 100644
@@ -1,3 +1,6 @@
+#include <linux/kernel.h>
+#include <linux/bits.h>
+#include <linux/bitfield.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <perf/cpumap.h>
@@ -10,15 +13,14 @@
 
 #define MIDR "/regs/identification/midr_el1"
 #define MIDR_SIZE 19
-#define MIDR_REVISION_MASK      0xf
-#define MIDR_VARIANT_SHIFT      20
-#define MIDR_VARIANT_MASK       (0xf << MIDR_VARIANT_SHIFT)
+#define MIDR_REVISION_MASK      GENMASK(3, 0)
+#define MIDR_VARIANT_MASK	GENMASK(23, 20)
 
 static int _get_cpuid(char *buf, size_t sz, struct perf_cpu_map *cpus)
 {
 	const char *sysfs = sysfs__mountpoint();
-	u64 midr = 0;
 	int cpu;
+	int ret = EINVAL;
 
 	if (!sysfs || sz < MIDR_SIZE)
 		return EINVAL;
@@ -44,22 +46,13 @@ static int _get_cpuid(char *buf, size_t sz, struct perf_cpu_map *cpus)
 		}
 		fclose(file);
 
-		/* Ignore/clear Variant[23:20] and
-		 * Revision[3:0] of MIDR
-		 */
-		midr = strtoul(buf, NULL, 16);
-		midr &= (~(MIDR_VARIANT_MASK | MIDR_REVISION_MASK));
-		scnprintf(buf, MIDR_SIZE, "0x%016lx", midr);
 		/* got midr break loop */
+		ret = 0;
 		break;
 	}
 
 	perf_cpu_map__put(cpus);
-
-	if (!midr)
-		return EINVAL;
-
-	return 0;
+	return ret;
 }
 
 int get_cpuid(char *buf, size_t sz)
@@ -99,3 +92,47 @@ char *get_cpuid_str(struct perf_pmu *pmu)
 
 	return buf;
 }
+
+/*
+ * Return 0 if idstr is a higher or equal to version of the same part as
+ * mapcpuid. Therefore, if mapcpuid has 0 for revision and variant then any
+ * version of idstr will match as long as it's the same CPU type.
+ *
+ * Return 1 if the CPU type is different or the version of idstr is lower.
+ */
+int strcmp_cpuid_str(const char *mapcpuid, const char *idstr)
+{
+	u64 map_id = strtoull(mapcpuid, NULL, 16);
+	char map_id_variant = FIELD_GET(MIDR_VARIANT_MASK, map_id);
+	char map_id_revision = FIELD_GET(MIDR_REVISION_MASK, map_id);
+	u64 id = strtoull(idstr, NULL, 16);
+	char id_variant = FIELD_GET(MIDR_VARIANT_MASK, id);
+	char id_revision = FIELD_GET(MIDR_REVISION_MASK, id);
+	u64 id_fields = ~(MIDR_VARIANT_MASK | MIDR_REVISION_MASK);
+
+	/* Compare without version first */
+	if ((map_id & id_fields) != (id & id_fields))
+		return 1;
+
+	/*
+	 * ID matches, now compare version.
+	 *
+	 * Arm revisions (like r0p0) are compared here like two digit semver
+	 * values eg. 1.3 < 2.0 < 2.1 < 2.2.
+	 *
+	 *  r = high value = 'Variant' field in MIDR
+	 *  p = low value  = 'Revision' field in MIDR
+	 *
+	 */
+	if (id_variant > map_id_variant)
+		return 0;
+
+	if (id_variant == map_id_variant && id_revision >= map_id_revision)
+		return 0;
+
+	/*
+	 * variant is less than mapfile variant or variants are the same but
+	 * the revision doesn't match. Return no match.
+	 */
+	return 1;
+}
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
reviewer Reviewed-by on v6 1/6
#2 lore raw

Re: [PATCH v6 1/6] perf arm64: Allow version comparisons of CPU IDs

Quoted 12 lines · James Clark

Currently variant and revision fields are masked out of the MIDR so it's not possible to compare different versions of the same CPU. In a later commit a workaround will be removed just for N2 r0p3, so enable comparisons on version.

This has the side effect of changing the MIDR stored in the header of the perf.data file to no longer have masked version fields. It also affects the lookups in mapfile.csv, but as that currently only has zeroed version fields, it has no actual effect. The mapfile.csv documentation also states to zero the version fields, so unless this isn't done it will continue to have no effect.

Quoted 4 lines

There is an existing weak default strcmp_cpuid_str() function, and an x86 version. This adds another version for arm64.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#3 lore raw

PATCH v6 2/6 perf test: Add a test for the new Arm CPU ID comparison behavior

In reply to James Clark

Now that variant and revision fields are taken into account the behavior is slightly more complicated so add a test to ensure that this behaves as expected.

---
 tools/perf/arch/arm64/include/arch-tests.h |  3 ++
 tools/perf/arch/arm64/tests/Build          |  1 +
 tools/perf/arch/arm64/tests/arch-tests.c   |  4 +++
 tools/perf/arch/arm64/tests/cpuid-match.c  | 38 ++++++++++++++++++++++
 4 files changed, 46 insertions(+)
 create mode 100644 tools/perf/arch/arm64/tests/cpuid-match.c
tools/perf/arch/arm64/include/arch-tests.h +3 -0
index 452b3d904521..474d7cf5afbd 100644
@@ -2,6 +2,9 @@
 #ifndef ARCH_TESTS_H
 #define ARCH_TESTS_H
 
+struct test_suite;
+
+int test__cpuid_match(struct test_suite *test, int subtest);
 extern struct test_suite *arch_tests[];
 
 #endif
tools/perf/arch/arm64/tests/Build +1 -0
index a61c06bdb757..e337c09e7f56 100644
@@ -2,3 +2,4 @@ perf-y += regs_load.o
 perf-$(CONFIG_DWARF_UNWIND) += dwarf-unwind.o
 
 perf-y += arch-tests.o
+perf-y += cpuid-match.o
tools/perf/arch/arm64/tests/arch-tests.c +4 -0
index ad16b4f8f63e..74932e72c727 100644
@@ -3,9 +3,13 @@
 #include "tests/tests.h"
 #include "arch-tests.h"
 
+
+DEFINE_SUITE("arm64 CPUID matching", cpuid_match);
+
 struct test_suite *arch_tests[] = {
 #ifdef HAVE_DWARF_UNWIND_SUPPORT
 	&suite__dwarf_unwind,
 #endif
+	&suite__cpuid_match,
 	NULL,
 };
tools/perf/arch/arm64/tests/cpuid-match.c new file +38 -0
new file mode 100644
index 000000000000..af0871b54ae7
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/compiler.h>
+
+#include "arch-tests.h"
+#include "tests/tests.h"
+#include "util/header.h"
+
+int test__cpuid_match(struct test_suite *test __maybe_unused,
+			     int subtest __maybe_unused)
+{
+	/* midr with no leading zeros matches */
+	if (strcmp_cpuid_str("0x410fd0c0", "0x00000000410fd0c0"))
+		return -1;
+	/* Upper case matches */
+	if (strcmp_cpuid_str("0x410fd0c0", "0x00000000410FD0C0"))
+		return -1;
+	/* r0p0 = r0p0 matches */
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000410fd480"))
+		return -1;
+	/* r0p1 > r0p0 matches */
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000410fd481"))
+		return -1;
+	/* r1p0 > r0p0 matches*/
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000411fd480"))
+		return -1;
+	/* r0p0 < r0p1 doesn't match */
+	if (!strcmp_cpuid_str("0x00000000410fd481", "0x00000000410fd480"))
+		return -1;
+	/* r0p0 < r1p0 doesn't match */
+	if (!strcmp_cpuid_str("0x00000000411fd480", "0x00000000410fd480"))
+		return -1;
+	/* Different CPU doesn't match */
+	if (!strcmp_cpuid_str("0x00000000410fd4c0", "0x00000000430f0af0"))
+		return -1;
+
+	return 0;
+}
+
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#4 lore raw

Re: [PATCH v6 2/6] perf test: Add a test for the new Arm CPU ID comparison behavior

Em Wed, Aug 16, 2023 at 12:47:44PM +0100, James Clark escreveu:

Quoted 43 lines

+++ b/tools/perf/arch/arm64/tests/cpuid-match.c

tools/perf/arch/arm64/tests/cpuid-match.c new file +38 -0
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/compiler.h>
+
+#include "arch-tests.h"
+#include "tests/tests.h"
+#include "util/header.h"
+
+int test__cpuid_match(struct test_suite *test __maybe_unused,
+			     int subtest __maybe_unused)
+{
+	/* midr with no leading zeros matches */
+	if (strcmp_cpuid_str("0x410fd0c0", "0x00000000410fd0c0"))
+		return -1;
+	/* Upper case matches */
+	if (strcmp_cpuid_str("0x410fd0c0", "0x00000000410FD0C0"))
+		return -1;
+	/* r0p0 = r0p0 matches */
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000410fd480"))
+		return -1;
+	/* r0p1 > r0p0 matches */
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000410fd481"))
+		return -1;
+	/* r1p0 > r0p0 matches*/
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000411fd480"))
+		return -1;
+	/* r0p0 < r0p1 doesn't match */
+	if (!strcmp_cpuid_str("0x00000000410fd481", "0x00000000410fd480"))
+		return -1;
+	/* r0p0 < r1p0 doesn't match */
+	if (!strcmp_cpuid_str("0x00000000411fd480", "0x00000000410fd480"))
+		return -1;
+	/* Different CPU doesn't match */
+	if (!strcmp_cpuid_str("0x00000000410fd4c0", "0x00000000430f0af0"))
+		return -1;
+
+	return 0;
+}
+
Signature · 1 line
2.34.1

⬢[acme@toolbox perf-tools-next]$ git am ./v6_20230816_james_clark_perf_vendor_events_arm64_update_n2_and_v2_metrics_and_events_using_arm_telem.mbx
Applying: perf test: Add a test for the new Arm CPU ID comparison behavior .git/rebase-apply/patch:93: new blank line at EOF.
+
warning: 1 line adds whitespace errors.
⬢[acme@toolbox perf-tools-next]$

I'm removing it

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#5 lore raw

Re: [PATCH v6 2/6] perf test: Add a test for the new Arm CPU ID comparison behavior

Quoted 51 lines · Arnaldo Carvalho de Melo

Em Wed, Aug 16, 2023 at 12:47:44PM +0100, James Clark escreveu:

Quoted 43 lines

+++ b/tools/perf/arch/arm64/tests/cpuid-match.c

tools/perf/arch/arm64/tests/cpuid-match.c new file +38 -0
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/compiler.h>
+
+#include "arch-tests.h"
+#include "tests/tests.h"
+#include "util/header.h"
+
+int test__cpuid_match(struct test_suite *test __maybe_unused,
+			     int subtest __maybe_unused)
+{
+	/* midr with no leading zeros matches */
+	if (strcmp_cpuid_str("0x410fd0c0", "0x00000000410fd0c0"))
+		return -1;
+	/* Upper case matches */
+	if (strcmp_cpuid_str("0x410fd0c0", "0x00000000410FD0C0"))
+		return -1;
+	/* r0p0 = r0p0 matches */
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000410fd480"))
+		return -1;
+	/* r0p1 > r0p0 matches */
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000410fd481"))
+		return -1;
+	/* r1p0 > r0p0 matches*/
+	if (strcmp_cpuid_str("0x00000000410fd480", "0x00000000411fd480"))
+		return -1;
+	/* r0p0 < r0p1 doesn't match */
+	if (!strcmp_cpuid_str("0x00000000410fd481", "0x00000000410fd480"))
+		return -1;
+	/* r0p0 < r1p0 doesn't match */
+	if (!strcmp_cpuid_str("0x00000000411fd480", "0x00000000410fd480"))
+		return -1;
+	/* Different CPU doesn't match */
+	if (!strcmp_cpuid_str("0x00000000410fd4c0", "0x00000000430f0af0"))
+		return -1;
+
+	return 0;
+}
+
Signature · 1 line
2.34.1

⬢[acme@toolbox perf-tools-next]$ git am ./v6_20230816_james_clark_perf_vendor_events_arm64_update_n2_and_v2_metrics_and_events_using_arm_telem.mbx
Applying: perf test: Add a test for the new Arm CPU ID comparison behavior .git/rebase-apply/patch:93: new blank line at EOF.
+
warning: 1 line adds whitespace errors.
⬢[acme@toolbox perf-tools-next]$

Quoted 1 line

I'm removing it

Interesting that checkpatch.pl doesn't see that. Thanks for the fix.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#6 lore raw

PATCH v6 3/6 perf vendor events arm64: Update scale units and descriptions of common topdown metrics

In reply to James Clark

Metrics will be published here [1] going forwards, but they have slightly different scale units. To allow autogenerated metrics to be added more easily, update the scale units to match.

The more detailed descriptions have also been taken and added to the common file.

[1]: https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/tree/main/data/pmu/cpu/

---
 tools/perf/pmu-events/arch/arm64/sbsa.json | 24 +++++++++++-----------
 1 file changed, 12 insertions(+), 12 deletions(-)
tools/perf/pmu-events/arch/arm64/sbsa.json +12 -12
index f90b338261ac..4eed79a28f6e 100644
@@ -1,34 +1,34 @@
 [
     {
-        "MetricExpr": "stall_slot_frontend / (#slots * cpu_cycles)",
-        "BriefDescription": "Frontend bound L1 topdown metric",
+        "MetricExpr": "100 * (stall_slot_frontend / (#slots * cpu_cycles))",
+        "BriefDescription": "This metric is the percentage of total slots that were stalled due to resource constraints in the frontend of the processor.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "frontend_bound",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     },
     {
-        "MetricExpr": "(1 - op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles))",
-        "BriefDescription": "Bad speculation L1 topdown metric",
+        "MetricExpr": "100 * ((1 - op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles)))",
+        "BriefDescription": "This metric is the percentage of total slots that executed operations and didn't retire due to a pipeline flush.\nThis indicates cycles that were utilized but inefficiently.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "bad_speculation",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     },
     {
-        "MetricExpr": "(op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles))",
-        "BriefDescription": "Retiring L1 topdown metric",
+        "MetricExpr": "100 * ((op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles)))",
+        "BriefDescription": "This metric is the percentage of total slots that retired operations, which indicates cycles that were utilized efficiently.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "retiring",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     },
     {
-        "MetricExpr": "stall_slot_backend / (#slots * cpu_cycles)",
-        "BriefDescription": "Backend Bound L1 topdown metric",
+        "MetricExpr": "100 * (stall_slot_backend / (#slots * cpu_cycles))",
+        "BriefDescription": "This metric is the percentage of total slots that were stalled due to resource constraints in the backend of the processor.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "backend_bound",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     }
 ]
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#7 lore raw

Re: [PATCH v6 3/6] perf vendor events arm64: Update scale units and descriptions of common topdown metrics

Em Wed, Aug 16, 2023 at 12:47:45PM +0100, James Clark escreveu:

Quoted 8 lines

Metrics will be published here [1] going forwards, but they have slightly different scale units. To allow autogenerated metrics to be added more easily, update the scale units to match.

The more detailed descriptions have also been taken and added to the common file.

[1]: https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/tree/main/data/pmu/cpu/

This one I had cherry-picked and is already in perf-tools-next, I applied the others locally, will wait till later to see if people ack/object.

- Arnaldo

Quoted 61 lines
---
 tools/perf/pmu-events/arch/arm64/sbsa.json | 24 +++++++++++-----------
 1 file changed, 12 insertions(+), 12 deletions(-)
tools/perf/pmu-events/arch/arm64/sbsa.json +12 -12
index f90b338261ac..4eed79a28f6e 100644
@@ -1,34 +1,34 @@
 [
     {
-        "MetricExpr": "stall_slot_frontend / (#slots * cpu_cycles)",
-        "BriefDescription": "Frontend bound L1 topdown metric",
+        "MetricExpr": "100 * (stall_slot_frontend / (#slots * cpu_cycles))",
+        "BriefDescription": "This metric is the percentage of total slots that were stalled due to resource constraints in the frontend of the processor.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "frontend_bound",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     },
     {
-        "MetricExpr": "(1 - op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles))",
-        "BriefDescription": "Bad speculation L1 topdown metric",
+        "MetricExpr": "100 * ((1 - op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles)))",
+        "BriefDescription": "This metric is the percentage of total slots that executed operations and didn't retire due to a pipeline flush.\nThis indicates cycles that were utilized but inefficiently.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "bad_speculation",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     },
     {
-        "MetricExpr": "(op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles))",
-        "BriefDescription": "Retiring L1 topdown metric",
+        "MetricExpr": "100 * ((op_retired / op_spec) * (1 - stall_slot / (#slots * cpu_cycles)))",
+        "BriefDescription": "This metric is the percentage of total slots that retired operations, which indicates cycles that were utilized efficiently.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "retiring",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     },
     {
-        "MetricExpr": "stall_slot_backend / (#slots * cpu_cycles)",
-        "BriefDescription": "Backend Bound L1 topdown metric",
+        "MetricExpr": "100 * (stall_slot_backend / (#slots * cpu_cycles))",
+        "BriefDescription": "This metric is the percentage of total slots that were stalled due to resource constraints in the backend of the processor.",
         "DefaultMetricgroupName": "TopdownL1",
         "MetricGroup": "Default;TopdownL1",
         "MetricName": "backend_bound",
-        "ScaleUnit": "100%"
+        "ScaleUnit": "1percent of slots"
     }
 ]
Signature · 1 line
2.34.1
Signature · 6 lines
- Arnaldo

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#8 lore raw

PATCH v6 4/6 perf jevents: Add a new expression builtin strcmp_cpuid_str()

In reply to James Clark

This will allow writing formulas that are conditional on a specific CPU type or CPU version. It calls through to the existing strcmp_cpuid_str() function in Perf which has a default weak version, and an arch specific version for x86 and arm64.

The function takes an 'ID' type value, which is a string. But in this case Arm CPU IDs are hex numbers prefixed with '0x'. metric.py assumes strings are only used by event names, and that they can't start with a number ('0'), so an additional change has to be made to the regex to convert hex numbers back to 'ID' types.

---
 tools/perf/arch/arm64/util/pmu.c | 18 +-----------------
 tools/perf/pmu-events/metric.py  | 17 +++++++++++++++--
 tools/perf/util/expr.c           | 18 ++++++++++++++++++
 tools/perf/util/expr.h           |  1 +
 tools/perf/util/expr.l           |  1 +
 tools/perf/util/expr.y           |  8 +++++++-
 tools/perf/util/pmu.c            | 17 +++++++++++++++++
 tools/perf/util/pmu.h            |  1 +
 8 files changed, 61 insertions(+), 20 deletions(-)
tools/perf/arch/arm64/util/pmu.c +1 -17
index 512a8f13c4de..615084eb88d8 100644
@@ -2,28 +2,12 @@
 
 #include <internal/cpumap.h>
 #include "../../../util/cpumap.h"
+#include "../../../util/header.h"
 #include "../../../util/pmu.h"
 #include "../../../util/pmus.h"
 #include <api/fs/fs.h>
 #include <math.h>
 
-static struct perf_pmu *pmu__find_core_pmu(void)
-{
-	struct perf_pmu *pmu = NULL;
-
-	while ((pmu = perf_pmus__scan_core(pmu))) {
-		/*
-		 * The cpumap should cover all CPUs. Otherwise, some CPUs may
-		 * not support some events or have different event IDs.
-		 */
-		if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)
-			return NULL;
-
-		return pmu;
-	}
-	return NULL;
-}
-
 const struct pmu_metrics_table *pmu_metrics_table__find(void)
 {
 	struct perf_pmu *pmu = pmu__find_core_pmu();
tools/perf/pmu-events/metric.py +15 -2
index 85a3545f5b6a..0e9ec65d92ae 100644
@@ -413,6 +413,10 @@ def has_event(event: Event) -> Function:
   # pylint: disable=invalid-name
   return Function('has_event', event)
 
+def strcmp_cpuid_str(event: str) -> Function:
+  # pylint: disable=redefined-builtin
+  # pylint: disable=invalid-name
+  return Function('strcmp_cpuid_str', event)
 
 class Metric:
   """An individual metric that will specifiable on the perf command line."""
@@ -541,14 +545,23 @@ def ParsePerfJson(orig: str) -> Expression:
   """
   # pylint: disable=eval-used
   py = orig.strip()
+  # First try to convert everything that looks like a string (event name) into Event(r"EVENT_NAME").
+  # This isn't very selective so is followed up by converting some unwanted conversions back again
   py = re.sub(r'([a-zA-Z][^-+/\* \\\(\),]*(?:\\.[^-+/\* \\\(\),]*)*)',
               r'Event(r"\1")', py)
+  # If it started with a # it should have been a literal, rather than an event name
   py = re.sub(r'#Event\(r"([^"]*)"\)', r'Literal("#\1")', py)
+  # Convert accidentally converted hex constants ("0Event(r"xDEADBEEF)"") back to a constant,
+  # but keep it wrapped in Event(), otherwise Python drops the 0x prefix and it gets interpreted as
+  # a double by the Bison parser
+  py = re.sub(r'0Event\(r"[xX]([0-9a-fA-F]*)"\)', r'Event("0x\1")', py)
+  # Convert accidentally converted scientific notation constants back
   py = re.sub(r'([0-9]+)Event\(r"(e[0-9]+)"\)', r'\1\2', py)
-  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event']
+  # Convert all the known keywords back from events to just the keyword
+  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event', 'strcmp_cpuid_str',
+              'cpuid_not_more_than']
   for kw in keywords:
     py = re.sub(rf'Event\(r"{kw}"\)', kw, py)
-
   try:
     parsed = ast.parse(py, mode='eval')
   except SyntaxError as e:
tools/perf/util/expr.c +18 -0
index 7410a165f68b..0985a3cbc6f9 100644
@@ -13,6 +13,8 @@
 #include <util/expr-bison.h>
 #include <util/expr-flex.h>
 #include "util/hashmap.h"
+#include "util/header.h"
+#include "util/pmu.h"
 #include "smt.h"
 #include "tsc.h"
 #include <api/fs/fs.h>
@@ -495,3 +497,19 @@ double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const
 	evlist__delete(tmp);
 	return ret;
 }
+
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx __maybe_unused,
+		       bool compute_ids __maybe_unused, const char *test_id)
+{
+	double ret;
+	struct perf_pmu *pmu = pmu__find_core_pmu();
+	char *cpuid = perf_pmu__getcpuid(pmu);
+
+	if (!cpuid)
+		return NAN;
+
+	ret = !strcmp_cpuid_str(test_id, cpuid);
+
+	free(cpuid);
+	return ret;
+}
tools/perf/util/expr.h +1 -0
index 3c1e49b3e35d..c0cec29ddc29 100644
@@ -55,5 +55,6 @@ double expr_id_data__value(const struct expr_id_data *data);
 double expr_id_data__source_count(const struct expr_id_data *data);
 double expr__get_literal(const char *literal, const struct expr_scanner_ctx *ctx);
 double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
 
 #endif
tools/perf/util/expr.l +1 -0
index dbb117414710..0feef0726c48 100644
@@ -114,6 +114,7 @@ if		{ return IF; }
 else		{ return ELSE; }
 source_count	{ return SOURCE_COUNT; }
 has_event	{ return HAS_EVENT; }
+strcmp_cpuid_str	{ return STRCMP_CPUID_STR; }
 {literal}	{ return literal(yyscanner, sctx); }
 {number}	{ return value(yyscanner); }
 {symbol}	{ return str(yyscanner, ID, sctx->runtime); }
tools/perf/util/expr.y +7 -1
index 65d54a6f29ad..6c93b358cc2d 100644
@@ -39,7 +39,7 @@ int expr_lex(YYSTYPE * yylval_param , void *yyscanner);
 	} ids;
 }
 
-%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT EXPR_ERROR
+%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT STRCMP_CPUID_STR EXPR_ERROR
 %left MIN MAX IF
 %left '|'
 %left '^'
@@ -207,6 +207,12 @@ expr: NUMBER
 	$$.ids = NULL;
 	free($3);
 }
+| STRCMP_CPUID_STR '(' ID ')'
+{
+	$$.val = expr__strcmp_cpuid_str(ctx, compute_ids, $3);
+	$$.ids = NULL;
+	free($3);
+}
 | expr '|' expr
 {
 	if (is_const($1.val) && is_const($3.val)) {
tools/perf/util/pmu.c +17 -0
index d5406effc169..a7f05e4dda97 100644
@@ -1790,3 +1790,20 @@ void perf_pmu__delete(struct perf_pmu *pmu)
 	zfree(&pmu->alias_name);
 	free(pmu);
 }
+
+struct perf_pmu *pmu__find_core_pmu(void)
+{
+	struct perf_pmu *pmu = NULL;
+
+	while ((pmu = perf_pmus__scan_core(pmu))) {
+		/*
+		 * The cpumap should cover all CPUs. Otherwise, some CPUs may
+		 * not support some events or have different event IDs.
+		 */
+		if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)
+			return NULL;
+
+		return pmu;
+	}
+	return NULL;
+}
tools/perf/util/pmu.h +1 -0
index 6b414cecbad2..7ff925224165 100644
@@ -289,5 +289,6 @@ int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename,
 struct perf_pmu *perf_pmu__lookup(struct list_head *pmus, int dirfd, const char *lookup_name);
 struct perf_pmu *perf_pmu__create_placeholder_core_pmu(struct list_head *core_pmus);
 void perf_pmu__delete(struct perf_pmu *pmu);
+struct perf_pmu *pmu__find_core_pmu(void);
 
 #endif /* __PMU_H */
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
reviewer Reviewed-by on v6 4/6
#9 lore raw

Re: [PATCH v6 4/6] perf jevents: Add a new expression builtin strcmp_cpuid_str()

Quoted 5 lines · James Clark

This will allow writing formulas that are conditional on a specific CPU type or CPU version. It calls through to the existing strcmp_cpuid_str() function in Perf which has a default weak version, and an arch specific version for x86 and arm64.

Quoted 7 lines

The function takes an 'ID' type value, which is a string. But in this case Arm CPU IDs are hex numbers prefixed with '0x'. metric.py assumes strings are only used by event names, and that they can't start with a number ('0'), so an additional change has to be made to the regex to convert hex numbers back to 'ID' types.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

reviewer
#10 lore raw

Re: [PATCH v6 4/6] perf jevents: Add a new expression builtin strcmp_cpuid_str()

In reply to James Clark
Quoted 148 lines · James Clark

This will allow writing formulas that are conditional on a specific CPU type or CPU version. It calls through to the existing strcmp_cpuid_str() function in Perf which has a default weak version, and an arch specific version for x86 and arm64.

The function takes an 'ID' type value, which is a string. But in this case Arm CPU IDs are hex numbers prefixed with '0x'. metric.py assumes strings are only used by event names, and that they can't start with a number ('0'), so an additional change has to be made to the regex to convert hex numbers back to 'ID' types.

---
 tools/perf/arch/arm64/util/pmu.c | 18 +-----------------
 tools/perf/pmu-events/metric.py  | 17 +++++++++++++++--
 tools/perf/util/expr.c           | 18 ++++++++++++++++++
 tools/perf/util/expr.h           |  1 +
 tools/perf/util/expr.l           |  1 +
 tools/perf/util/expr.y           |  8 +++++++-
 tools/perf/util/pmu.c            | 17 +++++++++++++++++
 tools/perf/util/pmu.h            |  1 +
 8 files changed, 61 insertions(+), 20 deletions(-)
tools/perf/arch/arm64/util/pmu.c +1 -17
index 512a8f13c4de..615084eb88d8 100644
@@ -2,28 +2,12 @@
 
 #include <internal/cpumap.h>
 #include "../../../util/cpumap.h"
+#include "../../../util/header.h"
 #include "../../../util/pmu.h"
 #include "../../../util/pmus.h"
 #include <api/fs/fs.h>
 #include <math.h>
 
-static struct perf_pmu *pmu__find_core_pmu(void)
-{
-       struct perf_pmu *pmu = NULL;
-
-       while ((pmu = perf_pmus__scan_core(pmu))) {
-               /*
-                * The cpumap should cover all CPUs. Otherwise, some CPUs may
-                * not support some events or have different event IDs.
-                */
-               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)
-                       return NULL;
-
-               return pmu;
-       }
-       return NULL;
-}
-
 const struct pmu_metrics_table *pmu_metrics_table__find(void)
 {
        struct perf_pmu *pmu = pmu__find_core_pmu();
tools/perf/pmu-events/metric.py +15 -2
index 85a3545f5b6a..0e9ec65d92ae 100644
@@ -413,6 +413,10 @@ def has_event(event: Event) -> Function:
   # pylint: disable=invalid-name
   return Function('has_event', event)
 
+def strcmp_cpuid_str(event: str) -> Function:
+  # pylint: disable=redefined-builtin
+  # pylint: disable=invalid-name
+  return Function('strcmp_cpuid_str', event)
 
 class Metric:
   """An individual metric that will specifiable on the perf command line."""
@@ -541,14 +545,23 @@ def ParsePerfJson(orig: str) -> Expression:
   """
   # pylint: disable=eval-used
   py = orig.strip()
+  # First try to convert everything that looks like a string (event name) into Event(r"EVENT_NAME").
+  # This isn't very selective so is followed up by converting some unwanted conversions back again
   py = re.sub(r'([a-zA-Z][^-+/\* \\\(\),]*(?:\\.[^-+/\* \\\(\),]*)*)',
               r'Event(r"\1")', py)
+  # If it started with a # it should have been a literal, rather than an event name
   py = re.sub(r'#Event\(r"([^"]*)"\)', r'Literal("#\1")', py)
+  # Convert accidentally converted hex constants ("0Event(r"xDEADBEEF)"") back to a constant,
+  # but keep it wrapped in Event(), otherwise Python drops the 0x prefix and it gets interpreted as
+  # a double by the Bison parser
+  py = re.sub(r'0Event\(r"[xX]([0-9a-fA-F]*)"\)', r'Event("0x\1")', py)
+  # Convert accidentally converted scientific notation constants back
   py = re.sub(r'([0-9]+)Event\(r"(e[0-9]+)"\)', r'\1\2', py)
-  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event']
+  # Convert all the known keywords back from events to just the keyword
+  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event', 'strcmp_cpuid_str',
+              'cpuid_not_more_than']
   for kw in keywords:
     py = re.sub(rf'Event\(r"{kw}"\)', kw, py)
-
   try:
     parsed = ast.parse(py, mode='eval')
   except SyntaxError as e:
tools/perf/util/expr.c +18 -0
index 7410a165f68b..0985a3cbc6f9 100644
@@ -13,6 +13,8 @@
 #include <util/expr-bison.h>
 #include <util/expr-flex.h>
 #include "util/hashmap.h"
+#include "util/header.h"
+#include "util/pmu.h"
 #include "smt.h"
 #include "tsc.h"
 #include <api/fs/fs.h>
@@ -495,3 +497,19 @@ double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const
        evlist__delete(tmp);
        return ret;
 }
+
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx __maybe_unused,
+                      bool compute_ids __maybe_unused, const char *test_id)
+{
+       double ret;
+       struct perf_pmu *pmu = pmu__find_core_pmu();
+       char *cpuid = perf_pmu__getcpuid(pmu);
+
+       if (!cpuid)
+               return NAN;
+
+       ret = !strcmp_cpuid_str(test_id, cpuid);
+
+       free(cpuid);
+       return ret;
+}
tools/perf/util/expr.h +1 -0
index 3c1e49b3e35d..c0cec29ddc29 100644
@@ -55,5 +55,6 @@ double expr_id_data__value(const struct expr_id_data *data);
 double expr_id_data__source_count(const struct expr_id_data *data);
 double expr__get_literal(const char *literal, const struct expr_scanner_ctx *ctx);
 double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
 
 #endif
tools/perf/util/expr.l +0 -0
index dbb117414710..0feef0726c48 100644

tools/perf/util/expr.l if { return IF; } · line 117

Quoted 4 lines
tools/perf/util/expr.l +1 -0
@@ -114,6 +114,7 @@ if          { return IF; }
 else           { return ELSE; }
 source_count   { return SOURCE_COUNT; }
 has_event      { return HAS_EVENT; }
+strcmp_cpuid_str       { return STRCMP_CPUID_STR; }

For the sake of coverage, it'd be nice to have a test case like:
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/expr.c?h=tmp.perf-tools-next#n257

Quoted 34 lines
tools/perf/util/expr.l +0 -0
 {literal}      { return literal(yyscanner, sctx); }
 {number}       { return value(yyscanner); }
 {symbol}       { return str(yyscanner, ID, sctx->runtime); }
tools/perf/util/expr.y +7 -1
index 65d54a6f29ad..6c93b358cc2d 100644
@@ -39,7 +39,7 @@ int expr_lex(YYSTYPE * yylval_param , void *yyscanner);
        } ids;
 }
 
-%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT EXPR_ERROR
+%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT STRCMP_CPUID_STR EXPR_ERROR
 %left MIN MAX IF
 %left '|'
 %left '^'
@@ -207,6 +207,12 @@ expr: NUMBER
        $$.ids = NULL;
        free($3);
 }
+| STRCMP_CPUID_STR '(' ID ')'
+{
+       $$.val = expr__strcmp_cpuid_str(ctx, compute_ids, $3);
+       $$.ids = NULL;
+       free($3);
+}
 | expr '|' expr
 {
        if (is_const($1.val) && is_const($3.val)) {
tools/perf/util/pmu.c +0 -0
index d5406effc169..a7f05e4dda97 100644

tools/perf/util/pmu.c void perf_pmu__delete(struct perf_pmu *pmu) · line 1794

Quoted 5 lines
tools/perf/util/pmu.c +2 -0
@@ -1790,3 +1790,20 @@ void perf_pmu__delete(struct perf_pmu *pmu)
        zfree(&pmu->alias_name);
        free(pmu);
 }
+
+struct perf_pmu *pmu__find_core_pmu(void)

I think as this is scanning all PMUs it more logically belongs in pmus.c than pmu.c.

tools/perf/util/pmu.c void perf_pmu__delete(struct perf_pmu *pmu) · line 1803

Quoted 9 lines
tools/perf/util/pmu.c +9 -0
+{
+       struct perf_pmu *pmu = NULL;
+
+       while ((pmu = perf_pmus__scan_core(pmu))) {
+               /*
+                * The cpumap should cover all CPUs. Otherwise, some CPUs may
+                * not support some events or have different event IDs.
+                */
+               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)

The RC_CHK_ACCESS can be avoided using perf_cpu_map__nr.

Not all CPUs need to be online. I think the number for the PMU is bound by the number online while cpu__max_cpu gives the maximum present. It is common to workaround SMT vulnerabilities by making cores offline in which case this would never be true. I think the following is nearly equivalent and simpler:

perf_pmus__num_core_pmus > 1 ? NULL : perf_pmus__scan_core(NULL);

Thanks,
Ian

Quoted 18 lines
tools/perf/util/pmu.c +6 -0
+                       return NULL;
+
+               return pmu;
+       }
+       return NULL;
+}
tools/perf/util/pmu.h +1 -1
index 6b414cecbad2..7ff925224165 100644
@@ -289,5 +289,6 @@ int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename,
 struct perf_pmu *perf_pmu__lookup(struct list_head *pmus, int dirfd, const char *lookup_name);
 struct perf_pmu *perf_pmu__create_placeholder_core_pmu(struct list_head *core_pmus);
 void perf_pmu__delete(struct perf_pmu *pmu);
+struct perf_pmu *pmu__find_core_pmu(void);
 
 #endif /* __PMU_H */
--
Quoted 2 lines

2.34.1

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#11 lore raw

Re: [PATCH v6 4/6] perf jevents: Add a new expression builtin strcmp_cpuid_str()

Quoted 212 lines · Ian Rogers
Quoted 148 lines · James Clark

This will allow writing formulas that are conditional on a specific CPU type or CPU version. It calls through to the existing strcmp_cpuid_str() function in Perf which has a default weak version, and an arch specific version for x86 and arm64.

The function takes an 'ID' type value, which is a string. But in this case Arm CPU IDs are hex numbers prefixed with '0x'. metric.py assumes strings are only used by event names, and that they can't start with a number ('0'), so an additional change has to be made to the regex to convert hex numbers back to 'ID' types.

---
 tools/perf/arch/arm64/util/pmu.c | 18 +-----------------
 tools/perf/pmu-events/metric.py  | 17 +++++++++++++++--
 tools/perf/util/expr.c           | 18 ++++++++++++++++++
 tools/perf/util/expr.h           |  1 +
 tools/perf/util/expr.l           |  1 +
 tools/perf/util/expr.y           |  8 +++++++-
 tools/perf/util/pmu.c            | 17 +++++++++++++++++
 tools/perf/util/pmu.h            |  1 +
 8 files changed, 61 insertions(+), 20 deletions(-)
tools/perf/arch/arm64/util/pmu.c +1 -17
index 512a8f13c4de..615084eb88d8 100644
@@ -2,28 +2,12 @@
 
 #include <internal/cpumap.h>
 #include "../../../util/cpumap.h"
+#include "../../../util/header.h"
 #include "../../../util/pmu.h"
 #include "../../../util/pmus.h"
 #include <api/fs/fs.h>
 #include <math.h>
 
-static struct perf_pmu *pmu__find_core_pmu(void)
-{
-       struct perf_pmu *pmu = NULL;
-
-       while ((pmu = perf_pmus__scan_core(pmu))) {
-               /*
-                * The cpumap should cover all CPUs. Otherwise, some CPUs may
-                * not support some events or have different event IDs.
-                */
-               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)
-                       return NULL;
-
-               return pmu;
-       }
-       return NULL;
-}
-
 const struct pmu_metrics_table *pmu_metrics_table__find(void)
 {
        struct perf_pmu *pmu = pmu__find_core_pmu();
tools/perf/pmu-events/metric.py +15 -2
index 85a3545f5b6a..0e9ec65d92ae 100644
@@ -413,6 +413,10 @@ def has_event(event: Event) -> Function:
   # pylint: disable=invalid-name
   return Function('has_event', event)
 
+def strcmp_cpuid_str(event: str) -> Function:
+  # pylint: disable=redefined-builtin
+  # pylint: disable=invalid-name
+  return Function('strcmp_cpuid_str', event)
 
 class Metric:
   """An individual metric that will specifiable on the perf command line."""
@@ -541,14 +545,23 @@ def ParsePerfJson(orig: str) -> Expression:
   """
   # pylint: disable=eval-used
   py = orig.strip()
+  # First try to convert everything that looks like a string (event name) into Event(r"EVENT_NAME").
+  # This isn't very selective so is followed up by converting some unwanted conversions back again
   py = re.sub(r'([a-zA-Z][^-+/\* \\\(\),]*(?:\\.[^-+/\* \\\(\),]*)*)',
               r'Event(r"\1")', py)
+  # If it started with a # it should have been a literal, rather than an event name
   py = re.sub(r'#Event\(r"([^"]*)"\)', r'Literal("#\1")', py)
+  # Convert accidentally converted hex constants ("0Event(r"xDEADBEEF)"") back to a constant,
+  # but keep it wrapped in Event(), otherwise Python drops the 0x prefix and it gets interpreted as
+  # a double by the Bison parser
+  py = re.sub(r'0Event\(r"[xX]([0-9a-fA-F]*)"\)', r'Event("0x\1")', py)
+  # Convert accidentally converted scientific notation constants back
   py = re.sub(r'([0-9]+)Event\(r"(e[0-9]+)"\)', r'\1\2', py)
-  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event']
+  # Convert all the known keywords back from events to just the keyword
+  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event', 'strcmp_cpuid_str',
+              'cpuid_not_more_than']
   for kw in keywords:
     py = re.sub(rf'Event\(r"{kw}"\)', kw, py)
-
   try:
     parsed = ast.parse(py, mode='eval')
   except SyntaxError as e:
tools/perf/util/expr.c +18 -0
index 7410a165f68b..0985a3cbc6f9 100644
@@ -13,6 +13,8 @@
 #include <util/expr-bison.h>
 #include <util/expr-flex.h>
 #include "util/hashmap.h"
+#include "util/header.h"
+#include "util/pmu.h"
 #include "smt.h"
 #include "tsc.h"
 #include <api/fs/fs.h>
@@ -495,3 +497,19 @@ double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const
        evlist__delete(tmp);
        return ret;
 }
+
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx __maybe_unused,
+                      bool compute_ids __maybe_unused, const char *test_id)
+{
+       double ret;
+       struct perf_pmu *pmu = pmu__find_core_pmu();
+       char *cpuid = perf_pmu__getcpuid(pmu);
+
+       if (!cpuid)
+               return NAN;
+
+       ret = !strcmp_cpuid_str(test_id, cpuid);
+
+       free(cpuid);
+       return ret;
+}
tools/perf/util/expr.h +1 -0
index 3c1e49b3e35d..c0cec29ddc29 100644
@@ -55,5 +55,6 @@ double expr_id_data__value(const struct expr_id_data *data);
 double expr_id_data__source_count(const struct expr_id_data *data);
 double expr__get_literal(const char *literal, const struct expr_scanner_ctx *ctx);
 double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
 
 #endif
tools/perf/util/expr.l +0 -0
index dbb117414710..0feef0726c48 100644

tools/perf/util/expr.l if { return IF; } · line 117

Quoted 4 lines
tools/perf/util/expr.l +1 -0
@@ -114,6 +114,7 @@ if          { return IF; }
 else           { return ELSE; }
 source_count   { return SOURCE_COUNT; }
 has_event      { return HAS_EVENT; }
+strcmp_cpuid_str       { return STRCMP_CPUID_STR; }

For the sake of coverage, it'd be nice to have a test case like:
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/expr.c?h=tmp.perf-tools-next#n257

Quoted 34 lines
tools/perf/util/expr.l +0 -0
 {literal}      { return literal(yyscanner, sctx); }
 {number}       { return value(yyscanner); }
 {symbol}       { return str(yyscanner, ID, sctx->runtime); }
tools/perf/util/expr.y +7 -1
index 65d54a6f29ad..6c93b358cc2d 100644
@@ -39,7 +39,7 @@ int expr_lex(YYSTYPE * yylval_param , void *yyscanner);
        } ids;
 }
 
-%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT EXPR_ERROR
+%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT STRCMP_CPUID_STR EXPR_ERROR
 %left MIN MAX IF
 %left '|'
 %left '^'
@@ -207,6 +207,12 @@ expr: NUMBER
        $$.ids = NULL;
        free($3);
 }
+| STRCMP_CPUID_STR '(' ID ')'
+{
+       $$.val = expr__strcmp_cpuid_str(ctx, compute_ids, $3);
+       $$.ids = NULL;
+       free($3);
+}
 | expr '|' expr
 {
        if (is_const($1.val) && is_const($3.val)) {
tools/perf/util/pmu.c +0 -0
index d5406effc169..a7f05e4dda97 100644

tools/perf/util/pmu.c void perf_pmu__delete(struct perf_pmu *pmu) · line 1794

Quoted 5 lines
tools/perf/util/pmu.c +2 -0
@@ -1790,3 +1790,20 @@ void perf_pmu__delete(struct perf_pmu *pmu)
        zfree(&pmu->alias_name);
        free(pmu);
 }
+
+struct perf_pmu *pmu__find_core_pmu(void)

I think as this is scanning all PMUs it more logically belongs in pmus.c than pmu.c.

tools/perf/util/pmu.c void perf_pmu__delete(struct perf_pmu *pmu) · line 1803

Quoted 9 lines
tools/perf/util/pmu.c +9 -0
+{
+       struct perf_pmu *pmu = NULL;
+
+       while ((pmu = perf_pmus__scan_core(pmu))) {
+               /*
+                * The cpumap should cover all CPUs. Otherwise, some CPUs may
+                * not support some events or have different event IDs.
+                */
+               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)

The RC_CHK_ACCESS can be avoided using perf_cpu_map__nr.

Quoted 8 lines

Not all CPUs need to be online. I think the number for the PMU is bound by the number online while cpu__max_cpu gives the maximum present. It is common to workaround SMT vulnerabilities by making cores offline in which case this would never be true. I think the following is nearly equivalent and simpler:

perf_pmus__num_core_pmus > 1 ? NULL : perf_pmus__scan_core(NULL);

I'm not sure about this one, I'm hesitant to change it because I only moved it out of the arm file and into the generic one and it seems like it could cause subtle issues.

From the comment above the if statement, it seems like it intentionally refuses to return a PMU on heterogeneous systems, and the original commit message (e126bef55f1d) from John lines up with that too. I can't see exactly why it should be like that though.

Although you could say that now I've pulled some Arm specific quirk into the generic code and we should re-think it. To me I don't see why it doesn't just unconditionally return the first core PMU it finds on any architecture, but I'd like to understand the original intention first.

Apart from this one, I've done the other changes and will send patches.

Quoted 19 lines

Thanks,
Ian

Quoted 16 lines
tools/perf/util/pmu.c +6 -0
+                       return NULL;
+
+               return pmu;
+       }
+       return NULL;
+}
tools/perf/util/pmu.h +1 -0
index 6b414cecbad2..7ff925224165 100644
@@ -289,5 +289,6 @@ int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename,
 struct perf_pmu *perf_pmu__lookup(struct list_head *pmus, int dirfd, const char *lookup_name);
 struct perf_pmu *perf_pmu__create_placeholder_core_pmu(struct list_head *core_pmus);
 void perf_pmu__delete(struct perf_pmu *pmu);
+struct perf_pmu *pmu__find_core_pmu(void);
 
Quoted 4 lines
Quoted 4 lines
tools/perf/util/pmu.h +0 -1
 #endif /* __PMU_H */
--

2.34.1

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

reviewer
#12 lore raw

Re: [PATCH v6 4/6] perf jevents: Add a new expression builtin strcmp_cpuid_str()

Quoted 234 lines · James Clark
Quoted 212 lines · Ian Rogers
Quoted 148 lines · James Clark

This will allow writing formulas that are conditional on a specific CPU type or CPU version. It calls through to the existing strcmp_cpuid_str() function in Perf which has a default weak version, and an arch specific version for x86 and arm64.

The function takes an 'ID' type value, which is a string. But in this case Arm CPU IDs are hex numbers prefixed with '0x'. metric.py assumes strings are only used by event names, and that they can't start with a number ('0'), so an additional change has to be made to the regex to convert hex numbers back to 'ID' types.

---
 tools/perf/arch/arm64/util/pmu.c | 18 +-----------------
 tools/perf/pmu-events/metric.py  | 17 +++++++++++++++--
 tools/perf/util/expr.c           | 18 ++++++++++++++++++
 tools/perf/util/expr.h           |  1 +
 tools/perf/util/expr.l           |  1 +
 tools/perf/util/expr.y           |  8 +++++++-
 tools/perf/util/pmu.c            | 17 +++++++++++++++++
 tools/perf/util/pmu.h            |  1 +
 8 files changed, 61 insertions(+), 20 deletions(-)
tools/perf/arch/arm64/util/pmu.c +1 -17
index 512a8f13c4de..615084eb88d8 100644
@@ -2,28 +2,12 @@
 
 #include <internal/cpumap.h>
 #include "../../../util/cpumap.h"
+#include "../../../util/header.h"
 #include "../../../util/pmu.h"
 #include "../../../util/pmus.h"
 #include <api/fs/fs.h>
 #include <math.h>
 
-static struct perf_pmu *pmu__find_core_pmu(void)
-{
-       struct perf_pmu *pmu = NULL;
-
-       while ((pmu = perf_pmus__scan_core(pmu))) {
-               /*
-                * The cpumap should cover all CPUs. Otherwise, some CPUs may
-                * not support some events or have different event IDs.
-                */
-               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)
-                       return NULL;
-
-               return pmu;
-       }
-       return NULL;
-}
-
 const struct pmu_metrics_table *pmu_metrics_table__find(void)
 {
        struct perf_pmu *pmu = pmu__find_core_pmu();
tools/perf/pmu-events/metric.py +15 -2
index 85a3545f5b6a..0e9ec65d92ae 100644
@@ -413,6 +413,10 @@ def has_event(event: Event) -> Function:
   # pylint: disable=invalid-name
   return Function('has_event', event)
 
+def strcmp_cpuid_str(event: str) -> Function:
+  # pylint: disable=redefined-builtin
+  # pylint: disable=invalid-name
+  return Function('strcmp_cpuid_str', event)
 
 class Metric:
   """An individual metric that will specifiable on the perf command line."""
@@ -541,14 +545,23 @@ def ParsePerfJson(orig: str) -> Expression:
   """
   # pylint: disable=eval-used
   py = orig.strip()
+  # First try to convert everything that looks like a string (event name) into Event(r"EVENT_NAME").
+  # This isn't very selective so is followed up by converting some unwanted conversions back again
   py = re.sub(r'([a-zA-Z][^-+/\* \\\(\),]*(?:\\.[^-+/\* \\\(\),]*)*)',
               r'Event(r"\1")', py)
+  # If it started with a # it should have been a literal, rather than an event name
   py = re.sub(r'#Event\(r"([^"]*)"\)', r'Literal("#\1")', py)
+  # Convert accidentally converted hex constants ("0Event(r"xDEADBEEF)"") back to a constant,
+  # but keep it wrapped in Event(), otherwise Python drops the 0x prefix and it gets interpreted as
+  # a double by the Bison parser
+  py = re.sub(r'0Event\(r"[xX]([0-9a-fA-F]*)"\)', r'Event("0x\1")', py)
+  # Convert accidentally converted scientific notation constants back
   py = re.sub(r'([0-9]+)Event\(r"(e[0-9]+)"\)', r'\1\2', py)
-  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event']
+  # Convert all the known keywords back from events to just the keyword
+  keywords = ['if', 'else', 'min', 'max', 'd_ratio', 'source_count', 'has_event', 'strcmp_cpuid_str',
+              'cpuid_not_more_than']
   for kw in keywords:
     py = re.sub(rf'Event\(r"{kw}"\)', kw, py)
-
   try:
     parsed = ast.parse(py, mode='eval')
   except SyntaxError as e:
tools/perf/util/expr.c +18 -0
index 7410a165f68b..0985a3cbc6f9 100644
@@ -13,6 +13,8 @@
 #include <util/expr-bison.h>
 #include <util/expr-flex.h>
 #include "util/hashmap.h"
+#include "util/header.h"
+#include "util/pmu.h"
 #include "smt.h"
 #include "tsc.h"
 #include <api/fs/fs.h>
@@ -495,3 +497,19 @@ double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const
        evlist__delete(tmp);
        return ret;
 }
+
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx __maybe_unused,
+                      bool compute_ids __maybe_unused, const char *test_id)
+{
+       double ret;
+       struct perf_pmu *pmu = pmu__find_core_pmu();
+       char *cpuid = perf_pmu__getcpuid(pmu);
+
+       if (!cpuid)
+               return NAN;
+
+       ret = !strcmp_cpuid_str(test_id, cpuid);
+
+       free(cpuid);
+       return ret;
+}
tools/perf/util/expr.h +1 -0
index 3c1e49b3e35d..c0cec29ddc29 100644
@@ -55,5 +55,6 @@ double expr_id_data__value(const struct expr_id_data *data);
 double expr_id_data__source_count(const struct expr_id_data *data);
 double expr__get_literal(const char *literal, const struct expr_scanner_ctx *ctx);
 double expr__has_event(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
+double expr__strcmp_cpuid_str(const struct expr_parse_ctx *ctx, bool compute_ids, const char *id);
 
 #endif
tools/perf/util/expr.l +0 -0
index dbb117414710..0feef0726c48 100644

tools/perf/util/expr.l if { return IF; } · line 117

Quoted 4 lines
tools/perf/util/expr.l +1 -0
@@ -114,6 +114,7 @@ if          { return IF; }
 else           { return ELSE; }
 source_count   { return SOURCE_COUNT; }
 has_event      { return HAS_EVENT; }
+strcmp_cpuid_str       { return STRCMP_CPUID_STR; }

For the sake of coverage, it'd be nice to have a test case like:
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/expr.c?h=tmp.perf-tools-next#n257

Quoted 34 lines
tools/perf/util/expr.l +0 -0
 {literal}      { return literal(yyscanner, sctx); }
 {number}       { return value(yyscanner); }
 {symbol}       { return str(yyscanner, ID, sctx->runtime); }
tools/perf/util/expr.y +7 -1
index 65d54a6f29ad..6c93b358cc2d 100644
@@ -39,7 +39,7 @@ int expr_lex(YYSTYPE * yylval_param , void *yyscanner);
        } ids;
 }
 
-%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT EXPR_ERROR
+%token ID NUMBER MIN MAX IF ELSE LITERAL D_RATIO SOURCE_COUNT HAS_EVENT STRCMP_CPUID_STR EXPR_ERROR
 %left MIN MAX IF
 %left '|'
 %left '^'
@@ -207,6 +207,12 @@ expr: NUMBER
        $$.ids = NULL;
        free($3);
 }
+| STRCMP_CPUID_STR '(' ID ')'
+{
+       $$.val = expr__strcmp_cpuid_str(ctx, compute_ids, $3);
+       $$.ids = NULL;
+       free($3);
+}
 | expr '|' expr
 {
        if (is_const($1.val) && is_const($3.val)) {
tools/perf/util/pmu.c +0 -0
index d5406effc169..a7f05e4dda97 100644

tools/perf/util/pmu.c void perf_pmu__delete(struct perf_pmu *pmu) · line 1794

Quoted 5 lines
tools/perf/util/pmu.c +2 -0
@@ -1790,3 +1790,20 @@ void perf_pmu__delete(struct perf_pmu *pmu)
        zfree(&pmu->alias_name);
        free(pmu);
 }
+
+struct perf_pmu *pmu__find_core_pmu(void)

I think as this is scanning all PMUs it more logically belongs in pmus.c than pmu.c.

tools/perf/util/pmu.c void perf_pmu__delete(struct perf_pmu *pmu) · line 1803

Quoted 9 lines
tools/perf/util/pmu.c +9 -0
+{
+       struct perf_pmu *pmu = NULL;
+
+       while ((pmu = perf_pmus__scan_core(pmu))) {
+               /*
+                * The cpumap should cover all CPUs. Otherwise, some CPUs may
+                * not support some events or have different event IDs.
+                */
+               if (RC_CHK_ACCESS(pmu->cpus)->nr != cpu__max_cpu().cpu)

The RC_CHK_ACCESS can be avoided using perf_cpu_map__nr.

Quoted 8 lines

Not all CPUs need to be online. I think the number for the PMU is bound by the number online while cpu__max_cpu gives the maximum present. It is common to workaround SMT vulnerabilities by making cores offline in which case this would never be true. I think the following is nearly equivalent and simpler:

perf_pmus__num_core_pmus > 1 ? NULL : perf_pmus__scan_core(NULL);

I'm not sure about this one, I'm hesitant to change it because I only moved it out of the arm file and into the generic one and it seems like it could cause subtle issues.

From the comment above the if statement, it seems like it intentionally refuses to return a PMU on heterogeneous systems, and the original commit message (e126bef55f1d) from John lines up with that too. I can't see exactly why it should be like that though.

Quoted 6 lines

Although you could say that now I've pulled some Arm specific quirk into the generic code and we should re-think it. To me I don't see why it doesn't just unconditionally return the first core PMU it finds on any architecture, but I'd like to understand the original intention first.

Apart from this one, I've done the other changes and will send patches.

Even on a homogeneous ARM system this will always return NULL/NAN if a core is taken offline. We didn't used to count core PMUs, so the approach I proposed wouldn't have been possible previously.

Thanks,
Ian

Quoted 19 lines
Quoted 19 lines

Thanks,
Ian

Quoted 16 lines
tools/perf/util/pmu.c +6 -0
+                       return NULL;
+
+               return pmu;
+       }
+       return NULL;
+}
tools/perf/util/pmu.h +1 -0
index 6b414cecbad2..7ff925224165 100644
@@ -289,5 +289,6 @@ int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename,
 struct perf_pmu *perf_pmu__lookup(struct list_head *pmus, int dirfd, const char *lookup_name);
 struct perf_pmu *perf_pmu__create_placeholder_core_pmu(struct list_head *core_pmus);
 void perf_pmu__delete(struct perf_pmu *pmu);
+struct perf_pmu *pmu__find_core_pmu(void);
 
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
tools/perf/util/pmu.h +0 -1
 #endif /* __PMU_H */
--

2.34.1

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#13 lore raw

PATCH v6 5/6 perf vendor events arm64: Update stall_slot workaround for N2 r0p3

In reply to James Clark

N2 r0p3 doesn't require the workaround [1], so gating on (#slots - 5) no longer works because all N2s have 5 slots. Use the new expression builtin that allows calling strcmp_cpuid_str() and comparing CPUIDs in metric formulas.

In this case, the commented formula looks like this:

strcmp_cpuid_str(0x410fd493) # greater than or equal to N2 r0p3
| strcmp_cpuid_str(0x410fd490) ^ 1 # OR NOT any version of N2

[1]: https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-n2-r0p3.json

---
 .../pmu-events/arch/arm64/arm/neoverse-n2-v2/metrics.json | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/metrics.json +4 -4
index 8ad15b726dca..9757cc178bdf 100644
@@ -1,15 +1,15 @@
 [
     {
         "ArchStdEvent": "FRONTEND_BOUND",
-        "MetricExpr": "((stall_slot_frontend) if (#slots - 5) else (stall_slot_frontend - cpu_cycles)) / (#slots * cpu_cycles)"
+        "MetricExpr": "((stall_slot_frontend) if (strcmp_cpuid_str(0x410fd493) | strcmp_cpuid_str(0x410fd490) ^ 1) else (stall_slot_frontend - cpu_cycles)) / (#slots * cpu_cycles)"
     },
     {
         "ArchStdEvent": "BAD_SPECULATION",
-        "MetricExpr": "(1 - op_retired / op_spec) * (1 - (stall_slot if (#slots - 5) else (stall_slot - cpu_cycles)) / (#slots * cpu_cycles))"
+        "MetricExpr": "(1 - op_retired / op_spec) * (1 - (stall_slot if (strcmp_cpuid_str(0x410fd493) | strcmp_cpuid_str(0x410fd490) ^ 1) else (stall_slot - cpu_cycles)) / (#slots * cpu_cycles))"
     },
     {
         "ArchStdEvent": "RETIRING",
-        "MetricExpr": "(op_retired / op_spec) * (1 - (stall_slot if (#slots - 5) else (stall_slot - cpu_cycles)) / (#slots * cpu_cycles))"
+        "MetricExpr": "(op_retired / op_spec) * (1 - (stall_slot if (strcmp_cpuid_str(0x410fd493) | strcmp_cpuid_str(0x410fd490) ^ 1) else (stall_slot - cpu_cycles)) / (#slots * cpu_cycles))"
     },
     {
         "ArchStdEvent": "BACKEND_BOUND"
@@ -201,7 +201,7 @@
         "ScaleUnit": "100%"
     },
     {
-        "MetricExpr": "OP_RETIRED / OP_SPEC * (1 - (STALL_SLOT if (#slots - 5) else (STALL_SLOT - CPU_CYCLES)) / (#slots * CPU_CYCLES))",
+        "MetricExpr": "OP_RETIRED / OP_SPEC * (1 - (STALL_SLOT if (strcmp_cpuid_str(0x410fd493) | strcmp_cpuid_str(0x410fd490) ^ 1) else (STALL_SLOT - CPU_CYCLES)) / (#slots * CPU_CYCLES))",
         "BriefDescription": "The truly effective ratio of micro-operations executed by the CPU, which means that misprediction and stall are not included",
         "MetricGroup": "PEutilization",
         "MetricName": "cpu_utilization",
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
reviewer Reviewed-by on v6 5/6
#14 lore raw

Re: [PATCH v6 5/6] perf vendor events arm64: Update stall_slot workaround for N2 r0p3

Quoted 5 lines · James Clark

N2 r0p3 doesn't require the workaround [1], so gating on (#slots - 5) no longer works because all N2s have 5 slots. Use the new expression builtin that allows calling strcmp_cpuid_str() and comparing CPUIDs in metric formulas.

Quoted 7 lines

In this case, the commented formula looks like this:

   strcmp_cpuid_str(0x410fd493)        # greater than or equal to N2 r0p3
   | strcmp_cpuid_str(0x410fd490) ^ 1  # OR NOT any version of N2
[1]:https://urldefense.com/v3/__https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-n2-r0p3.json__;!!ACWV5N9M2RV99hQ!JvpXOAN-7zsKcA750zS-vMlcJKvvIZ_H9zg6jitWJRMitlFGDrx8bXoMrhdK5ubhO8CtXObRFx7bDYM2R5Feqg$  Signed-off-by: James Clark<[email protected]>
---

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#15 lore raw

PATCH v6 6/6 perf vendor events arm64: Update N2 and V2 metrics and events using Arm telemetry repo

In reply to James Clark

Apart from some slight naming and grouping differences, the new metrics are functionally the same as the existing ones. Any missing metrics were manually appended to the end of the auto generated file.

For the events, the new data includes descriptions that may have product specific details and new groupings that will be consistent with other products.

After generating the metrics from the telemetry repo [1], the following manual steps were performed:

* Change the topdown expressions to compare on CPUID and use #slots so that the same data can be shared between N2 and V2. Apart from these modifications, the expressions now match more closely with the Arm telemetry data which will hopefully make future updates easier.

* Append some metrics from the old N2/V2 data that aren't present in the telemetry data. These will possibly be added to the telemetry-solution repo at a later time:

l3d_cache_mpki, l3d_cache_miss_rate, branch_pki, ipc_rate, spec_ipc, retired_rate, wasted_rate, branch_immed_spec_rate, branch_return_spec_rate, branch_indirect_spec_rate

[1]: https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-n2.json

---
 .../arch/arm64/arm/neoverse-n2-v2/branch.json |   8 -
 .../arch/arm64/arm/neoverse-n2-v2/bus.json    |  18 +-
 .../arch/arm64/arm/neoverse-n2-v2/cache.json  | 155 --------
 .../arm64/arm/neoverse-n2-v2/exception.json   |  45 ++-
 .../arm/neoverse-n2-v2/fp_operation.json      |  22 ++
 .../arm64/arm/neoverse-n2-v2/general.json     |  10 +
 .../arm64/arm/neoverse-n2-v2/instruction.json | 143 -------
 .../arm64/arm/neoverse-n2-v2/l1d_cache.json   |  54 +++
 .../arm64/arm/neoverse-n2-v2/l1i_cache.json   |  14 +
 .../arm64/arm/neoverse-n2-v2/l2_cache.json    |  50 +++
 .../arm64/arm/neoverse-n2-v2/l3_cache.json    |  22 ++
 .../arm64/arm/neoverse-n2-v2/ll_cache.json    |  10 +
 .../arch/arm64/arm/neoverse-n2-v2/memory.json |  39 +-
 .../arm64/arm/neoverse-n2-v2/metrics.json     | 365 ++++++++++--------
 .../arm64/arm/neoverse-n2-v2/pipeline.json    |  23 --
 .../arm64/arm/neoverse-n2-v2/retired.json     |  30 ++
 .../arch/arm64/arm/neoverse-n2-v2/spe.json    |  12 +-
 .../arm/neoverse-n2-v2/spec_operation.json    | 110 ++++++
 .../arch/arm64/arm/neoverse-n2-v2/stall.json  |  30 ++
 .../arch/arm64/arm/neoverse-n2-v2/sve.json    |  50 +++
 .../arch/arm64/arm/neoverse-n2-v2/tlb.json    |  66 ++++
 .../arch/arm64/arm/neoverse-n2-v2/trace.json  |  27 +-
 22 files changed, 752 insertions(+), 551 deletions(-)
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/branch.json
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/fp_operation.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/general.json
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/instruction.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l1d_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l1i_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l2_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l3_cache.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/ll_cache.json
 delete mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/pipeline.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/retired.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/spec_operation.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/stall.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/sve.json
 create mode 100644 tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/tlb.json
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/branch.json deleted +0 -8
deleted file mode 100644
index 79f2016c53b0..000000000000
@@ -1,8 +0,0 @@
-[
-    {
-        "ArchStdEvent": "BR_MIS_PRED"
-    },
-    {
-        "ArchStdEvent": "BR_PRED"
-    }
-]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/bus.json +8 -10
index 579c1c993d17..2e11a8c4a484 100644
@@ -1,20 +1,18 @@
 [
     {
-        "ArchStdEvent": "CPU_CYCLES"
+        "ArchStdEvent": "BUS_ACCESS",
+        "PublicDescription": "Counts memory transactions issued by the CPU to the external bus, including snoop requests and snoop responses. Each beat of data is counted individually."
     },
     {
-        "ArchStdEvent": "BUS_ACCESS"
+        "ArchStdEvent": "BUS_CYCLES",
+        "PublicDescription": "Counts bus cycles in the CPU. Bus cycles represent a clock cycle in which a transaction could be sent or received on the interface from the CPU to the external bus. Since that interface is driven at the same clock speed as the CPU, this event is a duplicate of CPU_CYCLES."
     },
     {
-        "ArchStdEvent": "BUS_CYCLES"
+        "ArchStdEvent": "BUS_ACCESS_RD",
+        "PublicDescription": "Counts memory read transactions seen on the external bus. Each beat of data is counted individually."
     },
     {
-        "ArchStdEvent": "BUS_ACCESS_RD"
-    },
-    {
-        "ArchStdEvent": "BUS_ACCESS_WR"
-    },
-    {
-        "ArchStdEvent": "CNT_CYCLES"
+        "ArchStdEvent": "BUS_ACCESS_WR",
+        "PublicDescription": "Counts memory write transactions seen on the external bus. Each beat of data is counted individually."
     }
 ]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/cache.json deleted +0 -155

155 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/exception.json +30 -15

77 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/fp_operation.json new file +22 -0
new file mode 100644
index 000000000000..cec3435ac766
@@ -0,0 +1,22 @@
+[
+    {
+        "ArchStdEvent": "FP_HP_SPEC",
+        "PublicDescription": "Counts speculatively executed half precision floating point operations."
+    },
+    {
+        "ArchStdEvent": "FP_SP_SPEC",
+        "PublicDescription": "Counts speculatively executed single precision floating point operations."
+    },
+    {
+        "ArchStdEvent": "FP_DP_SPEC",
+        "PublicDescription": "Counts speculatively executed double precision floating point operations."
+    },
+    {
+        "ArchStdEvent": "FP_SCALE_OPS_SPEC",
+        "PublicDescription": "Counts speculatively executed scalable single precision floating point operations."
+    },
+    {
+        "ArchStdEvent": "FP_FIXED_OPS_SPEC",
+        "PublicDescription": "Counts speculatively executed non-scalable single precision floating point operations."
+    }
+]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/general.json new file +10 -0
new file mode 100644
index 000000000000..428810f855b8
@@ -0,0 +1,10 @@
+[
+    {
+        "ArchStdEvent": "CPU_CYCLES",
+        "PublicDescription": "Counts CPU clock cycles (not timer cycles). The clock measured by this event is defined as the physical clock driving the CPU logic."
+    },
+    {
+        "ArchStdEvent": "CNT_CYCLES",
+        "PublicDescription": "Counts constant frequency cycles"
+    }
+]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/instruction.json deleted +0 -143

143 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l1d_cache.json new file +54 -0

54 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l1i_cache.json new file +14 -0
new file mode 100644
index 000000000000..633f1030359d
@@ -0,0 +1,14 @@
+[
+    {
+        "ArchStdEvent": "L1I_CACHE_REFILL",
+        "PublicDescription": "Counts cache line refills in the level 1 instruction cache caused by a missed instruction fetch. Instruction fetches may include accessing multiple instructions, but the single cache line allocation is counted once."
+    },
+    {
+        "ArchStdEvent": "L1I_CACHE",
+        "PublicDescription": "Counts instruction fetches which access the level 1 instruction cache. Instruction cache accesses caused by cache maintenance operations are not counted."
+    },
+    {
+        "ArchStdEvent": "L1I_CACHE_LMISS",
+        "PublicDescription": "Counts cache line refills into the level 1 instruction cache, that incurred additional latency."
+    }
+]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l2_cache.json new file +50 -0

50 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/l3_cache.json new file +22 -0
new file mode 100644
index 000000000000..45bfba532df7
@@ -0,0 +1,22 @@
+[
+    {
+        "ArchStdEvent": "L3D_CACHE_ALLOCATE",
+        "PublicDescription": "Counts level 3 cache line allocates that do not fetch data from outside the level 3 data or unified cache. For example, allocates due to streaming stores."
+    },
+    {
+        "ArchStdEvent": "L3D_CACHE_REFILL",
+        "PublicDescription": "Counts level 3 accesses that receive data from outside the L3 cache."
+    },
+    {
+        "ArchStdEvent": "L3D_CACHE",
+        "PublicDescription": "Counts level 3 cache accesses. level 3 cache is a unified cache for data and instruction accesses. Accesses are for misses in the lower level caches or translation resolutions due to accesses."
+    },
+    {
+        "ArchStdEvent": "L3D_CACHE_RD",
+        "PublicDescription": "TBD"
+    },
+    {
+        "ArchStdEvent": "L3D_CACHE_LMISS_RD",
+        "PublicDescription": "Counts any cache line refill into the level 3 cache from memory read operations that incurred additional latency."
+    }
+]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/ll_cache.json new file +10 -0
new file mode 100644
index 000000000000..bb712d57d58a
@@ -0,0 +1,10 @@
+[
+    {
+        "ArchStdEvent": "LL_CACHE_RD",
+        "PublicDescription": "Counts read transactions that were returned from outside the core cluster. This event counts when the system register CPUECTLR.EXTLLC bit is set. This event counts read transactions returned from outside the core if those transactions are either hit in the system level cache or missed in the SLC and are returned from any other external sources."
+    },
+    {
+        "ArchStdEvent": "LL_CACHE_MISS_RD",
+        "PublicDescription": "Counts read transactions that were returned from outside the core cluster but missed in the system level cache. This event counts when the system register CPUECTLR.EXTLLC bit is set. This event counts read transactions returned from outside the core if those transactions are missed in the System level Cache. The data source of the transaction is indicated by a field in the CHI transaction returning to the CPU. This event does not count reads caused by cache maintenance operations."
+    }
+]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/memory.json +22 -17
index 7b2b21ac150f..106a97f8b2e7 100644
@@ -1,41 +1,46 @@
 [
     {
-        "ArchStdEvent": "MEM_ACCESS"
+        "ArchStdEvent": "MEM_ACCESS",
+        "PublicDescription": "Counts memory accesses issued by the CPU load store unit, where those accesses are issued due to load or store operations. This event counts memory accesses no matter whether the data is received from any level of cache hierarchy or external memory. If memory accesses are broken up into smaller transactions than what were specified in the load or store instructions, then the event counts those smaller memory transactions."
     },
     {
-        "ArchStdEvent": "REMOTE_ACCESS"
+        "ArchStdEvent": "MEMORY_ERROR",
+        "PublicDescription": "Counts any detected correctable or uncorrectable physical memory errors (ECC or parity) in protected CPUs RAMs. On the core, this event counts errors in the caches (including data and tag rams). Any detected memory error (from either a speculative and abandoned access, or an architecturally executed access) is counted. Note that errors are only detected when the actual protected memory is accessed by an operation."
     },
     {
-        "ArchStdEvent": "MEM_ACCESS_RD"
+        "ArchStdEvent": "REMOTE_ACCESS",
+        "PublicDescription": "Counts accesses to another chip, which is implemented as a different CMN mesh in the system. If the CHI bus response back to the core indicates that the data source is from another chip (mesh), then the counter is updated. If no data is returned, even if the system snoops another chip/mesh, then the counter is not updated."
     },
     {
-        "ArchStdEvent": "MEM_ACCESS_WR"
+        "ArchStdEvent": "MEM_ACCESS_RD",
+        "PublicDescription": "Counts memory accesses issued by the CPU due to load operations. The event counts any memory load access, no matter whether the data is received from any level of cache hierarchy or external memory. The event also counts atomic load operations. If memory accesses are broken up by the load/store unit into smaller transactions that are issued by the bus interface, then the event counts those smaller transactions."
     },
     {
-        "ArchStdEvent": "UNALIGNED_LD_SPEC"
+        "ArchStdEvent": "MEM_ACCESS_WR",
+        "PublicDescription": "Counts memory accesses issued by the CPU due to store operations. The event counts any memory store access, no matter whether the data is located in any level of cache or external memory. The event also counts atomic load and store operations. If memory accesses are broken up by the load/store unit into smaller transactions that are issued by the bus interface, then the event counts those smaller transactions."
     },
     {
-        "ArchStdEvent": "UNALIGNED_ST_SPEC"
+        "ArchStdEvent": "LDST_ALIGN_LAT",
+        "PublicDescription": "Counts the number of memory read and write accesses in a cycle that incurred additional latency, due to the alignment of the address and the size of data being accessed, which results in store crossing a single cache line."
     },
     {
-        "ArchStdEvent": "UNALIGNED_LDST_SPEC"
+        "ArchStdEvent": "LD_ALIGN_LAT",
+        "PublicDescription": "Counts the number of memory read accesses in a cycle that incurred additional latency, due to the alignment of the address and size of data being accessed, which results in load crossing a single cache line."
     },
     {
-        "ArchStdEvent": "LDST_ALIGN_LAT"
+        "ArchStdEvent": "ST_ALIGN_LAT",
+        "PublicDescription": "Counts the number of memory write access in a cycle that incurred additional latency, due to the alignment of the address and size of data being accessed incurred additional latency."
     },
     {
-        "ArchStdEvent": "LD_ALIGN_LAT"
+        "ArchStdEvent": "MEM_ACCESS_CHECKED",
+        "PublicDescription": "Counts the number of memory read and write accesses in a cycle that are tag checked by the Memory Tagging Extension (MTE)."
     },
     {
-        "ArchStdEvent": "ST_ALIGN_LAT"
+        "ArchStdEvent": "MEM_ACCESS_CHECKED_RD",
+        "PublicDescription": "Counts the number of memory read accesses in a cycle that are tag checked by the Memory Tagging Extension (MTE)."
     },
     {
-        "ArchStdEvent": "MEM_ACCESS_CHECKED"
-    },
-    {
-        "ArchStdEvent": "MEM_ACCESS_CHECKED_RD"
-    },
-    {
-        "ArchStdEvent": "MEM_ACCESS_CHECKED_WR"
+        "ArchStdEvent": "MEM_ACCESS_CHECKED_WR",
+        "PublicDescription": "Counts the number of memory write accesses in a cycle that is tag checked by the Memory Tagging Extension (MTE)."
     }
 ]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/metrics.json +198 -167

470 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/pipeline.json deleted +0 -23
deleted file mode 100644
index f9fae15f7555..000000000000
@@ -1,23 +0,0 @@
-[
-    {
-        "ArchStdEvent": "STALL_FRONTEND"
-    },
-    {
-        "ArchStdEvent": "STALL_BACKEND"
-    },
-    {
-        "ArchStdEvent": "STALL"
-    },
-    {
-        "ArchStdEvent": "STALL_SLOT_BACKEND"
-    },
-    {
-        "ArchStdEvent": "STALL_SLOT_FRONTEND"
-    },
-    {
-        "ArchStdEvent": "STALL_SLOT"
-    },
-    {
-        "ArchStdEvent": "STALL_BACKEND_MEM"
-    }
-]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/retired.json new file +30 -0
new file mode 100644
index 000000000000..f297b049b62f
@@ -0,0 +1,30 @@
+[
+    {
+        "ArchStdEvent": "SW_INCR",
+        "PublicDescription": "Counts software writes to the PMSWINC_EL0 (software PMU increment) register. The PMSWINC_EL0 register is a manually updated counter for use by application software.\n\nThis event could be used to measure any user program event, such as accesses to a particular data structure (by writing to the PMSWINC_EL0 register each time the data structure is accessed).\n\nTo use the PMSWINC_EL0 register and event, developers must insert instructions that write to the PMSWINC_EL0 register into the source code.\n\nSince the SW_INCR event records writes to the PMSWINC_EL0 register, there is no need to do a read/increment/write sequence to the PMSWINC_EL0 register."
+    },
+    {
+        "ArchStdEvent": "INST_RETIRED",
+        "PublicDescription": "Counts instructions that have been architecturally executed."
+    },
+    {
+        "ArchStdEvent": "CID_WRITE_RETIRED",
+        "PublicDescription": "Counts architecturally executed writes to the CONTEXTIDR register, which usually contain the kernel PID and can be output with hardware trace."
+    },
+    {
+        "ArchStdEvent": "TTBR_WRITE_RETIRED",
+        "PublicDescription": "Counts architectural writes to TTBR0/1_EL1. If virtualization host extensions are enabled (by setting the HCR_EL2.E2H bit to 1), then accesses to TTBR0/1_EL1 that are redirected to TTBR0/1_EL2, or accesses to TTBR0/1_EL12, are counted. TTBRn registers are typically updated when the kernel is swapping user-space threads or applications."
+    },
+    {
+        "ArchStdEvent": "BR_RETIRED",
+        "PublicDescription": "Counts architecturally executed branches, whether the branch is taken or not. Instructions that explicitly write to the PC are also counted."
+    },
+    {
+        "ArchStdEvent": "BR_MIS_PRED_RETIRED",
+        "PublicDescription": "Counts branches counted by BR_RETIRED which were mispredicted and caused a pipeline flush."
+    },
+    {
+        "ArchStdEvent": "OP_RETIRED",
+        "PublicDescription": "Counts micro-operations that are architecturally executed. This is a count of number of micro-operations retired from the commit queue in a single cycle."
+    }
+]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/spe.json +8 -4
index 20f2165c85fe..5de8b0f3a440 100644
@@ -1,14 +1,18 @@
 [
     {
-        "ArchStdEvent": "SAMPLE_POP"
+        "ArchStdEvent": "SAMPLE_POP",
+        "PublicDescription": "Counts statistical profiling sample population, the count of all operations that could be sampled but may or may not be chosen for sampling."
     },
     {
-        "ArchStdEvent": "SAMPLE_FEED"
+        "ArchStdEvent": "SAMPLE_FEED",
+        "PublicDescription": "Counts statistical profiling samples taken for sampling."
     },
     {
-        "ArchStdEvent": "SAMPLE_FILTRATE"
+        "ArchStdEvent": "SAMPLE_FILTRATE",
+        "PublicDescription": "Counts statistical profiling samples taken which are not removed by filtering."
     },
     {
-        "ArchStdEvent": "SAMPLE_COLLISION"
+        "ArchStdEvent": "SAMPLE_COLLISION",
+        "PublicDescription": "Counts statistical profiling samples that have collided with a previous sample and so therefore not taken."
     }
 ]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/spec_operation.json new file +110 -0

110 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/stall.json new file +30 -0
new file mode 100644
index 000000000000..bbbebc805034
@@ -0,0 +1,30 @@
+[
+    {
+        "ArchStdEvent": "STALL_FRONTEND",
+        "PublicDescription": "Counts cycles when frontend could not send any micro-operations to the rename stage because of frontend resource stalls caused by fetch memory latency or branch prediction flow stalls. All the frontend slots were empty during the cycle when this event counts."
+    },
+    {
+        "ArchStdEvent": "STALL_BACKEND",
+        "PublicDescription": "Counts cycles whenever the rename unit is unable to send any micro-operations to the backend of the pipeline because of backend resource constraints. Backend resource constraints can include issue stage fullness, execution stage fullness, or other internal pipeline resource fullness. All the backend slots were empty during the cycle when this event counts."
+    },
+    {
+        "ArchStdEvent": "STALL",
+        "PublicDescription": "Counts cycles when no operations are sent to the rename unit from the frontend or from the rename unit to the backend for any reason (either frontend or backend stall)."
+    },
+    {
+        "ArchStdEvent": "STALL_SLOT_BACKEND",
+        "PublicDescription": "Counts slots per cycle in which no operations are sent from the rename unit to the backend due to backend resource constraints."
+    },
+    {
+        "ArchStdEvent": "STALL_SLOT_FRONTEND",
+        "PublicDescription": "Counts slots per cycle in which no operations are sent to the rename unit from the frontend due to frontend resource constraints."
+    },
+    {
+        "ArchStdEvent": "STALL_SLOT",
+        "PublicDescription": "Counts slots per cycle in which no operations are sent to the rename unit from the frontend or from the rename unit to the backend for any reason (either frontend or backend stall)."
+    },
+    {
+        "ArchStdEvent": "STALL_BACKEND_MEM",
+        "PublicDescription": "Counts cycles when the backend is stalled because there is a pending demand load request in progress in the last level core cache."
+    }
+]
tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/sve.json new file +50 -0

50 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/tlb.json new file +66 -0

66 lines not on this page · raw message

tools/perf/pmu-events/arch/arm64/arm/neoverse-n2-v2/trace.json +18 -9
index 3116135c59e2..98f6fabfebc7 100644
@@ -1,29 +1,38 @@
 [
     {
-        "ArchStdEvent": "TRB_WRAP"
+        "ArchStdEvent": "TRB_WRAP",
+        "PublicDescription": "This event is generated each time the current write pointer is wrapped to the base pointer."
     },
     {
-        "ArchStdEvent": "TRCEXTOUT0"
+        "ArchStdEvent": "TRCEXTOUT0",
+        "PublicDescription": "This event is generated each time an event is signaled by ETE external event 0."
     },
     {
-        "ArchStdEvent": "TRCEXTOUT1"
+        "ArchStdEvent": "TRCEXTOUT1",
+        "PublicDescription": "This event is generated each time an event is signaled by ETE external event 1."
     },
     {
-        "ArchStdEvent": "TRCEXTOUT2"
+        "ArchStdEvent": "TRCEXTOUT2",
+        "PublicDescription": "This event is generated each time an event is signaled by ETE external event 2."
     },
     {
-        "ArchStdEvent": "TRCEXTOUT3"
+        "ArchStdEvent": "TRCEXTOUT3",
+        "PublicDescription": "This event is generated each time an event is signaled by ETE external event 3."
     },
     {
-        "ArchStdEvent": "CTI_TRIGOUT4"
+        "ArchStdEvent": "CTI_TRIGOUT4",
+        "PublicDescription": "This event is generated each time an event is signaled on CTI output trigger 4."
     },
     {
-        "ArchStdEvent": "CTI_TRIGOUT5"
+        "ArchStdEvent": "CTI_TRIGOUT5",
+        "PublicDescription": "This event is generated each time an event is signaled on CTI output trigger 5."
     },
     {
-        "ArchStdEvent": "CTI_TRIGOUT6"
+        "ArchStdEvent": "CTI_TRIGOUT6",
+        "PublicDescription": "This event is generated each time an event is signaled on CTI output trigger 6."
     },
     {
-        "ArchStdEvent": "CTI_TRIGOUT7"
+        "ArchStdEvent": "CTI_TRIGOUT7",
+        "PublicDescription": "This event is generated each time an event is signaled on CTI output trigger 7."
     }
 ]
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
reviewer Reviewed-by on v6 6/6
#16 lore raw
Quoted 21 lines · James Clark

Apart from some slight naming and grouping differences, the new metrics are functionally the same as the existing ones. Any missing metrics were manually appended to the end of the auto generated file.

For the events, the new data includes descriptions that may have product specific details and new groupings that will be consistent with other products.

After generating the metrics from the telemetry repo [1], the following manual steps were performed:

* Change the topdown expressions to compare on CPUID and use #slots so that the same data can be shared between N2 and V2. Apart from these modifications, the expressions now match more closely with the Arm telemetry data which will hopefully make future updates easier.

* Append some metrics from the old N2/V2 data that aren't present in the telemetry data. These will possibly be added to the telemetry-solution repo at a later time:

Quoted 7 lines

l3d_cache_mpki, l3d_cache_miss_rate, branch_pki, ipc_rate, spec_ipc, retired_rate, wasted_rate, branch_immed_spec_rate, branch_return_spec_rate, branch_indirect_spec_rate

[1]:https://urldefense.com/v3/__https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-n2.json__;!!ACWV5N9M2RV99hQ!NW7DYe7T69u8RFn9WLyiCcoHm-8BtlxNK3PRw19udocwlwv0vQpjcSDT5XqGbWzvPQyFrG-eRkuu_VZt7cO8-Q$  

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#17 lore raw
Quoted 22 lines · John Garry
Quoted 21 lines · James Clark

Apart from some slight naming and grouping differences, the new metrics are functionally the same as the existing ones. Any missing metrics were manually appended to the end of the auto generated file.

For the events, the new data includes descriptions that may have product specific details and new groupings that will be consistent with other products.

After generating the metrics from the telemetry repo [1], the following manual steps were performed:

  * Change the topdown expressions to compare on CPUID and use #slots so that the same data can be shared between N2 and V2. Apart from these modifications, the expressions now match more closely with the Arm telemetry data which will hopefully make future updates easier.

  * Append some metrics from the old N2/V2 data that aren't present in the telemetry data. These will possibly be added to the telemetry-solution repo at a later time:

Quoted 8 lines
Quoted 6 lines

     l3d_cache_mpki, l3d_cache_miss_rate, branch_pki, ipc_rate, spec_ipc, retired_rate, wasted_rate, branch_immed_spec_rate, branch_return_spec_rate, branch_indirect_spec_rate

[1]:https://urldefense.com/v3/__https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-n2.json__;!!ACWV5N9M2RV99hQ!NW7DYe7T69u8RFn9WLyiCcoHm-8BtlxNK3PRw19udocwlwv0vQpjcSDT5XqGbWzvPQyFrG-eRkuu_VZt7cO8-Q$

Thanks for the reviews!

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#18 lore raw

Em Thu, Aug 17, 2023 at 03:41:11PM +0100, James Clark escreveu:

Quoted 29 lines · John Garry
Quoted 26 lines · James Clark
Quoted 25 lines

Apart from some slight naming and grouping differences, the new metrics are functionally the same as the existing ones. Any missing metrics were manually appended to the end of the auto generated file.

For the events, the new data includes descriptions that may have product specific details and new groupings that will be consistent with other products.

After generating the metrics from the telemetry repo [1], the following manual steps were performed:

  * Change the topdown expressions to compare on CPUID and use #slots so that the same data can be shared between N2 and V2. Apart from these modifications, the expressions now match more closely with the Arm telemetry data which will hopefully make future updates easier.

  * Append some metrics from the old N2/V2 data that aren't present in the telemetry data. These will possibly be added to the telemetry-solution repo at a later time:

     l3d_cache_mpki, l3d_cache_miss_rate, branch_pki, ipc_rate, spec_ipc, retired_rate, wasted_rate, branch_immed_spec_rate, branch_return_spec_rate, branch_indirect_spec_rate

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

[1]:https://urldefense.com/v3/__https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-n2.json__;!!ACWV5N9M2RV99hQ!NW7DYe7T69u8RFn9WLyiCcoHm-8BtlxNK3PRw19udocwlwv0vQpjcSDT5XqGbWzvPQyFrG-eRkuu_VZt7cO8-Q$

Thanks for the reviews!

Thanks everybody, patches merged, reviewed-by tags collected.

- Arnaldo

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel