Skip to contentSkip to conversation

perf test: Parse perf stat Topdown events for aarch64

PATCH Applied [email protected] 3 participants 6 messages Oct 29, 2024 – Dec 10, 2024 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf test: Parse perf stat Topdown events for aarch64

From: Veronika Molnarova <[email protected]>

The 'perf stat' output on aarch64 machines with topdown events wasn't counted for in the 'perf stat STD output linter' test case. Add the topdown metric to the skip_metric list as it is done for topdown events on other systems.

The Topdown events are also disabled on aarch64 KVM guests because the value of caps/slots is set to 0 due to the part of the system register being a stub. This prevents the metric for the topdown events from being computed, leaving the 'perf stat' topdown metric without any value at all. Add the "TopdownL1" to the skip_metric list as well to handle this possibility.

Before aarch64:

100: perf stat STD output linter:
--- start ---
test child forked, pid 403305
Checking STD output: no args Unknown event name in TopdownL1 # 4.3 percent of slots slots_lost_misspeculation_fraction
---- end(-1) ----
100: perf stat STD output linter : FAILED!

Before aarch64 KVM:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404671
Checking STD output: no args Unknown event name in TopdownL1 ---- end(-1) ----
100: perf stat STD output linter : FAILED!

After:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404777
Checking STD output: no args [Success]
Checking STD output: system wide [Success] Checking STD output: interval [Success] Checking STD output: per thread [Success] Checking STD output: per node [Success] Checking STD output: system wide no aggregation [Success] Checking STD output: per core [Success] Checking STD output: per cache instance [Success] Checking STD output: per cluster [Success] Checking STD output: per die [Success]
Checking STD output: per socket [Success]
---- end(0) ----
100: perf stat STD output linter : Ok

---
 tools/perf/tests/shell/stat+std_output.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/tests/shell/stat+std_output.sh +1 -1
index cbf2894b2c846a04..0f7967be60afdc7d 100755
@@ -13,7 +13,7 @@ stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX)
 
 event_name=(cpu-clock task-clock context-switches cpu-migrations page-faults stalled-cycles-frontend stalled-cycles-backend cycles instructions branches branch-misses)
 event_metric=("CPUs utilized" "CPUs utilized" "/sec" "/sec" "/sec" "frontend cycles idle" "backend cycles idle" "GHz" "insn per cycle" "/sec" "of all branches")
-skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound")
+skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound" "TopdownL1" "percent of slots")
 
 cleanup() {
   rm -f "${stat_output}"
Signature · 1 line
2.43.0
reviewer
#1 lore raw
Quoted 63 lines · On Tue, Oct 29, 2024 at 7:44 AM <[email protected]> wrote:

From: Veronika Molnarova <[email protected]>

The 'perf stat' output on aarch64 machines with topdown events wasn't counted for in the 'perf stat STD output linter' test case. Add the topdown metric to the skip_metric list as it is done for topdown events on other systems.

The Topdown events are also disabled on aarch64 KVM guests because the value of caps/slots is set to 0 due to the part of the system register being a stub. This prevents the metric for the topdown events from being computed, leaving the 'perf stat' topdown metric without any value at all. Add the "TopdownL1" to the skip_metric list as well to handle this possibility.

Before aarch64:

100: perf stat STD output linter:
--- start ---
test child forked, pid 403305
Checking STD output: no args Unknown event name in TopdownL1 # 4.3 percent of slots slots_lost_misspeculation_fraction
---- end(-1) ----
100: perf stat STD output linter : FAILED!

Before aarch64 KVM:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404671
Checking STD output: no args Unknown event name in TopdownL1 ---- end(-1) ----
100: perf stat STD output linter : FAILED!

After:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404777
Checking STD output: no args [Success]
Checking STD output: system wide [Success] Checking STD output: interval [Success] Checking STD output: per thread [Success] Checking STD output: per node [Success] Checking STD output: system wide no aggregation [Success] Checking STD output: per core [Success] Checking STD output: per cache instance [Success] Checking STD output: per cluster [Success] Checking STD output: per die [Success]
Checking STD output: per socket [Success]
---- end(0) ----
100: perf stat STD output linter : Ok

---
 tools/perf/tests/shell/stat+std_output.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/tests/shell/stat+std_output.sh +0 -0
index cbf2894b2c846a04..0f7967be60afdc7d 100755

tools/perf/tests/shell/stat+std_output.sh stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX) · line 16

Quoted 5 lines
tools/perf/tests/shell/stat+std_output.sh +1 -1
@@ -13,7 +13,7 @@ stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX)
 
 event_name=(cpu-clock task-clock context-switches cpu-migrations page-faults stalled-cycles-frontend stalled-cycles-backend cycles instructions branches branch-misses)
 event_metric=("CPUs utilized" "CPUs utilized" "/sec" "/sec" "/sec" "frontend cycles idle" "backend cycles idle" "GHz" "insn per cycle" "/sec" "of all branches")
-skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound")
+skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound" "TopdownL1" "percent of slots")

I saw similar issues and sent:
https://lore.kernel.org/lkml/[email protected]/
but got distracted. A difference there is automating the generation of the skip_metric list rather than hard coding it. Hard coding may be the most pragmatic thing to do in the near term.

Thanks,
Ian

Quoted 5 lines
tools/perf/tests/shell/stat+std_output.sh +0 -1
 cleanup() {
   rm -f "${stat_output}"
--

2.43.0

#2 lore raw
Quoted 76 lines · Ian Rogers
Quoted 63 lines · On Tue, Oct 29, 2024 at 7:44 AM <[email protected]> wrote:

From: Veronika Molnarova <[email protected]>

The 'perf stat' output on aarch64 machines with topdown events wasn't counted for in the 'perf stat STD output linter' test case. Add the topdown metric to the skip_metric list as it is done for topdown events on other systems.

The Topdown events are also disabled on aarch64 KVM guests because the value of caps/slots is set to 0 due to the part of the system register being a stub. This prevents the metric for the topdown events from being computed, leaving the 'perf stat' topdown metric without any value at all. Add the "TopdownL1" to the skip_metric list as well to handle this possibility.

Before aarch64:

100: perf stat STD output linter:
--- start ---
test child forked, pid 403305
Checking STD output: no args Unknown event name in TopdownL1 # 4.3 percent of slots slots_lost_misspeculation_fraction
---- end(-1) ----
100: perf stat STD output linter : FAILED!

Before aarch64 KVM:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404671
Checking STD output: no args Unknown event name in TopdownL1 ---- end(-1) ----
100: perf stat STD output linter : FAILED!

After:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404777
Checking STD output: no args [Success]
Checking STD output: system wide [Success] Checking STD output: interval [Success] Checking STD output: per thread [Success] Checking STD output: per node [Success] Checking STD output: system wide no aggregation [Success] Checking STD output: per core [Success] Checking STD output: per cache instance [Success] Checking STD output: per cluster [Success] Checking STD output: per die [Success]
Checking STD output: per socket [Success]
---- end(0) ----
100: perf stat STD output linter : Ok

---
 tools/perf/tests/shell/stat+std_output.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/tests/shell/stat+std_output.sh +0 -0
index cbf2894b2c846a04..0f7967be60afdc7d 100755

tools/perf/tests/shell/stat+std_output.sh stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX) · line 16

Quoted 5 lines
tools/perf/tests/shell/stat+std_output.sh +1 -1
@@ -13,7 +13,7 @@ stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX)
 
 event_name=(cpu-clock task-clock context-switches cpu-migrations page-faults stalled-cycles-frontend stalled-cycles-backend cycles instructions branches branch-misses)
 event_metric=("CPUs utilized" "CPUs utilized" "/sec" "/sec" "/sec" "frontend cycles idle" "backend cycles idle" "GHz" "insn per cycle" "/sec" "of all branches")
-skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound")
+skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound" "TopdownL1" "percent of slots")

I saw similar issues and sent:
https://lore.kernel.org/lkml/[email protected]/
but got distracted. A difference there is automating the generation of the skip_metric list rather than hard coding it. Hard coding may be the most pragmatic thing to do in the near term.

Quoted 3 lines

Thanks,
Ian

Automating the skip_metric list generation would be great, but for the time being, we can stick with hard coding until there is such possibility.

Thanks,
Veronika

Quoted 6 lines
Quoted 5 lines
tools/perf/tests/shell/stat+std_output.sh +0 -1
 cleanup() {
   rm -f "${stat_output}"
--

2.43.0

reviewer Acked-by
#3 lore raw
Quoted 80 lines · Veronika Molnarova
Quoted 76 lines · Ian Rogers
Quoted 63 lines · On Tue, Oct 29, 2024 at 7:44 AM <[email protected]> wrote:

From: Veronika Molnarova <[email protected]>

The 'perf stat' output on aarch64 machines with topdown events wasn't counted for in the 'perf stat STD output linter' test case. Add the topdown metric to the skip_metric list as it is done for topdown events on other systems.

The Topdown events are also disabled on aarch64 KVM guests because the value of caps/slots is set to 0 due to the part of the system register being a stub. This prevents the metric for the topdown events from being computed, leaving the 'perf stat' topdown metric without any value at all. Add the "TopdownL1" to the skip_metric list as well to handle this possibility.

Before aarch64:

100: perf stat STD output linter:
--- start ---
test child forked, pid 403305
Checking STD output: no args Unknown event name in TopdownL1 # 4.3 percent of slots slots_lost_misspeculation_fraction
---- end(-1) ----
100: perf stat STD output linter : FAILED!

Before aarch64 KVM:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404671
Checking STD output: no args Unknown event name in TopdownL1 ---- end(-1) ----
100: perf stat STD output linter : FAILED!

After:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404777
Checking STD output: no args [Success]
Checking STD output: system wide [Success] Checking STD output: interval [Success] Checking STD output: per thread [Success] Checking STD output: per node [Success] Checking STD output: system wide no aggregation [Success] Checking STD output: per core [Success] Checking STD output: per cache instance [Success] Checking STD output: per cluster [Success] Checking STD output: per die [Success]
Checking STD output: per socket [Success]
---- end(0) ----
100: perf stat STD output linter : Ok

---
 tools/perf/tests/shell/stat+std_output.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/tests/shell/stat+std_output.sh +0 -0
index cbf2894b2c846a04..0f7967be60afdc7d 100755

tools/perf/tests/shell/stat+std_output.sh stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX) · line 16

Quoted 5 lines
tools/perf/tests/shell/stat+std_output.sh +1 -1
@@ -13,7 +13,7 @@ stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX)
 
 event_name=(cpu-clock task-clock context-switches cpu-migrations page-faults stalled-cycles-frontend stalled-cycles-backend cycles instructions branches branch-misses)
 event_metric=("CPUs utilized" "CPUs utilized" "/sec" "/sec" "/sec" "frontend cycles idle" "backend cycles idle" "GHz" "insn per cycle" "/sec" "of all branches")
-skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound")
+skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound" "TopdownL1" "percent of slots")

I saw similar issues and sent:
https://lore.kernel.org/lkml/[email protected]/
but got distracted. A difference there is automating the generation of the skip_metric list rather than hard coding it. Hard coding may be the most pragmatic thing to do in the near term.

Quoted 6 lines
Quoted 3 lines

Thanks,
Ian

Automating the skip_metric list generation would be great, but for the time being, we can stick with hard coding until there is such possibility.

Thanks,
Ian

#4 lore raw
Quoted 84 lines · Ian Rogers
Quoted 83 lines · Veronika Molnarova
Quoted 79 lines · Ian Rogers
Quoted 63 lines · On Tue, Oct 29, 2024 at 7:44 AM <[email protected]> wrote:

From: Veronika Molnarova <[email protected]>

The 'perf stat' output on aarch64 machines with topdown events wasn't counted for in the 'perf stat STD output linter' test case. Add the topdown metric to the skip_metric list as it is done for topdown events on other systems.

The Topdown events are also disabled on aarch64 KVM guests because the value of caps/slots is set to 0 due to the part of the system register being a stub. This prevents the metric for the topdown events from being computed, leaving the 'perf stat' topdown metric without any value at all. Add the "TopdownL1" to the skip_metric list as well to handle this possibility.

Before aarch64:

100: perf stat STD output linter:
--- start ---
test child forked, pid 403305
Checking STD output: no args Unknown event name in TopdownL1 # 4.3 percent of slots slots_lost_misspeculation_fraction
---- end(-1) ----
100: perf stat STD output linter : FAILED!

Before aarch64 KVM:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404671
Checking STD output: no args Unknown event name in TopdownL1 ---- end(-1) ----
100: perf stat STD output linter : FAILED!

After:

100: perf stat STD output linter:
--- start ---
test child forked, pid 404777
Checking STD output: no args [Success]
Checking STD output: system wide [Success] Checking STD output: interval [Success] Checking STD output: per thread [Success] Checking STD output: per node [Success] Checking STD output: system wide no aggregation [Success] Checking STD output: per core [Success] Checking STD output: per cache instance [Success] Checking STD output: per cluster [Success] Checking STD output: per die [Success]
Checking STD output: per socket [Success]
---- end(0) ----
100: perf stat STD output linter : Ok

---
 tools/perf/tests/shell/stat+std_output.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/tests/shell/stat+std_output.sh +0 -0
index cbf2894b2c846a04..0f7967be60afdc7d 100755

tools/perf/tests/shell/stat+std_output.sh stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX) · line 16

Quoted 5 lines
tools/perf/tests/shell/stat+std_output.sh +1 -1
@@ -13,7 +13,7 @@ stat_output=$(mktemp /tmp/__perf_test.stat_output.std.XXXXX)
 
 event_name=(cpu-clock task-clock context-switches cpu-migrations page-faults stalled-cycles-frontend stalled-cycles-backend cycles instructions branches branch-misses)
 event_metric=("CPUs utilized" "CPUs utilized" "/sec" "/sec" "/sec" "frontend cycles idle" "backend cycles idle" "GHz" "insn per cycle" "/sec" "of all branches")
-skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound")
+skip_metric=("stalled cycles per insn" "tma_" "retiring" "frontend_bound" "bad_speculation" "backend_bound" "TopdownL1" "percent of slots")

I saw similar issues and sent:
https://lore.kernel.org/lkml/[email protected]/
but got distracted. A difference there is automating the generation of the skip_metric list rather than hard coding it. Hard coding may be the most pragmatic thing to do in the near term.

Thanks,
Ian

Quoted 9 lines
Quoted 3 lines

Automating the skip_metric list generation would be great, but for the time being, we can stick with hard coding until there is such possibility.

Thanks,
Ian

Ping on this patch. Please let me know if it's missing anything.

Thanks,
Veronika

maintainer Applied
#5 lore raw
Quoted 9 lines · Veronika Molnarova
Quoted 7 lines · Ian Rogers
Quoted 6 lines · Veronika Molnarova
Quoted 5 lines · Ian Rogers

I saw similar issues and sent:
https://lore.kernel.org/lkml/[email protected]/
but got distracted. A difference there is automating the generation of the skip_metric list rather than hard coding it. Hard coding may be the most pragmatic thing to do in the near term.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Automating the skip_metric list generation would be great, but for the time being, we can stick with hard coding until there is such possibility.

Ping on this patch. Please let me know if it's missing anything.

Thanks, applied to perf-tools-next.

- Arnaldo