Re: [PATCH] perf stat: Add Topdown metrics events as default events
From: Liang, Kan
Date: Thu Jan 21 2021 - 17:07:36 EST
On 1/21/2021 2:57 PM, Arnaldo Carvalho de Melo wrote:
Em Thu, Jan 21, 2021 at 05:37:52AM -0800, kan.liang@xxxxxxxxxxxxxxx escreveu:
From: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
The Topdown Microarchitecture Analysis (TMA) Method is a structured
analysis methodology to identify critical performance bottlenecks in
out-of-order processors. From the Ice Lake and later platforms, the
Topdown information can be retrieved from the dedicated "metrics"
register, which isn't impacted by other events. Also, the Topdown
metrics support both per thread/process and per core measuring.
Adding Topdown metrics events as default events can enrich the default
measuring information, and would not cost any extra multiplexing.
Introduce arch_evlist__add_default_attrs() to allow architecture
specific default events. Add the Topdown metrics events in the X86
specific arch_evlist__add_default_attrs(). Other architectures can
add their own default events later separately.
With the patch,
$perf stat sleep 1
Performance counter stats for 'sleep 1':
0.82 msec task-clock:u # 0.001 CPUs utilized
0 context-switches:u # 0.000 K/sec
0 cpu-migrations:u # 0.000 K/sec
61 page-faults:u # 0.074 M/sec
319,941 cycles:u # 0.388 GHz
242,802 instructions:u # 0.76 insn per cycle
54,380 branches:u # 66.028 M/sec
4,043 branch-misses:u # 7.43% of all branches
1,585,555 slots:u # 1925.189 M/sec
238,941 topdown-retiring:u # 15.0% retiring
410,378 topdown-bad-spec:u # 25.8% bad speculation
634,222 topdown-fe-bound:u # 39.9% frontend bound
304,675 topdown-be-bound:u # 19.2% backend bound
Shouldn't we be adding this to one of the -d levels?
The -d levels' events are architecture events, which should be available
on all architectures. The Topdown events are X86 specific for now.
I don't think we should add the Topdown events into any of the -d levels.
Different architectures should have their unique events. I think we
should allow architectures to default their unique and very useful events.
But you say that this is essentially for free, so you check if this
extra register is in place and if it is, hey, its for free, add it, is
that the rationale?
> I.e. it doesn't cause any impact to what we were getting before, so we
should default to use free stuff?
Free is just one of the benefits. The key motivation is that we think
the Topdown information is very useful information, which can help users
identify performance bottlenecks in modern out-of-order processors.
Here is an example that triggers lots of itlb misses.
With the patch, users can simply identify the performance issue with
default perf stat (98.0% frontend bound, which leads the users to check
the fetch bandwidth or the fetch latency related events, e.g., itlb
misses).
$perf stat ./itlb -b10000 -n200 -c0
Performance counter stats for './itlb -b10000 -n200 -c0':
14,567.03 msec task-clock:u # 1.000 CPUs utilized
0 context-switches:u # 0.000 K/sec
0 cpu-migrations:u # 0.000 K/sec
683 page-faults:u # 0.047 K/sec
55,890,568,522 cycles:u # 3.837 GHz
3,601,934,137 instructions:u # 0.06 insn per cycle
400,353,806 branches:u # 27.484 M/sec
24,078 branch-misses:u # 0.01% of all
branches
279,391,452,475 slots:u # 19179.715 M/sec
2,230,224,702 topdown-retiring:u # 0.8% retiring
3,286,958,264 topdown-bad-spec:u # 1.2% bad
speculation
273,953,648,626 topdown-fe-bound:u # 98.0% frontend
bound
4,579,804 topdown-be-bound:u # 0.0% backend bound
Without the patch, users cannot get any useful hints from the default
perf stat. Even with -d -d, users still cannot locate the issue to itlb
misses, because we don't have an architecture iTLB-loads event.
The -d -d also triggers the multiplexing.
$ perf stat -d -d ./itlb -b10000 -n200 -c0
Performance counter stats for './itlb -b10000 -n200 -c0':
14,633.58 msec task-clock:u # 1.000 CPUs utilized
0 context-switches:u # 0.000 K/sec
0 cpu-migrations:u # 0.000 K/sec
682 page-faults:u # 0.047 K/sec
56,198,668,109 cycles:u # 3.840 GHz
(38.41%)
3,602,460,111 instructions:u # 0.06 insn per
cycle (46.12%)
400,445,992 branches:u # 27.365 M/sec
(53.83%)
28,757 branch-misses:u # 0.01% of all
branches (61.54%)
761,047 L1-dcache-loads:u # 0.052 M/sec
(69.27%)
24,619,906 L1-dcache-load-misses:u # 3235.00% of all
L1-dcache accesses (69.28%)
905,610 LLC-loads:u # 0.062 M/sec
(69.28%)
603,464 LLC-load-misses:u # 66.64% of all
LL-cache accesses (69.28%)
<not supported> L1-icache-loads:u
3,870,274,747 L1-icache-load-misses:u
(30.73%)
7,341 dTLB-loads:u # 0.502 K/sec
(30.72%)
4,730 dTLB-load-misses:u # 64.43% of all
dTLB cache accesses (30.72%)
<not supported> iTLB-loads:u
196,493,394 iTLB-load-misses:u
(30.72%)
Thanks,
Kan
- Arnaldo
1.001791625 seconds time elapsed
0.000000000 seconds user
0.001572000 seconds sys
Signed-off-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
---
tools/perf/arch/x86/util/Build | 1 +
tools/perf/arch/x86/util/evlist.c | 15 +++++++++++++++
tools/perf/builtin-stat.c | 3 +++
tools/perf/util/evlist.c | 5 +++++
tools/perf/util/evlist.h | 2 ++
5 files changed, 26 insertions(+)
create mode 100644 tools/perf/arch/x86/util/evlist.c
diff --git a/tools/perf/arch/x86/util/Build b/tools/perf/arch/x86/util/Build
index 347c39b960eb..ce1ec92fecdc 100644
--- a/tools/perf/arch/x86/util/Build
+++ b/tools/perf/arch/x86/util/Build
@@ -6,6 +6,7 @@ perf-y += perf_regs.o
perf-y += topdown.o
perf-y += machine.o
perf-y += event.o
+perf-y += evlist.o
perf-$(CONFIG_DWARF) += dwarf-regs.o
perf-$(CONFIG_BPF_PROLOGUE) += dwarf-regs.o
diff --git a/tools/perf/arch/x86/util/evlist.c b/tools/perf/arch/x86/util/evlist.c
new file mode 100644
index 000000000000..8c6732cc7794
--- /dev/null
+++ b/tools/perf/arch/x86/util/evlist.c
@@ -0,0 +1,15 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <stdio.h>
+#include "util/pmu.h"
+#include "util/evlist.h"
+#include "util/parse-events.h"
+
+#define TOPDOWN_L1_EVENTS "{slots,topdown-retiring,topdown-bad-spec,topdown-fe-bound,topdown-be-bound}"
+
+int arch_evlist__add_default_attrs(struct evlist *evlist)
+{
+ if (!pmu_have_event("cpu", "slots"))
+ return 0;
+
+ return parse_events(evlist, TOPDOWN_L1_EVENTS, NULL);
+}
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 3c054b8d4677..abcdabaf1701 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1827,6 +1827,9 @@ static int add_default_attributes(void)
}
if (evlist__add_default_attrs(evsel_list, default_attrs1) < 0)
return -1;
+
+ if (arch_evlist__add_default_attrs(evsel_list) < 0)
+ return -1;
}
/* Detailed events get appended to the event list: */
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 05363a7247c4..b38589d8c027 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -303,6 +303,11 @@ int __evlist__add_default_attrs(struct evlist *evlist, struct perf_event_attr *a
return evlist__add_attrs(evlist, attrs, nr_attrs);
}
+__weak int arch_evlist__add_default_attrs(struct evlist *evlist __maybe_unused)
+{
+ return 0;
+}
+
struct evsel *evlist__find_tracepoint_by_id(struct evlist *evlist, int id)
{
struct evsel *evsel;
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 1aae75895dea..9eba4958a1e9 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -110,6 +110,8 @@ int __evlist__add_default_attrs(struct evlist *evlist,
#define evlist__add_default_attrs(evlist, array) \
__evlist__add_default_attrs(evlist, array, ARRAY_SIZE(array))
+int arch_evlist__add_default_attrs(struct evlist *evlist);
+
int evlist__add_dummy(struct evlist *evlist);
int evlist__add_sb_event(struct evlist *evlist, struct perf_event_attr *attr,
--
2.25.1