From: Jiri Olsa <jolsa@redhat.com>
To: Ingo Molnar <mingo@kernel.org>
Cc: linux-kernel@vger.kernel.org, Andi Kleen <ak@linux.intel.com>,
Jiri Olsa <jolsa@redhat.com>
Subject: [PATCH 14/17] perf tools: Fix metrics calculation with event qualifiers
Date: Mon, 14 Apr 2014 16:17:38 +0200 [thread overview]
Message-ID: <1397485061-17215-15-git-send-email-jolsa@redhat.com> (raw)
In-Reply-To: <1397485061-17215-1-git-send-email-jolsa@redhat.com>
From: Andi Kleen <ak@linux.intel.com>
Currently in perf IPC and other metrics cannot be directly shown
separately for both user and kernel in a single run. The problem was
that the metrics matching code did not check event qualifiers.
With this patch the following case works correctly.
% perf stat -e cycles:k,cycles:u,instructions:k,instructions:u true
Performance counter stats for 'true':
531,718 cycles:k
203,895 cycles:u
338,151 instructions:k # 0.64 insns per cycle
105,961 instructions:u # 0.52 insns per cycle
0.002989739 seconds time elapsed
Previously it would misreport the ratios because they were matching
the wrong value.
The patch is fairly big, but quite mechanic as it just
adds context indexes everywhere.
I didn't support Hypervisor. It's not clear it's worth it.
Reported-by: William Cohen <wcohen@redhat.com>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Link: http://lkml.kernel.org/r/1395778047-23111-1-git-send-email-andi@firstfloor.org
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
---
tools/perf/builtin-stat.c | 129 +++++++++++++++++++++++++++-------------------
1 file changed, 77 insertions(+), 52 deletions(-)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 65a151e..6687dab 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -246,21 +246,35 @@ out_free:
return -1;
}
+#define NUM_CTX 3
+
+enum { CTX_USER, CTX_KERNEL, CTX_ALL };
+
static struct stats runtime_nsecs_stats[MAX_NR_CPUS];
-static struct stats runtime_cycles_stats[MAX_NR_CPUS];
-static struct stats runtime_stalled_cycles_front_stats[MAX_NR_CPUS];
-static struct stats runtime_stalled_cycles_back_stats[MAX_NR_CPUS];
-static struct stats runtime_branches_stats[MAX_NR_CPUS];
-static struct stats runtime_cacherefs_stats[MAX_NR_CPUS];
-static struct stats runtime_l1_dcache_stats[MAX_NR_CPUS];
-static struct stats runtime_l1_icache_stats[MAX_NR_CPUS];
-static struct stats runtime_ll_cache_stats[MAX_NR_CPUS];
-static struct stats runtime_itlb_cache_stats[MAX_NR_CPUS];
-static struct stats runtime_dtlb_cache_stats[MAX_NR_CPUS];
-static struct stats runtime_cycles_in_tx_stats[MAX_NR_CPUS];
+static struct stats runtime_cycles_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_stalled_cycles_front_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_stalled_cycles_back_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_branches_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_cacherefs_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_l1_dcache_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_l1_icache_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_ll_cache_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_itlb_cache_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_dtlb_cache_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_cycles_in_tx_stats[NUM_CTX][MAX_NR_CPUS];
static struct stats walltime_nsecs_stats;
-static struct stats runtime_transaction_stats[MAX_NR_CPUS];
-static struct stats runtime_elision_stats[MAX_NR_CPUS];
+static struct stats runtime_transaction_stats[NUM_CTX][MAX_NR_CPUS];
+static struct stats runtime_elision_stats[NUM_CTX][MAX_NR_CPUS];
+
+static int evsel_context(struct perf_evsel *evsel)
+{
+ if (evsel->attr.exclude_kernel)
+ return CTX_USER;
+ if (evsel->attr.exclude_user)
+ return CTX_KERNEL;
+ /* Handle hypervisor too? */
+ return CTX_ALL;
+}
static void perf_stat__reset_stats(struct perf_evlist *evlist)
{
@@ -354,37 +368,39 @@ static struct perf_evsel *nth_evsel(int n)
*/
static void update_shadow_stats(struct perf_evsel *counter, u64 *count)
{
+ int ctx = evsel_context(counter);
+
if (perf_evsel__match(counter, SOFTWARE, SW_TASK_CLOCK))
update_stats(&runtime_nsecs_stats[0], count[0]);
else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES))
- update_stats(&runtime_cycles_stats[0], count[0]);
+ update_stats(&runtime_cycles_stats[ctx][0], count[0]);
else if (transaction_run &&
perf_evsel__cmp(counter, nth_evsel(T_CYCLES_IN_TX)))
- update_stats(&runtime_cycles_in_tx_stats[0], count[0]);
+ update_stats(&runtime_cycles_in_tx_stats[ctx][0], count[0]);
else if (transaction_run &&
perf_evsel__cmp(counter, nth_evsel(T_TRANSACTION_START)))
- update_stats(&runtime_transaction_stats[0], count[0]);
+ update_stats(&runtime_transaction_stats[ctx][0], count[0]);
else if (transaction_run &&
perf_evsel__cmp(counter, nth_evsel(T_ELISION_START)))
- update_stats(&runtime_elision_stats[0], count[0]);
+ update_stats(&runtime_elision_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND))
- update_stats(&runtime_stalled_cycles_front_stats[0], count[0]);
+ update_stats(&runtime_stalled_cycles_front_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_BACKEND))
- update_stats(&runtime_stalled_cycles_back_stats[0], count[0]);
+ update_stats(&runtime_stalled_cycles_back_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HARDWARE, HW_BRANCH_INSTRUCTIONS))
- update_stats(&runtime_branches_stats[0], count[0]);
+ update_stats(&runtime_branches_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HARDWARE, HW_CACHE_REFERENCES))
- update_stats(&runtime_cacherefs_stats[0], count[0]);
+ update_stats(&runtime_cacherefs_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1D))
- update_stats(&runtime_l1_dcache_stats[0], count[0]);
+ update_stats(&runtime_l1_dcache_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1I))
- update_stats(&runtime_l1_icache_stats[0], count[0]);
+ update_stats(&runtime_l1_icache_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_LL))
- update_stats(&runtime_ll_cache_stats[0], count[0]);
+ update_stats(&runtime_ll_cache_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_DTLB))
- update_stats(&runtime_dtlb_cache_stats[0], count[0]);
+ update_stats(&runtime_dtlb_cache_stats[ctx][0], count[0]);
else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB))
- update_stats(&runtime_itlb_cache_stats[0], count[0]);
+ update_stats(&runtime_itlb_cache_stats[ctx][0], count[0]);
}
/*
@@ -801,8 +817,9 @@ static void print_stalled_cycles_frontend(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_cycles_stats[cpu]);
+ total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -820,8 +837,9 @@ static void print_stalled_cycles_backend(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_cycles_stats[cpu]);
+ total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -839,8 +857,9 @@ static void print_branch_misses(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_branches_stats[cpu]);
+ total = avg_stats(&runtime_branches_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -858,8 +877,9 @@ static void print_l1_dcache_misses(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_l1_dcache_stats[cpu]);
+ total = avg_stats(&runtime_l1_dcache_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -877,8 +897,9 @@ static void print_l1_icache_misses(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_l1_icache_stats[cpu]);
+ total = avg_stats(&runtime_l1_icache_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -896,8 +917,9 @@ static void print_dtlb_cache_misses(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_dtlb_cache_stats[cpu]);
+ total = avg_stats(&runtime_dtlb_cache_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -915,8 +937,9 @@ static void print_itlb_cache_misses(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_itlb_cache_stats[cpu]);
+ total = avg_stats(&runtime_itlb_cache_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -934,8 +957,9 @@ static void print_ll_cache_misses(int cpu,
{
double total, ratio = 0.0;
const char *color;
+ int ctx = evsel_context(evsel);
- total = avg_stats(&runtime_ll_cache_stats[cpu]);
+ total = avg_stats(&runtime_ll_cache_stats[ctx][cpu]);
if (total)
ratio = avg / total * 100.0;
@@ -952,6 +976,7 @@ static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg)
double total, ratio = 0.0, total2;
double sc = evsel->scale;
const char *fmt;
+ int ctx = evsel_context(evsel);
if (csv_output) {
fmt = sc != 1.0 ? "%.2f%s" : "%.0f%s";
@@ -983,13 +1008,13 @@ static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg)
return;
if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) {
- total = avg_stats(&runtime_cycles_stats[cpu]);
+ total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
if (total) {
ratio = avg / total;
fprintf(output, " # %5.2f insns per cycle ", ratio);
}
- total = avg_stats(&runtime_stalled_cycles_front_stats[cpu]);
- total = max(total, avg_stats(&runtime_stalled_cycles_back_stats[cpu]));
+ total = avg_stats(&runtime_stalled_cycles_front_stats[ctx][cpu]);
+ total = max(total, avg_stats(&runtime_stalled_cycles_back_stats[ctx][cpu]));
if (total && avg) {
ratio = total / avg;
@@ -1000,46 +1025,46 @@ static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg)
}
} else if (perf_evsel__match(evsel, HARDWARE, HW_BRANCH_MISSES) &&
- runtime_branches_stats[cpu].n != 0) {
+ runtime_branches_stats[ctx][cpu].n != 0) {
print_branch_misses(cpu, evsel, avg);
} else if (
evsel->attr.type == PERF_TYPE_HW_CACHE &&
evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1D |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
- runtime_l1_dcache_stats[cpu].n != 0) {
+ runtime_l1_dcache_stats[ctx][cpu].n != 0) {
print_l1_dcache_misses(cpu, evsel, avg);
} else if (
evsel->attr.type == PERF_TYPE_HW_CACHE &&
evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1I |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
- runtime_l1_icache_stats[cpu].n != 0) {
+ runtime_l1_icache_stats[ctx][cpu].n != 0) {
print_l1_icache_misses(cpu, evsel, avg);
} else if (
evsel->attr.type == PERF_TYPE_HW_CACHE &&
evsel->attr.config == ( PERF_COUNT_HW_CACHE_DTLB |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
- runtime_dtlb_cache_stats[cpu].n != 0) {
+ runtime_dtlb_cache_stats[ctx][cpu].n != 0) {
print_dtlb_cache_misses(cpu, evsel, avg);
} else if (
evsel->attr.type == PERF_TYPE_HW_CACHE &&
evsel->attr.config == ( PERF_COUNT_HW_CACHE_ITLB |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
- runtime_itlb_cache_stats[cpu].n != 0) {
+ runtime_itlb_cache_stats[ctx][cpu].n != 0) {
print_itlb_cache_misses(cpu, evsel, avg);
} else if (
evsel->attr.type == PERF_TYPE_HW_CACHE &&
evsel->attr.config == ( PERF_COUNT_HW_CACHE_LL |
((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
- runtime_ll_cache_stats[cpu].n != 0) {
+ runtime_ll_cache_stats[ctx][cpu].n != 0) {
print_ll_cache_misses(cpu, evsel, avg);
} else if (perf_evsel__match(evsel, HARDWARE, HW_CACHE_MISSES) &&
- runtime_cacherefs_stats[cpu].n != 0) {
- total = avg_stats(&runtime_cacherefs_stats[cpu]);
+ runtime_cacherefs_stats[ctx][cpu].n != 0) {
+ total = avg_stats(&runtime_cacherefs_stats[ctx][cpu]);
if (total)
ratio = avg * 100 / total;
@@ -1059,15 +1084,15 @@ static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg)
}
} else if (transaction_run &&
perf_evsel__cmp(evsel, nth_evsel(T_CYCLES_IN_TX))) {
- total = avg_stats(&runtime_cycles_stats[cpu]);
+ total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
if (total)
fprintf(output,
" # %5.2f%% transactional cycles ",
100.0 * (avg / total));
} else if (transaction_run &&
perf_evsel__cmp(evsel, nth_evsel(T_CYCLES_IN_TX_CP))) {
- total = avg_stats(&runtime_cycles_stats[cpu]);
- total2 = avg_stats(&runtime_cycles_in_tx_stats[cpu]);
+ total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
+ total2 = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
if (total2 < avg)
total2 = avg;
if (total)
@@ -1077,8 +1102,8 @@ static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg)
} else if (transaction_run &&
perf_evsel__cmp(evsel, nth_evsel(T_TRANSACTION_START)) &&
avg > 0 &&
- runtime_cycles_in_tx_stats[cpu].n != 0) {
- total = avg_stats(&runtime_cycles_in_tx_stats[cpu]);
+ runtime_cycles_in_tx_stats[ctx][cpu].n != 0) {
+ total = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
if (total)
ratio = total / avg;
@@ -1087,8 +1112,8 @@ static void abs_printout(int cpu, int nr, struct perf_evsel *evsel, double avg)
} else if (transaction_run &&
perf_evsel__cmp(evsel, nth_evsel(T_ELISION_START)) &&
avg > 0 &&
- runtime_cycles_in_tx_stats[cpu].n != 0) {
- total = avg_stats(&runtime_cycles_in_tx_stats[cpu]);
+ runtime_cycles_in_tx_stats[ctx][cpu].n != 0) {
+ total = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
if (total)
ratio = total / avg;
--
1.8.3.1
next prev parent reply other threads:[~2014-04-14 14:21 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-04-14 14:17 [GIT PULL v3 00/17] perf/core fixes Jiri Olsa
2014-04-14 14:17 ` [PATCH 01/17] perf top: Fix documentation of invalid -s option Jiri Olsa
2014-04-14 14:17 ` [PATCH 02/17] tools lib traceevent: Print event name when show warning if possible Jiri Olsa
2014-04-14 14:17 ` [PATCH 03/17] tools lib traceevent: Do not call warning() directly Jiri Olsa
2014-04-14 14:17 ` [PATCH 04/17] perf callchains: Disable unwind libraries when libelf isn't found Jiri Olsa
2014-04-14 14:17 ` [PATCH 05/17] MAINTAINERS: Change e-mail to kernel.org one Jiri Olsa
2014-04-14 14:17 ` [PATCH 06/17] perf tools: Pick up libdw without explicit LIBDW_DIR Jiri Olsa
2014-04-14 14:17 ` [PATCH 07/17] perf probe: Fix --line option behavior Jiri Olsa
2014-04-14 14:17 ` [PATCH 08/17] perf probe: Fix to handle errors in line_range searching Jiri Olsa
2014-04-14 14:17 ` [PATCH 09/17] perf probe: Use dwarf_getcfi_elf() instead of dwarf_getcfi() Jiri Olsa
2014-04-14 14:17 ` [PATCH 10/17] perf bench: Update manpage to mention numa and futex Jiri Olsa
2014-04-14 14:17 ` [PATCH 11/17] perf bench: Fix segfault at the end of an 'all' execution Jiri Olsa
2014-04-14 14:17 ` [PATCH 12/17] perf bench: Set more defaults in the 'numa' suite Jiri Olsa
2014-04-14 14:17 ` [PATCH 13/17] perf stat: Initialize statistics correctly Jiri Olsa
2014-04-14 14:17 ` Jiri Olsa [this message]
2014-04-14 14:38 ` [PATCH 14/17] perf tools: Fix metrics calculation with event qualifiers Ingo Molnar
2014-04-14 14:17 ` [PATCH 15/17] perf tools: Fix double free in perf test 21 (code-reading.c) Jiri Olsa
2014-04-14 14:17 ` [PATCH 16/17] perf machine: Search for modules in %s/lib/modules/%s Jiri Olsa
2014-04-14 14:17 ` [PATCH 17/17] perf tools: Instead of redirecting flex output, use -o Jiri Olsa
2014-04-14 14:42 ` [GIT PULL v3 00/17] perf/core fixes Ingo Molnar
2014-04-14 14:52 ` Jiri Olsa
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1397485061-17215-15-git-send-email-jolsa@redhat.com \
--to=jolsa@redhat.com \
--cc=ak@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox