From: Jiri Olsa <jolsa@kernel.org>
To: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: lkml <linux-kernel@vger.kernel.org>,
Don Zickus <dzickus@redhat.com>, Joe Mario <jmario@redhat.com>,
Ingo Molnar <mingo@kernel.org>,
Peter Zijlstra <a.p.zijlstra@chello.nl>,
Namhyung Kim <namhyung@kernel.org>,
David Ahern <dsahern@gmail.com>, Andi Kleen <andi@firstfloor.org>
Subject: [PATCH 39/61] perf c2c report: Add node sort key
Date: Mon, 19 Sep 2016 15:09:48 +0200 [thread overview]
Message-ID: <1474290610-23241-40-git-send-email-jolsa@kernel.org> (raw)
In-Reply-To: <1474290610-23241-1-git-send-email-jolsa@kernel.org>
Adding node dimension key wrapper.
It is to be displayed in the single cacheline output:
node
It displays nodes hits related to cacheline accesses.
The node filed comes in 3 flavors:
- node IDs separated by ','
- node IDs with stats for each ID, in following format:
Node{cpus %hitms %stores}
- node IDs with list of affected CPUs in following format:
Node{cpu list}
User can switch the flavor with -N option (-NN,-NNN).
It will be available in TUI to switch this with 'n' key.
Link: http://lkml.kernel.org/n/tip-6742e6g0r7n63y5wc4rrgxx5@git.kernel.org
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
tools/perf/builtin-c2c.c | 219 +++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 219 insertions(+)
diff --git a/tools/perf/builtin-c2c.c b/tools/perf/builtin-c2c.c
index b3dcd590e97a..6b4224764ae4 100644
--- a/tools/perf/builtin-c2c.c
+++ b/tools/perf/builtin-c2c.c
@@ -1,6 +1,7 @@
#include <linux/compiler.h>
#include <linux/kernel.h>
#include <linux/stringify.h>
+#include <asm/bug.h>
#include "util.h"
#include "debug.h"
#include "builtin.h"
@@ -22,6 +23,8 @@ struct c2c_hists {
struct c2c_hist_entry {
struct c2c_hists *hists;
struct c2c_stats stats;
+ unsigned long *cpuset;
+ struct c2c_stats *node_stats;
/*
* must be at the end,
* because of its callchain dynamic entry
@@ -32,6 +35,12 @@ struct c2c_hist_entry {
struct perf_c2c {
struct perf_tool tool;
struct c2c_hists hists;
+
+ unsigned long **nodes;
+ int nodes_cnt;
+ int cpus_cnt;
+ int *cpu2node;
+ int node_info;
};
static struct perf_c2c c2c;
@@ -44,6 +53,14 @@ static void *c2c_he_zalloc(size_t size)
if (!c2c_he)
return NULL;
+ c2c_he->cpuset = bitmap_alloc(c2c.cpus_cnt);
+ if (!c2c_he->cpuset)
+ return NULL;
+
+ c2c_he->node_stats = zalloc(c2c.nodes_cnt * sizeof(*c2c_he->node_stats));
+ if (!c2c_he->node_stats)
+ return NULL;
+
return &c2c_he->he;
}
@@ -57,6 +74,8 @@ static void c2c_he_free(void *he)
free(c2c_he->hists);
}
+ free(c2c_he->cpuset);
+ free(c2c_he->node_stats);
free(c2c_he);
}
@@ -91,6 +110,16 @@ he__get_c2c_hists(struct hist_entry *he,
return hists;
}
+static void c2c_he__set_cpu(struct c2c_hist_entry *c2c_he,
+ struct perf_sample *sample)
+{
+ if (WARN_ONCE(sample->cpu == (unsigned int) -1,
+ "WARNING: no sample cpu value"))
+ return;
+
+ set_bit(sample->cpu, c2c_he->cpuset);
+}
+
static int process_sample_event(struct perf_tool *tool __maybe_unused,
union perf_event *event,
struct perf_sample *sample,
@@ -131,10 +160,23 @@ static int process_sample_event(struct perf_tool *tool __maybe_unused,
c2c_add_stats(&c2c_he->stats, &stats);
c2c_add_stats(&c2c_hists->stats, &stats);
+ c2c_he__set_cpu(c2c_he, sample);
+
hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
ret = hist_entry__append_callchain(he, sample);
if (!ret) {
+ /*
+ * There's already been warning about missing
+ * sample's cpu value. Let's account all to
+ * node 0 in this case, without any further
+ * warning.
+ *
+ * Doing node stats only for single callchain data.
+ */
+ int cpu = sample->cpu == (unsigned int) -1 ? 0 : sample->cpu;
+ int node = c2c.cpu2node[cpu];
+
mi = mi_dup;
mi_dup = memdup(mi, sizeof(*mi));
@@ -154,6 +196,9 @@ static int process_sample_event(struct perf_tool *tool __maybe_unused,
c2c_he = container_of(he, struct c2c_hist_entry, he);
c2c_add_stats(&c2c_he->stats, &stats);
c2c_add_stats(&c2c_hists->stats, &stats);
+ c2c_add_stats(&c2c_he->node_stats[node], &stats);
+
+ c2c_he__set_cpu(c2c_he, sample);
hists__inc_nr_samples(&c2c_hists->hists, he->filtered);
ret = hist_entry__append_callchain(he, sample);
@@ -826,6 +871,97 @@ pid_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
return left->thread->pid_ - right->thread->pid_;
}
+static int64_t
+empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
+ struct hist_entry *left __maybe_unused,
+ struct hist_entry *right __maybe_unused)
+{
+ return 0;
+}
+
+static int
+node_entry(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *hpp,
+ struct hist_entry *he)
+{
+ struct c2c_hist_entry *c2c_he;
+ bool first = true;
+ int node;
+ int ret = 0;
+
+ c2c_he = container_of(he, struct c2c_hist_entry, he);
+
+ for (node = 0; node < c2c.nodes_cnt; node++) {
+ DECLARE_BITMAP(set, c2c.cpus_cnt);
+
+ bitmap_zero(set, c2c.cpus_cnt);
+ bitmap_and(set, c2c_he->cpuset, c2c.nodes[node], c2c.cpus_cnt);
+
+ if (!bitmap_weight(set, c2c.cpus_cnt)) {
+ if (c2c.node_info == 1) {
+ ret = scnprintf(hpp->buf, hpp->size, "%21s", " ");
+ advance_hpp(hpp, ret);
+ }
+ continue;
+ }
+
+ if (!first) {
+ ret = scnprintf(hpp->buf, hpp->size, " ");
+ advance_hpp(hpp, ret);
+ }
+
+ switch (c2c.node_info) {
+ case 0:
+ ret = scnprintf(hpp->buf, hpp->size, "%2d", node);
+ advance_hpp(hpp, ret);
+ break;
+ case 1:
+ {
+ int num = bitmap_weight(c2c_he->cpuset, c2c.cpus_cnt);
+ struct c2c_stats *stats = &c2c_he->node_stats[node];
+
+ ret = scnprintf(hpp->buf, hpp->size, "%2d{%2d ", node, num);
+ advance_hpp(hpp, ret);
+
+
+ if (c2c_he->stats.rmt_hitm > 0) {
+ ret = scnprintf(hpp->buf, hpp->size, "%5.1f%% ",
+ percent(stats->rmt_hitm, c2c_he->stats.rmt_hitm));
+ } else {
+ ret = scnprintf(hpp->buf, hpp->size, "%6s ", "n/a");
+ }
+
+ advance_hpp(hpp, ret);
+
+ if (c2c_he->stats.store > 0) {
+ ret = scnprintf(hpp->buf, hpp->size, "%5.1f%%}",
+ percent(stats->store, c2c_he->stats.store));
+ } else {
+ ret = scnprintf(hpp->buf, hpp->size, "%6s}", "n/a");
+ }
+
+ advance_hpp(hpp, ret);
+ break;
+ }
+ case 2:
+ ret = scnprintf(hpp->buf, hpp->size, "%2d{", node);
+ advance_hpp(hpp, ret);
+
+ ret = bitmap_scnprintf(set, c2c.cpus_cnt, hpp->buf, hpp->size);
+ advance_hpp(hpp, ret);
+
+ ret = scnprintf(hpp->buf, hpp->size, "}");
+ advance_hpp(hpp, ret);
+ break;
+ default:
+ break;
+ }
+
+ first = false;
+ }
+
+ return 0;
+}
+
#define HEADER_LOW(__h) \
{ \
.line[1] = { \
@@ -1115,6 +1251,19 @@ static struct c2c_dimension dim_dso = {
.se = &sort_dso,
};
+static struct c2c_header header_node[3] = {
+ HEADER_LOW("Node"),
+ HEADER_LOW("Node{cpus %hitms %stores}"),
+ HEADER_LOW("Node{cpu list}"),
+};
+
+static struct c2c_dimension dim_node = {
+ .name = "node",
+ .cmp = empty_cmp,
+ .entry = node_entry,
+ .width = 4,
+};
+
#undef HEADER_LOW
#undef HEADER_BOTH
#undef HEADER_SPAN
@@ -1153,6 +1302,7 @@ static struct c2c_dimension *dimensions[] = {
&dim_tid,
&dim_symbol,
&dim_dso,
+ &dim_node,
NULL,
};
@@ -1379,6 +1529,68 @@ static int resort_cl_cb(struct hist_entry *he)
return 0;
}
+static void setup_nodes_header(void)
+{
+ dim_node.header = header_node[c2c.node_info];
+}
+
+static int setup_nodes(struct perf_session *session)
+{
+ struct numa_node *n;
+ unsigned long **nodes;
+ int node, cpu;
+ int *cpu2node;
+
+ if (c2c.node_info > 2)
+ c2c.node_info = 2;
+
+ c2c.nodes_cnt = session->header.env.nr_numa_nodes;
+ c2c.cpus_cnt = session->header.env.nr_cpus_online;
+
+ n = session->header.env.numa_nodes;
+ if (!n)
+ return -EINVAL;
+
+ nodes = zalloc(sizeof(unsigned long *) * c2c.nodes_cnt);
+ if (!nodes)
+ return -ENOMEM;
+
+ c2c.nodes = nodes;
+
+ cpu2node = zalloc(sizeof(int) * c2c.cpus_cnt);
+ if (!cpu2node)
+ return -ENOMEM;
+
+ for (cpu = 0; cpu < c2c.cpus_cnt; cpu++)
+ cpu2node[cpu] = -1;
+
+ c2c.cpu2node = cpu2node;
+
+ for (node = 0; node < c2c.nodes_cnt; node++) {
+ struct cpu_map *map = n[node].map;
+ unsigned long *set;
+
+ set = bitmap_alloc(c2c.cpus_cnt);
+ if (!set)
+ return -ENOMEM;
+
+ for (cpu = 0; cpu < map->nr; cpu++) {
+ set_bit(map->map[cpu], set);
+
+ if (WARN_ONCE(cpu2node[map->map[cpu]] != -1, "node/cpu topology bug"))
+ return -EINVAL;
+
+ cpu2node[map->map[cpu]] = node;
+ }
+
+ nodes[node] = set;
+ }
+
+ setup_nodes_header();
+ return 0;
+}
+
+
static int perf_c2c__report(int argc, const char **argv)
{
struct perf_session *session;
@@ -1393,6 +1605,8 @@ static int perf_c2c__report(int argc, const char **argv)
"be more verbose (show counter open errors, etc)"),
OPT_STRING('i', "input", &input_name, "file",
"the input file to process"),
+ OPT_INCR('N', "node-info", &c2c.node_info,
+ "show extra node info in report (repeat for more info)"),
OPT_END()
};
int err = 0;
@@ -1418,6 +1632,11 @@ static int perf_c2c__report(int argc, const char **argv)
pr_debug("No memory for session\n");
goto out;
}
+ err = setup_nodes(session);
+ if (err) {
+ pr_err("Failed setup nodes\n");
+ goto out;
+ }
if (symbol__init(&session->header.env) < 0)
goto out_session;
--
2.7.4
next prev parent reply other threads:[~2016-09-19 13:12 UTC|newest]
Thread overview: 85+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-09-19 13:09 [PATCHv3 00/61] perf c2c: Add new tool to analyze cacheline contention on NUMA systems Jiri Olsa
2016-09-19 13:09 ` [PATCH 01/61] perf symbols: Do not open device files again Jiri Olsa
2016-09-20 15:28 ` Arnaldo Carvalho de Melo
2016-09-20 15:36 ` Jiri Olsa
2016-09-20 16:12 ` [PATCHv2 01/61] perf symbols: Do not open device files Jiri Olsa
2016-09-20 21:45 ` [tip:perf/core] " tip-bot for Jiri Olsa
2016-09-19 13:09 ` [PATCH 02/61] perf tools: Remove superfluous initialization of weight Jiri Olsa
2016-09-21 15:15 ` Arnaldo Carvalho de Melo
2016-09-23 5:24 ` [tip:perf/core] perf evsel: " tip-bot for Jiri Olsa
2016-09-19 13:09 ` [PATCH 03/61] perf tools: Make hist_entry__snprintf work over struct perf_hpp_list Jiri Olsa
2016-09-21 15:14 ` Arnaldo Carvalho de Melo
2016-09-21 15:30 ` Jiri Olsa
2016-09-19 13:09 ` [PATCH 04/61] perf tools: Use bigger buffer for stdio headers Jiri Olsa
2016-09-21 15:15 ` Arnaldo Carvalho de Melo
2016-09-23 5:25 ` [tip:perf/core] perf hists: " tip-bot for Jiri Olsa
2016-09-19 13:09 ` [PATCH 05/61] perf tools: Introduce c2c_decode_stats function Jiri Olsa
2016-09-19 17:15 ` Nilay Vaish
2016-09-19 18:04 ` Joe Mario
[not found] ` <CACDz1GupJi3kcDx6zBK68KtpL=Q9hJvUFvHCdtMirMyuuuyMOQ@mail.gmail.com>
2016-09-21 9:18 ` Jiri Olsa
2016-09-21 15:16 ` Don Zickus
2016-09-21 15:32 ` Jiri Olsa
2016-09-19 13:09 ` [PATCH 06/61] perf tools: Introduce c2c_add_stats function Jiri Olsa
2016-09-19 13:09 ` [PATCH 07/61] perf tools: Make reset_dimensions global Jiri Olsa
2016-09-19 13:09 ` [PATCH 08/61] perf tools: Make output_field_add and sort_dimension__add global Jiri Olsa
2016-09-19 13:09 ` [PATCH 09/61] perf tools: Make several sorting functions global Jiri Olsa
2016-09-19 13:09 ` [PATCH 10/61] perf tools: Make several display " Jiri Olsa
2016-09-19 13:09 ` [PATCH 11/61] perf tools: Make hist_entry__snprintf function global Jiri Olsa
2016-09-19 13:09 ` [PATCH 12/61] perf tools: Make hists__fprintf_headers " Jiri Olsa
2016-09-19 13:09 ` [PATCH 13/61] perf c2c: Add c2c command Jiri Olsa
2016-09-19 13:09 ` [PATCH 14/61] perf c2c: Add record subcommand Jiri Olsa
2016-09-19 13:09 ` [PATCH 15/61] perf c2c: Add report subcommand Jiri Olsa
2016-09-19 13:09 ` [PATCH 16/61] perf c2c report: Add dimension support Jiri Olsa
2016-09-19 13:09 ` [PATCH 17/61] perf c2c report: Add sort_entry " Jiri Olsa
2016-09-19 13:09 ` [PATCH 18/61] perf c2c report: Fallback to standard dimensions Jiri Olsa
2016-09-19 13:09 ` [PATCH 19/61] perf c2c report: Add sample processing Jiri Olsa
2016-09-19 13:09 ` [PATCH 20/61] perf c2c report: Add cacheline hists processing Jiri Olsa
2016-09-19 13:09 ` [PATCH 21/61] perf c2c report: Decode c2c_stats for hist entries Jiri Olsa
2016-09-19 13:09 ` [PATCH 22/61] perf c2c report: Add header macros Jiri Olsa
2016-09-19 13:09 ` [PATCH 23/61] perf c2c report: Add dcacheline dimension key Jiri Olsa
2016-09-19 13:09 ` [PATCH 24/61] perf c2c report: Add offset " Jiri Olsa
2016-09-19 13:09 ` [PATCH 25/61] perf c2c report: Add iaddr " Jiri Olsa
2016-09-19 13:09 ` [PATCH 26/61] perf c2c report: Add hitm related dimension keys Jiri Olsa
2016-09-19 13:09 ` [PATCH 27/61] perf c2c report: Add stores " Jiri Olsa
2016-09-19 13:09 ` [PATCH 28/61] perf c2c report: Add loads " Jiri Olsa
2016-09-19 13:09 ` [PATCH 29/61] perf c2c report: Add llc and remote " Jiri Olsa
2016-09-19 13:09 ` [PATCH 30/61] perf c2c report: Add llc load miss dimension key Jiri Olsa
2016-09-19 13:09 ` [PATCH 31/61] perf c2c report: Add total record sort key Jiri Olsa
2016-09-19 13:09 ` [PATCH 32/61] perf c2c report: Add total loads " Jiri Olsa
2016-09-19 13:09 ` [PATCH 33/61] perf c2c report: Add hitm percent " Jiri Olsa
2016-09-19 13:09 ` [PATCH 34/61] perf c2c report: Add hitm/store percent related sort keys Jiri Olsa
2016-09-19 13:09 ` [PATCH 35/61] perf c2c report: Add dram " Jiri Olsa
2016-09-19 13:09 ` [PATCH 36/61] perf c2c report: Add pid sort key Jiri Olsa
2016-09-19 13:09 ` [PATCH 37/61] perf c2c report: Add tid " Jiri Olsa
2016-09-19 13:09 ` [PATCH 38/61] perf c2c report: Add symbol and dso sort keys Jiri Olsa
2016-09-19 13:09 ` Jiri Olsa [this message]
2016-09-19 13:09 ` [PATCH 40/61] perf c2c report: Add stats related " Jiri Olsa
2016-09-19 13:09 ` [PATCH 41/61] perf c2c report: Add cpu cnt sort key Jiri Olsa
2016-09-19 13:09 ` [PATCH 42/61] perf c2c report: Add src line " Jiri Olsa
2016-09-19 13:09 ` [PATCH 43/61] perf c2c report: Setup number of header lines for hists Jiri Olsa
2016-09-19 13:09 ` [PATCH 44/61] perf c2c report: Set final resort fields Jiri Olsa
2016-09-19 13:09 ` [PATCH 45/61] perf c2c report: Add stdio output support Jiri Olsa
2016-09-19 13:09 ` [PATCH 46/61] perf c2c report: Add main browser Jiri Olsa
2016-09-19 13:09 ` [PATCH 47/61] perf c2c report: Add cacheline browser Jiri Olsa
2016-09-20 20:10 ` Kim Phillips
2016-09-21 8:21 ` Jiri Olsa
2016-09-21 12:55 ` Jiri Olsa
2016-09-21 19:35 ` Kim Phillips
2016-09-19 13:09 ` [PATCH 48/61] perf c2c report: Add global stats stdio output Jiri Olsa
2016-09-19 13:09 ` [PATCH 49/61] perf c2c report: Add shared cachelines " Jiri Olsa
2016-09-19 13:09 ` [PATCH 50/61] perf c2c report: Add c2c related " Jiri Olsa
2016-09-19 13:10 ` [PATCH 51/61] perf c2c report: Allow to report callchains Jiri Olsa
2016-09-19 13:10 ` [PATCH 52/61] perf c2c report: Limit the cachelines table entries Jiri Olsa
2016-09-19 13:10 ` [PATCH 53/61] perf c2c report: Add support to choose local HITMs Jiri Olsa
2016-09-19 13:10 ` [PATCH 54/61] perf c2c report: Allow to set cacheline sort fields Jiri Olsa
2016-09-19 13:10 ` [PATCH 55/61] perf c2c report: Recalc width of global sort entries Jiri Olsa
2016-09-19 13:10 ` [PATCH 56/61] perf c2c report: Add cacheline index entry Jiri Olsa
2016-09-19 13:10 ` [PATCH 57/61] perf c2c report: Add support to manage symbol name length Jiri Olsa
2016-09-19 13:10 ` [PATCH 58/61] perf c2c report: Iterate node display in browser Jiri Olsa
2016-09-19 13:10 ` [PATCH 59/61] perf c2c report: Add help windows Jiri Olsa
2016-09-19 13:10 ` [PATCH 60/61] perf c2c: Add man page and credits Jiri Olsa
2016-09-19 13:10 ` [PATCH 61/61] perf tools: Fix width computation for srcline sort entry Jiri Olsa
2016-09-19 14:33 ` Arnaldo Carvalho de Melo
2016-09-19 14:49 ` Jiri Olsa
2016-09-19 14:57 ` Arnaldo Carvalho de Melo
2016-09-20 21:43 ` [tip:perf/core] perf hists: " tip-bot for 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=1474290610-23241-40-git-send-email-jolsa@kernel.org \
--to=jolsa@kernel.org \
--cc=a.p.zijlstra@chello.nl \
--cc=acme@kernel.org \
--cc=andi@firstfloor.org \
--cc=dsahern@gmail.com \
--cc=dzickus@redhat.com \
--cc=jmario@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@kernel.org \
--cc=namhyung@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;
as well as URLs for NNTP newsgroup(s).