[PATCH v3 0/3] Improvements for kallsyms__parse

[PATCH v3 0/3] Improvements for kallsyms__parse

[Ian Rogers]

kallsyms__parse is called 4 times during perf record startup. Add a
benchmark to measure its performance. Transition it to using the api
io.h buffered reading, improving performance by ~8% or saving ~5% from
perf record start up time.

v3. Fixed a link error report in hex2u64 by jolsa@redhat.com.
v2. Fix the err return value for success, error reported by
    jolsa@redhat.com. Add clean-up for hex2u64.

Ian Rogers (3):
  perf bench: add kallsyms parsing
  lib kallsyms: parse using io api
  lib kallsyms: move hex2u64 out of header

 tools/lib/api/io.h                |  3 ++
 tools/lib/symbol/kallsyms.c       | 86 ++++++++++++++-----------------
 tools/lib/symbol/kallsyms.h       |  2 -
 tools/perf/bench/Build            |  1 +
 tools/perf/bench/bench.h          |  1 +
 tools/perf/bench/kallsyms-parse.c | 75 +++++++++++++++++++++++++++
 tools/perf/builtin-bench.c        |  1 +
 tools/perf/util/symbol.c          | 14 +++++
 8 files changed, 133 insertions(+), 50 deletions(-)
 create mode 100644 tools/perf/bench/kallsyms-parse.c

-- 
2.26.2.526.g744177e7f7-goog

[PATCH v3 1/3] perf bench: add kallsyms parsing

[Ian Rogers]

Add a benchmark for kallsyms parsing. Example output:

  Running 'internals/kallsyms-parse' benchmark:
  Average kallsyms__parse took: 103.971 ms (+- 0.121 ms)

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/bench/Build            |  1 +
 tools/perf/bench/bench.h          |  1 +
 tools/perf/bench/kallsyms-parse.c | 75 +++++++++++++++++++++++++++++++
 tools/perf/builtin-bench.c        |  1 +
 4 files changed, 78 insertions(+)
 create mode 100644 tools/perf/bench/kallsyms-parse.c

diff --git a/tools/perf/bench/Build b/tools/perf/bench/Build
index 042827385c87..768e408757a0 100644
--- a/tools/perf/bench/Build
+++ b/tools/perf/bench/Build
@@ -9,6 +9,7 @@ perf-y += futex-lock-pi.o
 perf-y += epoll-wait.o
 perf-y += epoll-ctl.o
 perf-y += synthesize.o
+perf-y += kallsyms-parse.o
 
 perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-lib.o
 perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-asm.o
diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h
index 4d669c803237..61cae4966cae 100644
--- a/tools/perf/bench/bench.h
+++ b/tools/perf/bench/bench.h
@@ -44,6 +44,7 @@ int bench_futex_lock_pi(int argc, const char **argv);
 int bench_epoll_wait(int argc, const char **argv);
 int bench_epoll_ctl(int argc, const char **argv);
 int bench_synthesize(int argc, const char **argv);
+int bench_kallsyms_parse(int argc, const char **argv);
 
 #define BENCH_FORMAT_DEFAULT_STR	"default"
 #define BENCH_FORMAT_DEFAULT		0
diff --git a/tools/perf/bench/kallsyms-parse.c b/tools/perf/bench/kallsyms-parse.c
new file mode 100644
index 000000000000..2b0d0f980ae9
--- /dev/null
+++ b/tools/perf/bench/kallsyms-parse.c
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Benchmark of /proc/kallsyms parsing.
+ *
+ * Copyright 2020 Google LLC.
+ */
+#include <stdlib.h>
+#include "bench.h"
+#include "../util/stat.h"
+#include <linux/time64.h>
+#include <subcmd/parse-options.h>
+#include <symbol/kallsyms.h>
+
+static unsigned int iterations = 100;
+
+static const struct option options[] = {
+	OPT_UINTEGER('i', "iterations", &iterations,
+		"Number of iterations used to compute average"),
+	OPT_END()
+};
+
+static const char *const bench_usage[] = {
+	"perf bench internals kallsyms-parse <options>",
+	NULL
+};
+
+static int bench_process_symbol(void *arg __maybe_unused,
+				const char *name __maybe_unused,
+				char type __maybe_unused,
+				u64 start __maybe_unused)
+{
+	return 0;
+}
+
+static int do_kallsyms_parse(void)
+{
+	struct timeval start, end, diff;
+	u64 runtime_us;
+	unsigned int i;
+	double time_average, time_stddev;
+	int err;
+	struct stats time_stats;
+
+	init_stats(&time_stats);
+
+	for (i = 0; i < iterations; i++) {
+		gettimeofday(&start, NULL);
+		err = kallsyms__parse("/proc/kallsyms", NULL,
+				bench_process_symbol);
+		if (err)
+			return err;
+
+		gettimeofday(&end, NULL);
+		timersub(&end, &start, &diff);
+		runtime_us = diff.tv_sec * USEC_PER_SEC + diff.tv_usec;
+		update_stats(&time_stats, runtime_us);
+	}
+
+	time_average = avg_stats(&time_stats) / USEC_PER_MSEC;
+	time_stddev = stddev_stats(&time_stats) / USEC_PER_MSEC;
+	printf("  Average kallsyms__parse took: %.3f ms (+- %.3f ms)\n",
+		time_average, time_stddev);
+	return 0;
+}
+
+int bench_kallsyms_parse(int argc, const char **argv)
+{
+	argc = parse_options(argc, argv, options, bench_usage, 0);
+	if (argc) {
+		usage_with_options(bench_usage, options);
+		exit(EXIT_FAILURE);
+	}
+
+	return do_kallsyms_parse();
+}
diff --git a/tools/perf/builtin-bench.c b/tools/perf/builtin-bench.c
index 11c79a8d85d6..083273209c88 100644
--- a/tools/perf/builtin-bench.c
+++ b/tools/perf/builtin-bench.c
@@ -78,6 +78,7 @@ static struct bench epoll_benchmarks[] = {
 
 static struct bench internals_benchmarks[] = {
 	{ "synthesize", "Benchmark perf event synthesis",	bench_synthesize	},
+	{ "kallsyms-parse", "Benchmark kallsyms parsing",	bench_kallsyms_parse	},
 	{ NULL,		NULL,					NULL			}
 };
 
-- 
2.26.2.526.g744177e7f7-goog

[PATCH v3 2/3] lib kallsyms: parse using io api

[Ian Rogers]

Perf record will call kallsyms__parse 4 times during startup and process
megabytes of data. This changes kallsyms__parse to use the io library
rather than fgets to improve performance of the user code by over 8%.

Before:
  Running 'internals/kallsyms-parse' benchmark:
  Average kallsyms__parse took: 103.988 ms (+- 0.203 ms)
After:
  Running 'internals/kallsyms-parse' benchmark:
  Average kallsyms__parse took: 95.571 ms (+- 0.006 ms)

For a workload like:
$ perf record /bin/true
Run under 'perf record -e cycles:u -g' the time goes from:
Before
30.10%     1.67%  perf     perf                [.] kallsyms__parse
After
25.55%    20.04%  perf     perf                [.] kallsyms__parse
So a little under 5% of the start-up time is removed. A lot of what
remains is on the kernel side, but caching kallsyms within perf would
at least impact memory footprint.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/lib/api/io.h          |  3 ++
 tools/lib/symbol/kallsyms.c | 81 +++++++++++++++++++------------------
 2 files changed, 45 insertions(+), 39 deletions(-)

diff --git a/tools/lib/api/io.h b/tools/lib/api/io.h
index b7e55b5f8a4a..777c20f6b604 100644
--- a/tools/lib/api/io.h
+++ b/tools/lib/api/io.h
@@ -7,6 +7,9 @@
 #ifndef __API_IO__
 #define __API_IO__
 
+#include <stdlib.h>
+#include <unistd.h>
+
 struct io {
 	/* File descriptor being read/ */
 	int fd;
diff --git a/tools/lib/symbol/kallsyms.c b/tools/lib/symbol/kallsyms.c
index 1a7a9f877095..1f2f519a4e3f 100644
--- a/tools/lib/symbol/kallsyms.c
+++ b/tools/lib/symbol/kallsyms.c
@@ -1,7 +1,9 @@
 // SPDX-License-Identifier: GPL-2.0
 #include "symbol/kallsyms.h"
+#include "api/io.h"
 #include <stdio.h>
-#include <stdlib.h>
+#include <sys/stat.h>
+#include <fcntl.h>
 
 u8 kallsyms2elf_type(char type)
 {
@@ -28,61 +30,62 @@ int hex2u64(const char *ptr, u64 *long_val)
 	return p - ptr;
 }
 
+static void read_to_eol(struct io *io)
+{
+	int ch;
+
+	for (;;) {
+		ch = io__get_char(io);
+		if (ch < 0 || ch == '\n')
+			return;
+	}
+}
+
 int kallsyms__parse(const char *filename, void *arg,
 		    int (*process_symbol)(void *arg, const char *name,
 					  char type, u64 start))
 {
-	char *line = NULL;
-	size_t n;
-	int err = -1;
-	FILE *file = fopen(filename, "r");
+	struct io io;
+	char bf[BUFSIZ];
+	int err;
 
-	if (file == NULL)
-		goto out_failure;
+	io.fd = open(filename, O_RDONLY, 0);
 
-	err = 0;
-
-	while (!feof(file)) {
-		u64 start;
-		int line_len, len;
-		char symbol_type;
-		char *symbol_name;
-
-		line_len = getline(&line, &n, file);
-		if (line_len < 0 || !line)
-			break;
+	if (io.fd < 0)
+		return -1;
 
-		line[--line_len] = '\0'; /* \n */
+	io__init(&io, io.fd, bf, sizeof(bf));
 
-		len = hex2u64(line, &start);
+	err = 0;
+	while (!io.eof) {
+		__u64 start;
+		int ch;
+		size_t i;
+		char symbol_type;
+		char symbol_name[KSYM_NAME_LEN + 1];
 
-		/* Skip the line if we failed to parse the address. */
-		if (!len)
+		if (io__get_hex(&io, &start) != ' ') {
+			read_to_eol(&io);
 			continue;
-
-		len++;
-		if (len + 2 >= line_len)
+		}
+		symbol_type = io__get_char(&io);
+		if (io__get_char(&io) != ' ') {
+			read_to_eol(&io);
 			continue;
-
-		symbol_type = line[len];
-		len += 2;
-		symbol_name = line + len;
-		len = line_len - len;
-
-		if (len >= KSYM_NAME_LEN) {
-			err = -1;
-			break;
 		}
+		for (i = 0; i < sizeof(symbol_name); i++) {
+			ch = io__get_char(&io);
+			if (ch < 0 || ch == '\n')
+				break;
+			symbol_name[i]  = ch;
+		}
+		symbol_name[i]  = '\0';
 
 		err = process_symbol(arg, symbol_name, symbol_type, start);
 		if (err)
 			break;
 	}
 
-	free(line);
-	fclose(file);
+	close(io.fd);
 	return err;
-
-out_failure:
-	return -1;
 }
-- 
2.26.2.526.g744177e7f7-goog

[PATCH v3 3/3] lib kallsyms: move hex2u64 out of header

[Ian Rogers]

hex2u64 is a helper that's out of place in kallsyms.h as not being kallsyms
related. Move from kallsyms.h to the only user.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/lib/symbol/kallsyms.c | 13 -------------
 tools/lib/symbol/kallsyms.h |  2 --
 tools/perf/util/symbol.c    | 14 ++++++++++++++
 3 files changed, 14 insertions(+), 15 deletions(-)

diff --git a/tools/lib/symbol/kallsyms.c b/tools/lib/symbol/kallsyms.c
index 1f2f519a4e3f..e335ac2b9e19 100644
--- a/tools/lib/symbol/kallsyms.c
+++ b/tools/lib/symbol/kallsyms.c
@@ -17,19 +17,6 @@ bool kallsyms__is_function(char symbol_type)
 	return symbol_type == 'T' || symbol_type == 'W';
 }
 
-/*
- * While we find nice hex chars, build a long_val.
- * Return number of chars processed.
- */
-int hex2u64(const char *ptr, u64 *long_val)
-{
-	char *p;
-
-	*long_val = strtoull(ptr, &p, 16);
-
-	return p - ptr;
-}
-
 static void read_to_eol(struct io *io)
 {
 	int ch;
diff --git a/tools/lib/symbol/kallsyms.h b/tools/lib/symbol/kallsyms.h
index bd988f7b18d4..72ab9870454b 100644
--- a/tools/lib/symbol/kallsyms.h
+++ b/tools/lib/symbol/kallsyms.h
@@ -18,8 +18,6 @@ static inline u8 kallsyms2elf_binding(char type)
 	return isupper(type) ? STB_GLOBAL : STB_LOCAL;
 }
 
-int hex2u64(const char *ptr, u64 *long_val);
-
 u8 kallsyms2elf_type(char type);
 
 bool kallsyms__is_function(char symbol_type);
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index 8f4300492dc7..381da6b39f89 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -566,6 +566,20 @@ void dso__sort_by_name(struct dso *dso)
 	return symbols__sort_by_name(&dso->symbol_names, &dso->symbols);
 }
 
+/*
+ * While we find nice hex chars, build a long_val.
+ * Return number of chars processed.
+ */
+static int hex2u64(const char *ptr, u64 *long_val)
+{
+	char *p;
+
+	*long_val = strtoull(ptr, &p, 16);
+
+	return p - ptr;
+}
+
+
 int modules__parse(const char *filename, void *arg,
 		   int (*process_module)(void *arg, const char *name,
 					 u64 start, u64 size))
-- 
2.26.2.526.g744177e7f7-goog