Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1319075 > unrolled thread

[PATCHSET 00/10] perf tools: Apply percent-limit to callchains (v2)

Back to article view | Back to linux.kernel

Contents

  [PATCHSET 00/10] perf tools: Apply percent-limit to callchains (v2) Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
    [PATCH 07/10] perf hists browser: Fix dump to show correct callchain style Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
    [PATCH 09/10] perf hists browser: Fix percent display in callchains Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
    [PATCH 04/10] perf report: Get rid of hist_entry__callchain_fprintf() Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
    [PATCH 01/10] perf hists: Fix min callchain hits calculation Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
    [PATCH 03/10] perf report: Apply --percent-limit to callchains also Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
    [PATCH 06/10] perf report: Fix percent display in callchains on --stdio Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
    [RFC/PATCH 10/10] perf tools: Change default calchain percent limit to 0.005% Namhyung Kim <namhyung@kernel.org> - 2016-01-27 16:50 +0100
      Re: [RFC/PATCH 10/10] perf tools: Change default calchain percent  limit to 0.005% Andi Kleen <andi@firstfloor.org> - 2016-01-27 17:10 +0100

#1319075 — [PATCHSET 00/10] perf tools: Apply percent-limit to callchains (v2)

Hello,

This patchset tries to implement percent limit to callchains which was
requested by Andi Kleen.  For some reason, limiting callchains by
(overhead) percentage didn't work well.  This patch fixes it and make
--percent-limit also works for callchains as well as hist entries.

 * Changes from v1)
  - fix insertion path instead of changing all UI code
  - show percent value even on single path (if needed)
  - change default callchain percent limit
  
This is available on 'perf/callchain-limit-v2' branch in my tree:

  git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git

Any comments are welcome,

Thanks,
Namhyung
*** BLURB HERE ***

Namhyung Kim (10):
  perf hists: Fix min callchain hits calculation
  perf hists: Update hists' total period when adding entries
  perf report: Apply --percent-limit to callchains also
  perf report: Get rid of hist_entry__callchain_fprintf()
  perf tools: Pass parent_samples to __callchain__fprintf_graph()
  perf report: Fix percent display in callchains on --stdio
  perf hists browser: Fix dump to show correct callchain style
  perf hists browser: Pass parent_total to callchain print functions
  perf hists browser: Fix percent display in callchains
  perf tools: Change default calchain percent limit to 0.005%

 tools/perf/Documentation/perf-report.txt |   2 +-
 tools/perf/builtin-report.c              |  11 ++-
 tools/perf/builtin-top.c                 |   2 +-
 tools/perf/ui/browsers/hists.c           | 113 +++++++++++++++++++------------
 tools/perf/ui/stdio/hist.c               |  72 ++++++++++----------
 tools/perf/util/hist.c                   |  24 +++++--
 tools/perf/util/util.c                   |   2 +-
 7 files changed, 136 insertions(+), 90 deletions(-)

-- 
2.6.4

[toc] | [next] | [standalone]

#1319078 — [PATCH 07/10] perf hists browser: Fix dump to show correct callchain style

The commit 8c430a348699 ("perf hists browser: Support folded
callchains") missed to update hist_browser__dump() so it always shows
graph-style callchains regardless of current setting.

To fix that, factor out callchain printing code and rename the existing
function which prints graph-style callchain.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/ui/browsers/hists.c | 73 ++++++++++++++++++++++++------------------
 1 file changed, 41 insertions(+), 32 deletions(-)

diff --git a/tools/perf/ui/browsers/hists.c b/tools/perf/ui/browsers/hists.c
index d07e6be42ab1..91804c5b0130 100644
--- a/tools/perf/ui/browsers/hists.c
+++ b/tools/perf/ui/browsers/hists.c
@@ -844,7 +844,7 @@ next:
 	return row - first_row;
 }
 
-static int hist_browser__show_callchain(struct hist_browser *browser,
+static int hist_browser__show_callchain_graph(struct hist_browser *browser,
 					struct rb_root *root, int level,
 					unsigned short row, u64 total,
 					print_callchain_entry_fn print,
@@ -898,7 +898,7 @@ static int hist_browser__show_callchain(struct hist_browser *browser,
 			else
 				new_total = total;
 
-			row += hist_browser__show_callchain(browser, &child->rb_root,
+			row += hist_browser__show_callchain_graph(browser, &child->rb_root,
 							    new_level, row, new_total,
 							    print, arg, is_output_full);
 		}
@@ -910,6 +910,43 @@ out:
 	return row - first_row;
 }
 
+static int hist_browser__show_callchain(struct hist_browser *browser,
+					struct hist_entry *entry, int level,
+					unsigned short row,
+					print_callchain_entry_fn print,
+					struct callchain_print_arg *arg,
+					check_output_full_fn is_output_full)
+{
+	u64 total = hists__total_period(entry->hists);
+	int printed;
+
+	if (callchain_param.mode == CHAIN_GRAPH_REL) {
+		if (symbol_conf.cumulate_callchain)
+			total = entry->stat_acc->period;
+		else
+			total = entry->stat.period;
+	}
+
+	if (callchain_param.mode == CHAIN_FLAT) {
+		printed = hist_browser__show_callchain_flat(browser,
+						&entry->sorted_chain, row, total,
+						print, arg, is_output_full);
+	} else if (callchain_param.mode == CHAIN_FOLDED) {
+		printed = hist_browser__show_callchain_folded(browser,
+						&entry->sorted_chain, row, total,
+						print, arg, is_output_full);
+	} else {
+		printed = hist_browser__show_callchain_graph(browser,
+						&entry->sorted_chain, level, row, total,
+						print, arg, is_output_full);
+	}
+
+	if (arg->is_current_entry)
+		browser->he_selection = entry;
+
+	return printed;
+}
+
 struct hpp_arg {
 	struct ui_browser *b;
 	char folded_sign;
@@ -1084,38 +1121,14 @@ static int hist_browser__show_entry(struct hist_browser *browser,
 		--row_offset;
 
 	if (folded_sign == '-' && row != browser->b.rows) {
-		u64 total = hists__total_period(entry->hists);
 		struct callchain_print_arg arg = {
 			.row_offset = row_offset,
 			.is_current_entry = current_entry,
 		};
 
-		if (callchain_param.mode == CHAIN_GRAPH_REL) {
-			if (symbol_conf.cumulate_callchain)
-				total = entry->stat_acc->period;
-			else
-				total = entry->stat.period;
-		}
-
-		if (callchain_param.mode == CHAIN_FLAT) {
-			printed += hist_browser__show_callchain_flat(browser,
-					&entry->sorted_chain, row, total,
+		printed += hist_browser__show_callchain(browser, entry, 1, row,
 					hist_browser__show_callchain_entry, &arg,
 					hist_browser__check_output_full);
-		} else if (callchain_param.mode == CHAIN_FOLDED) {
-			printed += hist_browser__show_callchain_folded(browser,
-					&entry->sorted_chain, row, total,
-					hist_browser__show_callchain_entry, &arg,
-					hist_browser__check_output_full);
-		} else {
-			printed += hist_browser__show_callchain(browser,
-					&entry->sorted_chain, 1, row, total,
-					hist_browser__show_callchain_entry, &arg,
-					hist_browser__check_output_full);
-		}
-
-		if (arg.is_current_entry)
-			browser->he_selection = entry;
 	}
 
 	return printed;
@@ -1380,15 +1393,11 @@ do_offset:
 static int hist_browser__fprintf_callchain(struct hist_browser *browser,
 					   struct hist_entry *he, FILE *fp)
 {
-	u64 total = hists__total_period(he->hists);
 	struct callchain_print_arg arg  = {
 		.fp = fp,
 	};
 
-	if (symbol_conf.cumulate_callchain)
-		total = he->stat_acc->period;
-
-	hist_browser__show_callchain(browser, &he->sorted_chain, 1, 0, total,
+	hist_browser__show_callchain(browser, he, 1, 0,
 				     hist_browser__fprintf_callchain_entry, &arg,
 				     hist_browser__check_dump_full);
 	return arg.printed;
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1319081 — [PATCH 09/10] perf hists browser: Fix percent display in callchains

When there's only a single callchain, perf doesn't print its percentage
in front of the symbols.  This is because it assumes that the percentage
is same as parents.  But if a percent limit is applied, it's possible
that there are actually a couple of child nodes but only one of them is
shown.  In this case it should display the percent to prevent
misunderstanding of its percentage is same as the parent's.

For example, let's see the following callchain.

  $ perf report --percent-limit 0.01 --tui
  ...
  -    0.06%  sleep    [kernel.vmlinux]    [k] kmem_cache_alloc_trace
       kmem_cache_alloc_trace
     - perf_event_mmap
        - 0.04% mmap_region
             do_mmap_pgoff
           - vm_mmap_pgoff
              + 0.02% sys_mmap_pgoff
              + 0.02% vm_mmap
           + 0.02% mprotect_fixup

Current code omits the percent if 'mmap_region' becomes the only node
when percent limit is set to 0.03%, its percent is not 0.06% but users
will assume it incorrectly.

Before:

  $ perf report --percent-limit 0.03 --tui
  ...
     0.06%  sleep    [kernel.vmlinux]    [k] kmem_cache_alloc_trace
       kmem_cache_alloc_trace
     - perf_event_mmap
        - mmap_region
          do_mmap_pgoff
          vm_mmap_pgoff

After:

  $ perf report --percent-limit 0.03 --stdio
  ...
     0.06%  sleep    [kernel.vmlinux]    [k] kmem_cache_alloc_trace
       kmem_cache_alloc_trace
     - perf_event_mmap
        - 0.04% mmap_region
             do_mmap_pgoff
             vm_mmap_pgoff

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/ui/browsers/hists.c | 24 +++++++++++++++++++-----
 1 file changed, 19 insertions(+), 5 deletions(-)

diff --git a/tools/perf/ui/browsers/hists.c b/tools/perf/ui/browsers/hists.c
index 726c79def88d..4019221f6755 100644
--- a/tools/perf/ui/browsers/hists.c
+++ b/tools/perf/ui/browsers/hists.c
@@ -657,10 +657,24 @@ static int hist_browser__show_callchain_list(struct hist_browser *browser,
 	return 1;
 }
 
+static bool check_percent_display(struct rb_node *node, u64 parent_total)
+{
+	struct callchain_node *child;
+
+	if (node == NULL)
+		return false;
+
+	if (rb_next(node))
+		return true;
+
+	child = rb_entry(node, struct callchain_node, rb_node);
+	return callchain_cumul_hits(child) != parent_total;
+}
+
 static int hist_browser__show_callchain_flat(struct hist_browser *browser,
 					     struct rb_root *root,
 					     unsigned short row, u64 total,
-					     u64 parent_total __maybe_unused,
+					     u64 parent_total,
 					     print_callchain_entry_fn print,
 					     struct callchain_print_arg *arg,
 					     check_output_full_fn is_output_full)
@@ -670,7 +684,7 @@ static int hist_browser__show_callchain_flat(struct hist_browser *browser,
 	bool need_percent;
 
 	node = rb_first(root);
-	need_percent = node && rb_next(node);
+	need_percent = check_percent_display(node, parent_total);
 
 	while (node) {
 		struct callchain_node *child = rb_entry(node, struct callchain_node, rb_node);
@@ -764,7 +778,7 @@ static char *hist_browser__folded_callchain_str(struct hist_browser *browser,
 static int hist_browser__show_callchain_folded(struct hist_browser *browser,
 					       struct rb_root *root,
 					       unsigned short row, u64 total,
-					       u64 parent_total __maybe_unused,
+					       u64 parent_total,
 					       print_callchain_entry_fn print,
 					       struct callchain_print_arg *arg,
 					       check_output_full_fn is_output_full)
@@ -774,7 +788,7 @@ static int hist_browser__show_callchain_folded(struct hist_browser *browser,
 	bool need_percent;
 
 	node = rb_first(root);
-	need_percent = node && rb_next(node);
+	need_percent = check_percent_display(node, parent_total);
 
 	while (node) {
 		struct callchain_node *child = rb_entry(node, struct callchain_node, rb_node);
@@ -863,7 +877,7 @@ static int hist_browser__show_callchain_graph(struct hist_browser *browser,
 		percent_total = parent_total;
 
 	node = rb_first(root);
-	need_percent = node && rb_next(node);
+	need_percent = check_percent_display(node, parent_total);
 
 	while (node) {
 		struct callchain_node *child = rb_entry(node, struct callchain_node, rb_node);
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1319084 — [PATCH 04/10] perf report: Get rid of hist_entry__callchain_fprintf()

It's just a wrapper function to align the start position ofcallchains to
'comm' of each thread if it's a first sort key.  But it doesn't not work
with tracepoint events and also with upcoming hierarchy view.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/ui/stdio/hist.c | 27 ++-------------------------
 1 file changed, 2 insertions(+), 25 deletions(-)

diff --git a/tools/perf/ui/stdio/hist.c b/tools/perf/ui/stdio/hist.c
index 387110d50b00..8e25f7dd6e84 100644
--- a/tools/perf/ui/stdio/hist.c
+++ b/tools/perf/ui/stdio/hist.c
@@ -349,30 +349,6 @@ static size_t hist_entry_callchain__fprintf(struct hist_entry *he,
 	return 0;
 }
 
-static size_t hist_entry__callchain_fprintf(struct hist_entry *he,
-					    struct hists *hists,
-					    FILE *fp)
-{
-	int left_margin = 0;
-	u64 total_period = hists->stats.total_period;
-
-	if (field_order == NULL && (sort_order == NULL ||
-				    !prefixcmp(sort_order, "comm"))) {
-		struct perf_hpp_fmt *fmt;
-
-		perf_hpp__for_each_format(fmt) {
-			if (!perf_hpp__is_sort_entry(fmt))
-				continue;
-
-			/* must be 'comm' sort entry */
-			left_margin = fmt->width(fmt, NULL, hists_to_evsel(hists));
-			left_margin -= thread__comm_len(he->thread);
-			break;
-		}
-	}
-	return hist_entry_callchain__fprintf(he, total_period, left_margin, fp);
-}
-
 static int hist_entry__snprintf(struct hist_entry *he, struct perf_hpp *hpp)
 {
 	const char *sep = symbol_conf.field_sep;
@@ -418,6 +394,7 @@ static int hist_entry__fprintf(struct hist_entry *he, size_t size,
 		.buf		= bf,
 		.size		= size,
 	};
+	u64 total_period = hists->stats.total_period;
 
 	if (size == 0 || size > bfsz)
 		size = hpp.size = bfsz;
@@ -427,7 +404,7 @@ static int hist_entry__fprintf(struct hist_entry *he, size_t size,
 	ret = fprintf(fp, "%s\n", bf);
 
 	if (symbol_conf.use_callchain)
-		ret += hist_entry__callchain_fprintf(he, hists, fp);
+		ret += hist_entry_callchain__fprintf(he, total_period, 0, fp);
 
 	return ret;
 }
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1319085 — [PATCH 01/10] perf hists: Fix min callchain hits calculation

The total period should be get using hists__total_period() since it
takes filtered entries into account.  In addition, if callchain mode is
'fractal', the total period should be the entry's period.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/util/hist.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c
index 81ce0aff69d1..b96194676c91 100644
--- a/tools/perf/util/hist.c
+++ b/tools/perf/util/hist.c
@@ -1163,9 +1163,18 @@ static void __hists__insert_output_entry(struct rb_root *entries,
 	struct rb_node *parent = NULL;
 	struct hist_entry *iter;
 
-	if (use_callchain)
+	if (use_callchain) {
+		if (callchain_param.mode == CHAIN_GRAPH_REL) {
+			u64 total = he->stat.period;
+
+			if (symbol_conf.cumulate_callchain)
+				total = he->stat_acc->period;
+
+			min_callchain_hits = total * (callchain_param.min_percent / 100);
+		}
 		callchain_param.sort(&he->sorted_chain, he->callchain,
 				      min_callchain_hits, &callchain_param);
+	}
 
 	while (*p != NULL) {
 		parent = *p;
@@ -1195,7 +1204,7 @@ void hists__output_resort(struct hists *hists, struct ui_progress *prog)
 	else
 		use_callchain = symbol_conf.use_callchain;
 
-	min_callchain_hits = hists->stats.total_period * (callchain_param.min_percent / 100);
+	min_callchain_hits = hists__total_period(hists) * (callchain_param.min_percent / 100);
 
 	if (sort__need_collapse)
 		root = &hists->entries_collapsed;
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1319086 — [PATCH 03/10] perf report: Apply --percent-limit to callchains also

Currently --percent-limit option only works for hist entries.  However
it'd be better to have same effect to callchains as well

Requested-by: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/builtin-report.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 2bf537f190a0..72ed0b46d5a1 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -75,7 +75,10 @@ static int report__config(const char *var, const char *value, void *cb)
 		return 0;
 	}
 	if (!strcmp(var, "report.percent-limit")) {
-		rep->min_percent = strtof(value, NULL);
+		double pcnt = strtof(value, NULL);
+
+		rep->min_percent = pcnt;
+		callchain_param.min_percent = pcnt;
 		return 0;
 	}
 	if (!strcmp(var, "report.children")) {
@@ -633,8 +636,10 @@ parse_percent_limit(const struct option *opt, const char *str,
 		    int unset __maybe_unused)
 {
 	struct report *rep = opt->value;
+	double pcnt = strtof(str, NULL);
 
-	rep->min_percent = strtof(str, NULL);
+	rep->min_percent = pcnt;
+	callchain_param.min_percent = pcnt;
 	return 0;
 }
 
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1319087 — [PATCH 06/10] perf report: Fix percent display in callchains on --stdio

When there's only a single callchain, perf doesn't print its percentage
in front of the symbols.  This is because it assumes that the percentage
is same as parents.  But if a percent limit is applied, it's possible
that there are actually a couple of child nodes but only one of them is
shown.  In this case it should display the percent to prevent
misunderstanding of its percentage is same as the parent's.

For example, let's see the following callchain.

  $ perf report -s comm --percent-limit 0.01 --stdio
  ...
     9.95%  swapper
            |
            |--7.57%--intel_idle
            |          cpuidle_enter_state
            |          cpuidle_enter
            |          call_cpuidle
            |          cpu_startup_entry
            |          |
            |          |--4.89%--start_secondary
            |          |
            |           --2.68%--rest_init
            |                     start_kernel
            |                     x86_64_start_reservations
            |                     x86_64_start_kernel
	    |
	    |--0.15%--__schedule
	    |          |
	    |          |--0.13%--schedule
	    |          |          schedule_preempt_disable
	    |          |          cpu_startup_entry
            |          |          |
            |          |          |--0.09%--start_secondary
            |          |          |
            |          |           --0.04%--rest_init
            |          |                     start_kernel
            |          |                     x86_64_start_reservations
            |          |                     x86_64_start_kernel
            |          |
            |           --0.01%--schedule_preempt_disabled
            |                     cpu_startup_entry
  ...

Current code omits the percent if 'intel_idle' becomes the only node
when percent limit is set to 0.5%, its percent is not 9.95% but users
will assume it incorrectly.

Before:

  $ perf report --percent-limit 0.5 --stdio
  ...
     9.95%  swapper
            |
            ---intel_idle
               cpuidle_enter_state
               cpuidle_enter
               call_cpuidle
               cpu_startup_entry
               |
               |--4.89%--start_secondary
               |
                --2.68%--rest_init
                          start_kernel
                          x86_64_start_reservations
                          x86_64_start_kernel

After:

  $ perf report --percent-limit 0.5 --stdio
  ...
     9.95%  swapper
            |
             --7.57%--intel_idle
                       cpuidle_enter_state
                       cpuidle_enter
                       call_cpuidle
                       cpu_startup_entry
                       |
                       |--4.89%--start_secondary
                       |
                        --2.68%--rest_init
                                  start_kernel
                                  x86_64_start_reservations
                                  x86_64_start_kernel

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/ui/stdio/hist.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)

diff --git a/tools/perf/ui/stdio/hist.c b/tools/perf/ui/stdio/hist.c
index 96188ea12771..76ff46becac8 100644
--- a/tools/perf/ui/stdio/hist.c
+++ b/tools/perf/ui/stdio/hist.c
@@ -165,6 +165,25 @@ static size_t __callchain__fprintf_graph(FILE *fp, struct rb_root *root,
 	return ret;
 }
 
+/*
+ * If have one single callchain root, don't bother printing
+ * its percentage (100 % in fractal mode and the same percentage
+ * than the hist in graph mode). This also avoid one level of column.
+ *
+ * However when percent-limit applied, it's possible that single callchain
+ * node have different (non-100% in fractal mode) percentage.
+ */
+static bool need_percent_display(struct rb_node *node, u64 parent_samples)
+{
+	struct callchain_node *cnode;
+
+	if (rb_next(node))
+		return true;
+
+	cnode = rb_entry(node, struct callchain_node, rb_node);
+	return callchain_cumul_hits(cnode) != parent_samples;
+}
+
 static size_t callchain__fprintf_graph(FILE *fp, struct rb_root *root,
 				       u64 total_samples, u64 parent_samples,
 				       int left_margin)
@@ -178,13 +197,8 @@ static size_t callchain__fprintf_graph(FILE *fp, struct rb_root *root,
 	int ret = 0;
 	char bf[1024];
 
-	/*
-	 * If have one single callchain root, don't bother printing
-	 * its percentage (100 % in fractal mode and the same percentage
-	 * than the hist in graph mode). This also avoid one level of column.
-	 */
 	node = rb_first(root);
-	if (node && !rb_next(node)) {
+	if (node && !need_percent_display(node, parent_samples)) {
 		cnode = rb_entry(node, struct callchain_node, rb_node);
 		list_for_each_entry(chain, &cnode->val, list) {
 			/*
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1319088 — [RFC/PATCH 10/10] perf tools: Change default calchain percent limit to 0.005%

The current default limit of 0.5% is (mostly) for 'fractal' mode which
calculates the percentage relatively.  As we changed the default
callchain mode to 'graph', the existing limit is too high IMHO.
Normally there're many entries under 0.5% overhead, it'd be supprising
that they don't show callchains.

Cc: Taeung Song <treeze.taeung@gmail.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/Documentation/perf-report.txt | 2 +-
 tools/perf/builtin-report.c              | 2 +-
 tools/perf/builtin-top.c                 | 2 +-
 tools/perf/util/util.c                   | 2 +-
 4 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
index 8a301f6afb37..1591b45c79e5 100644
--- a/tools/perf/Documentation/perf-report.txt
+++ b/tools/perf/Documentation/perf-report.txt
@@ -209,7 +209,7 @@ OPTIONS
 	- none: disable call chain display.
 
 	threshold is a percentage value which specifies a minimum percent to be
-	included in the output call graph.  Default is 0.5 (%).
+	included in the output call graph.  Default is 0.005 (%).
 
 	print_limit is only applied when stdio interface is used.  It's to limit
 	number of call graph entries in a single hist entry.  Note that it needs
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 72ed0b46d5a1..ba6eba5f5d69 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -643,7 +643,7 @@ parse_percent_limit(const struct option *opt, const char *str,
 	return 0;
 }
 
-#define CALLCHAIN_DEFAULT_OPT  "graph,0.5,caller,function,percent"
+#define CALLCHAIN_DEFAULT_OPT  "graph,0.005,caller,function,percent"
 
 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
 				     CALLCHAIN_REPORT_HELP
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index bf01cbb0ef23..1b471135ae2d 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -1086,7 +1086,7 @@ parse_percent_limit(const struct option *opt, const char *arg,
 }
 
 const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
-	"\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
+	"\n\t\t\t\tDefault: fp,graph,0.005,caller,function";
 
 int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
 {
diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
index 7a2da7ef556e..25ae6989c89d 100644
--- a/tools/perf/util/util.c
+++ b/tools/perf/util/util.c
@@ -20,7 +20,7 @@
 
 struct callchain_param	callchain_param = {
 	.mode	= CHAIN_GRAPH_ABS,
-	.min_percent = 0.5,
+	.min_percent = 0.005,
 	.order  = ORDER_CALLEE,
 	.key	= CCKEY_FUNCTION,
 	.value	= CCVAL_PERCENT,
-- 
2.6.4

[toc] | [prev] | [next] | [standalone]

#1319112 — Re: [RFC/PATCH 10/10] perf tools: Change default calchain percent limit to 0.005%

On Thu, Jan 28, 2016 at 12:40:57AM +0900, Namhyung Kim wrote:
> The current default limit of 0.5% is (mostly) for 'fractal' mode which
> calculates the percentage relatively.  As we changed the default
> callchain mode to 'graph', the existing limit is too high IMHO.
> Normally there're many entries under 0.5% overhead, it'd be supprising
> that they don't show callchains.

It's too small imho. On more complex workloads with a lot of user space
code we often end up with default perf report --stdio callgraph output of several
hundred KB, and most of it is useless as it is only very small.

-Andi

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web