Skip to contentSkip to conversation

perf mem: Describe overhead calculation in brief

PATCH Applied Namhyung Kim 3 participants 5 messages May 23, 2025 – May 28, 2025 perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

PATCH 1/3 perf mem: Describe overhead calculation in brief

From: Ravi Bangoria <[email protected]>

Unlike perf-report which uses sample period for overhead calculation, perf-mem overhead is calculated using sample weight. Describe perf-mem overhead calculation method in it's man page.

---
 tools/perf/Documentation/perf-mem.txt | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
tools/perf/Documentation/perf-mem.txt +19 -0
index a9e3c71a220557d4..965e73d377724607 100644
@@ -137,6 +137,25 @@ REPORT OPTIONS
 In addition, for report all perf report options are valid, and for record
 all perf record options.
 
+OVERHEAD CALCULATION
+--------------------
+Unlike linkperf:perf-report[1], which calculates overhead from the actual
+sample period, perf-mem overhead is calculated using sample weight. E.g.
+there are two samples in perf.data file, both with the same sample period,
+but one sample with weight 180 and the other with weight 20:
+
+  $ perf script -F period,data_src,weight,ip,sym
+  100000    629080842 |OP LOAD|LVL L3 hit|...     20       7e69b93ca524 strcmp
+  100000   1a29081042 |OP LOAD|LVL RAM hit|...   180   ffffffff82429168 memcpy
+
+  $ perf report -F overhead,symbol
+  50%   [.] strcmp
+  50%   [k] memcpy
+
+  $ perf mem report -F overhead,symbol
+  90%   [k] memcpy
+  10%   [.] strcmp
+
 SEE ALSO
 --------
 linkperf:perf-record[1], linkperf:perf-report[1], linkperf:perf-arm-spe[1]
Signature · 1 line
2.49.0.1151.ga128411c76-goog
author maintainer
#1 lore raw

PATCH 2/3 perf mem: Display sort order only if it's available

IOW it's not used when -F option is used alone. Let's make it conditional to skip printing incorrect information.

---
 tools/perf/builtin-report.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/builtin-report.c +4 -1
index f0299c7ee0254a37..e662e1c3a7c6b6ec 100644
@@ -529,7 +529,10 @@ static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report
 
 	if (rep->mem_mode) {
 		ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
-		ret += fprintf(fp, "\n# Sort order   : %s", sort_order ? : default_mem_sort_order);
+		if (sort_order || !field_order) {
+			ret += fprintf(fp, "\n# Sort order   : %s",
+				       sort_order ? : default_mem_sort_order);
+		}
 	} else
 		ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
 
Signature · 1 line
2.49.0.1151.ga128411c76-goog
author maintainer
#2 lore raw

PATCH 3/3 perf mem: Show absolute percent in mem_stat output

In reply to Namhyung Kim

Currently the output sums up to 100% for each entry. But it can be confusing when it's displayed with 'overhead'.

Before:

  $ perf mem report -F overhead,sample,cache,comm
  ...
  #                         -------------- Cache --------------
  # Overhead       Samples       L1     L2     L3 L1-buf  Other  Command
  # ........  ............  ...................................  ...............
  #
      25.38%           517    34.6%   0.0%  15.8%  23.3%  26.2%  swapper
       9.03%           239    35.4%   0.8%   9.1%  22.1%  32.6%  chrome
       8.61%           233    45.3%   1.2%   8.9%  22.7%  21.9%  Chrome_ChildIOT
       7.81%           189    33.6%   0.4%   5.5%  35.9%  24.6%  Isolated Web Co
       3.73%           103    40.4%   0.3%   2.7%  39.4%  17.2%  gnome-shell

Let's convert it to use absolute percent value so that it can add up to the overhead for that entry.

After:
# -------------- Cache --------------
# Overhead Samples L1 L2 L3 L1-buf Other Command

  # ........  ............  ...................................  ...............
  #
      25.38%           517     8.8%   0.0%   4.0%   5.9%   6.7%  swapper
       9.03%           239     3.2%   0.1%   0.8%   2.0%   2.9%  chrome
       8.61%           233     3.9%   0.1%   0.8%   2.0%   1.9%  Chrome_ChildIOT
       7.81%           189     2.6%   0.0%   0.4%   2.8%   1.9%  Isolated Web Co
       3.73%           103     1.5%   0.0%   0.1%   1.5%   0.6%  gnome-shell

This aligns well with the existing 'mem' sort key.

  $ perf mem report -s comm,mem -H
  ...
  #
  #    Overhead       Samples  Command / Memory access
  # .........................  ..........................................
  #
      25.38%           517     swapper
          8.78%           150     L1 hit
          6.66%            72     RAM hit
          5.92%           137     LFB/MAB hit
          4.02%           157     L3 hit
          0.00%             1     L3 miss
       9.03%           239     chrome
          3.19%           117     L1 hit
          2.94%            35     RAM hit
          1.99%            48     LFB/MAB hit
          0.82%            32     L3 hit
          0.08%             5     L2 hit
          0.00%             2     L3 miss

We can add an option or a config to change the setting later.

---
 tools/perf/ui/hist.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/ui/hist.c +1 -1
index ed5c40ebd906f076..b085eb0de84997f4 100644
@@ -172,7 +172,7 @@ int hpp__fmt_mem_stat(struct perf_hpp_fmt *fmt __maybe_unused, struct perf_hpp *
 	assert(mem_stat_idx != -1);
 
 	for (int i = 0; i < MEM_STAT_LEN; i++)
-		total += he->mem_stat[mem_stat_idx].entries[i];
+		total += hists->mem_stat_total[mem_stat_idx].entries[i];
 	assert(total != 0);
 
 	for (int i = 0; i < MEM_STAT_LEN; i++) {
Signature · 1 line
2.49.0.1151.ga128411c76-goog
Reviewed-by on v1 1/3
#3 lore raw
In reply to Namhyung Kim
Quoted 8 lines · Namhyung Kim

From: Ravi Bangoria <[email protected]>

Unlike perf-report which uses sample period for overhead calculation, perf-mem overhead is calculated using sample weight. Describe perf-mem overhead calculation method in it's man page.

For whole series:

Quoted 37 lines
---
 tools/perf/Documentation/perf-mem.txt | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
tools/perf/Documentation/perf-mem.txt +19 -0
index a9e3c71a220557d4..965e73d377724607 100644
@@ -137,6 +137,25 @@ REPORT OPTIONS
 In addition, for report all perf report options are valid, and for record
 all perf record options.
 
+OVERHEAD CALCULATION
+--------------------
+Unlike linkperf:perf-report[1], which calculates overhead from the actual
+sample period, perf-mem overhead is calculated using sample weight. E.g.
+there are two samples in perf.data file, both with the same sample period,
+but one sample with weight 180 and the other with weight 20:
+
+  $ perf script -F period,data_src,weight,ip,sym
+  100000    629080842 |OP LOAD|LVL L3 hit|...     20       7e69b93ca524 strcmp
+  100000   1a29081042 |OP LOAD|LVL RAM hit|...   180   ffffffff82429168 memcpy
+
+  $ perf report -F overhead,symbol
+  50%   [.] strcmp
+  50%   [k] memcpy
+
+  $ perf mem report -F overhead,symbol
+  90%   [k] memcpy
+  10%   [.] strcmp
+
 SEE ALSO
 --------
 linkperf:perf-record[1], linkperf:perf-report[1], linkperf:perf-arm-spe[1]
Signature · 1 line
2.49.0.1151.ga128411c76-goog
maintainer Applied
#4 lore raw
Quoted 7 lines · Leo Yan
Quoted 6 lines · Namhyung Kim

From: Ravi Bangoria <[email protected]>

Unlike perf-report which uses sample period for overhead calculation, perf-mem overhead is calculated using sample weight. Describe perf-mem overhead calculation method in it's man page.

Quoted 6 lines
Quoted 2 lines

For whole series:

Thanks, applied to perf-tools-next,

- Arnaldo