Skip to contentSkip to conversation

perf jevents: Sort list of input files

PATCH In review Bernhard M. Wiedemann 3 participants 7 messages Mar 20, 2023 – Mar 21, 2023 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf jevents: Sort list of input files

Without this, pmu-events.c would be generated with variations in ordering depending on non-deterministic filesystem readdir order.

I tested that pmu-events.c still has the same number of lines and that perf list output works.

This patch was done while working on reproducible builds for openSUSE, but also solves issues in Debian [1] and other distributions.

[1] https://tests.reproducible-builds.org/debian/rb-pkg/unstable/i386/linux.html

---
 tools/perf/pmu-events/jevents.py | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
tools/perf/pmu-events/jevents.py +7 -4
index 2bcd07ce609f..f06e1abac7c7 100755
@@ -381,10 +381,13 @@ def read_json_events(path: str, topic: str) -> Sequence[JsonEvent]:
 
   return events
 
+def sorted_scandir(path: str) -> list[os.DirEntry]:
+  return sorted(os.scandir(path), key=lambda e: e.name)
+
 def preprocess_arch_std_files(archpath: str) -> None:
   """Read in all architecture standard events."""
   global _arch_std_events
-  for item in os.scandir(archpath):
+  for item in sorted_scandir(archpath):
     if item.is_file() and item.name.endswith('.json'):
       for event in read_json_events(item.path, topic=''):
         if event.name:
@@ -497,7 +500,7 @@ def preprocess_one_file(parents: Sequence[str], item: os.DirEntry) -> None:
 def process_one_file(parents: Sequence[str], item: os.DirEntry) -> None:
   """Process a JSON file during the main walk."""
   def is_leaf_dir(path: str) -> bool:
-    for item in os.scandir(path):
+    for item in sorted_scandir(path):
       if item.is_dir():
         return False
     return True
@@ -889,7 +892,7 @@ def main() -> None:
   def ftw(path: str, parents: Sequence[str],
           action: Callable[[Sequence[str], os.DirEntry], None]) -> None:
     """Replicate the directory/file walking behavior of C's file tree walk."""
-    for item in os.scandir(path):
+    for item in sorted_scandir(path):
       if _args.model != 'all' and item.is_dir():
         # Check if the model matches one in _args.model.
         if len(parents) == _args.model.split(',')[0].count('/'):
@@ -930,7 +933,7 @@ struct compact_pmu_event {
 
 """)
   archs = []
-  for item in os.scandir(_args.starting_dir):
+  for item in sorted_scandir(_args.starting_dir):
     if not item.is_dir():
       continue
     if item.name == _args.arch or _args.arch == 'all' or item.name == 'test':
Signature · 1 line
2.35.3
reviewer
#1 lore raw
Quoted 10 lines · Bernhard M. Wiedemann

Without this, pmu-events.c would be generated with variations in ordering depending on non-deterministic filesystem readdir order.

I tested that pmu-events.c still has the same number of lines and that perf list output works.

This patch was done while working on reproducible builds for openSUSE, but also solves issues in Debian [1] and other distributions.

Quoted 7 lines

[1] https://tests.reproducible-builds.org/debian/rb-pkg/unstable/i386/linux.html

---
 tools/perf/pmu-events/jevents.py | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

Thanks Bernhard,

I think this may already be addressed by sorting prior to output:
https://lore.kernel.org/r/[email protected]

Could you confirm?

Thanks,
Ian

Quoted 50 lines
tools/perf/pmu-events/jevents.py +7 -5
index 2bcd07ce609f..f06e1abac7c7 100755
@@ -381,10 +381,13 @@ def read_json_events(path: str, topic: str) -> Sequence[JsonEvent]:
 
   return events
 
+def sorted_scandir(path: str) -> list[os.DirEntry]:
+  return sorted(os.scandir(path), key=lambda e: e.name)
+
 def preprocess_arch_std_files(archpath: str) -> None:
   """Read in all architecture standard events."""
   global _arch_std_events
-  for item in os.scandir(archpath):
+  for item in sorted_scandir(archpath):
     if item.is_file() and item.name.endswith('.json'):
       for event in read_json_events(item.path, topic=''):
         if event.name:
@@ -497,7 +500,7 @@ def preprocess_one_file(parents: Sequence[str], item: os.DirEntry) -> None:
 def process_one_file(parents: Sequence[str], item: os.DirEntry) -> None:
   """Process a JSON file during the main walk."""
   def is_leaf_dir(path: str) -> bool:
-    for item in os.scandir(path):
+    for item in sorted_scandir(path):
       if item.is_dir():
         return False
     return True
@@ -889,7 +892,7 @@ def main() -> None:
   def ftw(path: str, parents: Sequence[str],
           action: Callable[[Sequence[str], os.DirEntry], None]) -> None:
     """Replicate the directory/file walking behavior of C's file tree walk."""
-    for item in os.scandir(path):
+    for item in sorted_scandir(path):
       if _args.model != 'all' and item.is_dir():
         # Check if the model matches one in _args.model.
         if len(parents) == _args.model.split(',')[0].count('/'):
@@ -930,7 +933,7 @@ struct compact_pmu_event {
 
 """)
   archs = []
-  for item in os.scandir(_args.starting_dir):
+  for item in sorted_scandir(_args.starting_dir):
     if not item.is_dir():
       continue
     if item.name == _args.arch or _args.arch == 'all' or item.name == 'test':
--

2.35.3

#2 lore raw
Quoted 4 lines · Ian Rogers

I think this may already be addressed by sorting prior to output:
https://lore.kernel.org/r/[email protected]

Could you confirm?

Hi Ian,

I was testing on 6.2.6 which includes that patch and it was still affected.
The trouble with sorting at the end is, that there can be influences of ordering in earlier processing steps, that don't get ironed out by the sort later.

Some more experimenting showed that only the ftw scandir needed sorting, which allows to further simplify the patch to

"""Replicate the directory/file walking behavior of C's ...
- for item in os.scandir(path):
+ for item in sorted(os.scandir(path), key=lambda e: e.name):
action(parents, item)

Without the patch, a random diff in pmu-events.c starts with -static const struct compact_pmu_event pme_amdzen2[] = { +static const struct compact_pmu_event pme_silvermont[] = {

While I'm testing on scratch ext4 filesystems where dirindex causes randomness, you could also use the disorderfs FUSE-filesystem with its shuffle mode to give you random order.

Ciao
Bernhard M.

reviewer Acked-by
#3 lore raw
Quoted 28 lines · Bernhard M. Wiedemann
Quoted 4 lines · Ian Rogers

I think this may already be addressed by sorting prior to output:
https://lore.kernel.org/r/[email protected]

Could you confirm?

Hi Ian,

I was testing on 6.2.6 which includes that patch and it was still affected. The trouble with sorting at the end is, that there can be influences of ordering in earlier processing steps, that don't get ironed out by the sort later.

Some more experimenting showed that only the ftw scandir needed sorting, which allows to further simplify the patch to

"""Replicate the directory/file walking behavior of C's ...

(quoted diff) +1 -1
-    for item in os.scandir(path):
+    for item in sorted(os.scandir(path), key=lambda e: e.name):
        action(parents, item)

Without the patch, a random diff in pmu-events.c starts with

(quoted diff) +1 -1
-static const struct compact_pmu_event pme_amdzen2[] = {
+static const struct compact_pmu_event pme_silvermont[] = {
Quoted 6 lines

While I'm testing on scratch ext4 filesystems where dirindex causes randomness, you could also use the disorderfs FUSE-filesystem with its shuffle mode to give you random order.

Ciao
Bernhard M.

Thanks Bernhard!

I wonder about determining the order from the mapfile.csv, but that's another problem for another day.

Thanks,
Ian

#4 lore raw
In reply to Bernhard M. Wiedemann

Em Mon, Mar 20, 2023 at 10:30:00PM +0100, Bernhard M. Wiedemann escreveu:

Quoted 29 lines · Ian Rogers
Quoted 4 lines

I think this may already be addressed by sorting prior to output:
https://lore.kernel.org/r/[email protected]

Could you confirm?

Hi Ian,

I was testing on 6.2.6 which includes that patch and it was still affected. The trouble with sorting at the end is, that there can be influences of ordering in earlier processing steps, that don't get ironed out by the sort later.

Some more experimenting showed that only the ftw scandir needed sorting, which allows to further simplify the patch to

"""Replicate the directory/file walking behavior of C's ...

(quoted diff) +1 -1
-    for item in os.scandir(path):
+    for item in sorted(os.scandir(path), key=lambda e: e.name):
       action(parents, item)

Without the patch, a random diff in pmu-events.c starts with

(quoted diff) +1 -1
-static const struct compact_pmu_event pme_amdzen2[] = {
+static const struct compact_pmu_event pme_silvermont[] = {
Quoted 3 lines

While I'm testing on scratch ext4 filesystems where dirindex causes randomness, you could also use the disorderfs FUSE-filesystem with its shuffle mode to give you random order.

So, Ian acked the current patch, but you found some further simplification, can you please resubmit?

- Arnaldo

#5 lore raw

PATCH 1/1 perf jevents: Sort list of input files

Without this, pmu-events.c would be generated with variations in ordering depending on non-deterministic filesystem readdir order.

I tested that pmu-events.c still has the same number of lines and that perf list output works.

This patch was done while working on reproducible builds for openSUSE, but also solves issues in Debian [1] and other distributions.

[1] https://tests.reproducible-builds.org/debian/rb-pkg/unstable/i386/linux.html

---
 tools/perf/pmu-events/jevents.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/pmu-events/jevents.py +1 -1
index 2bcd07ce609f..736ee0a75cf8 100755
@@ -889,7 +889,7 @@ def main() -> None:
   def ftw(path: str, parents: Sequence[str],
           action: Callable[[Sequence[str], os.DirEntry], None]) -> None:
     """Replicate the directory/file walking behavior of C's file tree walk."""
-    for item in os.scandir(path):
+    for item in sorted(os.scandir(path), key=lambda e: e.name):
       if _args.model != 'all' and item.is_dir():
         # Check if the model matches one in _args.model.
         if len(parents) == _args.model.split(',')[0].count('/'):
Signature · 1 line
2.35.3
maintainer Applied
#6 lore raw

Em Tue, Mar 21, 2023 at 07:30:32AM +0100, Bernhard M. Wiedemann escreveu:

Quoted 10 lines

Without this, pmu-events.c would be generated with variations in ordering depending on non-deterministic filesystem readdir order.

I tested that pmu-events.c still has the same number of lines and that perf list output works.

This patch was done while working on reproducible builds for openSUSE, but also solves issues in Debian [1] and other distributions.

[1] https://tests.reproducible-builds.org/debian/rb-pkg/unstable/i386/linux.html

Thanks, applied.

- Arnaldo

Quoted 22 lines
---
 tools/perf/pmu-events/jevents.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/pmu-events/jevents.py +1 -1
index 2bcd07ce609f..736ee0a75cf8 100755
@@ -889,7 +889,7 @@ def main() -> None:
   def ftw(path: str, parents: Sequence[str],
           action: Callable[[Sequence[str], os.DirEntry], None]) -> None:
     """Replicate the directory/file walking behavior of C's file tree walk."""
-    for item in os.scandir(path):
+    for item in sorted(os.scandir(path), key=lambda e: e.name):
       if _args.model != 'all' and item.is_dir():
         # Check if the model matches one in _args.model.
         if len(parents) == _args.model.split(',')[0].count('/'):
Signature · 1 line
2.35.3
Signature · 1 line
- Arnaldo