* [PATCH v1] perf script python: Improve physical mem type resolution
@ 2024-11-19 18:01 Ian Rogers
2024-11-19 20:05 ` Liang, Kan
0 siblings, 1 reply; 3+ messages in thread
From: Ian Rogers @ 2024-11-19 18:01 UTC (permalink / raw)
To: Kan Liang
Cc: Peter Zijlstra, Ingo Molnar, Arnaldo Carvalho de Melo,
Namhyung Kim, Mark Rutland, Alexander Shishkin, Jiri Olsa,
Ian Rogers, Adrian Hunter, linux-perf-users, linux-kernel
Previously system RAM and persistent memory were hard code matched,
change so that the label of the memory region is just read from
/proc/iomem. This avoids frequent N/A samples.
Change the /proc/iomem reading, event processing and output so that
nested entries appear and their counts count toward their parent. As
labels may be repeated, include the memory ranges in the output to
make it clear why, for example, "System RAM" appears twice.
Before:
```
Event: mem_inst_retired.all_loads:P
Memory type count percentage
---------------------------------------- ---------- ----------
System RAM 9460 96.5%
N/A 998 3.5%
```
After:
```
Event: mem_inst_retired.all_loads:P
Memory type count percentage
---------------------------------------- ---------- ----------
100000000-105f7fffff : System RAM 36741 96.5
841400000-8416599ff : Kernel data 89 0.2
840800000-8412a6fff : Kernel rodata 60 0.2
841ebe000-8423fffff : Kernel bss 34 0.1
0-fff : Reserved 1345 3.5
100000-89dd9fff : System RAM 2 0.0
```
Before:
```
Event: mem_inst_retired.any:P
Memory type count percentage
---------------------------------------- ----------- -----------
System RAM 9460 90.5%
N/A 998 9.5%
```
After:
```
Event: mem_inst_retired.any:P
Memory type count percentage
---------------------------------------- ---------- ----------
100000000-105f7fffff : System RAM 9460 90.5
841400000-8416599ff : Kernel data 45 0.4
840800000-8412a6fff : Kernel rodata 19 0.2
841ebe000-8423fffff : Kernel bss 12 0.1
0-fff : Reserved 998 9.5
```
The code has been updated to python 3 with type hints and resolving
issues reported by mypy and pylint. Tabs are swapped to spaces as
preferred in PEP8, because most lines of code were modified (of this
small file) and this makes pylint significantly less noisy.
Signed-off-by: Ian Rogers <irogers@google.com>
---
tools/perf/scripts/python/mem-phys-addr.py | 177 ++++++++++++---------
1 file changed, 102 insertions(+), 75 deletions(-)
diff --git a/tools/perf/scripts/python/mem-phys-addr.py b/tools/perf/scripts/python/mem-phys-addr.py
index 1f332e72b9b0..5e237a5a5f1b 100644
--- a/tools/perf/scripts/python/mem-phys-addr.py
+++ b/tools/perf/scripts/python/mem-phys-addr.py
@@ -3,98 +3,125 @@
#
# Copyright (c) 2018, Intel Corporation.
-from __future__ import division
-from __future__ import print_function
-
import os
import sys
-import struct
import re
import bisect
import collections
+from dataclasses import dataclass
+from typing import (Dict, Optional)
sys.path.append(os.environ['PERF_EXEC_PATH'] + \
- '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
+ '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
+
+@dataclass(frozen=True)
+class IomemEntry:
+ """Read from a line in /proc/iomem"""
+ begin: int
+ end: int
+ indent: int
+ label: str
-#physical address ranges for System RAM
-system_ram = []
-#physical address ranges for Persistent Memory
-pmem = []
-#file object for proc iomem
-f = None
-#Count for each type of memory
-load_mem_type_cnt = collections.Counter()
-#perf event name
-event_name = None
+# Physical memory layout from /proc/iomem. Key is the indent and then
+# a list of ranges.
+iomem: Dict[int, list[IomemEntry]] = collections.defaultdict(list)
+# Child nodes from the iomem parent.
+children: Dict[IomemEntry, set[IomemEntry]] = collections.defaultdict(set)
+# Maximum indent seen before an entry in the iomem file.
+max_indent: int = 0
+# Count for each range of memory.
+load_mem_type_cnt: Dict[IomemEntry, int] = collections.Counter()
+# Perf event name set from the first sample in the data.
+event_name: Optional[str] = None
def parse_iomem():
- global f
- f = open('/proc/iomem', 'r')
- for i, j in enumerate(f):
- m = re.split('-|:',j,2)
- if m[2].strip() == 'System RAM':
- system_ram.append(int(m[0], 16))
- system_ram.append(int(m[1], 16))
- if m[2].strip() == 'Persistent Memory':
- pmem.append(int(m[0], 16))
- pmem.append(int(m[1], 16))
+ """Populate iomem from /proc/iomem file"""
+ global iomem
+ global max_indent
+ global children
+ with open('/proc/iomem', 'r', encoding='ascii') as f:
+ for line in f:
+ indent = 0
+ while line[indent] == ' ':
+ indent += 1
+ if indent > max_indent:
+ max_indent = indent
+ m = re.split('-|:', line, 2)
+ begin = int(m[0], 16)
+ end = int(m[1], 16)
+ label = m[2].strip()
+ entry = IomemEntry(begin, end, indent, label)
+ # Before adding entry, search for a parent node using its begin.
+ if indent > 0:
+ parent = find_memory_type(begin)
+ assert parent, f"Given indent expected a parent for {label}"
+ children[parent].add(entry)
+ iomem[indent].append(entry)
-def print_memory_type():
- print("Event: %s" % (event_name))
- print("%-40s %10s %10s\n" % ("Memory type", "count", "percentage"), end='')
- print("%-40s %10s %10s\n" % ("----------------------------------------",
- "-----------", "-----------"),
- end='');
- total = sum(load_mem_type_cnt.values())
- for mem_type, count in sorted(load_mem_type_cnt.most_common(), \
- key = lambda kv: (kv[1], kv[0]), reverse = True):
- print("%-40s %10d %10.1f%%\n" %
- (mem_type, count, 100 * count / total),
- end='')
+def find_memory_type(phys_addr) -> Optional[IomemEntry]:
+ """Search iomem for the range containing phys_addr with the maximum indent"""
+ for i in range(max_indent, -1, -1):
+ if i not in iomem:
+ continue
+ position = bisect.bisect_right(iomem[i], phys_addr,
+ key=lambda entry: entry.begin)
+ if position is None:
+ continue
+ iomem_entry = iomem[i][position-1]
+ if iomem_entry.begin <= phys_addr <= iomem_entry.end:
+ return iomem_entry
+ print(f"Didn't find {phys_addr}")
+ return None
-def trace_begin():
- parse_iomem()
+def print_memory_type():
+ print(f"Event: {event_name}")
+ print(f"{'Memory type':<40} {'count':>10} {'percentage':>10}")
+ print(f"{'-' * 40:<40} {'-' * 10:>10} {'-' * 10:>10}")
+ total = sum(load_mem_type_cnt.values())
+ # Add count from children into the parent.
+ for i in range(max_indent, -1, -1):
+ if i not in iomem:
+ continue
+ for entry in iomem[i]:
+ global children
+ for child in children[entry]:
+ if load_mem_type_cnt[child] > 0:
+ load_mem_type_cnt[entry] += load_mem_type_cnt[child]
-def trace_end():
- print_memory_type()
- f.close()
+ def print_entries(entries):
+ """Print counts from parents down to their children"""
+ global children
+ for entry in sorted(entries,
+ key = lambda entry: load_mem_type_cnt[entry],
+ reverse = True):
+ count = load_mem_type_cnt[entry]
+ if count > 0:
+ mem_type = ' ' * entry.indent + f"{entry.begin:x}-{entry.end:x} : {entry.label}"
+ percent = 100 * count / total
+ print(f"{mem_type:<40} {count:>10} {percent:>10.1f}")
+ print_entries(children[entry])
-def is_system_ram(phys_addr):
- #/proc/iomem is sorted
- position = bisect.bisect(system_ram, phys_addr)
- if position % 2 == 0:
- return False
- return True
+ print_entries(iomem[0])
-def is_persistent_mem(phys_addr):
- position = bisect.bisect(pmem, phys_addr)
- if position % 2 == 0:
- return False
- return True
+def trace_begin():
+ parse_iomem()
-def find_memory_type(phys_addr):
- if phys_addr == 0:
- return "N/A"
- if is_system_ram(phys_addr):
- return "System RAM"
+def trace_end():
+ print_memory_type()
- if is_persistent_mem(phys_addr):
- return "Persistent Memory"
+def process_event(param_dict):
+ if "sample" not in param_dict:
+ return
- #slow path, search all
- f.seek(0, 0)
- for j in f:
- m = re.split('-|:',j,2)
- if int(m[0], 16) <= phys_addr <= int(m[1], 16):
- return m[2]
- return "N/A"
+ sample = param_dict["sample"]
+ if "phys_addr" not in sample:
+ return
-def process_event(param_dict):
- name = param_dict["ev_name"]
- sample = param_dict["sample"]
- phys_addr = sample["phys_addr"]
+ phys_addr = sample["phys_addr"]
+ entry = find_memory_type(phys_addr)
+ if entry:
+ load_mem_type_cnt[entry] += 1
- global event_name
- if event_name == None:
- event_name = name
- load_mem_type_cnt[find_memory_type(phys_addr)] += 1
+ global event_name
+ if event_name is None:
+ event_name = param_dict["ev_name"]
--
2.47.0.338.g60cca15819-goog
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH v1] perf script python: Improve physical mem type resolution
2024-11-19 18:01 [PATCH v1] perf script python: Improve physical mem type resolution Ian Rogers
@ 2024-11-19 20:05 ` Liang, Kan
2024-12-04 21:14 ` Arnaldo Carvalho de Melo
0 siblings, 1 reply; 3+ messages in thread
From: Liang, Kan @ 2024-11-19 20:05 UTC (permalink / raw)
To: Ian Rogers
Cc: Peter Zijlstra, Ingo Molnar, Arnaldo Carvalho de Melo,
Namhyung Kim, Mark Rutland, Alexander Shishkin, Jiri Olsa,
Adrian Hunter, linux-perf-users, linux-kernel
On 2024-11-19 1:01 p.m., Ian Rogers wrote:
> Previously system RAM and persistent memory were hard code matched,
> change so that the label of the memory region is just read from
> /proc/iomem. This avoids frequent N/A samples.
>
> Change the /proc/iomem reading, event processing and output so that
> nested entries appear and their counts count toward their parent. As
> labels may be repeated, include the memory ranges in the output to
> make it clear why, for example, "System RAM" appears twice.
>
> Before:
> ```
> Event: mem_inst_retired.all_loads:P
> Memory type count percentage
> ---------------------------------------- ---------- ----------
> System RAM 9460 96.5%
> N/A 998 3.5%
> ```
>
> After:
> ```
> Event: mem_inst_retired.all_loads:P
> Memory type count percentage
> ---------------------------------------- ---------- ----------
> 100000000-105f7fffff : System RAM 36741 96.5
> 841400000-8416599ff : Kernel data 89 0.2
> 840800000-8412a6fff : Kernel rodata 60 0.2
> 841ebe000-8423fffff : Kernel bss 34 0.1
> 0-fff : Reserved 1345 3.5
> 100000-89dd9fff : System RAM 2 0.0
> ```
>
> Before:
> ```
> Event: mem_inst_retired.any:P
> Memory type count percentage
> ---------------------------------------- ----------- -----------
> System RAM 9460 90.5%
> N/A 998 9.5%
> ```
>
> After:
> ```
> Event: mem_inst_retired.any:P
> Memory type count percentage
> ---------------------------------------- ---------- ----------
> 100000000-105f7fffff : System RAM 9460 90.5
> 841400000-8416599ff : Kernel data 45 0.4
> 840800000-8412a6fff : Kernel rodata 19 0.2
> 841ebe000-8423fffff : Kernel bss 12 0.1
> 0-fff : Reserved 998 9.5
> ```
>
> The code has been updated to python 3 with type hints and resolving
> issues reported by mypy and pylint. Tabs are swapped to spaces as
> preferred in PEP8, because most lines of code were modified (of this
> small file) and this makes pylint significantly less noisy.
>
Thanks Ian. A very nice improvement!
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Thanks,
Kan
> Signed-off-by: Ian Rogers <irogers@google.com>
> ---
> tools/perf/scripts/python/mem-phys-addr.py | 177 ++++++++++++---------
> 1 file changed, 102 insertions(+), 75 deletions(-)
>
> diff --git a/tools/perf/scripts/python/mem-phys-addr.py b/tools/perf/scripts/python/mem-phys-addr.py
> index 1f332e72b9b0..5e237a5a5f1b 100644
> --- a/tools/perf/scripts/python/mem-phys-addr.py
> +++ b/tools/perf/scripts/python/mem-phys-addr.py
> @@ -3,98 +3,125 @@
> #
> # Copyright (c) 2018, Intel Corporation.
>
> -from __future__ import division
> -from __future__ import print_function
> -
> import os
> import sys
> -import struct
> import re
> import bisect
> import collections
> +from dataclasses import dataclass
> +from typing import (Dict, Optional)
>
> sys.path.append(os.environ['PERF_EXEC_PATH'] + \
> - '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
> + '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
> +
> +@dataclass(frozen=True)
> +class IomemEntry:
> + """Read from a line in /proc/iomem"""
> + begin: int
> + end: int
> + indent: int
> + label: str
>
> -#physical address ranges for System RAM
> -system_ram = []
> -#physical address ranges for Persistent Memory
> -pmem = []
> -#file object for proc iomem
> -f = None
> -#Count for each type of memory
> -load_mem_type_cnt = collections.Counter()
> -#perf event name
> -event_name = None
> +# Physical memory layout from /proc/iomem. Key is the indent and then
> +# a list of ranges.
> +iomem: Dict[int, list[IomemEntry]] = collections.defaultdict(list)
> +# Child nodes from the iomem parent.
> +children: Dict[IomemEntry, set[IomemEntry]] = collections.defaultdict(set)
> +# Maximum indent seen before an entry in the iomem file.
> +max_indent: int = 0
> +# Count for each range of memory.
> +load_mem_type_cnt: Dict[IomemEntry, int] = collections.Counter()
> +# Perf event name set from the first sample in the data.
> +event_name: Optional[str] = None
>
> def parse_iomem():
> - global f
> - f = open('/proc/iomem', 'r')
> - for i, j in enumerate(f):
> - m = re.split('-|:',j,2)
> - if m[2].strip() == 'System RAM':
> - system_ram.append(int(m[0], 16))
> - system_ram.append(int(m[1], 16))
> - if m[2].strip() == 'Persistent Memory':
> - pmem.append(int(m[0], 16))
> - pmem.append(int(m[1], 16))
> + """Populate iomem from /proc/iomem file"""
> + global iomem
> + global max_indent
> + global children
> + with open('/proc/iomem', 'r', encoding='ascii') as f:
> + for line in f:
> + indent = 0
> + while line[indent] == ' ':
> + indent += 1
> + if indent > max_indent:
> + max_indent = indent
> + m = re.split('-|:', line, 2)
> + begin = int(m[0], 16)
> + end = int(m[1], 16)
> + label = m[2].strip()
> + entry = IomemEntry(begin, end, indent, label)
> + # Before adding entry, search for a parent node using its begin.
> + if indent > 0:
> + parent = find_memory_type(begin)
> + assert parent, f"Given indent expected a parent for {label}"
> + children[parent].add(entry)
> + iomem[indent].append(entry)
>
> -def print_memory_type():
> - print("Event: %s" % (event_name))
> - print("%-40s %10s %10s\n" % ("Memory type", "count", "percentage"), end='')
> - print("%-40s %10s %10s\n" % ("----------------------------------------",
> - "-----------", "-----------"),
> - end='');
> - total = sum(load_mem_type_cnt.values())
> - for mem_type, count in sorted(load_mem_type_cnt.most_common(), \
> - key = lambda kv: (kv[1], kv[0]), reverse = True):
> - print("%-40s %10d %10.1f%%\n" %
> - (mem_type, count, 100 * count / total),
> - end='')
> +def find_memory_type(phys_addr) -> Optional[IomemEntry]:
> + """Search iomem for the range containing phys_addr with the maximum indent"""
> + for i in range(max_indent, -1, -1):
> + if i not in iomem:
> + continue
> + position = bisect.bisect_right(iomem[i], phys_addr,
> + key=lambda entry: entry.begin)
> + if position is None:
> + continue
> + iomem_entry = iomem[i][position-1]
> + if iomem_entry.begin <= phys_addr <= iomem_entry.end:
> + return iomem_entry
> + print(f"Didn't find {phys_addr}")
> + return None
>
> -def trace_begin():
> - parse_iomem()
> +def print_memory_type():
> + print(f"Event: {event_name}")
> + print(f"{'Memory type':<40} {'count':>10} {'percentage':>10}")
> + print(f"{'-' * 40:<40} {'-' * 10:>10} {'-' * 10:>10}")
> + total = sum(load_mem_type_cnt.values())
> + # Add count from children into the parent.
> + for i in range(max_indent, -1, -1):
> + if i not in iomem:
> + continue
> + for entry in iomem[i]:
> + global children
> + for child in children[entry]:
> + if load_mem_type_cnt[child] > 0:
> + load_mem_type_cnt[entry] += load_mem_type_cnt[child]
>
> -def trace_end():
> - print_memory_type()
> - f.close()
> + def print_entries(entries):
> + """Print counts from parents down to their children"""
> + global children
> + for entry in sorted(entries,
> + key = lambda entry: load_mem_type_cnt[entry],
> + reverse = True):
> + count = load_mem_type_cnt[entry]
> + if count > 0:
> + mem_type = ' ' * entry.indent + f"{entry.begin:x}-{entry.end:x} : {entry.label}"
> + percent = 100 * count / total
> + print(f"{mem_type:<40} {count:>10} {percent:>10.1f}")
> + print_entries(children[entry])
>
> -def is_system_ram(phys_addr):
> - #/proc/iomem is sorted
> - position = bisect.bisect(system_ram, phys_addr)
> - if position % 2 == 0:
> - return False
> - return True
> + print_entries(iomem[0])
>
> -def is_persistent_mem(phys_addr):
> - position = bisect.bisect(pmem, phys_addr)
> - if position % 2 == 0:
> - return False
> - return True
> +def trace_begin():
> + parse_iomem()
>
> -def find_memory_type(phys_addr):
> - if phys_addr == 0:
> - return "N/A"
> - if is_system_ram(phys_addr):
> - return "System RAM"
> +def trace_end():
> + print_memory_type()
>
> - if is_persistent_mem(phys_addr):
> - return "Persistent Memory"
> +def process_event(param_dict):
> + if "sample" not in param_dict:
> + return
>
> - #slow path, search all
> - f.seek(0, 0)
> - for j in f:
> - m = re.split('-|:',j,2)
> - if int(m[0], 16) <= phys_addr <= int(m[1], 16):
> - return m[2]
> - return "N/A"
> + sample = param_dict["sample"]
> + if "phys_addr" not in sample:
> + return
>
> -def process_event(param_dict):
> - name = param_dict["ev_name"]
> - sample = param_dict["sample"]
> - phys_addr = sample["phys_addr"]
> + phys_addr = sample["phys_addr"]
> + entry = find_memory_type(phys_addr)
> + if entry:
> + load_mem_type_cnt[entry] += 1
>
> - global event_name
> - if event_name == None:
> - event_name = name
> - load_mem_type_cnt[find_memory_type(phys_addr)] += 1
> + global event_name
> + if event_name is None:
> + event_name = param_dict["ev_name"]
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH v1] perf script python: Improve physical mem type resolution
2024-11-19 20:05 ` Liang, Kan
@ 2024-12-04 21:14 ` Arnaldo Carvalho de Melo
0 siblings, 0 replies; 3+ messages in thread
From: Arnaldo Carvalho de Melo @ 2024-12-04 21:14 UTC (permalink / raw)
To: Liang, Kan
Cc: Ian Rogers, Peter Zijlstra, Ingo Molnar, Namhyung Kim,
Mark Rutland, Alexander Shishkin, Jiri Olsa, Adrian Hunter,
linux-perf-users, linux-kernel
On Tue, Nov 19, 2024 at 03:05:41PM -0500, Liang, Kan wrote:
> On 2024-11-19 1:01 p.m., Ian Rogers wrote:
> > After:
> > ```
> > Event: mem_inst_retired.any:P
> > Memory type count percentage
> > ---------------------------------------- ---------- ----------
> > 100000000-105f7fffff : System RAM 9460 90.5
> > 841400000-8416599ff : Kernel data 45 0.4
> > 840800000-8412a6fff : Kernel rodata 19 0.2
> > 841ebe000-8423fffff : Kernel bss 12 0.1
> > 0-fff : Reserved 998 9.5
> > ```
> > The code has been updated to python 3 with type hints and resolving
> > issues reported by mypy and pylint. Tabs are swapped to spaces as
> > preferred in PEP8, because most lines of code were modified (of this
> > small file) and this makes pylint significantly less noisy.
> Thanks Ian. A very nice improvement!
> Acked-by: Kan Liang <kan.liang@linux.intel.com>
Thanks, tested on a:
root@number:/tmp# grep -m1 "model name" /proc/cpuinfo
model name : Intel(R) Core(TM) i7-14700K
root@number:/tmp#
and applied to perf-tools-next:
root@number:/tmp# perf script mem-phys-addr -a find /
<SNIP>
/bin
/lib
/lib64
/sbin
Warning:
744 out of order events recorded.
Event: cpu_core/mem_inst_retired.all_loads/P
Memory type count percentage
---------------------------------------- ---------- ----------
100000000-8bfbfffff : System RAM 364561 76.5
621400000-6223a6fff : Kernel rodata 10474 2.2
622400000-62283d4bf : Kernel data 4828 1.0
623304000-6237fffff : Kernel bss 1063 0.2
620000000-6213fffff : Kernel code 98 0.0
0-fff : Reserved 111480 23.4
100000-2b0ca017 : System RAM 337 0.1
2fbad000-30d92fff : System RAM 44 0.0
2c79d000-2fbabfff : System RAM 30 0.0
30d94000-316d5fff : System RAM 16 0.0
2b131a58-2c71dfff : System RAM 7 0.0
root@number:/tmp#
- Arnaldo
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2024-12-04 21:14 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2024-11-19 18:01 [PATCH v1] perf script python: Improve physical mem type resolution Ian Rogers
2024-11-19 20:05 ` Liang, Kan
2024-12-04 21:14 ` Arnaldo Carvalho de Melo
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).