There could be different types of memory in the system. E.g normal System Memory, Persistent Memory. To understand how the workload maps to those memories, it's important to know the I/O statistics of them. Perf can collect physical addresses, but those are raw data. It still needs extra work to resolve the physical addresses. Provide a script to facilitate the physical addresses resolving and I/O statistics. Profile with MEM_INST_RETIRED.ALL_LOADS or MEM_UOPS_RETIRED.ALL_LOADS event if any of them is available. Look up the /proc/iomem and resolve the physical address. Provide memory type summary. Here is an example output: # perf script report mem-phys-addr Event: mem_inst_retired.all_loads:P Memory type count percentage ---------------------------------------- ----------- ----------- System RAM 74 53.2% Persistent Memory 55 39.6% N/A --- Changes since V2: - Apply the new license rules. - Add comments for globals Changes since V1: - Do not mix DLA and Load Latency. Do not compare the loads and stores. Only profile the loads. - Use event name to replace the RAW event
Showing the changes to tools/perf/scripts/python (3 of 4 files). Show all
Unified · Split
| @@ -0,0 +1,19 @@ | |||
| 1 | #!/bin/bash | ||
| 2 | |||
| 3 | # | ||
| 4 | # Profiling physical memory by all retired load instructions/uops event | ||
| 5 | # MEM_INST_RETIRED.ALL_LOADS or MEM_UOPS_RETIRED.ALL_LOADS | ||
| 6 | # | ||
| 7 | |||
| 8 | load=`perf list | grep mem_inst_retired.all_loads` | ||
| 9 | if [ -z "$load" ]; then | ||
| 10 | load=`perf list | grep mem_uops_retired.all_loads` | ||
| 11 | fi | ||
| 12 | if [ -z "$load" ]; then | ||
| 13 | echo "There is no event to count all retired load instructions/uops." | ||
| 14 | exit 1 | ||
| 15 | fi | ||
| 16 | |||
| 17 | arg=$(echo $load | tr -d ' ') | ||
| 18 | arg="$arg:P" | ||
| 19 | perf record --phys-data -e $arg $@ | ||
| @@ -0,0 +1,3 @@ | |||
| 1 | #!/bin/bash | ||
| 2 | # description: resolve physical address samples | ||
| 3 | perf script $@ -s "$PERF_EXEC_PATH"/scripts/python/mem-phys-addr.py | ||
| @@ -0,0 +1,95 @@ | |||
| 1 | # mem-phys-addr.py: Resolve physical address samples | ||
| 2 | # SPDX-License-Identifier: GPL-2.0 | ||
| 3 | # | ||
| 4 | # Copyright (c) 2018, Intel Corporation. | ||
| 5 | |||
| 6 | from __future__ import division | ||
| 7 | import os | ||
| 8 | import sys | ||
| 9 | import struct | ||
| 10 | import re | ||
| 11 | import bisect | ||
| 12 | import collections | ||
| 13 | |||
| 14 | sys.path.append(os.environ['PERF_EXEC_PATH'] + \ | ||
| 15 | '/scripts/python/Perf-Trace-Util/lib/Perf/Trace') | ||
| 16 | |||
| 17 | #physical address ranges for System RAM | ||
| 18 | system_ram = [] | ||
| 19 | #physical address ranges for Persistent Memory | ||
| 20 | pmem = [] | ||
| 21 | #file object for proc iomem | ||
| 22 | f = None | ||
| 23 | #Count for each type of memory | ||
| 24 | load_mem_type_cnt = collections.Counter() | ||
| 25 | #perf event name | ||
| 26 | event_name = None | ||
| 27 | |||
| 28 | def parse_iomem(): | ||
| 29 | global f | ||
| 30 | f = open('/proc/iomem', 'r') | ||
| 31 | for i, j in enumerate(f): | ||
| 32 | m = re.split('-|:',j,2) | ||
| 33 | if m[2].strip() == 'System RAM': | ||
| 34 | system_ram.append(long(m[0], 16)) | ||
| 35 | system_ram.append(long(m[1], 16)) | ||
| 36 | if m[2].strip() == 'Persistent Memory': | ||
| 37 | pmem.append(long(m[0], 16)) | ||
| 38 | pmem.append(long(m[1], 16)) | ||
| 39 | |||
| 40 | def print_memory_type(): | ||
| 41 | print "Event: %s" % (event_name) | ||
| 42 | print "%-40s %10s %10s\n" % ("Memory type", "count", "percentage"), | ||
| 43 | print "%-40s %10s %10s\n" % ("----------------------------------------", \ | ||
| 44 | "-----------", "-----------"), | ||
| 45 | total = sum(load_mem_type_cnt.values()) | ||
| 46 | for mem_type, count in sorted(load_mem_type_cnt.most_common(), \ | ||
| 47 | key = lambda(k, v): (v, k), reverse = True): | ||
| 48 | print "%-40s %10d %10.1f%%\n" % (mem_type, count, 100 * count / total), | ||
| 49 | |||
| 50 | def trace_begin(): | ||
| 51 | parse_iomem() | ||
| 52 | |||
| 53 | def trace_end(): | ||
| 54 | print_memory_type() | ||
| 55 | f.close() | ||
| 56 | |||
| 57 | def is_system_ram(phys_addr): | ||
| 58 | #/proc/iomem is sorted | ||
| 59 | position = bisect.bisect(system_ram, phys_addr) | ||
| 60 | if position % 2 == 0: | ||
| 61 | return False | ||
| 62 | return True | ||
| 63 | |||
| 64 | def is_persistent_mem(phys_addr): | ||
| 65 | position = bisect.bisect(pmem, phys_addr) | ||
| 66 | if position % 2 == 0: | ||
| 67 | return False | ||
| 68 | return True | ||
| 69 | |||
| 70 | def find_memory_type(phys_addr): | ||
| 71 | if phys_addr == 0: | ||
| 72 | return "N/A" | ||
| 73 | if is_system_ram(phys_addr): | ||
| 74 | return "System RAM" | ||
| 75 | |||
| 76 | if is_persistent_mem(phys_addr): | ||
| 77 | return "Persistent Memory" | ||
| 78 | |||
| 79 | #slow path, search all | ||
| 80 | f.seek(0, 0) | ||
| 81 | for j in f: | ||
| 82 | m = re.split('-|:',j,2) | ||
| 83 | if long(m[0], 16) <= phys_addr <= long(m[1], 16): | ||
| 84 | return m[2] | ||
| 85 | return "N/A" | ||
| 86 | |||
| 87 | def process_event(param_dict): | ||
| 88 | name = param_dict["ev_name"] | ||
| 89 | sample = param_dict["sample"] | ||
| 90 | phys_addr = sample["phys_addr"] | ||
| 91 | |||
| 92 | global event_name | ||
| 93 | if event_name == None: | ||
| 94 | event_name = name | ||
| 95 | load_mem_type_cnt[find_memory_type(phys_addr)] += 1 | ||