- 14 Aug, 2024 9 commits
-
-
Kan Liang authored
Display the branch counter histogram in the annotation view. Press 'B' to display the branch counter's abbreviation list as well. Samples: 1M of events 'anon group { branch-instructions:ppp, branch-misses }', 4000 Hz, Event count (approx.): f3 /home/sdp/test/tchain_edit [Percent: local period] Percent │ IPC Cycle Branch Counter (Average IPC: 1.39, IPC Coverage: 29.4%) │ 0000000000401755 <f3>: 0.00 0.00 │ endbr64 │ push %rbp │ mov %rsp,%rbp │ movl $0x0,-0x4(%rbp) 0.00 0.00 │1.33 3 |A |- | ↓ jmp 25 11.03 11.03 │ 11: mov -0x4(%rbp),%eax │ and $0x1,%eax │ test %eax,%eax 17.13 17.13 │2.41 1 |A |- | ↓ je 21 │ addl $0x1,-0x4(%rbp) 21.84 21.84 │2.22 2 |AA |- | ↓ jmp 25 17.13 17.13 │ 21: addl $0x1,-0x4(%rbp) 21.84 21.84 │ 25: cmpl $0x270f,-0x4(%rbp) 11.03 11.03 │0.61 3 |A |- | ↑ jle 11 │ nop │ pop %rbp 0.00 0.00 │0.24 20 |AA |B | ← ret Originally-by:Tinghao Zhang <tinghao.zhang@intel.com> Reviewed-by:
Andi Kleen <ak@linux.intel.com> Signed-off-by:
Kan Liang <kan.liang@linux.intel.com> Acked-by:
Namhyung Kim <namhyung@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: https://lore.kernel.org/r/20240813160208.2493643-8-kan.liang@linux.intel.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Kan Liang authored
Reusing the existing --total-cycles option to display the branch counters. Add a new PERF_HPP_REPORT__BLOCK_BRANCH_COUNTER to display the logged branch counter events. They are shown right after all the cycle-related annotations. Extend the 'struct block_info' to store and pass the branch counter related information. The annotation_br_cntr_entry() is to print the histogram of each branch counter event. If the number of logged events is less than 4, the exact number of the abbr name is printed. Otherwise, using '+' to stands for more than 3 events. Assume the number of logged events is less than 4. The annotation_br_cntr_abbr_list() prints the branch counter's abbreviation list. Press 'B' to display the list in the TUI mode. $ perf record -e "{branch-instructions:ppp,branch-misses}:S" -j any,counter $ perf report --total-cycles --stdio # To display the perf.data header info, please use --header/--header-only options. # # # Total Lost Samples: 0 # # Samples: 1M of events 'anon group { branch-instructions:ppp, branch-misses }' # Event count (approx.): 1610046 # # Branch counter abbr list: # branch-instructions:ppp = A # branch-misses = B # '-' No event occurs # '+' Event occurrences may be lost due to branch counter saturated # # Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles Branch Counter [Program Block Range] # ............... .............. ........... .......... .............. .................. # 57.55% 2.5M 0.00% 3 |A |- | ... 25.27% 1.1M 0.00% 2 |AA |- | ... 15.61% 667.2K 0.00% 1 |A |- | ... 0.16% 6.9K 0.81% 575 |A |- | ... 0.16% 6.8K 1.38% 977 |AA |- | ... 0.16% 6.8K 0.04% 28 |AA |B | ... 0.15% 6.6K 1.33% 946 |A |- | ... 0.11% 4.5K 0.06% 46 |AAA+|- | ... 0.10% 4.4K 0.88% 624 |A |- | ... 0.09% 3.7K 0.74% 524 |AAA+|B | ... With -v applied, # Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles Branch Counter [Program Block Range] # ............... .............. ........... .......... .............. .................. # 57.55% 2.5M 0.00% 3 A=1 ,B=- ... 25.27% 1.1M 0.00% 2 A=2 ,B=- ... 15.61% 667.2K 0.00% 1 A=1 ,B=- ... 0.16% 6.9K 0.81% 575 A=1 ,B=- ... 0.16% 6.8K 1.38% 977 A=2 ,B=- ... 0.16% 6.8K 0.04% 28 A=2 ,B=1 ... 0.15% 6.6K 1.33% 946 A=1 ,B=- ... 0.11% 4.5K 0.06% 46 A=3+,B=- ... 0.10% 4.4K 0.88% 624 A=1 ,B=- ... 0.09% 3.7K 0.74% 524 A=3+,B=1 ... Reviewed-by: -
Kan Liang authored
There could be several branch counter events. If perf tool output the result via the format "event name + a number", the line could be very long and hard to read. An abbreviation is introduced to replace the full event name in the display. The abbreviation starts from 'A' to 'Z9', which can support up to 286 events. The same abbreviation will be assigned if the same events are found in the evlist. The next patch will utilize the abbreviation name to show the branch counter events in the output. Reviewed-by:
-
Kan Liang authored
When annotating a basic block, it's useful to display the occurrences of other events in the block. The branch counter feature is only available for newer Intel platforms. So a dedicated option to display the branch counters is not introduced. Reuse the existing --total-cycles option, which triggers the annotation of a basic block and displays the cycle-related annotation. When the branch counters information is available, the branch counters are automatically appended after all the cycle-related annotation. Accounting the branch counters as well when accounting the cycles in hist__account_cycles(). In 'struct annotated_branch', introduce a br_cntr array to save the accumulation of each branch counter. In a sample, all the branch counters for a branch are saved in a u64 space. Because the saturation of a branch counter is small, e.g., for Intel Sierra Forest, the saturation is only 3. Add ANNOTATION__BR_CNTR_SATURATED_FLAG to indicate if a branch counter once saturated. That can be used to indicate a potential event lost because of the saturation. Reviewed-by:
-
Kan Liang authored
The branch counters logging (A.K.A LBR event logging) introduces a per-counter indication of precise event occurrences in LBRs. The kernel only dumps the number of occurrences into a record. The perf tool has to map the number to the corresponding event. Add evlist__update_br_cntr() to go through the evlist to pick the events that are configured to be logged. Assign a logical idx to track them, and add the total number of the events in the leader event. The total number will be used to allocate the space to save the branch counters for a block. The logical idx will be used to locate the corresponding event quickly in the following patches. It only needs to iterate the evlist once. The evsel__has_branch_counters() is also optimized. Reviewed-by:
-
Kan Liang authored
A false overflow warning is triggered if a sample doesn't have any LBRs recorded and the branch counters feature is enabled. The current code does OVERFLOW_CHECK_u64() at the very beginning when reading the information of branch counters. It assumes that there is at least one LBR in the PEBS record. But it is a valid case that 0 LBR is recorded especially in a high context switch. Remove the OVERFLOW_CHECK_u64(). The later OVERFLOW_CHECK() should be good enough to check the overflow when reading the information of the branch counters. Fixes: 9fbb4b02 ("perf tools: Add branch counter knob") Reviewed-by:
-
Kan Liang authored
The --total-cycles may output wrong information with the --stdio. For example: # perf record -e "{cycles,instructions}",cache-misses -b sleep 1 # perf report --total-cycles --stdio The total cycles output of {cycles,instructions} and cache-misses are almost the same. # Samples: 938 of events 'anon group { cycles, instructions }' # Event count (approx.): 938 # # Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Program Block Range] # ............... .............. ........... .......... ..................................................> # 11.19% 2.6K 0.10% 21 [perf_iterate_ctx+48 -> > 5.79% 1.4K 0.45% 97 [__intel_pmu_enable_all.constprop.0+80 -> __intel_> 5.11% 1.2K 0.33% 71 [native_write_msr+0 ->> # Samples: 293 of event 'cache-misses' # Event count (approx.): 293 # # Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Program Block Range] # ............... .............. ........... .......... ..................................................> # 11.19% 2.6K 0.13% 21 [perf_iterate_ctx+48 -> > 5.79% 1.4K 0.59% 97 [__intel_pmu_enable_all.constprop.0+80 -> __intel_> 5.11% 1.2K 0.43% 71 [native_write_msr+0 ->> With the symbol_conf.event_group, the 'perf report' should only report the block information of the leader event in a group. However, the current implementation retrieves the next event's block information, rather than the next group leader's block information. Make sure the index is updated even if the event is skipped. With the patch, # Samples: 293 of event 'cache-misses' # Event count (approx.): 293 # # Sampled Cycles% Sampled Cycles Avg Cycles% Avg Cycles [Program Block Range] # ............... .............. ........... .......... ..................................................> # 37.98% 9.0K 4.05% 299 [perf_event_addr_filters_exec+0 -> perf_event_a> 11.19% 2.6K 0.28% 21 [perf_iterate_ctx+48 -> > 5.79% 1.4K 1.32% 97 [__intel_pmu_enable_all.constprop.0+80 -> __intel_> Fixes: 6f7164fa ("perf report: Sort by sampled cycles percent per block for stdio") Reviewed-by: -
Ian Rogers authored
Help to better identify the location of test failures but dumping the failing test in the trap handler. Signed-off-by:
Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Richter <tmricht@linux.ibm.com> Link: https://lore.kernel.org/r/20240813040613.882075-2-irogers@google.comSigned-off-by:
-
Ian Rogers authored
lock__parse() calls disasm_line__parse() passing &ops->locked.ins.name that will use strdup() to populate it. Ensure ops->locked.ins.name is freed in lock__delete(). Found with address/leak sanitizer. Signed-off-by:
-
- 13 Aug, 2024 7 commits
-
-
Ian Rogers authored
The DSO build id is injected when the dso is first encountered but the checking for first encountered only looks at the sample->ip not the entire callchain. Use the callchain logic to ensure all build ids are inserted. Fixes: 454c407e ("perf: add perf-inject builtin") Signed-off-by:
-
Ian Rogers authored
Add a for_each callback style API to callchain with sample__for_each_callchain_node(). Possibly in the future such an API can avoid the overhead of constructing the call chain list. Signed-off-by:
-
Weilin Wang authored
Intel TPEBS sampling mode is supported through perf record. The counting mode code uses perf record to capture retire_latency value and use it in metric calculation. This test checks the counting mode code on Intel platforms. Committer testing: root@x1:~# perf test tpebs 123: test Intel TPEBS counting mode : Ok root@x1:~# set -o vi root@x1:~# perf test tpebs 123: test Intel TPEBS counting mode : Ok root@x1:~# perf test -v tpebs 123: test Intel TPEBS counting mode : Ok root@x1:~# perf test -vvv tpebs 123: test Intel TPEBS counting mode: --- start --- test child forked, pid 16603 Testing without --record-tpebs Testing with --record-tpebs ---- end(0) ---- 123: test Intel TPEBS counting mode : Ok root@x1:~# Reviewed-by:
Weilin Wang <weilin.wang@intel.com> Acked-by:
-
Weilin Wang authored
TPEBS (Timed PEBS(Precise Event-Based Sampling)) is a new feature Intel PMU from Granite Rapids microarchitecture. It will be used in new TMA (Top-Down Microarchitecture Analysis) releases. Add related introduction to documents while adding new code to support it in 'perf stat'. Reviewed-by:
-
Weilin Wang authored
With this command line option, TPEBS recording is turned off in 'perf stat' on default. It will only be turned on when this option is given in 'perf stat' command. Example with --record-tpebs: perf stat -M tma_split_loads -C1-4 --record-tpebs sleep 1 [ perf record: Woken up 2 times to write data ] [ perf record: Captured and wrote 0.044 MB - ] Performance counter stats for 'CPU(s) 1-4': 53,259,156,071 cpu_core/TOPDOWN.SLOTS/ # 1.6 % tma_split_loads (50.00%) 15,867,565,250 cpu_core/topdown-retiring/ (50.00%) 15,655,580,731 cpu_core/topdown-mem-bound/ (50.00%) 11,738,022,218 cpu_core/topdown-bad-spec/ (50.00%) 6,151,265,424 cpu_core/topdown-fe-bound/ (50.00%) 20,445,917,581 cpu_core/topdown-be-bound/ (50.00%) 6,925,098,013 cpu_core/L1D_PEND_MISS.PENDING/ (50.00%) 3,838,653,421 cpu_core/MEMORY_ACTIVITY.STALLS_L1D_MISS/ (50.00%) 4,797,059,783 cpu_core/EXE_ACTIVITY.BOUND_ON_LOADS/ (50.00%) 11,931,916,714 cpu_core/CPU_CLK_UNHALTED.THREAD/ (50.00%) 102,576,164 cpu_core/MEM_LOAD_COMPLETED.L1_MISS_ANY/ (50.00%) 64,071,854 cpu_core/MEM_INST_RETIRED.SPLIT_LOADS/ (50.00%) 3 cpu_core/MEM_INST_RETIRED.SPLIT_LOADS/R 1.003049679 seconds time elapsed Example without --record-tpebs: perf stat -M tma_contested_accesses -C1 sleep 1 Performance counter stats for 'CPU(s) 1': 50,203,891 cpu_core/TOPDOWN.SLOTS/ # 0.0 % tma_contested_accesses (63.60%) 10,040,777 cpu_core/topdown-retiring/ (63.60%) 6,890,729 cpu_core/topdown-mem-bound/ (63.60%) 2,756,463 cpu_core/topdown-bad-spec/ (63.60%) 10,828,288 cpu_core/topdown-fe-bound/ (63.60%) 28,350,432 cpu_core/topdown-be-bound/ (63.60%) 98 cpu_core/OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM/ (63.70%) 577,520 cpu_core/MEMORY_ACTIVITY.STALLS_L2_MISS/ (54.62%) 313,339 cpu_core/MEMORY_ACTIVITY.STALLS_L3_MISS/ (54.62%) 14,155 cpu_core/MEM_LOAD_RETIRED.L1_MISS/ (45.54%) 0 cpu_core/OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD/ (36.30%) 8,468,077 cpu_core/CPU_CLK_UNHALTED.THREAD/ (45.38%) 198 cpu_core/MEM_LOAD_L3_HIT_RETIRED.XSNP_MISS/ (45.38%) 8,324 cpu_core/MEM_LOAD_RETIRED.FB_HIT/ (45.38%) 3,388,031,520 TSC 23,226,785 cpu_core/CPU_CLK_UNHALTED.REF_TSC/ (54.46%) 80 cpu_core/MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD/ (54.46%) 0 cpu_core/MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD/R 0 cpu_core/MEM_LOAD_L3_HIT_RETIRED.XSNP_MISS/R 1,006,816,667 ns duration_time 1.002537737 seconds time elapsed Reviewed-by: -
Weilin Wang authored
Add MTL metric JSON file for TMA4.8. Some of the metrics' formulas use TPEBS retire_latency in MTL. This also includes lated E-Core TMA3.6 changes. Reviewed-by:
-
Weilin Wang authored
perf stat: Fork and launch 'perf record' when 'perf stat' needs to get retire latency value for a metric. When retire_latency value is used in a metric formula, evsel would fork a 'perf record' process with "-e" and "-W" options. 'perf record' will collect required retire_latency values in parallel while 'perf stat' is collecting counting values. At the point of time that 'perf stat' stops counting, evsel would stop 'perf record' by sending sigterm signal to 'perf record' process. Sampled data will be processed to get retire latency value. Another thread is required to synchronize between 'perf stat' and 'perf record' when we pass data through pipe. Retire_latency evsel is not opened for 'perf stat' so that there is no counter wasted on it. This commit includes code suggested by Namhyung to adjust reading size for groups that include retire_latency evsels. In current :R parsing implementation, the parser would recognize events with retire_latency modifier and insert them into the evlist like a normal event. Ideally, we need to avoid counting these events. In this commit, at the time when a retire_latency evsel is read, set the retire latency value processed from the sampled data to count value. This sampled retire latency value will be used for metric calculation and final event count print out. No special metric calculation and event print out code required for retire_latency events. Reviewed-by:
-
- 12 Aug, 2024 24 commits
-
-
Weilin Wang authored
When in PIPE mode, allow to use fd dynamically opened and asigned to data->file.fd instead of STDIN_FILENO or STDOUT_FILENO. Reviewed-by:
-
Ian Rogers authored
Retirement latency is a separate sampled count used on newer Intel CPUs. Reviewed-by:
-
Ian Rogers authored
Make tool const now that all uses are const and perf_tool__fill_defaults() won't be used. The aim is to better capture that sessions don't mutate tools. Signed-off-by:
-
Ian Rogers authored
Now all tools are fully initialized prior to use it has no use so remove. Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Ensure tool is initialized to avoid lazy initialization pattern so that more uses of struct perf_tool can be made const. Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Use perf_tool__init() so that more uses of 'struct perf_tool' can be const and not relying on perf_tool__fill_defaults(). Signed-off-by:
-
Ian Rogers authored
Reduce scope of build_id__mark_dso_hit_ops() to the scope of function perf_session__list_build_ids, its only use, and use perf_tool__init() for the default values. Move perf_event__exit_del_thread() to event.[ch] so it can be used in builtin-buildid-list.c. Signed-off-by:
-
Ian Rogers authored
Reduce the scope of the tool from global/static to just that of the cmd_kmem function where the session is scoped. Use the perf_tool__init() to initialize default values. Signed-off-by:
-
