Messages in this thread Patch in this message | | | From | kan.liang@linux ... | Subject | [PATCH 48/49] perf Documentation: Document intel-hybrid support | Date | Mon, 8 Feb 2021 07:25:45 -0800 |
| |
From: Jin Yao <yao.jin@linux.intel.com>
Add some words and examples to help understanding of Intel hybrid perf support.
Reviewed-by: Andi Kleen <ak@linux.intel.com> Signed-off-by: Jin Yao <yao.jin@linux.intel.com> --- tools/perf/Documentation/intel-hybrid.txt | 335 ++++++++++++++++++++++++++++++ tools/perf/Documentation/perf-record.txt | 1 + tools/perf/Documentation/perf-stat.txt | 2 + 3 files changed, 338 insertions(+) create mode 100644 tools/perf/Documentation/intel-hybrid.txt
diff --git a/tools/perf/Documentation/intel-hybrid.txt b/tools/perf/Documentation/intel-hybrid.txt new file mode 100644 index 0000000..bdf38d0 --- /dev/null +++ b/tools/perf/Documentation/intel-hybrid.txt @@ -0,0 +1,335 @@ +Intel hybrid support +-------------------- +Support for Intel hybrid events within perf tools. + +For some Intel platforms, such as AlderLake, which is hybrid platform and +it consists of atom cpu and core cpu. Each cpu has dedicated event list. +Part of events are available on core cpu, part of events are available +on atom cpu and even part of events are available on both. + +Kernel exports two new cpu pmus via sysfs: +/sys/devices/cpu_core +/sys/devices/cpu_atom + +The 'cpus' files are created under the directories. For example, + +cat /sys/devices/cpu_core/cpus +0-15 + +cat /sys/devices/cpu_atom/cpus +16-23 + +It indicates cpu0-cpu15 are core cpus and cpu16-cpu23 are atom cpus. + +Quickstart + +List hybrid event +----------------- + +As before, use perf-list to list the symbolic event. + +perf list + +inst_retired.any + [Fixed Counter: Counts the number of instructions retired. Unit: cpu_atom] +inst_retired.any + [Number of instructions retired. Fixed Counter - architectural event. Unit: cpu_core] + +The 'Unit: xxx' is added to brief description to indicate which pmu +the event is belong to. Same event name but with different pmu can +be supported. + +Use '--cputype' option to list core only event or atom only event. + +perf list --cputype atom + +inst_retired.any + [Fixed Counter: Counts the number of instructions retired. Unit: cpu_atom] + +Enable hybrid event with a specific pmu +--------------------------------------- + +To enable a core only event or atom only event, following syntax is supported: + + cpu_core/<event name>/ +or + cpu_atom/<event name>/ +or + --cputype core -e <event name> +or + --cputype atom -e <event name> + +For example, count the 'cycles' event on core cpus. + + perf stat -e cpu_core/cycles/ +or + perf stat --cputype core -e cycles + +If '--cputype' value conflicts with pmu prefix, '--cputype' is ignored and +a warning will be displayed. + +Create two events for one hardware event automatically +------------------------------------------------------ + +When creating one event and the event is available on both atom and core, +two events are created automatically. One is for atom, the other is for +core. Most of hardware events and cache events are available on both +cpu_core and cpu_atom. + +For hardware events, they have pre-defined configs (e.g. 0 for cycles). +But on hybrid platform, kernel needs to know where the event comes from +(from atom or from core). The original perf event type PERF_TYPE_HARDWARE +can't carry pmu information. So a new type PERF_TYPE_HARDWARE_PMU is +introduced. + +The new attr.config layout for PERF_TYPE_HARDWARE_PMU: + +0xDD000000AA +AA: original hardware event ID +DD: PMU type ID + +Cache event is similar. A new type PERF_TYPE_HW_CACHE_PMU is introduced. + +The new attr.config layout for PERF_TYPE_HW_CACHE_PMU: + +0xDD00CCBBAA +AA: original hardware cache ID +BB: original hardware cache op ID +CC: original hardware cache op result ID +DD: PMU type ID + +PMU type ID is retrieved from sysfs + +cat /sys/devices/cpu_atom/type +10 + +cat /sys/devices/cpu_core/type +4 + +When enabling a hardware event without specified pmu, such as, +perf stat -e cycles -a (use system-wide in this example), two events +are created automatically. + + ------------------------------------------------------------ + perf_event_attr: + type 6 + size 120 + config 0x400000000 + sample_type IDENTIFIER + read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING + disabled 1 + inherit 1 + exclude_guest 1 + ------------------------------------------------------------ + +and + + ------------------------------------------------------------ + perf_event_attr: + type 6 + size 120 + config 0xa00000000 + sample_type IDENTIFIER + read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING + disabled 1 + inherit 1 + exclude_guest 1 + ------------------------------------------------------------ + +type 6 is PERF_TYPE_HARDWARE_PMU. +0x4 in 0x400000000 indicates it's cpu_core pmu. +0xa in 0xa00000000 indicates it's cpu_atom pmu (atom pmu type id is random). + +The kernel creates 'cycles' (0x400000000) on cpu0-cpu15 (core cpus), +and create 'cycles' (0xa00000000) on cpu16-cpu23 (atom cpus). + +For perf-stat result, it displays two events: + + Performance counter stats for 'system wide': + + 14,240,632 cycles + 1,556,789 cycles + +The first 'cycles' is core event, the second 'cycles' is atom event. + +In event list, for the events with same event name, the first one is core +event, and the second one is atom event. + +Thread mode example: +-------------------- + +perf-stat reports the scaled counts for hybrid event and with a percentage +displayed. The percentage is the event's running time/enabling time. + +One example, 'triad_loop' runs on cpu16 (atom core), while we can see the +scaled value for core cycles is 160,444,092 and the percentage is 0.47%. + +perf stat -e cycles -- taskset -c 16 ./triad_loop + +As previous, two events are created. + +------------------------------------------------------------ +perf_event_attr: + type 6 + size 120 + config 0x400000000 + sample_type IDENTIFIER + read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING + disabled 1 + inherit 1 + enable_on_exec 1 + exclude_guest 1 +------------------------------------------------------------ + +and + +------------------------------------------------------------ +perf_event_attr: + type 6 + size 120 + config 0xa00000000 + sample_type IDENTIFIER + read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING + disabled 1 + inherit 1 + enable_on_exec 1 + exclude_guest 1 +------------------------------------------------------------ + +cycles: 0: 751035 206505423 966647 +cycles: 0: 601381075 206505423 205538776 +cycles: 160444092 206505423 966647 +cycles: 604209364 206505423 205538776 + + Performance counter stats for 'taskset -c 16 ./triad_loop': + + 160,444,092 cycles (0.47%) + 604,209,364 cycles (99.53%) + + 0.207494637 seconds time elapsed + +We can use '--no-scale' to see the original data. + +perf stat -e cycles --no-scale -- taskset -c 16 ./triad_loop + +cycles: 0: 755604 206642115 956961 +cycles: 0: 601433846 206642115 205685154 +cycles: 755604 206642115 956961 +cycles: 601433846 206642115 205685154 + + Performance counter stats for 'taskset -c 16 ./triad_loop': + + 755,604 cycles (0.46%) + 601,433,846 cycles (99.54%) + +Suppert metrics with hybrid events: +----------------------------------- + +For hybrid platform, the same metric is probably made of the events +with different pmus. + +For example "CPI". + +Atom: +CPI = cpu_clk_unhalted.core / inst_retired.any_p + +Core: +CPI = cpu_clk_unhalted.thread / inst_retired.any + +inst_retired.any_p and inst_retired.any are available on both core +and atom CPU. But cpu_clk_unhalted.core is only available on atom and +cpu_clk_unhalted.thread is only available on core CPU. + +The metric group string is expanded to: +"{inst_retired.any,cpu_clk_unhalted.thread}:W,{cpu_clk_unhalted.core,inst_retired.any_p}:W". + +"{inst_retired.any,cpu_clk_unhalted.thread}:W" applies CPI on core. +"{cpu_clk_unhalted.core,inst_retired.any_p}:W" applies CPI on atom. + +The difficulty is "inst_retired.any_p" and "inst_retired.any" +can be available on both. So add pmu suffixes "cpu_atom"/"cpu_core" to +limit the event group on a specified PMU. + +"{inst_retired.any,cpu_clk_unhalted.thread}:W#cpu_core,{cpu_clk_unhalted.core,inst_retired.any_p}:W#cpu_atom" + +That means the group "{inst_retired.any,cpu_clk_unhalted.thread}:W" is only +enabled on core and "{cpu_clk_unhalted.core,inst_retired.any_p}:W" is only +enabled on atom. + +For perf-stat result, it reports two 'CPI'. + +perf stat -M CPI -a -- sleep 1 + + Performance counter stats for 'system wide': + + 3,654,776 inst_retired.any # 4.76 CPI + 17,409,944 cpu_clk_unhalted.thread + 771,568 cpu_clk_unhalted.core # 6.83 CPI + 112,943 inst_retired.any_p + +First CPI is for core, the second CPI is for atom. +Of course, we can get core CPI only by enabling '--cputype' option. + +perf stat --cputype core -M CPI --no-merge -a -- sleep 1 + + Performance counter stats for 'system wide': + + 2,496,815 inst_retired.any [cpu_core] # 3.66 CPI + 9,149,133 cpu_clk_unhalted.thread [cpu_core] + +perf-record: +------------ + +If there is no '-e' specified in perf record, on hybrid platform, +it creates two default 'cycles' and adds them to event list. One +is for core, the other is for atom. + +Support the '--cputype' option for perf-record. It can only enable +the events on specific cpus with specified pmu type. + +perf record --cputype core + +Only the core 'cycles' are enabled on core cpus. + +perf-stat: +---------- + +If there is no '-e' specified in perf stat, on hybrid platform, +besides of software events, following events are created and +added to event list in order. + +core 'cycles', +atom 'cycles', +core 'instructions', +atom 'instructions', +core 'branches', +atom 'branches', +core 'branch-misses', +atom 'branch-misses' + +Support the '--cputype' option for perf-stat. It can only enable +the events on specific cpus with specified pmu type. + +perf stat --cputype core + +Only the core events are enabled on core cpus. + +core 'cycles', +core 'instructions', +core 'branches', +core 'branch-misses', + +Of course, both perf-stat and perf-record support to enable +hybrid event with a specific pmu. + +e.g. +perf stat -e cpu_core/cycles/ +perf stat -e cpu_atom/cycles/ +perf stat -e cpu_core/r1a/ +perf stat -e cpu_atom/L1-icache-loads/ +perf stat -e cpu_core/cycles/,cpu_atom/instructions/ +perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' + +But '{cpu_core/cycles/,cpu_atom/instructions/}' will return +"<not supported>" for 'instructions', because the pmus in +group are not matched (cpu_core vs. cpu_atom). \ No newline at end of file diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt index 34cf651..6fc01f0 100644 --- a/tools/perf/Documentation/perf-record.txt +++ b/tools/perf/Documentation/perf-record.txt @@ -680,6 +680,7 @@ measurements: wait -n ${perf_pid} exit $? +include::intel-hybrid.txt[] SEE ALSO -------- diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt index 3d083a3..43bdeaa 100644 --- a/tools/perf/Documentation/perf-stat.txt +++ b/tools/perf/Documentation/perf-stat.txt @@ -520,6 +520,8 @@ The fields are in this order: Additional metrics may be printed with all earlier fields being empty. +include::intel-hybrid.txt[] + SEE ALSO -------- linkperf:perf-top[1], linkperf:perf-list[1] -- 2.7.4
| |