[Backport][devel-5.4] AMD Turin Genoa IBS skew sampling patches · Pull Request !501 · OpenCloudOS/OpenCloudOS-Kernel

**IBS skew sampling patches for AMD Turin and Genoa systems**

Base patch:
perf record ibs: Warn about sampling period skew

Dependencies:
perf/aux: Allow using AUX data in perf samples
perf evsel: Move and globalize perf_evsel__find_pmu() and perf_evsel__is_aux_event()
perf env: Add perf_env__numa_node()
perf env: Add perf_env__cpuid, perf_env__{nr_}pmu_mappings
perf evsel: Make evsel__env() always return a valid env
perf record: Move sb_evlist to 'struct record'
perf top: Move sb_evlist to 'struct perf_top'
perf bpf: Decouple creating the evlist from adding the SB event
libperf: Add perf_mmap__init() function
libperf: Add 'struct perf_mmap_param'
libperf: Adopt perf_mmap__mmap_len() function from tools/perf
libperf: Adopt perf_mmap__mmap() function from tools/perf
libperf: Adopt perf_mmap__get() function from tools/perf
libperf: Adopt perf_mmap__unmap() function from tools/perf
libperf: Adopt perf_mmap__put() function from tools/perf
perf tools: Use perf_mmap way to detect aux mmap
libperf: Adopt perf_mmap__consume() function from tools/perf
libperf: Adopt perf_mmap__read_init() from tools/perf
libperf: Adopt perf_mmap__read_done() from tools/perf
libperf: Adopt perf_mmap__read_event() from tools/perf
perf evlist: Move the sideband thread routines to separate object
perf tools: Avoid 'sample_reg_masks' being const + weak
tools: Fix off-by 1 relative directory includes
perf evlist: Factor out asprintf routine to build a tracepoint pid filter
perf evlist: Introduce append_tp_filter() method
perf string: Export asprintf__tp_filter_pids()
perf tools: Move routines that probe for perf API features to separate file
perf evlist: Allow reusing the side band thread for more purposes
perf pmu: When using default config, record which bits of config were changed by the user
perf parse: Refactor 'struct perf_evsel_config_term'
tools headers UAPI: Update tools's copy of linux/perf_event.h
perf evsel: Support PERF_SAMPLE_BRANCH_HW_INDEX
perf/core: Add PERF_RECORD_CGROUP event
perf/core: Add PERF_SAMPLE_CGROUP feature
tools headers UAPI: Update tools's copy of linux/perf_event.h
perf tool: Provide an option to print perf_event_open args and return value
perf record: Add --all-cgroups option
perf evsel: Rename *perf_evsel__*set_sample_*() to *evsel__*set_sample_*()
perf/core: Add PERF_SAMPLE_DATA_PAGE_SIZE
tools headers UAPI: Update tools's copy of linux/perf_event.h
perf tools: Basic support for CGROUP event
perf record: Support new sample type for data page size
perf evsel: Emit warning about kernel not supporting the data page size sample_type bit
perf record: Add support for PERF_SAMPLE_CODE_PAGE_SIZE
perf/core: Add support for PERF_SAMPLE_CODE_PAGE_SIZE
perf/x86/intel: Support PERF_SAMPLE_DATA_PAGE_SIZE
perf/core: Add PERF_SAMPLE_WEIGHT_STRUCT
tools headers uapi: Update tools's copy of linux/perf_event.h
perf tools: Support PERF_SAMPLE_WEIGHT_STRUCT
perf evsel: Rename perf_evsel__new*() to evsel__new*()
perf record: Add dummy event during system wide synthesis
perf evlist: Fix the class prefix for 'struct evlist' 'add' evsel methods
perf pmu: Add pmu_id()
perf pmu: Use file system cache to optimize sysfs access
perf pmu: Save detected hybrid pmus to a global pmu list
perf pmu: Add support for PMU capabilities
perf pmu: Validate raw event with sysfs exported format bits
perf pmu: Add hybrid helper functions
perf/x86/intel: Support CPUID 10.ECX to disable fixed counters
perf/x86: Factor out x86_pmu_show_pmu_cap
perf/core: Factor out functions to allocate/free the task_ctx_data
perf/core: Use kmem_cache to allocate the PMU specific data
perf/x86/intel/lbr: Create kmem_cache for the LBR context data
tools headers uapi: Update tools's copy of linux/perf_event.h
perf record: Create two hybrid 'cycles' events by default
perf evsel: Don't set exclude_guest by default
perf stat: Factor out open error handling
perf affinity: Add infrastructure to save/restore affinity
perf evsel: Add iterator to iterate over events ordered by CPU
perf evlist: Maintain evlist->all_cpus
perf evsel: Avoid close(-1)
perf evsel: Add functions to close evsel on a CPU
perf stat: Use affinity for opening events
perf evsel: Rename perf_evsel__find_pmu() to evsel__find_pmu()
perf tools: Add hw_idx in struct branch_stack
perf record ibs: Warn about sampling period skew

Unit test
Without patch:
```
[root@hostos amd]# perf record -c 10000 -C 0 -e ibs_op/l3missonly=1/
WARNING: multiple event parsing errors
event syntax error: 'ibs_op/l3missonly=1/'
                            \___ unknown term
 
valid terms: cnt_ctl,config,config1,config2,name,period,freq,branch_type,time,call-graph,stack-size,no-inherit,inherit,max-stack,nr,no-overwrite,overwrite,driver-config,percore,aux-output,aux-sample-size
Run 'perf list' for a list of valid events
 
Usage: perf record [<options>] [<command>]
    or: perf record [<options>] -- <command> [<options>]
 
    -e, --event <event>   event selector. use 'perf list' to list available events
 
```
With patch:(will print the warning message)

```
[root@hostos perf]# ./perf record -c 10000 -C 0 -e ibs_op/l3missonly=1/ sleep 5
WARNING: Hw internally resets sampling period when L3 Miss Filtering is enabled
and tagged operation does not cause L3 Miss. This causes sampling period skew.
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.676 MB perf.data (11 samples) ]

```

类型	指派人员	状态
审查	chinaljp030 刘诗	已完成 (0/0人)
测试	chinaljp030 刘诗	已完成 (0/0人)

GVPOpenCloudOS/OpenCloudOS-Kernel

搜索帮助

GVP OpenCloudOS/OpenCloudOS-Kernel