- 13 Aug, 2022 18 commits
-
-
Zhengjun Xing authored
Update the events to v25, the metrics are based on TMA 4.4 full, update events and metrics for haswellx by the latest event converter tools. Use script at: https://github.com/intel/event-converter-for-linux-perf/blob/master/download_and_gen.py to download and generate the latest events and metrics. Manually copy the haswellx files into perf. Signed-off-by:
Xing Zhengjun <zhengjun.xing@linux.intel.com> Tested-by:
Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: https://lore.kernel.org/r/20220812085239.3089231-5-zhengjun.xing@linux.intel.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Zhengjun Xing authored
Update to v16, the metrics are based on TMA 4.4 full, update events and add new metrics “UNCORE_FREQ” for cascadelakex. Use script at: https://github.com/intel/event-converter-for-linux-perf/blob/master/download_and_gen.py to download and generate the latest events and metrics. Manually copy the cascadelakex files into perf. Signed-off-by:
-
Zhengjun Xing authored
Update to v19, the metrics are based on TMA 4.4 full, update events and add new metrics “UNCORE_FREQ” for broadwellx. Use script at: https://github.com/intel/event-converter-for-linux-perf/blob/master/download_and_gen.py to download and generate the latest events and metrics. Manually copy the broadwellx files into perf. Signed-off-by:
-
Zhengjun Xing authored
The metrics are based on TMA 4.4 full, add new metrics “UNCORE_FREQ” for broadwellde. Use script at: https://github.com/intel/event-converter-for-linux-perf/blob/master/download_and_gen.py to download and generate the latest events and metrics. Manually copy the broadwellde files into perf. Signed-off-by:
-
Ian Rogers authored
If a shorter string ends a longer string then the shorter string may reuse the longer string at an offset. For example, on x86 the event arith.cycles_div_busy and cycles_div_busy can be folded, even though they have difference names the strings are identical after 6 characters. cycles_div_busy can reuse the arith.cycles_div_busy string at an offset of 6. In pmu-events.c this looks like the following where the 'also:' lists folded strings: /* offset=177541 */ "arith.cycles_div_busy\000\000pipeline\000Cycles the divider is busy\000\000\000event=0x14,period=2000000,umask=0x1\000\000\000\000\000\000\000\000\000" /* also: cycles_div_busy\000\000pipeline\000Cycles the divider is busy\000\000\000event=0x14,period=2000000,umask=0x1\000\000\000\000\000\000\000\000\000 */ As jevents.py combines multiple strings for an event into a larger string, the amount of folding is minimal as all parts of the event must align. Other organizations can benefit more from folding, but lose space by say recording more offsets. Signed-off-by:
-
Ian Rogers authored
The pmu_events array requires 15 pointers per entry which in position independent code need relocating. Change the array to be an array of offsets within a big C string. Only the offset of the first variable is required, subsequent variables are stored in order after the \0 terminator (requiring a byte per variable rather than 4 bytes per offset). The file size savings are: no jevents - the same 19,788,464bytes x86 jevents - ~16.7% file size saving 23,744,288bytes vs 28,502,632bytes all jevents - ~19.5% file size saving 24,469,056bytes vs 30,379,920bytes default build options plus NO_LIBBFD=1. For example, the x86 build savings come from .rela.dyn and .data.rel.ro becoming smaller by 3,157,032bytes and 3,042,016bytes respectively. .rodata increases by 1,432,448bytes, giving an overall 4,766,600bytes saving. To make metric strings more shareable, the topic is changed from say 'skx metrics' to just 'metrics'. To try to help with the memory layout the pmu_events are ordered as used by perf qsort comparator functions. Signed-off-by:
-
Ian Rogers authored
The pmu_event passed to the pmu_events_table_for_each_event is invalid after the loop. Copy the entire struct in metricgroup__find_metric. Reduce the scope of this function to static. Signed-off-by:
-
Ian Rogers authored
Hide that the pmu_event structs are an array with a new wrapper struct. Signed-off-by:
-
Ian Rogers authored
The current code assumes that a struct pmu_event can be iterated over forward until a NULL pmu_event is encountered. This makes it difficult to refactor pmu_event. Add a loop function taking a callback function that's passed the struct pmu_event. This way the pmu_event is only needed for one element and not an entire array. Switch existing code iterating over the pmu_event arrays to use the new loop function pmu_events_table_for_each_event. Signed-off-by:
-
Ian Rogers authored
Move arrays of pmu_events into the JSON code so that it may be regenerated and modified by the jevents.py script. Signed-off-by:
-
Ian Rogers authored
The simple metric resolution doesn't handle recursion properly, switch to use the full resolution as with the parse-metric tests which also increases coverage. Don't set the values for the metric backward as failures to generate a result are ignored. Signed-off-by:
-
Ian Rogers authored
Move usage of the table to pmu-events.c so it may be hidden. By abstracting the table the implementation can later be changed. Signed-off-by:
-
Ian Rogers authored
Preparation for hiding pmu_events_map as an implementation detail. While the map is passed, the table of events is all that is normally wanted. While modifying the function's types, rename pmu_events_map__find to pmu_events_table__find to match later encapsulation. Similarly rename pmu_add_cpu_aliases_map to pmu_add_cpu_aliases_table. Signed-off-by:
-
Ian Rogers authored
Move usage of the table to pmu-events.c so it may be hidden. By abstracting the table the implementation can later be changed. Signed-off-by:
-
Ian Rogers authored
Sort the JSON files entries on conversion to C. The sort order tries to replicated cmp_sevent from pmu.c so that the input there is already sorted except for sysfs events. Specifically, the sort order is given by the tuple: (not j.desc is None, fix_none(j.topic), fix_none(j.name), fix_none(j.pmu), fix_none(j.metric_name)) which is putting events with descriptions and topics before those without, then sorting by name, then pmu and finally metric_name Add the topic to JsonEvent on reading to simplify. Remove an unnecessary lambda in the JSON reading. Signed-off-by:
-
Ian Rogers authored
If a JSONDecoderError or similar is raised then it is useful to know the path. Print this and then raise the exception agan. Signed-off-by:
-
Ian Rogers authored
pmu_events_map has a type variable that is always initialized to "core" and a version variable that is never read. Remove these from the API as it is straightforward to add them back when necessary. Signed-off-by:
-
Ian Rogers authored
When 'all' is passed as the architecture generate a mapping table for all architectures. This simplifies testing. To identify the table for an architecture add an arch variable to the pmu_events_map. Signed-off-by:
-
- 12 Aug, 2022 9 commits
-
-
Yang Li authored
util/topdown.h is included twice in builtin-stat.c, remove one of them. Reported-by:
Abaci Robot <abaci@linux.alibaba.com> Signed-off-by:
Yang Li <yang.lee@linux.alibaba.com> Tested-by:
-
shaomin Deng authored
Delete the repeated word "into" in comments. Signed-off-by:
shaomin Deng <dengshaomin@cdjrlc.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lore.kernel.org/lkml/20220807160239.474-1-dengshaomin@cdjrlc.comSigned-off-by:
-
shaomin Deng authored
Delete the repeated word "to" in comments. Signed-off-by:
-
shaomin Deng authored
Delete repeated word "and" in comments. Signed-off-by:
-
shaomin Deng authored
Delete the repeated word "from" in code. Signed-off-by:
-
shaomin Deng authored
Delete the redundant word "then" in comments. Signed-off-by:
-
Martin Liška authored
When one uses -p $not_existing_pid, the output of --help is printed: $ perf record -p 123456789 2>&1 | head -n3 Usage: perf record [<options>] [<command>] or: perf record [<options>] -- <command> [<options>] Let's change it something similar what perf top -p $not_existing_pid prints: $ ./perf top -p 123456789 --stdio Error: Couldn't create thread/CPU maps: No such process Newly suggested error message: $ ./perf record -p 123456789 Couldn't create thread/CPU maps: No such process Signed-off-by:Martin Liška <mliska@suse.cz> Tested-by:
-
Martin Liška authored
When ending a 'perf record' session, the querying of a debuginfod server can take quite some time. Inform a user about it when -v options is used. Signed-off-by:
-
Martin Liška authored
Use tabs instead of 8 spaces for the indentation. Signed-off-by:
-
- 11 Aug, 2022 13 commits
-
-
Leo Yan authored
Since the new display option 'peer' is introduced, this patch is to update the documentation to reflect it. Reviewed-by:
Ali Saidi <alisaidi@amazon.com> Signed-off-by:
Leo Yan <leo.yan@linaro.org> Acked-by:
-
Leo Yan authored
Since Arm64 arch doesn't support HITMs flags, this patch changes to use 'peer' as default display if user doesn't specify any type; for other arches, it still uses 'tot' as default display type if user doesn't specify it. This patch changes to call perf_session__new() in an earlier place, so session environment can be initialized ahead and arch info can be used for setting display type. Suggested-by:
-
Leo Yan authored
This patch adds a new option 'peer' so can sort on the cache hit for peer snooping. For displaying with option 'peer', the "Shared Data Cache Line Table" and "Shared Cache Line Distribution Pareto" both sort with the metrics "tot_peer". As result, we can get the 'peer' display: # perf c2c report -d peer --coalesce tid,pid,iaddr,dso -N --stdio ================================================= Shared Data Cache Line Table ================================================= # # ----------- Cacheline ---------- Peer ------- Load Peer ------- Total Total Total --------- Stores -------- ----- Core Load Hit ----- - LLC Load Hit -- - RMT Load Hit -- --- Load Dram ---- # Index Address Node PA cnt Snoop Total Local Remote records Loads Stores L1Hit L1Miss N/A FB L1 L2 LclHit LclHitm RmtHit RmtHitm Lcl Rmt # ..... .................. .... ...... ....... ....... ....... ....... ....... ....... ....... ....... ....... ....... ....... ....... ....... ........ ....... ........ ....... ........ ........ # 0 0xaaaac17d6000 N/A 0 100.00% 99 99 0 18851 18851 0 0 0 0 0 18752 0 99 0 0 0 0 0 ================================================= Shared Cache Line Distribution Pareto ================================================= # # -- Peer Snoop -- ------- Store Refs ------ --------- Data address --------- ---------- cycles ---------- Total cpu Shared # Num Rmt Lcl L1 Hit L1 Miss N/A Offset Node PA cnt Pid Tid Code address rmt peer lcl peer load records cnt Symbol Object Source:Line Node{cpus %peers %stores} # ..... ....... ....... ....... ....... ....... .................. .... ...... ....... ................. .................. ........ ........ ........ ....... ........ ...................... ................ ............... .... # ---------------------------------------------------------------------- 0 0 99 0 0 0 0xaaaac17d6000 ---------------------------------------------------------------------- 0.00% 3.03% 0.00% 0.00% 0.00% 0x20 N/A 0 3603 3603:memstress 0xaaaac17c25ac 0 376 41 9314 2 [.] 0x00000000000025ac memstress memstress[25ac] 0{ 2 100.0% n/a} 0.00% 3.03% 0.00% 0.00% 0.00% 0x20 N/A 0 3603 3606:memstress 0xaaaac17c25ac 0 375 44 9155 1 [.] 0x00000000000025ac memstress memstress[25ac] 0{ 1 100.0% n/a} 0.00% 48.48% 0.00% 0.00% 0.00% 0x29 N/A 0 3603 3606:memstress 0xaaaac17c3e88 0 180 170 65 1 [.] 0x0000000000003e88 memstress memstress[3e88] 0{ 1 100.0% n/a} 0.00% 45.45% 0.00% 0.00% 0.00% 0x29 N/A 0 3603 3603:memstress 0xaaaac17c3e88 0 180 175 70 2 [.] 0x0000000000003e88 memstress memstress[3e88] 0{ 2 100.0% n/a} Reviewed-by: -
Leo Yan authored
The display type is shown by combination the display string array and a suffix string "HITMs", which is not friendly to extend display for other sorting type (e.g. extension for peer operations). This patch moves the suffix string "HITMs" into display string array for HITM types, so it can allow us to not necessarily to output string "HITMs" for new incoming display type. Reviewed-by:
-
Leo Yan authored
The node header array contains 3 items, each item is used for one of the 3 flavors for node accessing info. To extend sorting on other snooping type and not always stick to HITMs, the second header string "Node{cpus %hitms %stores}" should be adjusted (e.g. it's changed as "Node{cpus %peer %stores}"). For this reason, this patch changes the node header array to three flat variables and uses switch-case in function setup_nodes_header(), thus it is easier for altering the header string. Reviewed-by: -
Leo Yan authored
Use more general naming for the main sort dimension, this can allow us not to sort only on HITM snoop type, so it can be extended to support other costly snooping operations. So rename the dimension to the prefix 'percent_costly_". Reviewed-by:
-
Leo Yan authored
Perf c2c tool has an assumption that it heavily depends on HITM snoop type to detect cache false sharing, unfortunately, HITM is not supported on some architectures. Essentially, perf c2c tool wants to find some very costly snooping operations for false cache sharing, this means it's not necessarily to stick using HITM tags and we can explore other snooping types (e.g. SNOOPX_PEER). For this reason, this patch renames HITM related display macros with suffix '_HITM', so it can be distinct if later add more display types for on other snooping type. Reviewed-by:
-
Leo Yan authored
This patch adds two dimensions for the mean value of peer operations. Reviewed-by:
-
Leo Yan authored
This patch adds dimensions of peer ops, which will be used for Shared cache line distribution pareto. It adds the percentage dimensions for local and remote peer operations, and the dimensions for accounting operation numbers which is used for stdio mode. Reviewed-by:
-
Leo Yan authored
This patch adds three dimensions for peer load operations of 'lcl_peer', 'rmt_peer' and 'tot_peer'. These three dimensions will be used in the shared data cache line table. Reviewed-by:
-
Leo Yan authored
This patch outputs statistics for peer snooping for whole trace events and global shared cache line. Reviewed-by:
-
Leo Yan authored
Since the flag PERF_MEM_SNOOPX_PEER is added to support cache snooping from peer cache line, it can come from a peer core, a peer cluster, or a remote NUMA node. This patch adds statistics for the flag PERF_MEM_SNOOPX_PEER. Note, we take PERF_MEM_SNOOPX_PEER as an affiliated info, it needs to cooperate with cache level statistics. Therefore, we account the load operations for both the cache level's metrics (e.g. ld_l2hit, ld_llchit, etc.) and peer related metrics when flag PERF_MEM_SNOOPX_PEER is set. So three new metrics are introduced: 'lcl_peer' is for local cache access, the metric 'rmt_peer' is for remote access (includes remote DRAM and any caches in remote node), and the metric 'tot_peer' is accounting the sum value of 'lcl_peer' and 'rmt_peer'. Reviewed-by:
-
Ali Saidi authored
When synthesizing data from SPE, augment the type with source information for Arm Neoverse cores. The field is IMPLDEF but the Neoverse cores all use the same encoding. I can't find encoding information for any other SPE implementations to unify their choices with Arm's thus that is left for future work. This change populates the mem_lvl_num for Neoverse cores as well as the deprecated mem_lvl namespace. Reviewed-by:
German Gomez <german.gomez@arm.com> Reviewed-by:
-
