- 04 Apr, 2023 1 commit
-
-
Artem Savkov authored
Append information about inlined functions to FP and LBR callchains from DWARF debuginfo when available. Do so by calling append_inlines() from add_callchain_ip(). Testing it: Frame-pointer mode recorded with 'perf record --call-graph=fp --freq=max -- ./a.out' #include <stdio.h> #include <stdint.h> static __attribute__((noinline)) uint32_t func5(uint32_t i) { return i + 10; } static uint32_t func4(uint32_t i) { return func5(i + 5); } static inline uint32_t func3(uint32_t i) { return func4(i + 4); } static __attribute__((noinline)) uint32_t func2(uint32_t i) { return func3(i + 3); } static uint32_t func1(uint32_t i) { return func2(i + 2); } __attribute__((noinline)) uint64_t entry(void) { uint64_t ret = 0; uint32_t i = 0; for (i = 0; i < 1000000; i++) { ret += func1(i); ret -= func2(i); ret += func3(i); ret += func4(i); ret -= func5(i); } return ret; } int main(int argc, char **argv) { printf("%s\n", __func__); return entry(); } ====== Here is the output I get with '--call-graph callee --no-children' ====== # To display the perf.data header info, please use --header/--header-only options. # # # Total Lost Samples: 0 # # Samples: 250 of event 'cycles:u' # Event count (approx.): 26819859 # # Overhead Command Shared Object Symbol # ........ ....... .................... ..................................... # 43.58% a.out a.out [.] func5 | |--28.93%--entry | main | __libc_start_call_main | --14.65%--func4 (inlined) | |--10.45%--entry | main | __libc_start_call_main | --4.20%--func3 (inlined) entry main __libc_start_call_main 38.80% a.out a.out [.] entry | |--23.27%--func4 (inlined) | | | |--20.28%--func3 (inlined) | | func2 | | main | | __libc_start_call_main | | | --2.99%--entry | main | __libc_start_call_main | |--8.17%--func5 | main | __libc_start_call_main | |--3.89%--func1 (inlined) | entry | main | __libc_start_call_main | --3.48%--entry main __libc_start_call_main 13.07% a.out a.out [.] func2 | ---func5 main __libc_start_call_main 1.54% a.out [unknown] [k] 0xffffffff81e011b7 1.16% a.out [unknown] [k] 0xffffffff81e00193 | --0.57%--__mmap64 (inlined) __mmap64 (inlined) 0.34% a.out ld-linux-x86-64.so.2 [.] __tunable_get_val 0.34% a.out ld-linux-x86-64.so.2 [.] strcmp 0.32% a.out libc.so.6 [.] strchr 0.31% a.out ld-linux-x86-64.so.2 [.] _dl_relocate_object 0.22% a.out ld-linux-x86-64.so.2 [.] _dl_init_paths 0.18% a.out ld-linux-x86-64.so.2 [.] get_common_cache_info.constprop.0 0.14% a.out ld-linux-x86-64.so.2 [.] __GI___tunables_init # # (Tip: Show individual samples with: perf script) # ====== It does not seem to be out of order, or at least it is consistent with what I get with dwarf unwinders. Committer notes: Adrian Hunter pointed out that this breaks --branch-history, so don't do it for branches, see the second Link below. Suggested-by:Andrii Nakryiko <andrii.nakryiko@gmail.com> Signed-off-by: <asavkov@redhat.com> Acked-by:
Namhyung Kim <namhyung@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Milian Wolff <milian.wolff@kdab.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: https://lore.kernel.org/r/20230316133557.868731-2-asavkov@redhat.com Link: https://lore.kernel.org/r/54129783-2960-84e1-05e9-97ac70ffb432@intel.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
- 21 Mar, 2023 5 commits
-
-
Rob Herring authored
perf_event_attr has gained a new field, config3, so add support for it extending the existing configN support. Signed-off-by:
Rob Herring <robh@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lore.kernel.org/lkml/20220914-arm-perf-tool-spe1-2-v2-v5-2-2cf5210b2f77@kernel.orgSigned-off-by:
-
James Clark authored
Generated from the telemetry solution repo[1] with this command: ./generate.py <linux-repo>/tools/perf/ --telemetry-files \ ../../data/pmu/cpu/neoverse/neoverse-n1.json Since this data source now includes the SPE events for N1, it has diverged from A76 which means the folder has to be split. The new data also uses more fine grained grouping, but this will be consistent for all future products. Long PublicDescriptions are now included even for common events because this can include product specific details. For non verbose mode the common BriefDescriptions remain the same. [1]: https://gitlab.arm.com/telemetry-solution/telemetry-solutionSigned-off-by:James Clark <james.clark@arm.com> Acked-by:
Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: John Garry <john.g.garry@oracle.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Will Deacon <will@kernel.org> Cc: linux-arm-kernel@lists.infradead.org Cc: renyu.zj@linux.alibaba.com Link: https://lore.kernel.org/r/20230320114601.524958-1-james.clark@arm.comSigned-off-by:
-
Bernhard M. Wiedemann authored
Without this, pmu-events.c would be generated with variations in ordering depending on non-deterministic filesystem readdir order. I tested that pmu-events.c still has the same number of lines and that perf list output works. This patch was done while working on reproducible builds for openSUSE, but also solves issues in Debian [1] and other distributions. [1] https://tests.reproducible-builds.org/debian/rb-pkg/unstable/i386/linux.htmlSigned-off-by:
Bernhard M. Wiedemann <bwiedemann@suse.de> Cc: Ian Rogers <irogers@google.com> Link: https://lore.kernel.org/r/20230321063032.19804-1-bwiedemann@suse.deSigned-off-by:
-
Leo Yan authored
It's good not to release resources for a program when kernel cleans up memory space, this patch explicitly releases histograms entries with hists__delete_entries(). Committer notice: This helps with memory leak checkers, but may delay exiting a tool by doing needless linked list traversals freeing lots of objects. Signed-off-by:
Leo Yan <leo.yan@linaro.org> Acked-by:
-
Leo Yan authored
hists__add_entry_ops() doesn't allocate a new histogram entry if it has an existing entry for a KVM event, in this case, find_create_kvm_event() allocates a 'struct kvm_info' but it's not used by any histograms and never freed. To fix the memory leak, this patch first introduces a refcnt and a set of functions for refcnt operations on 'struct kvm_info'. When the data structure is not anymore used (the refcnt hits zero) kvm_info__zput() will free the memory used. Committer: Provide a nop version of kvm_info__zput() to be used when HAVE_KVM_STAT_SUPPORT isn't defined as it is used unconditionally in hists__findnew_entry() and hist_entry__delete(). Signed-off-by:
-
- 20 Mar, 2023 11 commits
-
-
German Gomez authored
Add 'simd' sort field to visualize SIMD ops in 'perf report'. Rows are labeled with the SIMD ISA, and the type of predicate (if any): - [p] partial predicate - [e] empty predicate (no elements in the vector being used) Example with Arm SPE and SVE (Scalable Vector Extension): #include <arm_sve.h> double src[1025], dst[1025]; int main(void) { svfloat64_t vc = svdup_f64(1); for(;;) for(int i = 0; i < 1025; i += svcntd()) { svbool_t pg = svwhilelt_b64(i, 1025); svfloat64_t vsrc = svld1(pg, &src[i]); svfloat64_t vdst = svadd_x(pg, vsrc, vc); svst1(pg, &dst[i], vdst); } return 0; } ... compiled using "gcc-11 -march=armv8-a+sve -O3" Profiling on a platform that implements FEAT_SVE and FEAT_SPEv1p1: $ perf record -e arm_spe_0// -- ./a.out $ perf report --itrace=i1i -s overhead,pid,simd,sym Overhead Pid:Command Simd Symbol ........ ................ ....... ...................... 53.76% 10758:program [.] main 46.14% 10758:program [.] SVE [.] main 0.09% 10758:program [p] SVE [.] main The report shows 0.09% of the sampled SVE operations use partial predicates due to src and dst arrays not being multiples of the vector register lengths. Signed-off-by:German Gomez <german.gomez@arm.com> Acked-by:
-
German Gomez authored
Add flags from the Scalable Vector Extension (SVE) to the SPE samples which are available from Armv8.3 (FEAT_SPEv1p1). These will be displayed in a new SIMD sort field in a later commit. Signed-off-by:
-
German Gomez authored
Extend the decoder of Arm SPE records to support more fields from the operation packet type. Not all fields are being decoded by this commit. Only those needed to support the use-case SVE load/store/other operations. Suggested-by:
-
German Gomez authored
Add new field to 'struct perf_sample' to store flags related to SIMD ops. It will be used to store SIMD information from SVE and NEON when profiling using ARM SPE. Signed-off-by:
-
Adrian Hunter authored
Intel Flexible Return and Event Delivery (FRED) adds instructions ERETS (return to supervisor) and ERETU (return to user). Intel PT instruction decoder needs to know about these instructions because they are branch instructions. Similar to IRET instructions, when the decoder encounters one of these instructions it will match it to a TIP (target instruction pointer) packet that informs what the branch destination is. The existing "x86 instruction decoder - new instructions" test can be used to test the result e.g. $ perf test -v ins |& grep eret Decoded ok: f2 0f 01 ca erets Decoded ok: f3 0f 01 ca eretu Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Acked-by:
-
Adrian Hunter authored
UINTR and UIRET are listed in table 32-50 "CFE Packet Type and Vector Fields Details" in the Intel Processor Trace chapter of The Intel SDM Volume 3 version 078. The codes are for "User interrupt delivered" and "Exiting from user interrupt routine" respectively. Signed-off-by:
-
Ian Rogers authored
If finding a name doesn't find the sorted names then they are allocated and sorted. This shouldn't be done under a read lock as another reader may access it. Release the read lock and acquire the write lock, then release the write lock and reacquire the read lock. Signed-off-by:
-
Ian Rogers authored
machine__delete() doesn't delete threads. Add call to delete threads ahead of deleting the machine. Signed-off-by:
-
Ian Rogers authored
A later change will enforce that the map is put on this path regardless of success or error. Signed-off-by:
-
Ian Rogers authored
Avoid the use of internal apis via the cpumap accessor functions. Signed-off-by:
-
Ian Rogers authored
Rather than allocate memory, allow abi::__cxa_demangle to do that. This avoids a problem where on error NULL was returned triggering a memory leak. Fixes: 3b4e4efe ("perf symbol: Add abi::__cxa_demangle C++ demangling support") Signed-off-by:
-
- 15 Mar, 2023 23 commits
-
-
Leo Yan authored
Update documentation for new sorting and option '--stdio'. Reviewed-by:
-
Leo Yan authored
Since we have supported histograms list and prepared the dimensions in the tool, this patch adds TUI mode for stat report. It also adds UI progress for sorting for better user experience. Committer notes: kvm_display() is only used by functions enclosed in: #if defined(HAVE_KVM_STAT_SUPPORT) && defined(HAVE_LIBTRACEEVENT) So do it with this new function as well. Signed-off-by:
-
Leo Yan authored
Add dimensions for count and time percentages, it would be useful for user to review percentage statistics. Reviewed-by:
-
Leo Yan authored
This patch adds header, entry callback and width for every dimension, thus in TUI mode the tool can print items with the defined attributions. Reviewed-by:
-
Leo Yan authored
Since histograms supports sorting, the tool doesn't need to maintain the mapping between the sorting keys and the corresponding comparison callbacks, therefore, this patch removes structure kvm_event_key. But we still need to validate the sorting key, this patch uses an array for sorting keys and renames function select_key() to is_valid_key() to validate the sorting key passed by user. Signed-off-by:
-
Leo Yan authored
perf kvm tool defines its own cached list which is managed with RB tree, histograms also provide RB tree to manage data entries. Since now we have introduced histograms in the tool, it's not necessary to use the self defined list and we can directly use histograms list to manage KVM events. This patch changes to use histograms list to track KVM events, and it invokes the common function hists__output_resort_cb() to sort result, this also give us flexibility to extend more sorting key words easily. After histograms list supported, the cached list is redundant so remove the relevant code for it. Committer notes: kvm_hists__reinit() is only used by functions enclosed in: #if defined(HAVE_KVM_STAT_SUPPORT) && defined(HAVE_LIBTRACEEVENT) So do it with this new function as well. Signed-off-by:
-
Leo Yan authored
To support KVM event statistics, this patch firstly registers histograms columns and sorting fields; every column or field has its own format structure, the format structure is dereferenced to access the dimension, finally the dimension provides the comparison callback for sorting result. Signed-off-by:
-
Leo Yan authored
__hists__add_entry() creates a temporary entry and compare it with existed histograms entries, if any existed entry equals to the temporary entry it skips to allocation to avoid duplication. The problem for support KVM event in histograms is it doesn't contain any info to identify KVM event and can be used for comparison entries. This patch adds 'kvm_info' field in the histograms entry which contains the KVM event's key, this identifier will be used for comparison histograms entries in later change. Signed-off-by:
-
Leo Yan authored
Parse address location for samples and save it into the structure 'perf_kvm_stat', it is to be used by histograms entry. Reviewed-by:
-
Leo Yan authored
This patch adds an argument 'sample' for kvm_alloc_init_event(), and its caller functions are updated as well for passing down the 'sample' pointer. This is a preparation change to allow later patch to create histograms entries for kvm event, no any functionality changes. Reviewed-by:
-
Leo Yan authored
This is a preparation to support histograms in perf kvm tool. As first step, this patch defines histograms data structures and initialize them. Committer notes: Those are only used by functions enclosed in: #if efined(HAVE_KVM_STAT_SUPPORT) && defined(HAVE_LIBTRACEEVENT) So do this for these new functions and struct as well. Reviewed-by:
-
Leo Yan authored
The variable 'decode_str_len' defines the string length for KVM event name and every arch defines its own values. This introduces complexity that the variable definition are spreading in multiple source files under arch folder. This patch refactors code to use a macro KVM_EVENT_NAME_LEN to define event name length and thus remove the definitions in arch files. Signed-off-by:
-
Leo Yan authored
Currently the metrics comparison uses greater operator (>), it returns the boolean value (0 or 1). This patch changes to use subtraction as comparison result, which can be used by histograms sorting. Since the subtraction result is u64 type, we change key_cmp_fun's return type to int64_t to avoid overflow. Signed-off-by:
-
Leo Yan authored
This patch moves up the helper functions of event's metrics for later adding code to call them. No any functionality changes, but has a function renaming from compare_kvm_event_{metric}() to cmp_event_{metric}(). Committer notes: Those helper functions are only used if this is true: if defined(HAVE_KVM_STAT_SUPPORT) && defined(HAVE_LIBTRACEEVENT) So keep them enclosed with that. Reviewed-by: -
Leo Yan authored
Sometimes, handling kvm events needs to base on global variables, e.g. when read event counts we need to know the target vcpu ID; the global variables are stored in structure perf_kvm_stat. This patch adds add a 'perf_kvm_stat' pointer in kvm event structure, it is to be used by later refactoring. Reviewed-by:
-
Leo Yan authored
Currently the tool computes overall statistics when sort the results. This patch refactors overall statistics during events processing, therefore, the function update_total_coun() is not needed anymore, an extra benefit is we can de-couple code between the statistics and the sorting. This patch is not expected any functionality changes. Reviewed-by:
-
Namhyung Kim authored
Add more description and examples. Signed-off-by:
-
Namhyung Kim authored
For a BPF filter to work properly, users need to provide appropriate options to enable the sample types. Otherwise the BPF program would see an invalid value (i.e. always 0) and filter won't work well. Show a warning message if sample types are missing like below. $ sudo ./perf record -e cycles --filter 'addr < 100' true Error: cycles event does not have PERF_SAMPLE_ADDR Hint: please add -d option to perf record. failed to set filter "BPF" on event cycles with 22 (Invalid argument) Signed-off-by:
-
Namhyung Kim authored
It supports two or more expressions connected as a group and the group result is considered true when one of them returns true. The new group operators (GROUP_BEGIN and GROUP_END) are added to setup and check the condition. As it doesn't allow nested groups, the condition is saved in local variables. For example, the following is to get samples only if the data source memory level is L2 cache or the weight value is greater than 30. $ sudo ./perf record -adW -e cpu/mem-loads/pp \ > --filter 'mem_lvl == l2 || weight > 30' -- sleep 1 $ sudo ./perf script -F data_src,weight 10668100842 |OP LOAD|LVL L3 or L3 hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A 47 11868100242 |OP LOAD|LVL LFB/MAB or LFB/MAB hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A 57 10668100842 |OP LOAD|LVL L3 or L3 hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A 56 10650100842 |OP LOAD|LVL L3 or L3 hit|SNP None|TLB L2 miss|LCK No|BLK N/A 144 10468100442 |OP LOAD|LVL L2 or L2 hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A 16 10468100442 |OP LOAD|LVL L2 or L2 hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A 20 11868100242 |OP LOAD|LVL LFB/MAB or LFB/MAB hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A 189 1026a100142 |OP LOAD|LVL L1 or L1 hit|SNP None|TLB L1 or L2 hit|LCK Yes|BLK N/A 193 10468100442 |OP LOAD|LVL L2 or L2 hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A 18 ... Signed-off-by:Jiri Olsa <jolsa@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Hao Luo <haoluo@google.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: James Clark <james.clark@arm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ravi Bangoria <ravi.bangoria@amd.com> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: bpf@vger.kernel.org Link: https://lore.kernel.org/r/20230314234237.3008956-2-namhyung@kernel.orgSigned-off-by:
-
Namhyung Kim authored
The data_src has many entries to express memory behaviors. Add each term separately so that users can combine them for their purpose. I didn't add prefix for the constants for simplicity as they are mostly distinguishable but I had to use l1_miss and l2_hit for mem_dtlb since mem_lvl has different values for the same names. Note that I decided mem_lvl to be used as an alias of mem_lvlnum as it's deprecated now. According to the comment in the UAPI header, users should use the mix of mem_lvlnum, mem_remote and mem_snoop. Also the SNOOPX bits are concatenated to mem_snoop for simplicity. The following terms are used for data_src and the corresponding perf sample data fields: * mem_op : { load, store, pfetch, exec } * mem_lvl: { l1, l2, l3, l4, cxl, io, any_cache, lfb, ram, pmem } * mem_snoop: { none, hit, miss, hitm, fwd, peer } * mem_remote: { remote } * mem_lock: { locked } * mem_dtlb { l1_hit, l1_miss, l2_hit, l2_miss, any_hit, any_miss, walk, fault } * mem_blk { by_data, by_addr } * mem_hops { hops0, hops1, hops2, hops3 } We can now use a filter expression like below: 'mem_op == load, mem_lvl <= l2, mem_dtlb == l1_hit' 'mem_dtlb == l2_miss, mem_hops > hops1' 'mem_lvl == ram, mem_remote == 1' Note that 'na' is shared among the terms as it has the same value except for mem_lvl. I don't have a good idea to handle that for now. Signed-off-by: -
Namhyung Kim authored
The weight data consists of a couple of fields with the PERF_SAMPLE_WEIGHT_STRUCT. Add weight{1,2,3} term to select them separately. Also add their aliases like 'ins_lat', 'p_stage_cyc' and 'retire_lat'. Signed-off-by: -
Namhyung Kim authored
The pid is special because it's saved in the PERF_SAMPLE_TID together. So it needs to differenciate tid and pid using the 'part' field in the perf bpf filter entry struct. Signed-off-by:
-
Namhyung Kim authored
When it uses bpf filters, event might drop some samples. It'd be nice if it can report how many samples it lost. As LOST_SAMPLES event can carry the similar information, let's use it for bpf filters. To indicate it's from BPF filters, add a new misc flag for that and do not display cpu load warnings. Signed-off-by:
-
