- 10 Jun, 2019 38 commits
-
-
Adrian Hunter authored
Intel PT decoding is done in time order. In order to support efficient time interval filtering, add a facility to "fast forward" towards a particular timestamp. That involves finding the right buffer, stepping to that buffer, and then stepping forward PSBs. Because decoding must begin at a PSB, "fast forward" stops at the last PSB that has a timestamp before the target timestamp. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Jin Yao <yao.jin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Link: http://lkml.kernel.org/r/20190604130017.31207-9-adrian.hunter@intel.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Adrian Hunter authored
When the decoder gets the next trace buffer, some state is reset if the buffer is not consecutive to the previous buffer. Add a parameter 'reposition' so that can be done also to support a "fast forward" facility. Signed-off-by:
-
Adrian Hunter authored
Factor out intel_pt_reposition() so it can be reused. Signed-off-by:
-
Adrian Hunter authored
Factor out intel_pt_8b_tsc() so it can be reused. Signed-off-by:
-
Adrian Hunter authored
Add a callback function to enable the decoder to lookahead at subsequent trace buffers. This will be used to implement a "fast forward" facility which will be needed to support efficient time interval filtering. Signed-off-by:
-
Adrian Hunter authored
Instruction trace decoders can optimize output based on what time intervals will be filtered, so pass that information in itrace_synth_ops. Signed-off-by:
-
Adrian Hunter authored
Instruction trace decoders can optimize output based on what time intervals will be filtered, so pass that information in itrace_synth_ops. Signed-off-by:
-
Adrian Hunter authored
Instruction trace decoders can optimize output based on what time intervals will be filtered, so pass that information in itrace_synth_ops. Signed-off-by:
-
Leo Yan authored
The clang bpf cmdline template has defined default value in the file tools/perf/util/llvm-utils.c, which has been changed for several times. This patch updates the documentation to reflect the latest default value for the configuration llvm.clang-bpf-cmd-template. Signed-off-by:
Leo Yan <leo.yan@linaro.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@kernel.org> Cc: Andi Kleen <ak@linux.intel.com> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mark Drayton <mbd@fb.com> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Song Liu <songliubraving@fb.com> Cc: Yonghong Song <yhs@fb.com> Cc: bpf@vger.kernel.org Cc: netdev@vger.kernel.org Fixes: d35b168c ("perf bpf: Give precedence to bpf header dir") Fixes: cb763714 ("perf llvm: Allow passing options to llc in addition to clang") Fixes: 1b16fffa ("perf llvm-utils: Add bpf include path to clang command line") Link: http://lkml.kernel.org/r/20190607143508.18141-1-leo.yan@linaro.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Suzuki noticed that this should be more useful in a generic header, and after looking I noticed we have it already in our copy of include/linux/bits.h in tools/include, so just use it, test built on x86-64 and ubuntu 19.04 with: perfbuilder@46646c9e848e:/$ aarch64-linux-gnu-gcc --version |& head -1 aarch64-linux-gnu-gcc (Ubuntu/Linaro 8.3.0-6ubuntu1) 8.3.0 perfbuilder@46646c9e848e:/$ Suggested-by:
Suzuki K Poulose <suzuki.poulose@arm.com> Link: https://lkml.kernel.org/r/68c1c548-33cd-31e8-100d-7ffad008c7b2@arm.com Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org, Link: https://lkml.kernel.org/n/tip-69pd3mqvxdlh2shddsc7yhyv@git.kernel.orgSigned-off-by:
-
Mathieu Poirier authored
This patch adds the necessary intelligence to properly compute the value of 'old' and 'head' when operating in snapshot mode. That way we can get the latest information in the AUX buffer and be compatible with the generic AUX ring buffer mechanic. Tester notes: > Leo, have you had the chance to test/review this one? Suzuki? Sure. I applied this patch on the perf/core branch (with latest commit 3e4fbf36c1e3 'perf augmented_raw_syscalls: Move reading filename to the loop') and passed testing with below steps: # perf record -e cs_etm/@tmc_etr0/ -S -m,64 --per-thread ./sort & [1] 19097 Bubble sorting array of 30000 elements # kill -USR2 19097 # kill -USR2 19097 # kill -USR2 19097 [ perf record: Woken up 4 times to write data ] [ perf record: Captured and wrote 0.753 MB perf.data ] Signed-off-by:
Mathieu Poirier <mathieu.poirier@linaro.org> Tested-by:
-
Arnaldo Carvalho de Melo authored
The 'die' info isn't in the same array as core and socket ids, and we missed the 'dies' string list, that comes right after the 'core' + 'socket' id variable length array, followed by the VLA for the dies. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Andi Kleen <ak@linux.intel.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> Fixes: c9cb12c5ba08 ("perf header: Add die information in CPU topology") Link: https://lkml.kernel.org/n/tip-nubi6mxp2n8ofvlx7ph6k3h6@git.kernel.orgSigned-off-by: -
Kan Liang authored
The existing "thread_siblings" and "thread_siblings_list" attribute will be deprecated. Use the new CPU topology sysfs attributes, "core_cpus" and "core_cpus_list", which are synonymous with the deprecated attributes. Check the new name first. If not available, use the deprecated name to be compatible with old kernel. Signed-off-by:
Kan Liang <kan.liang@linux.intel.com> Reviewed-by:
Jiri Olsa <jolsa@kernel.org> Cc: Andi Kleen <ak@linux.intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1559688644-106558-5-git-send-email-kan.liang@linux.intel.comSigned-off-by:
-
Kan Liang authored
The "sibling cores" actually shows the sibling CPUs of a socket. The name "sibling cores" is very misleading. Rename "sibling cores" to "sibling sockets" Signed-off-by:
-
Kan Liang authored
It is useful to aggregate counts per die. E.g. Uncore becomes die-scope on Xeon Cascade Lake-AP. Introduce a new option "--per-die" to support per-die aggregation. The global id for each core has been changed to socket + die id + core id. The global id for each die is socket + die id. Add die information for per-core aggregation. The output of per-core aggregation will be changed from "S0-C0" to "S0-D0-C0". Any scripts which rely on the output format of per-core aggregation probably be broken. For 'perf stat record/report', there is no die information when processing the old perf.data. The per-die result will be the same as per-socket. Committer notes: Renamed 'die' variable to 'die_id' to fix the build in some systems: CC /tmp/build/perf/builtin-script.o cc1: warnings being treated as errors builtin-stat.c: In function 'perf_env__get_die': builtin-stat.c:963: error: declaration of 'die' shadows a global declaration util/util.h:19: error: shadowed declaration is here mv: cannot stat `/tmp/build/perf/.builtin-stat.o.tmp': No such file or directory Signed-off-by: -
Kan Liang authored
With the new CPUID.1F, a new level type of CPU topology, 'die', is introduced. The 'die' information in CPU topology should be added in perf header. To be compatible with old perf.data, the patch checks the section size before reading the die information. The new info is added at the end of the cpu_topology section, the old perf tool ignores the extra data. It never reads data crossing the section boundary. The new perf tool with the patch can be used on legacy kernel. Add a new function has_die_topology() to check if die topology information is supported by kernel. The function only check X86 and CPU 0. Assuming other CPUs have same topology. Use similar method for core and socket to support die id and sibling dies string. Signed-off-by:
-
Kan Liang authored
There is no function to retrieve die id information of a given CPU. Add cpu_map__get_die_id() to retrieve die id information. Signed-off-by:
-
Mathieu Poirier authored
Add support for CPU-wide trace scenarios by correlating range packets with timestamp packets. That way range packets received on different ETMQ/traceID channels can be processed and synthesized in chronological order. Signed-off-by:
-
Mathieu Poirier authored
This patch deals with timestamp packets received from the decoding library in order to give the front end packet processing loop a handle on the time instruction conveyed by range packets have been executed at. Signed-off-by:
-
Mathieu Poirier authored
Link contextID packets received from the decoder with the perf tool thread mechanic so that we know the specifics of the process currently executing. Signed-off-by:
-
Mathieu Poirier authored
When operating in CPU-wide trace mode with a source/sink topology of N:1 packets with multiple traceID will end up in the same cs_etm_queue. In order to properly decode packets they need to be split in different queues, i.e one queue per traceID. As such add support for multiple traceID per cs_etm_queue by adding a new cs_etm_traceid_queue every time a new traceID is discovered in the trace stream. Signed-off-by:
-
Mathieu Poirier authored
When working with CPU-wide traces different traceID may be found in the same stream. As such we need to use the decoder callback that provides the traceID in order to know the thread context being decoded. Signed-off-by:
-
Mathieu Poirier authored
The tid/pid fields of structure cs_etm_queue are CPU dependent and as such need to be part of the cs_etm_traceid_queue in order to support CPU-wide trace scenarios. Signed-off-by:
-
Mathieu Poirier authored
The thread field of structure cs_etm_queue is CPU dependent and as such need to be part of the cs_etm_traceid_queue in order to support CPU-wide trace scenarios. Signed-off-by:
-
Mathieu Poirier authored
Nowadays the synthesize code is using the packet's cpu information, making cs_etm_queue::cpu useless. As such simply remove it. Signed-off-by:
-
Mathieu Poirier authored
In an ideal world there is one CPU per cs_etm_queue and as such, one trace ID per cs_etm_queue. In the real world CoreSight topologies allow multiple CPUs to use the same sink, which translates to multiple trace IDs per cs_etm_queue. To deal with this a new cs_etm_traceid_queue structure is introduced to enclose all the information related to a single trace ID, allowing a cs_etm_queue to handle traces generated by any number of CPUs. Signed-off-by:
-
Mathieu Poirier authored
Fixing wrong indentation of the while() loop - no change of functionality. Signed-off-by:
-
Mathieu Poirier authored
The decoder needs to work with more than one traceID queue if we want to support CPU-wide scenarios with N:1 source/sink topologies. As such move the packet buffer and related fields out of the decoder structure and into the cs_etm_queue structure. Signed-off-by:
-
Mathieu Poirier authored
There is no point in having two different error goto statement since the openCSD API to free a decoder handles NULL pointers. As such function cs_etm_decoder__free() can be called to deal with all aspect of freeing decoder memory. Signed-off-by:
-
Mathieu Poirier authored
Add handling of SWITCH-CPU-WIDE events in order to add the tid/pid of the incoming process to the perf tools machine infrastructure. This information is later retrieved when a contextID packet is found in the trace stream. Signed-off-by:
-
Mathieu Poirier authored
Add handling of ITRACE events in order to add the tid/pid of the executing process to the perf tools machine infrastructure. This information is later retrieved when a contextID packet is found in the trace stream. Signed-off-by:
-
Mathieu Poirier authored
Ask the perf core to generate an event when processes are swapped in/out of context. That way proper action can be taken by the decoding code when faced with such event. Signed-off-by:
-
Mathieu Poirier authored
When operating in CPU-wide mode tracers need to generate timestamps in order to correlate the code being traced on one CPU with what is executed on other CPUs. Signed-off-by:
-
Mathieu Poirier authored
When operating in CPU-wide mode being notified of contextID changes is required so that the decoding mechanic is aware of the process context switch. Signed-off-by:
-
Jiri Olsa authored
It's already setup in the only caller of this method in perf_evsel__open(), right before calling perf_evsel__alloc_fd(), no need to do it again. Also it's better to have it out of the function before we move it to libperf. Signed-off-by:
-
yuzhoujian authored
One can just record callchains in the kernel or user space with this new options. We can use it together with "--all-kernel" options. This two options is used just like print_stack(sys) or print_ustack(usr) for systemtap. Shown below is the usage of this new option combined with "--all-kernel" options: 1. Configure all used events to run in kernel space and just collect kernel callchains. $ perf record -a -g --all-kernel --kernel-callchains 2. Configure all used events to run in kernel space and just collect user callchains. $ perf record -a -g --all-kernel --user-callchains Committer notes: Improved documentation to state that asking for kernel callchains really is asking for excluding user callchains, and vice versa. Further mentioned that using both won't get both, but nothing, as both will be excluded. Signed-off-by:
yuzhoujian <yuzhoujian@didichuxing.com> Acked-by:
-
Arnaldo Carvalho de Melo authored
So perf_config() uses: int ret = 0; perf_config_set__for_each_entry(config_set, section, item) { ... ret = fn(); if (ret < 0) break; } return ret; Expecting that that break will imediatelly go to function exit to return that error value (ret). The problem is that perf_config_set__for_each_entry() expands into two nested for() loops, one traversing the sections in a config and the second the items in each of those sections, so we have to change that 'break' to a goto label right before that final 'return ret'. With that, for instance 'perf trace' now correctly bails out when a event that is requested to be added via its 'trace.add_events' ~/.perfconfig entry gets rejected by the kernel BPF verifier: # perf trace ls event syntax error: '/home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.o' \___ Kernel verifier blocks program loading (add -v to see detail) Run 'perf list' for a list of valid events Error: wrong config key-value pair trace.add_events=/home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.o # While before it would continue and explode later, when trying to find maps that would have been in place had that augmented_raw_syscalls.o precompiled BPF proggie been accepted by the, humm, bast... rigorous kernel BPF verifier 8-) Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@kernel.org> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Song Liu <songliubraving@fb.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Taeung Song <treeze.taeung@gmail.com> Cc: Yonghong Song <yhs@fb.com> Fixes: 8a0a9c7e ("perf config: Introduce new init() and exit()") Link: https://lkml.kernel.org/n/tip-qvqxfk9d0rn1l7lcntwiezrr@git.kernel.orgSigned-off-by: -
Leo Yan authored
On my Juno board with ARM64 CPUs, perf trace command reports the eBPF program building failure but the command will not exit and continue to run. If we define an eBPF event in config file, the event will be parsed with below flow: perf_config() `> trace__config() `> parse_events_option() `> parse_events__scanner() `-> parse_events_parse() `> parse_events_load_bpf() `> llvm__compile_bpf() Though the low level functions return back error values when detect eBPF building failure, but parse_events_option() returns 1 for this case and trace__config() passes 1 to perf_config(); perf_config() doesn't treat the returned value 1 as failure and it continues to parse other configurations. Thus the perf command continues to run even without enabling eBPF event successfully. This patch changes error handling in trace__config(), when it detects failure it will return -1 rather than directly pass error value (1); finally, perf_config() will directly bail out and perf will exit for this case. Committer notes: Simplified the patch to just check directly the return of parse_events_option() and it it is non-zero, change err from its initial zero value to -1. Signed-off-by:
-
- 05 Jun, 2019 2 commits
-
-
Arnaldo Carvalho de Melo authored
For instance, the rename* family uses "oldname", "newname", so check if "name" is at the end and treat it as a filename. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lkml.kernel.org/n/tip-wjy7j4bk06g7atzwoz1mid24@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
To support the SCA_FILENAME beautifier in more than one syscall arg, as needed for syscalls such as the rename* family, we need to, after processing one such arg, bump the augmented pointers so that the next augmented arg don't reuse data for the previous augmented arguments. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Brendan Gregg <brendan.d.gregg@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Luis Cláudio Gonçalves <lclaudio@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lkml.kernel.org/n/tip-4e4cmzyjxb3wkonfo1x9a27y@git.kernel.orgSigned-off-by:
-
