- 07 Jun, 2016 11 commits
-
-
He Kuang authored
This patch extracts common unwind-libunwind APIs out of unwind-libunwind-local.c, this part will be used by both local and remote libunwind. Signed-off-by:
He Kuang <hekuang@huawei.com> Acked-by:
Jiri Olsa <jolsa@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ekaterina Tumanova <tumanova@linux.vnet.ibm.com> Cc: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Kan Liang <kan.liang@intel.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Pekka Enberg <penberg@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/r/1464924803-22214-9-git-send-email-hekuang@huawei.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
He Kuang authored
Since unwind-libunwind.c contains code for specific arithecture, we change it's name to unwind-libunwind-local.c, and let it only be built if local libunwind is supported. Signed-off-by:
-
He Kuang authored
CONFIG_LIBUNWIND/NO_LIBUNWIND are changed to CONFIG_LOCAL_LIBUNWIND/ NO_LOCAL_LIBUNWIND for retaining local unwind features. The new CONFIG_LIBUNWIND stands for either local or remote or both unwind are supported, and NO_LIBUNWIND means that neither local nor remote unwind is supported. LIBUNWIND_LIBS is eliminated in LDFLAGS if local libunwind is not supported. Signed-off-by:
-
He Kuang authored
LIBUNWIND_LIBS contains libunwind libraries used for local only, don't mix this into LIBUNWIND_LDFLAGS so we can later use LIBUNWIND_LDFLAGS both for local and remote libunwind. Signed-off-by:
-
He Kuang authored
To determine the libunwind methods to use, we should get the 32bit/64bit information from maps of a thread. When a thread is newly created, the information is not prepared. This patch moves unwind__prepare_access() into thread__insert_map() so we can get the information we need from maps. Meanwhile, let thread__insert_map() return value and show messages on error. Signed-off-by:
-
He Kuang authored
Currently, libunwind operations are fixed, and they are chosen according to the host architecture. This will lead to a problem that if a thread is run as x86_32 on a x86_64 machine, perf will use libunwind methods for x86_64 to parse the callchain and get wrong results. This patch changes the fixed methods of libunwind operations to be thread/map related, and each thread can have individual libunwind operations. Local libunwind methods are registered as default value. Signed-off-by:
-
He Kuang authored
Currently, the type of thread->addr_space is unw_addr_space_t, which is a pointer defined in libunwind headers. For local libunwind, we can simple include "libunwind.h", but for remote libunwind, the header file is depends on the target libunwind platform. This patch uses 'void *' instead to decouple the dependence on libunwind. Signed-off-by:
-
He Kuang authored
Pass LIBUNWIND_DIR to feature check flags for remote libunwind tests. So perf can be able to detect remote libunwind libraries from arbitrary directory. Signed-off-by:
-
Taeung Song authored
Instead of perf_config(), this function initializes config set by reading various files: user config ~/.perfconfig and system config $(sysconfdir)/perfconfig). If there are the same config variable in both user and system config files, user config has higher priority than system config. Signed-off-by:
Taeung Song <treeze.taeung@gmail.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1465291577-20973-3-git-send-email-treeze.taeung@gmail.comSigned-off-by:
-
Taeung Song authored
Because of die() at perf_parse_file() a config set was freed in collect_config(), if failed. But it is natural to free a config set after collect_config() is done when some problems happened. So, in case of failure, lastly free a config set at perf_config_set__new() instead of freeing the config set in collect_config(). Signed-off-by:
-
Wang Nan authored
build_id_cache__kallsyms_path() accepts a string buffer but also allocs a buffer using asnprintf. Unfortunately, the its only user passes it a stack-allocated buffer. Freeing it causes crashes like this: $ perf script *** Error in `/home/wangnan/perf': free(): invalid pointer: 0x00007fffffff9630 *** ======= Backtrace: ========= lib64/libc.so.6(+0x6eeef)[0x7ffff5dbaeef] lib64/libc.so.6(+0x78cae)[0x7ffff5dc4cae] lib64/libc.so.6(+0x79987)[0x7ffff5dc5987] /home/w00229757/perf(build_id_cache__kallsyms_path+0x6b)[0x49681b] /home/w00229757/perf[0x4bdd40] /home/w00229757/perf(dso__load+0xa3a)[0x4c048a] /home/w00229757/perf(map__load+0x6f)[0x4d561f] /home/w00229757/perf(thread__find_addr_map+0x235)[0x49e935] /home/w00229757/perf(machine__resolve+0x7d)[0x49ec6d] /home/w00229757/perf[0x4555a8] /home/w00229757/perf[0x4d9507] /home/w00229757/perf[0x4d9e80] /home/w00229757/perf(ordered_events__flush+0x354)[0x4dd444] /home/w00229757/perf(perf_session__process_events+0x3d0)[0x4dc140] /home/w00229757/perf(cmd_script+0x12b0)[0x4592e0] /home/w00229757/perf[0x4911f1] /home/w00229757/perf(main+0x68f)[0x4352ef] /lib64/libc.so.6(__libc_start_main+0xf5)[0x7ffff5d6dbd5] /home/w00229757/perf[0x435415] ======= Memory map: ======== This patch simplifies build_id_cache__kallsyms_path(), not even considering allocating a string buffer, so never frees anything. Its caller should manage memory allocation. Signed-off-by:
Wang Nan <wangnan0@huawei.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Fixes: 01412261 ("perf buildid-cache: Use path/to/bin/buildid/elf instead of path/to/bin/buildid") Link: http://lkml.kernel.org/r/1465271678-7392-1-git-send-email-wangnan0@huawei.comSigned-off-by:
-
- 06 Jun, 2016 14 commits
-
-
Arnaldo Carvalho de Melo authored
For consistency with class__priv() elsewhere, and with the callback typedef for clearing those areas (e.g. bpf_map_clear_priv_t). Acked-by:
-
Arnaldo Carvalho de Melo authored
For consistency with bpf_map__priv() and elsewhere. Acked-by:
-
Arnaldo Carvalho de Melo authored
The use of this term is not warranted here, we use it in the kernel sources and in tools/ for refcounting, so, for consistency, rename them. Acked-bu: Wang Nan <wangnan0@huawei.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Milian Wolff <milian.wolff@kdab.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/n/tip-4ya1ot2e2fkrz48ws9ebiofs@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
For consistency, leaving "get" for reference counting. Acked-by:
-
Arnaldo Carvalho de Melo authored
And for consistency, rename it to bpf_map__def(), leaving "get" for reference counting. Also make it return a const pointer, as suggested by Wang. Acked-by:
-
Arnaldo Carvalho de Melo authored
For consistency, leaving "get" for reference counting. Acked-by:
-
Arnaldo Carvalho de Melo authored
To try to, over time, consistently use the IS_ERR() interface instead of using two return values, i.e. the integer return value for an error and the pointer address to return the bpf_map->priv pointer. Also rename it to bpf__priv(), to leave the "get" term for reference counting. Noticed while working on using BPF for collecting non-integer syscall argument payloads (struct sockaddr in calls such as connect(), for instance), where we need to use BPF maps and thus generalise bpf__setup_stdout() to connect bpf_output events with maps in a bpf proggie. Acked-by:
-
Taeung Song authored
collect_config() collect all config key-value pairs from config files and put each config info in config set. But if config set (i.e. 'set' variable at collect_config()) is NULL, this is wrong so handle it. Signed-off-by:
-
Taeung Song authored
If a config file has wrong key-value pairs, the perf process will be forcibly terminated by die() at perf_parse_file() called by perf_config() so terminal settings can be crushed because of unusual termination. For example: If user config file has a wrong value 'red;default' instead of a normal value like 'red, default' for a key 'colors.top', # cat ~/.perfconfig [colors] medium = red;default # wrong value and if running sub-command 'top', # perf top perf process is dead by force and terminal setting is broken with a messge like below. Fatal: bad config file line 2 in /root/.perfconfig So fix it. If perf_config() can return on failure without calling die() at perf_parse_file(), this problem can be solved. And if a config file has wrong values, show the error message and then use default config values instead of wrong config values. Signed-off-by: -
Andi Kleen authored
When in CSV mode --metric-only outputs an header, unlike the other modes. Previously it did not properly print headers for the aggregation columns, so the headers were actually shifted against the real values. Fix this here by outputting the correct headers for CSV. v2: Indent array. Signed-off-by:
Andi Kleen <ak@linux.intel.com> Acked-by:
-
Andi Kleen authored
When --metric-only is enabled there were no headers for the topology in interval mode. Also when headers were printed they were on a separate line. Before: $ perf stat --metric-only -A -I 1000 -a 1.001038376 frontend cycles idle insn per cycle stalled cycles per insn branch-misses of all branches 1.001038376 CPU0 123.54% 0.23 5.29 7.61% 1.001038376 CPU1 137.78% 0.24 5.13 10.07% 1.001038376 CPU2 64.48% 0.22 5.50 6.84% After: $ perf stat --metric-only -A -I 1000 -a 1.001111114 CPU0 82.46% 0.32 2.60 7.64% 1.001111114 CPU1 126.63% 0.02 42.83 0.15% 1.001111114 CPU2 193.54% 0.32 2.59 6.92% v2: Move all headers on a single line Reported-by: -
Andi Kleen authored
Implement the TopDown formulas in 'perf stat'. The topdown basic metrics reported by the kernel are collected, and the formulas are computed and output as normal metrics. See the kernel commit exporting the events for details on the used metrics. Committer note: Output example: # perf stat --topdown -a usleep 1 Performance counter stats for 'system wide': retiring bad speculation frontend bound backend bound S0-C0 2 23.8% 11.6% 28.3% 36.3% S0-C1 2 16.2% 15.7% 36.5% 31.6% 0.000579956 seconds time elapsed # v2: Always print all metrics, only use thresholds for coloring. v3: Mark retiring over threshold green, not red. v4: Only print one decimal digit Fix color printing of one metric v5: Avoid printing -0.0 v6: Remove extra frontend event lookup Signed-off-by: -
Andi Kleen authored
Add basic plumbing for TopDown in perf stat TopDown is intended to replace the frontend cycles idle/ backend cycles idle metrics in standard perf stat output. These metrics are not reliable in many workloads, due to out of order effects. This implements a new --topdown mode in perf stat (similar to --transaction) that measures the pipe line bottlenecks using standardized formulas. The measurement can be all done with 5 counters (one fixed counter) The result are four metrics: FrontendBound, BackendBound, BadSpeculation, Retiring that describe the CPU pipeline behavior on a high level. The full top down methology has many hierarchical metrics. This implementation only supports level 1 which can be collected without multiplexing. A full implementation of top down on top of perf is available in pmu-tools toplev. (http://github.com/andikleen/pmu-tools) The current version works on Intel Core CPUs starting with Sandy Bridge, and Atom CPUs starting with Silvermont. In principle the generic metrics should be also implementable on other out of order CPUs. TopDown level 1 uses a set of abstracted metrics which are generic to out of order CPU cores (although some CPUs may not implement all of them): topdown-total-slots Available slots in the pipeline topdown-slots-issued Slots issued into the pipeline topdown-slots-retired Slots successfully retired topdown-fetch-bubbles Pipeline gaps in the frontend topdown-recovery-bubbles Pipeline gaps during recovery from misspeculation These metrics then allow to compute four useful metrics: FrontendBound, BackendBound, Retiring, BadSpeculation. Add a new --topdown options to enable events. When --topdown is specified set up events for all topdown events supported by the kernel. Add topdown-* as a special case to the event parser, as is needed for all events containing -. The actual code to compute the metrics is in follow-on patches. v2: Use standard sysctl read function. v3: Move x86 specific code to arch/ v4: Enable --metric-only implicitly for topdown. v5: Add --single-thread option to not force per core mode v6: Fix output order of topdown metrics v7: Allow combining with -d v8: Remove --single-thread again v9: Rename functions, adding arch_ and topdown_. v10: Expand man page and describe TopDown better Paste intro into commit description. Print error when malloc fails. Signed-off-by:
-
Andi Kleen authored
'perf test' tries to parse all entries in /sys/devices/cpu/events/. Ignore the special entries like '.scale', which cannot be directly parsed as an event. This patch assumes all files containing a '.' are special and can be ignored. Reported-by:
-
- 03 Jun, 2016 15 commits
-
-
He Kuang authored
There's a display inconsistency when there are multiple tracepoint events, some of which have the 'call-graph' config option set but the first one hasn't, i.e. the whole logic for call graph processing is enabled only if the first tracepoint event has call-graph set. For instance, if we record signal_deliver with call-graph and signal_generate without: $ perf record -g -a -e signal:signal_deliver -e signal:signal_generate/call-graph=no/ [ perf record: Captured and wrote 0.017 MB perf.data (2 samples) ] $ perf script kworker/u2:1 13 [000] 6563.875949: signal:signal_generate: sig=2 errno=0 code=128 comm=perf pid=1313 grp=1 res=0 ff61cc __send_signal+0x3ec ([kernel.kallsyms]) perf 1313 [000] 6563.877584: signal:signal_deliver: sig=2 errno=0 code=128 sa_handler=43115e sa_flags=14000000 7ffff314 get_signal+0x80007f0023a4 ([kernel.kallsyms]) 7fffe358 do_signal+0x80007f002028 ([kernel.kallsyms]) 7fffa5e8 exit_to_usermode_loop+0x80007f002053 ([kernel.kallsyms]) ... Then we exchange the order of these two events in commandline, and keep signal_generate without call-graph. $ perf record -g -a -e signal:signal_generate/call-graph=no/ -e signal:signal_deliver [ perf record: Captured and wrote 0.017 MB perf.data (2 samples) ] $ perf script kworker/u2:2 1314 [000] 6933.353060: signal:signal_generate: sig=2 errno=0 code=128 comm=perf pid=1321 grp=1 res=0 perf 1321 [000] 6933.353872: signal:signal_deliver: sig=2 errno=0 code=128 sa_handler=43115e sa_flags=14000000 This time, the callchain of the event signal_deliver disappeared. The problem is caused by that perf only checks for the first evsel in evlist and decides if callchain should be printed. This patch traverses all evsels in evlist to see if any of them have callchains, and shows the right result: $ perf script kworker/u2:2 1314 [000] 6933.353060: signal:signal_generate: sig=2 errno=0 code=128 comm=perf pid=1321 grp=1 res=0 ff61cc __send_signal+0x3ec ([kernel.kallsyms]) perf 1321 [000] 6933.353872: signal:signal_deliver: sig=2 errno=0 code=128 sa_handler=43115e sa_flags=14000000 7ffff314 get_signal+0x80007f0023a4 ([kernel.kallsyms]) 7fffe358 do_signal+0x80007f002028 ([kernel.kallsyms]) 7fffa5e8 exit_to_usermode_loop+0x80007f002053 ([kernel.kallsyms]) ... Signed-off-by: -
Lucas Stach authored
This fixes cross compilation of libapi. Signed-off-by:
Lucas Stach <l.stach@pengutronix.de> Cc: Jiri Olsa <jolsa@kernel.org> Cc: kernel@pengutronix.de Cc: patchwork-lst@pengutronix.de Link: http://lkml.kernel.org/r/1458235670-27341-1-git-send-email-l.stach@pengutronix.deSigned-off-by:
-
Wang Nan authored
If zalloc fail, setting evlist->mmap[i].fd is unsafe and perf_evlist__alloc_mmap() should bail out right after that. Signed-off-by:
Masami Hiramatsu <mhiramat@kernel.org> Cc: He Kuang <hekuang@huawei.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Fixes: d4c6fb36 ("perf evsel: Record fd into perf_mmap") Link: http://lkml.kernel.org/r/1464699975-230440-1-git-send-email-wangnan0@huawei.comSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Out of perf_evsel__intval(), that requires passing the variable name, that will then be searched in the list of tracepoint variables for the given evsel. In cases such as syscall file descriptor ("fd") tracking, this is wasteful, we need just to use perf_evsel__field() and cache the format_field. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Milian Wolff <milian.wolff@kdab.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-r6f89jx9j5nkx037d0naviqy@git.kernel.orgSigned-off-by: -
Andi Kleen authored
Now that we have topology_max_smt_threads() use it to detect the HT workarounds for older CPUs. Signed-off-by:
Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vince Weaver <vincent.weaver@maine.edu> Cc: acme@kernel.org Cc: jolsa@kernel.org Link: http://lkml.kernel.org/r/1463703002-19686-6-git-send-email-andi@firstfloor.orgSigned-off-by:
Ingo Molnar <mingo@kernel.org>
-
Andi Kleen authored
Add topdown event declarations to Silvermont / Airmont. These cores do not support the full Top Down metrics, but an useful subset (FrontendBound, Retiring, Backend Bound/Bad Speculation). The perf stat tool automatically handles the missing events and combines the available metrics. Signed-off-by:
-
Andi Kleen authored
Add declarations for the events needed for topdown to the Intel big core CPUs starting with Sandy Bridge. We need to report different values if HyperThreading is on or off. The only thing this patch does is to export some events in sysfs. topdown level 1 uses a set of abstracted metrics which are generic to out of order CPU cores (although some CPUs may not implement all of them): topdown-total-slots Available slots in the pipeline topdown-slots-issued Slots issued into the pipeline topdown-slots-retired Slots successfully retired topdown-fetch-bubbles Pipeline gaps in the frontend topdown-recovery-bubbles Pipeline gaps during recovery from misspeculation A slot is a single operation in the CPU pipe line. These metrics then allow to compute four useful metrics: FrontendBound, BackendBound, Retiring, BadSpeculation. The formulas to compute the metrics are generic, they only change based on the availability on the abstracted input values. The kernel declares the events supported by the current CPU and their scaling factors (such as the pipeline width) and perf stat then computes the formulas based on the available metrics. This is similar how existing perf metrics, such as TSC metrics or IPC, are implemented. This abstracts all CPU pipe line specific knowledge in the kernel driver, but still avoids the need for larger scale perf interface changes. For HyperThreading the any bit is needed to get accurate values when both threads are executing. This implies that the events can only be collected as root or with perf_event_paranoid=-1 for now. The basic scheme is based on the following paper: Yasin, A Top Down Method for Performance analysis and Counter architecture ISPASS14 (pdf available via google) Signed-off-by:
-
Andi Kleen authored
Add a way to show different sysfs events attributes depending on HyperThreading is on or off. This is difficult to determine early at boot, so we just do it dynamically when the sysfs attribute is read. Signed-off-by:
-
Andi Kleen authored
For SMT specific workarounds it is useful to know if SMT is active on any online CPU in the system. This currently requires a loop over all online CPUs. Add a global variable that is updated with the maximum number of smt threads on any CPU on online/offline, and use it for topology_max_smt_threads() The single call is easier to use than a loop. Not exported to user space because user space already can use the existing sibling interfaces to find this out. Signed-off-by:
-
Vineet Gupta authored
This allows (with a previous change to the perf error return ABI) for calling out in userspace the exact reason for perf record failing when PMU doesn't support overflow interrupts. Note that this needs to be put ahead of existing precise_ip check as that gets hit otherwise for the sampling fail case as well. Signed-off-by:
Vineet Gupta <vgupta@synopsys.com> Signed-off-by:
-
Vineet Gupta authored
Change the return code for sampling event not supported from -ENOTSUPP to -EOPNOTSUPP. This allows userspace to identify this case specifically, instead of printing the catch-all error message it did previously. Technically this is an ABI change, but we think we can get away with it. Old behavior: ------- | # perf record ls | Error: | The sys_perf_event_open() syscall returned with 524 (Unknown error 524) | for event (cycles:ppp). | /bin/dmesg may provide additional information. | No CONFIG_PERF_EVENTS=y kernel support configured? New behavior: ------- | # perf record ls | Error: | PMU Hardware doesn't support sampling/overflow-interrupts. Signed-off-by:
-
Kan Liang authored
Some platforms, e.g. Knights Landing, use a common PCI device ID for multiple instances of an uncore PMU device type. So it is impossible to locate the specific instances only by PCI device ID. The current code specially handles Knights Landing by arbitrarily pointing an instance to an unused uncore box. However, we still have no idea which uncore device is mapped to which box. Furthermore, there could be more platforms which use a common PCI device ID for uncore devices. We have to specially handle them one by one. This patch records full device information (slot, func, and device ID) in id_table[]. So the probe function can point the instance to a specific uncore box by checking the full device information. Tested-by:
Lukasz Odzioba <lukasz.odzioba@intel.com> Signed-off-by:
Kan Liang <kan.liang@intel.com> Signed-off-by:
-
Lukasz Odzioba authored
Due to change in register definition we need to update OCR mask: MSR_OFFCORE_RESP0 reserved bits: 3,4,18,29,30,33,34, 8,11,14 MSR_OFFCORE_RESP1 reserved bits: 3,4,18,29,30,33,34, 38 Reported-by:
-
Lukasz Odzioba authored
Add the 'static' keyword to intel_bdw_event_constraints[], snb_events_attrs[], nhm_events_attrs[] and intel_skl_event_constraints arrays[], because they are only used locally. Signed-off-by:
-
Kan Liang authored
This patch fixes an issue which was introduced by commit: 91a612ee ("perf/core: Fix dynamic interrupt throttle") ... which commit unconditionally sets the perf_sample_allowed_ns value to !0. But that could trigger a bug in the following corner case: The user can disable the dynamic interrupt throttle mechanism by setting perf_cpu_time_max_percent to 0. Then they change perf_event_max_sample_rate. For this case, the mechanism will be enabled implicitly, because perf_sample_allowed_ns becomes !0 - which is not what we want. This patch only updates perf_sample_allowed_ns when the dynamic interrupt throttle mechanism is enabled. Signed-off-by:
-
