- 27 Nov, 2013 24 commits
-
-
Stanislav Fomichev authored
Add -g flag to `perf timechart record` which saves callchain info in the perf.data. When generating SVG, add backtrace information to the figure details, so now it's possible to see which code path woke up the task and why some task went to sleep. Signed-off-by:
Stanislav Fomichev <stfomichev@yandex-team.ru> Acked-by:
Namhyung Kim <namhyung@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1383323151-19810-8-git-send-email-stfomichev@yandex-team.ruSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Stanislav Fomichev authored
If we don't want either power or task events we may use -T or -P with the `perf timechart record` command to filter out events while recording to keep perf.data small. Signed-off-by:
-
Stanislav Fomichev authored
Add titles to figures so we can run SVG interactively in Firefox and check event details in the tooltips. This also aids exploring SVG with Inkscape because when user clicks on one part of logical figure, all parts are selected. It's also possible to read titles with Inkscape in the object details. Signed-off-by:
-
Stanislav Fomichev authored
In order to make SVG smaller and faster to browse add possibility to switch off power related information with -T switch. Signed-off-by:
-
Stanislav Fomichev authored
Don't use special flag to indicate power-only mode, just set proc_num to 0. Signed-off-by:
-
Stanislav Fomichev authored
Add -n option to specify min. number of tasks to print. Signed-off-by:
-
Stanislav Fomichev authored
Always try to print at least 15 tasks no matter how long they run. Signed-off-by:
-
Adrian Hunter authored
The change to per-cpu mmaps causes the -p, -t and -u options now to have inheritance enabled by default. Change that back to no inheritance but for the -t option only. Requested-by:
Peter Zijlstra <peterz@infradead.org> Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1384768557-23331-5-git-send-email-adrian.hunter@intel.comSigned-off-by:
-
Adrian Hunter authored
OPT_BOOLEAN_SET records whether a boolean option was set by the user. That information can be used to change the default value for the option after the options have been parsed. Signed-off-by:
-
Adrian Hunter authored
Long options can be negated by prefixing them with 'no-'. However options that already start with 'no-', such as '--no-inherit' result in ugly double 'no's. Avoid that by accepting that the removal of 'no-' also negates the long option. Suggested-by:
-
Adrian Hunter authored
This affects the -p, -t and -u options that previously defaulted to per-thread mmaps. Consequently add an option to select per-thread mmaps to support the old behaviour. Note that per-thread can be used with a workload-only (i.e. none of -p, -t, -u, -a or -C is selected) to get a per-thread mmap with no inheritance. Signed-off-by:
Ingo Molnar <mingo@kernel.org> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/5286271D.3020808@intel.comSigned-off-by:
-
Namhyung Kim authored
The print_sample_start() will be reused by other printing routine for internal events like COMM, FORK and EXIT from next patch. And because they're not tied to a specific event, move the evname print code to its caller. Signed-off-by:
David Ahern <dsahern@gmail.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1384752894-10974-1-git-send-email-namhyung@kernel.orgSigned-off-by:
-
Ramkumar Ramachandra authored
Signed-off-by:
Ramkumar Ramachandra <artagnon@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Link: http://lkml.kernel.org/r/1384704807-15779-6-git-send-email-artagnon@gmail.com [ Fix 'make install' target ] Signed-off-by:
-
Ramkumar Ramachandra authored
__perfcomp(), __perfcomp_colon(), and _perf() have to be overridden. Inspired by the way the git.git completion system is structured. Signed-off-by:
-
Ramkumar Ramachandra authored
In our sole callsite, __ltrim_colon_completions is called after __perfcomp, to modify the COMPREPLY set by the invocation. This is problematic, because in the zsh equivalent (using compset/ compadd), we'll have to generate completions in one-shot. So factor out this entire callsite into a special override'able __perfcomp_colon function; we will override it when introducing zsh support. Signed-off-by:
-
Ramkumar Ramachandra authored
compgen is a bash-builtin; factor out the invocations into a separate function to give us a chance to override it with a zsh equivalent in future patches. Signed-off-by:
-
Ramkumar Ramachandra authored
Define the variables cur, words, cword, and prev outside the main completion function so that we have a chance to override it when we introduce zsh support. Signed-off-by:
-
David Ahern authored
In most commands -g is used for callchains. Make perf-top follow suit. Move group to just --group with no short cut making it similar to perf-record. Signed-off-by:
-
Pekka Enberg authored
Thread summary line coloring looks ugly. It doesn't add much value so remove coloring completely. Signed-off-by:
Pekka Enberg <penberg@kernel.org> Acked-by:
-
Steven Rostedt authored
Jiri Olsa reported that his plugin for scsi was chopping off part of the output. Investigating this, I found that Jiri used the same functions as what is in the kernel, which adds the following: trace_seq_putc(p, 0); This adds a '\0' to the output string. The reason this works in the kernel is that the "p" that is passed to the function helper is a temporary trace_seq. But in the libtraceevent library, it's the pointer to the trace_seq used to output. By adding the '\0', it truncates the line and nothing added after that will be printed. We can solve this in two ways. One is to have the helper functions for the library not add the unnecessary '\0'. The other is to change the library to also use a helper trace_seq structure that gets copied to the main trace_seq just like the kernel does. The latter allows the helper functions in the plugins to be the same as the kernel, which is the better solution. Signed-off-by:
Steven Rostedt <rostedt@goodmis.org> Reported-by:
Jiri Olsa <jolsa@redhat.com> Tested-by:
-
Stephane Eranian authored
The RAPL PMU counters do not interrupt on overflow. Therefore, the kernel needs to poll the counters to avoid missing an overflow. This patch adds the hrtimer code to do this. The timer interval is calculated at boot time based on the power unit used by the HW. There is one hrtimer per-cpu to handle the case of multiple simultaneous use across cores on the same package + hotplug CPU. Thanks to Maria Dimakopoulou for her contributions to this patch especially on the math aspects. Signed-off-by:
Stephane Eranian <eranian@google.com> Reviewed-by:
Maria Dimakopoulou <maria.n.dimakopoulou@gmail.com> Reviewed-by:
Andi Kleen <ak@linux.intel.com> [ Applied 32-bit build fix. ] Signed-off-by:
-
Stephane Eranian authored
This patch adds a new uncore PMU to expose the Intel RAPL energy consumption counters. Up to 3 counters, each counting a particular RAPL event are exposed. The RAPL counters are available on Intel SandyBridge, IvyBridge, Haswell. The server skus add a 3rd counter. The following events are available and exposed in sysfs: - power/energy-cores: power consumption of all cores on socket - power/energy-pkg: power consumption of all cores + LLc cache - power/energy-dram: power consumption of DRAM (servers only) For each event both the unit (Joules) and scale (2^-32 J) is exposed in sysfs for use by perf stat and other tools. The files are: /sys/devices/power/events/energy-*.unit /sys/devices/power/events/energy-*.scale The RAPL PMU is uncore by nature and is implemented such that it only works in system-wide mode. Measuring only one CPU per socket is sufficient. The /sys/devices/power/cpumask file can be used by tools to figure out which CPUs to monitor by default. For instance, on a 2-socket system, 2 CPUs (one on each socket) will be shown. All the counters measure in the same unit (exposed via sysfs). The perf_events API exposes all RAPL counters as 64-bit integers counting in unit of 1/2^32 Joules (about 0.23 nJ). User level tools must convert the counts by multiplying them by 2^-32 to obtain Joules. The reason for this is that the kernel avoids doing floating point math whenever possible because it is expensive (user floating-point state must be saved). The method used avoids kernel floating-point usage. There is no loss of precision. Thanks to PeterZ for suggesting this approach. To convert the raw count in Watt: W = C * 2.3 / (1e10 * time) or ldexp(C, -32). RAPL PMU is a new standalone PMU which registers with the perf_event core subsystem. The PMU type (attr->type) is dynamically allocated and is available from /sys/device/power/type. Sampling is not supported by the RAPL PMU. There is no privilege level filtering either. Signed-off-by:
-
Stephane Eranian authored
This patch adds perf stat support for handling event units and scales as exported by the kernel. The kernel can export PMU events actual unit and scaling factor via sysfs: $ ls -1 /sys/devices/power/events/energy-* /sys/devices/power/events/energy-cores /sys/devices/power/events/energy-cores.scale /sys/devices/power/events/energy-cores.unit /sys/devices/power/events/energy-pkg /sys/devices/power/events/energy-pkg.scale /sys/devices/power/events/energy-pkg.unit $ cat /sys/devices/power/events/energy-cores.scale 2.3283064365386962890625e-10 $ cat cat /sys/devices/power/events/energy-cores.unit Joules This patch modifies the pmu event alias code to check for the presence of the .unit and .scale files to load the corresponding values. They are then used by perf stat transparently: # perf stat -a -e power/energy-pkg/,power/energy-cores/,cycles -I 1000 sleep 1000 # time counts unit events 1.000214717 3.07 Joules power/energy-pkg/ [100.00%] 1.000214717 0.53 Joules power/energy-cores/ 1.000214717 12965028 cycles [100.00%] 2.000749289 3.01 Joules power/energy-pkg/ 2.000749289 0.52 Joules power/energy-cores/ 2.000749289 15817043 cycles When the event does not have an explicit unit exported by the kernel, nothing is printed. In csv output mode, there will be an empty field. Special thanks to Jiri for providing the supporting code in the parser to trigger reading of the scale and unit files. Signed-off-by: -
Stephane Eranian authored
This patch adds a new field to the struct perf_event. It is intended to be used to chain events which are active (enabled). It helps in the hardware layer for PMUs which do not have actual counter restrictions, i.e., free running read-only counters. Active events are chained as opposed to being tracked via the counter they use. To save space we use a union with hlist_entry as both are mutually exclusive (suggested by Jiri Olsa). Signed-off-by:
-
- 21 Nov, 2013 1 commit
-
-
Ingo Molnar authored
Merge branch 'uprobes/core' of git://git.kernel.org/pub/scm/linux/kernel/git/oleg/misc into perf/core Pull uprobes cleanups from Oleg Nesterov. Signed-off-by:
-
- 20 Nov, 2013 6 commits
-
-
Oleg Nesterov authored
Document xol_area and arch_uprobe. Signed-off-by:
Oleg Nesterov <oleg@redhat.com> Acked-by:
Srikar Dronamraju <srikar@linux.vnet.ibm.com>
-
Oleg Nesterov authored
1. Don't include asm/uprobes.h unconditionally, we only need it if CONFIG_UPROBES. 2. Move the definition of "struct xol_area" into uprobes.c. Perhaps we should simply kill struct uprobes_state, it buys nothing. 3. Kill the dummy definition of uprobe_get_swbp_addr(), nobody except handle_swbp() needs it. 4. Purely cosmetic, but move the decl of uprobe_get_swbp_addr() up, close to other __weak helpers. Signed-off-by:
-
Oleg Nesterov authored
powerpc has both arch_uprobe->insn and arch_uprobe->ainsn to make the generic code happy. This is no longer needed after the previous change, powerpc can just use "u32 insn". Signed-off-by:
Ananth N Mavinakayanahalli <ananth@in.ibm.com>
-
Oleg Nesterov authored
arch_uprobe should be opaque as much as possible to the generic code, but currently it assumes that insn/ixol must be u8[] of the known size. Remove this unnecessary dependency, we can use "&" and and sizeof() with the same effect. Signed-off-by:
-
Oleg Nesterov authored
uprobe_task->vaddr is a bit strange. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange. And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line, so it is safe to reuse both in UTASK_RUNNING state. This all means that logically ->vaddr belongs to arch_uprobe_task and we should probably move it there, arch_uprobe_pre_xol() can record instruction_pointer() itself. OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task. This patch adds the union with 2 anonymous structs into uprobe_task. The first struct is autask + vaddr, this way we "almost" move vaddr into autask. The second struct has 2 new members for uprobe_copy_process() paths: ->dup_xol_addr which can be used instead ->vaddr, and ->dup_xol_work which can be used to avoid kmalloc() and simplify the code. Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods. Signed-off-by:
Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Acked-by:
-
Ingo Molnar authored
Merge tag 'perf-urgent-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent Pull perf/urgent fixes from Arnaldo Carvalho de Melo: * Tag thread comm as overriden, showing the right comm for threads after forks. (Frederic Weisbecker) * Fix memory leak when processing perf.data file header. (Namhyung Kim.) * Don't try to free string constant used for anonymous event groups. (Namhyung Kim) * Fix use of multiple options in processing field in libtraceevent. (Steven Rostedt) * Fix conversion of pointer to integer of different size in libtraceevent. (Arnaldo Carvalho de Melo) Signed-off-by:
-
- 19 Nov, 2013 9 commits
-
-
Arnaldo Carvalho de Melo authored
gcc complaint on 32-bit system: /home/acme/git/linux/tools/lib/traceevent/event-parse.c: In function ‘eval_num_arg’: /home/acme/git/linux/tools/lib/traceevent/event-parse.c:3468:9: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast] This is because the eval_num_arg returns everything as an 'unsigned long long', so it converts a void pointer to a wider integer, fix it by converting the void pointer to an integer of the same size, 'unsigned long', before casting it to 'unsigned long long'. Acked-by:
-
Vince Weaver authored
The 64-bit attr.config value for perf trace events was being copied into an "int" before doing a comparison, meaning the top 32 bits were being truncated. As far as I can tell this didn't cause any errors, but it did mean it was possible to create valid aliases for all the tracepoint ids which I don't think was intended. (For example, 0xffffffff00000018 and 0x18 both enable the same tracepoint). Signed-off-by:
Vince Weaver <vincent.weaver@maine.edu> Signed-off-by:
-
Peter Zijlstra authored
Currently we only allocate a single cpu hashtable for per-cpu swevents; do away with this optimization for it is fragile in the face of things like perf_pmu_migrate_context(). The easiest thing is to make sure all CPUs are consistent wrt state. Signed-off-by:
-
Peter Zijlstra authored
Vince's perf-trinity fuzzer found yet another 'interesting' problem. When we sample the irq_work_exit tracepoint with period==1 (or PERF_SAMPLE_PERIOD) and we add an fasync SIGNAL handler we create an infinite event generation loop: ,-> <IPI> | irq_work_exit() -> | trace_irq_work_exit() -> | ... | __perf_event_overflow() -> (due to fasync) | irq_work_queue() -> (irq_work_list must be empty) '--------- arch_irq_work_raise() Similar things can happen due to regular poll() wakeups if we exceed the ring-buffer wakeup watermark, or have an event_limit. To avoid this, dis-allow sampling this particular tracepoint. In order to achieve this, create a special perf_perm function pointer for each event and call this (when set) on trying to create a tracepoint perf event. [ roasted: use expr... to allow for ',' in your expression ] Reported-by:
-
Steven Rostedt authored
Jiri Olsa reported that the scsi_dispatch_cmd_done event failed to parse with: Error: expected type 5 but read 4 Error: expected type 5 but read 4 The problem is with this part of the print_fmt: __print_symbolic(((REC->result) >> 24) & 0xff, ... The __print_symbolic() helper function's first parameter is the field to use to determine what symbol to print based on the value of the result. The parser can handle one operation, but it can not handle multiple operations ('>>' and '&'). Add code to process all operations for the field argument for __print_symbolic() as well as __print_flags(). Reported-by: -
Namhyung Kim authored
After processing all group descriptors or encountering an error, it frees all descriptors. However, current logic can leak memory since it might not traverse all descriptors. Note that the 'i' can have different value than nr_groups when an error occurred and it's safe to call free(desc[i].name) for every desc since we already make it NULL when it's reused for group names. Signed-off-by:
-
Namhyung Kim authored
When processing event group descriptor in perf file header, we reuse an allocated group name but forgot to prevent it from freeing. Reported-by:
-
Frederic Weisbecker authored
The problem is that when a thread overrides its default ":%pid" comm, we forget to tag the thread comm as overriden. Hence, this overriden comm is not inherited on future forks. Fix it. Signed-off-by:
Frederic Weisbecker <fweisbec@gmail.com> Tested-by:
-
Al Viro authored
Once we'd freed m->buf, m->count should become zero - we have no valid contents reachable via m->buf. Reported-by:
Charley (Hao Chuan) Chu <charley.chu@broadcom.com> Signed-off-by:
Al Viro <viro@zeniv.linux.org.uk> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
