| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| | |
|
| |
|
|
|
|
|
|
|
| |
So that LD_PRELOAD-ing doesn't force loading libpthread.so which may
slow down some single-threaded apps.
tcmalloc already has maybe_threads facility that can detect if
libpthread.so is loaded (via weak symbols) and provide 'simulations' of
some pthread functions that tcmalloc needs.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Default mode of operation of cpu profiler uses itimer and
SIGPROF. This timer is by definition per-process and no spec defines
which thread is going to receive SIGPROF. And it provides correct
profiles only if we assume that probability of picking threads will be
proportional to cpu time spent by threads.
It is easy to see, that recent Linux (at least on common SMP hardware)
doesn't satisfy that assumption. Quite big skews of SIGPROF ticks
between threads is visible. I.e. I could see as big as 70%/20%
division instead of 50%/50% for pair of cpu-hog threads. (And I do see
it become 50/50 with new mode)
Fortunately POSIX provides mechanism to track per-thread cpu time via
posix timers facility. And even more fortunately, Linux also provides
mechanism to deliver timer ticks to specific threads.
Interestingly, it looks like FreeBSD also has very similar facility
and seems to suffer from same skew. But due to difference in a way
how threads are identified, I haven't bothered to try to support this
mode on FreeBSD.
This commit implements new profiling mode where every thread creates
posix timer which tracks thread's cpu time. Threads also also set up
signal delivery to itself on overflows of that timer.
This new mode requires every thread to be registered in cpu
profiler. Existing ProfilerRegisterThread function is used for that.
Because registering threads requires application support (or suitable
LD_PRELOAD-able wrapper for thread creation API), new mode is off by
default. And it has to be manually activated by setting environment
variable CPUPROFILE_PER_THREAD_TIMERS.
New mode also requires librt symbols to be available. Which we do not
link to due to librt's dependency on libpthread. Which we avoid due
to perf impact of bringing in libpthread to otherwise single-threaded
programs. So it has to be either already loaded by profiling program
or LD_PRELOAD-ed.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* google-perftools: version 0.98 release
* Add ProfilerStartWithOptions() (cgd)
* Change tcmalloc_minimal to not do any stack-tracing at all (csilvers)
* Prefer mmap to sbrk for 64-buit debug mode (sanjay)
* Fix accounting for some tcmalloc stats (sanjay)
* Use setrlimit() to keep unittests from killing the machine (odo)
* Fix a bug when sbrk-ing near address 4G (csilvers)
* Make MallocHook thread-safe (jyasskin)
* Fix windows build for MemoryBarrier (jyasskin)
* Fix CPU-profiler docs to mention correct libs (csilvers)
* Fix for GetHeapProfile() when heap-profiling is off (maxim)
* Avoid realloc resizing ping-pongs using hysteresis (csilvers)
* Add --callgrind output support to pprof (klimek)
* Fix profiler.h and heap-profiler.h to be C-compatible (csilvers)
* Break malloc_hook.h into two parts to reduce dependencies (csilvers)
* Better handle systems that don't implement mmap (csilvers)
* PORTING: disable system_alloc_unittest for msvc (csilvers)
* PORTING: Makefile tweaks to build better on cygwin (csilvers)
git-svn-id: http://gperftools.googlecode.com/svn/trunk@52 6b5cf1ce-ec42-a296-1ba9-69fdba395a50
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* google-perftools: version 0.92 release
* PERFORMANCE: use a packed cache to speed up tcmalloc
* PORTING: preliminary windows support! (see README.windows)
* PORTING: better support for solaris, OS X, FreeBSD (see INSTALL)
* Envvar support for running the heap-checker under gdb
* Add weak declarations to maybe_threads to fix no-pthreads compile bugs
* Some 64bit fixes, especially with pprof
* Better heap-checker support for some low-level allocations
* Fix bug where heap-profiles would sometimes get truncated
* New documentation about how to handle common heap leak situations
* Use computed includes for hash_map/set: easier config
* Added all used .m4 templates to the distribution
git-svn-id: http://gperftools.googlecode.com/svn/trunk@36 6b5cf1ce-ec42-a296-1ba9-69fdba395a50
|
|
|
* google-perftools: version 0.2 release
* Use mmap2() instead of mmap(), to map more memory (menage)
* Do correct pthread-local checking in heap-checker! (maxim)
* Avoid overflow on 64-bit machines in pprof (sanjay)
* Add a few more GetPC() functions, including for AMD (csilvers)
* Better method for overriding pthread functions (menage)
* (Hacky) fix to avoid overwriting profile files after fork() (csilvers)
* Crashing bugfix involving dumping heaps on small-stack threads (tudor)
* Allow library versions with letters at the end (csilvers)
* Config fixes for systems that don't define PATH_MAX (csilvers)
* Confix fixes so we no longer need config.h after install (csilvers)
* Fix to pprof to correctly read very big cpu profiles (csilvers)
* Fix to pprof to deal with new commandline flags in modern gv's
* Better error reporting when we can't access /proc/maps (etune)
* Get rid of the libc-preallocate code (which could crash on some
systems); no longer needed with local-threads fix (csilvers)
git-svn-id: http://gperftools.googlecode.com/svn/trunk@11 6b5cf1ce-ec42-a296-1ba9-69fdba395a50
|
