)]}'
{
  "log": [
    {
      "commit": "5a9fa73072854981a5c05eb7ba18a96d49c2804f",
      "tree": "a73c0eba84744c112b794657a03d1a5159978528",
      "parents": [
        "ef864c958801768fb28bd3603cd0b098b394671c"
      ],
      "author": {
        "name": "Oleg Nesterov",
        "email": "oleg@tv-sign.ru",
        "time": "Mon Sep 22 14:42:50 2008 -0700"
      },
      "committer": {
        "name": "Thomas Gleixner",
        "email": "tglx@linutronix.de",
        "time": "Wed Sep 24 15:45:48 2008 +0200"
      },
      "message": "posix-timers: kill -\u003eit_sigev_signo and -\u003eit_sigev_value\n\nWith the recent changes -\u003eit_sigev_signo and -\u003eit_sigev_value are only\nused in sys_timer_create(), kill them.\n\nSigned-off-by: Oleg Nesterov \u003coleg@tv-sign.ru\u003e\nCc: mingo@elte.hu\nCc: Roland McGrath \u003croland@redhat.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Thomas Gleixner \u003ctglx@linutronix.de\u003e\n"
    },
    {
      "commit": "f06febc96ba8e0af80bcc3eaec0a109e88275fac",
      "tree": "46dba9432ef25d2eae9434ff2df638c7a268c0f1",
      "parents": [
        "6bfb09a1005193be5c81ebac9f3ef85210142650"
      ],
      "author": {
        "name": "Frank Mayhar",
        "email": "fmayhar@google.com",
        "time": "Fri Sep 12 09:54:39 2008 -0700"
      },
      "committer": {
        "name": "Ingo Molnar",
        "email": "mingo@elte.hu",
        "time": "Sun Sep 14 16:25:35 2008 +0200"
      },
      "message": "timers: fix itimer/many thread hang\n\nOverview\n\nThis patch reworks the handling of POSIX CPU timers, including the\nITIMER_PROF, ITIMER_VIRT timers and rlimit handling.  It was put together\nwith the help of Roland McGrath, the owner and original writer of this code.\n\nThe problem we ran into, and the reason for this rework, has to do with using\na profiling timer in a process with a large number of threads.  It appears\nthat the performance of the old implementation of run_posix_cpu_timers() was\nat least O(n*3) (where \"n\" is the number of threads in a process) or worse.\nEverything is fine with an increasing number of threads until the time taken\nfor that routine to run becomes the same as or greater than the tick time, at\nwhich point things degrade rather quickly.\n\nThis patch fixes bug 9906, \"Weird hang with NPTL and SIGPROF.\"\n\nCode Changes\n\nThis rework corrects the implementation of run_posix_cpu_timers() to make it\nrun in constant time for a particular machine.  (Performance may vary between\none machine and another depending upon whether the kernel is built as single-\nor multiprocessor and, in the latter case, depending upon the number of\nrunning processors.)  To do this, at each tick we now update fields in\nsignal_struct as well as task_struct.  The run_posix_cpu_timers() function\nuses those fields to make its decisions.\n\nWe define a new structure, \"task_cputime,\" to contain user, system and\nscheduler times and use these in appropriate places:\n\nstruct task_cputime {\n\tcputime_t utime;\n\tcputime_t stime;\n\tunsigned long long sum_exec_runtime;\n};\n\nThis is included in the structure \"thread_group_cputime,\" which is a new\nsubstructure of signal_struct and which varies for uniprocessor versus\nmultiprocessor kernels.  For uniprocessor kernels, it uses \"task_cputime\" as\na simple substructure, while for multiprocessor kernels it is a pointer:\n\nstruct thread_group_cputime {\n\tstruct task_cputime totals;\n};\n\nstruct thread_group_cputime {\n\tstruct task_cputime *totals;\n};\n\nWe also add a new task_cputime substructure directly to signal_struct, to\ncache the earliest expiration of process-wide timers, and task_cputime also\nreplaces the it_*_expires fields of task_struct (used for earliest expiration\nof thread timers).  The \"thread_group_cputime\" structure contains process-wide\ntimers that are updated via account_user_time() and friends.  In the non-SMP\ncase the structure is a simple aggregator; unfortunately in the SMP case that\nsimplicity was not achievable due to cache-line contention between CPUs (in\none measured case performance was actually _worse_ on a 16-cpu system than\nthe same test on a 4-cpu system, due to this contention).  For SMP, the\nthread_group_cputime counters are maintained as a per-cpu structure allocated\nusing alloc_percpu().  The timer functions update only the timer field in\nthe structure corresponding to the running CPU, obtained using per_cpu_ptr().\n\nWe define a set of inline functions in sched.h that we use to maintain the\nthread_group_cputime structure and hide the differences between UP and SMP\nimplementations from the rest of the kernel.  The thread_group_cputime_init()\nfunction initializes the thread_group_cputime structure for the given task.\nThe thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the\nout-of-line function thread_group_cputime_alloc_smp() to allocate and fill\nin the per-cpu structures and fields.  The thread_group_cputime_free()\nfunction, also a no-op for UP, in SMP frees the per-cpu structures.  The\nthread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls\nthread_group_cputime_alloc() if the per-cpu structures haven\u0027t yet been\nallocated.  The thread_group_cputime() function fills the task_cputime\nstructure it is passed with the contents of the thread_group_cputime fields;\nin UP it\u0027s that simple but in SMP it must also safely check that tsk-\u003esignal\nis non-NULL (if it is it just uses the appropriate fields of task_struct) and,\nif so, sums the per-cpu values for each online CPU.  Finally, the three\nfunctions account_group_user_time(), account_group_system_time() and\naccount_group_exec_runtime() are used by timer functions to update the\nrespective fields of the thread_group_cputime structure.\n\nNon-SMP operation is trivial and will not be mentioned further.\n\nThe per-cpu structure is always allocated when a task creates its first new\nthread, via a call to thread_group_cputime_clone_thread() from copy_signal().\nIt is freed at process exit via a call to thread_group_cputime_free() from\ncleanup_signal().\n\nAll functions that formerly summed utime/stime/sum_sched_runtime values from\nfrom all threads in the thread group now use thread_group_cputime() to\nsnapshot the values in the thread_group_cputime structure or the values in\nthe task structure itself if the per-cpu structure hasn\u0027t been allocated.\n\nFinally, the code in kernel/posix-cpu-timers.c has changed quite a bit.\nThe run_posix_cpu_timers() function has been split into a fast path and a\nslow path; the former safely checks whether there are any expired thread\ntimers and, if not, just returns, while the slow path does the heavy lifting.\nWith the dedicated thread group fields, timers are no longer \"rebalanced\" and\nthe process_timer_rebalance() function and related code has gone away.  All\nsumming loops are gone and all code that used them now uses the\nthread_group_cputime() inline.  When process-wide timers are set, the new\ntask_cputime structure in signal_struct is used to cache the earliest\nexpiration; this is checked in the fast path.\n\nPerformance\n\nThe fix appears not to add significant overhead to existing operations.  It\ngenerally performs the same as the current code except in two cases, one in\nwhich it performs slightly worse (Case 5 below) and one in which it performs\nvery significantly better (Case 2 below).  Overall it\u0027s a wash except in those\ntwo cases.\n\nI\u0027ve since done somewhat more involved testing on a dual-core Opteron system.\n\nCase 1: With no itimer running, for a test with 100,000 threads, the fixed\n\tkernel took 1428.5 seconds, 513 seconds more than the unfixed system,\n\tall of which was spent in the system.  There were twice as many\n\tvoluntary context switches with the fix as without it.\n\nCase 2: With an itimer running at .01 second ticks and 4000 threads (the most\n\tan unmodified kernel can handle), the fixed kernel ran the test in\n\teight percent of the time (5.8 seconds as opposed to 70 seconds) and\n\thad better tick accuracy (.012 seconds per tick as opposed to .023\n\tseconds per tick).\n\nCase 3: A 4000-thread test with an initial timer tick of .01 second and an\n\tinterval of 10,000 seconds (i.e. a timer that ticks only once) had\n\tvery nearly the same performance in both cases:  6.3 seconds elapsed\n\tfor the fixed kernel versus 5.5 seconds for the unfixed kernel.\n\nWith fewer threads (eight in these tests), the Case 1 test ran in essentially\nthe same time on both the modified and unmodified kernels (5.2 seconds versus\n5.8 seconds).  The Case 2 test ran in about the same time as well, 5.9 seconds\nversus 5.4 seconds but again with much better tick accuracy, .013 seconds per\ntick versus .025 seconds per tick for the unmodified kernel.\n\nSince the fix affected the rlimit code, I also tested soft and hard CPU limits.\n\nCase 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer\n\trunning), the modified kernel was very slightly favored in that while\n\tit killed the process in 19.997 seconds of CPU time (5.002 seconds of\n\twall time), only .003 seconds of that was system time, the rest was\n\tuser time.  The unmodified kernel killed the process in 20.001 seconds\n\tof CPU (5.014 seconds of wall time) of which .016 seconds was system\n\ttime.  Really, though, the results were too close to call.  The results\n\twere essentially the same with no itimer running.\n\nCase 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds\n\t(where the hard limit would never be reached) and an itimer running,\n\tthe modified kernel exhibited worse tick accuracy than the unmodified\n\tkernel: .050 seconds/tick versus .028 seconds/tick.  Otherwise,\n\tperformance was almost indistinguishable.  With no itimer running this\n\ttest exhibited virtually identical behavior and times in both cases.\n\nIn times past I did some limited performance testing.  those results are below.\n\nOn a four-cpu Opteron system without this fix, a sixteen-thread test executed\nin 3569.991 seconds, of which user was 3568.435s and system was 1.556s.  On\nthe same system with the fix, user and elapsed time were about the same, but\nsystem time dropped to 0.007 seconds.  Performance with eight, four and one\nthread were comparable.  Interestingly, the timer ticks with the fix seemed\nmore accurate:  The sixteen-thread test with the fix received 149543 ticks\nfor 0.024 seconds per tick, while the same test without the fix received 58720\nfor 0.061 seconds per tick.  Both cases were configured for an interval of\n0.01 seconds.  Again, the other tests were comparable.  Each thread in this\ntest computed the primes up to 25,000,000.\n\nI also did a test with a large number of threads, 100,000 threads, which is\nimpossible without the fix.  In this case each thread computed the primes only\nup to 10,000 (to make the runtime manageable).  System time dominated, at\n1546.968 seconds out of a total 2176.906 seconds (giving a user time of\n629.938s).  It received 147651 ticks for 0.015 seconds per tick, still quite\naccurate.  There is obviously no comparable test without the fix.\n\nSigned-off-by: Frank Mayhar \u003cfmayhar@google.com\u003e\nCc: Roland McGrath \u003croland@redhat.com\u003e\nCc: Alexey Dobriyan \u003cadobriyan@gmail.com\u003e\nCc: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n"
    },
    {
      "commit": "1711ef3866b0360e102327389fe4b76c849bbe83",
      "tree": "b74a2cb6167840563d450859a571d6685966b771",
      "parents": [
        "9c4751fd0eab5b8ebbfafb28cbcc8e03b0da5933"
      ],
      "author": {
        "name": "Toyo Abe",
        "email": "toyoa@mvista.com",
        "time": "Fri Sep 29 02:00:28 2006 -0700"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@g5.osdl.org",
        "time": "Fri Sep 29 09:18:15 2006 -0700"
      },
      "message": "[PATCH] posix-timers: Fix clock_nanosleep() doesn\u0027t return the remaining time in compatibility mode\n\nThe clock_nanosleep() function does not return the time remaining when the\nsleep is interrupted by a signal.\n\nThis patch creates a new call out, compat_clock_nanosleep_restart(), which\nhandles returning the remaining time after a sleep is interrupted.  This\npatch revives clock_nanosleep_restart().  It is now accessed via the new\ncall out.  The compat_clock_nanosleep_restart() is used for compatibility\naccess.\n\nSince this is implemented in compatibility mode the normal path is\nvirtually unaffected - no real performance impact.\n\nSigned-off-by: Toyo Abe \u003ctoyoa@mvista.com\u003e\nCc: Thomas Gleixner \u003ctglx@linutronix.de\u003e\nCc: Ingo Molnar \u003cmingo@elte.hu\u003e\nCc: Roland McGrath \u003croland@redhat.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n"
    },
    {
      "commit": "493f01d1d0699ddafc30067d33fcc18d0b95b624",
      "tree": "0e003b5e3ae425bccb30165aaeb21e128696fc1c",
      "parents": [
        "952bbc87f01f552ef091a62ea2a721b5b2670e74"
      ],
      "author": {
        "name": "Adrian Bunk",
        "email": "bunk@stusta.de",
        "time": "Wed Feb 01 03:05:14 2006 -0800"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@g5.osdl.org",
        "time": "Wed Feb 01 08:53:13 2006 -0800"
      },
      "message": "[PATCH] kernel/posix-timers.c: remove do_posix_clock_notimer_create()\n\nThis function is neither used nor has any real contents.\n\nSigned-off-by: Adrian Bunk \u003cbunk@stusta.de\u003e\nAcked-by: Thomas Gleixner \u003ctglx@linutronix.de\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n"
    },
    {
      "commit": "becf8b5d00f4b47e847f98322cdaf8cd16243861",
      "tree": "152ba7583324c64d34ecc70d5401957ca7225761",
      "parents": [
        "97735f25d2ba898ec5e13746451525580631c834"
      ],
      "author": {
        "name": "Thomas Gleixner",
        "email": "tglx@linutronix.de",
        "time": "Mon Jan 09 20:52:38 2006 -0800"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@g5.osdl.org",
        "time": "Tue Jan 10 08:01:39 2006 -0800"
      },
      "message": "[PATCH] hrtimer: convert posix timers completely\n\n- convert posix-timers.c to use hrtimers\n\n- remove the now obsolete abslist code\n\nSigned-off-by: Thomas Gleixner \u003ctglx@linutronix.de\u003e\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\nSigned-off-by: Miklos Szeredi \u003cmiklos@szeredi.hu\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n"
    },
    {
      "commit": "97735f25d2ba898ec5e13746451525580631c834",
      "tree": "a6da1078f04bec45967bd83f340b6f7bde731f5d",
      "parents": [
        "6ba1b91213e81aa92b5cf7539f7d2a94ff54947c"
      ],
      "author": {
        "name": "Thomas Gleixner",
        "email": "tglx@linutronix.de",
        "time": "Mon Jan 09 20:52:37 2006 -0800"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@g5.osdl.org",
        "time": "Tue Jan 10 08:01:38 2006 -0800"
      },
      "message": "[PATCH] hrtimer: switch clock_nanosleep to hrtimer nanosleep API\n\nSwitch clock_nanosleep to use the new nanosleep functions in hrtimer.c\n\nSigned-off-by: Thomas Gleixner \u003ctglx@linutronix.de\u003e\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n"
    },
    {
      "commit": "2a698971941bf5e6ebe96275f7d5318b2cf91ccf",
      "tree": "89440360d61ec88a52023682fb670aa7f77ed40b",
      "parents": [
        "a924b04ddea9788e09f387fe19ccbede5f09ddd8"
      ],
      "author": {
        "name": "Thomas Gleixner",
        "email": "tglx@linutronix.de",
        "time": "Mon Jan 09 20:52:28 2006 -0800"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@g5.osdl.org",
        "time": "Tue Jan 10 08:01:36 2006 -0800"
      },
      "message": "[PATCH] hrtimer: coding style and white space cleanup 2\n\nstyle/whitespace/macro cleanups of posix-timers.h\n\nSigned-off-by: Thomas Gleixner \u003ctglx@linutronix.de\u003e\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n"
    },
    {
      "commit": "a924b04ddea9788e09f387fe19ccbede5f09ddd8",
      "tree": "aa519a2427b7686d35ceae764d3e37d69a6ae5e8",
      "parents": [
        "57a558757bdbb877b54ed5ea15bd0892e02a707d"
      ],
      "author": {
        "name": "Thomas Gleixner",
        "email": "tglx@linutronix.de",
        "time": "Mon Jan 09 20:52:27 2006 -0800"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@g5.osdl.org",
        "time": "Tue Jan 10 08:01:36 2006 -0800"
      },
      "message": "[PATCH] hrtimer: make clockid_t arguments const\n\nadd const arguments to the posix-timers.h API functions\n\nSigned-off-by: Thomas Gleixner \u003ctglx@linutronix.de\u003e\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n"
    },
    {
      "commit": "1da177e4c3f41524e886b7f1b8a0c1fc7321cac2",
      "tree": "0bba044c4ce775e45a88a51686b5d9f90697ea9d",
      "parents": [],
      "author": {
        "name": "Linus Torvalds",
        "email": "torvalds@ppc970.osdl.org",
        "time": "Sat Apr 16 15:20:36 2005 -0700"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@ppc970.osdl.org",
        "time": "Sat Apr 16 15:20:36 2005 -0700"
      },
      "message": "Linux-2.6.12-rc2\n\nInitial git repository build. I\u0027m not bothering with the full history,\neven though we have it. We can create a separate \"historical\" git\narchive of that later if we want to, and in the meantime it\u0027s about\n3.2GB when imported into git - space that would just make the early\ngit days unnecessarily complicated, when we don\u0027t have a lot of good\ninfrastructure for it.\n\nLet it rip!\n"
    }
  ]
}