)]}' { "log": [ { "commit": "d6c1c49de577fa292af2449817364b7d89b574d8", "tree": "e97cf88a318f50eb23c18d790e8b0bcf7bb47169", "parents": [ "9dde9dc0a81c7aeb863b35121d09011f09b4897c", "ddf6e0e50723b62ac76ed18eb53e9417c6eefba7" ], "author": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Tue Dec 06 06:42:35 2011 +0100" }, "committer": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Tue Dec 06 06:43:49 2011 +0100" }, "message": "Merge branch \u0027perf/urgent\u0027 into perf/core\n\nMerge reason: Add these cherry-picked commits so that future changes\n on perf/core don\u0027t conflict.\n\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n" }, { "commit": "10c6db110d0eb4466b59812c49088ab56218fc2e", "tree": "d1d4e8debcf7415df49ce691b4c3da7443919f11", "parents": [ "16e5294e5f8303756a179cf218e37dfb9ed34417" ], "author": { "name": "Peter Zijlstra", "email": "a.p.zijlstra@chello.nl", "time": "Sat Nov 26 02:47:31 2011 +0100" }, "committer": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Mon Dec 05 09:33:03 2011 +0100" }, "message": "perf: Fix loss of notification with multi-event\n\nWhen you do:\n $ perf record -e cycles,cycles,cycles noploop 10\n\nYou expect about 10,000 samples for each event, i.e., 10s at\n1000samples/sec. However, this is not what\u0027s happening. You\nget much fewer samples, maybe 3700 samples/event:\n\n$ perf report -D | tail -15\nAggregated stats:\n TOTAL events: 10998\n MMAP events: 66\n COMM events: 2\n SAMPLE events: 10930\ncycles stats:\n TOTAL events: 3644\n SAMPLE events: 3644\ncycles stats:\n TOTAL events: 3642\n SAMPLE events: 3642\ncycles stats:\n TOTAL events: 3644\n SAMPLE events: 3644\n\nOn a Intel Nehalem or even AMD64, there are 4 counters capable\nof measuring cycles, so there is plenty of space to measure those\nevents without multiplexing (even with the NMI watchdog active).\nAnd even with multiplexing, we\u0027d expect roughly the same number\nof samples per event.\n\nThe root of the problem was that when the event that caused the buffer\nto become full was not the first event passed on the cmdline, the user\nnotification would get lost. The notification was sent to the file\ndescriptor of the overflowed event but the perf tool was not polling\non it. The perf tool aggregates all samples into a single buffer,\ni.e., the buffer of the first event. Consequently, it assumes\nnotifications for any event will come via that descriptor.\n\nThe seemingly straight forward solution of moving the waitq into the\nringbuffer object doesn\u0027t work because of life-time issues. One could\nperf_event_set_output() on a fd that you\u0027re also blocking on and cause\nthe old rb object to be freed while its waitq would still be\nreferenced by the blocked thread -\u003e FAIL.\n\nTherefore link all events to the ringbuffer and broadcast the wakeup\nfrom the ringbuffer object to all possible events that could be waited\nupon. This is rather ugly, and we\u0027re open to better solutions but it\nworks for now.\n\nReported-by: Stephane Eranian \u003ceranian@google.com\u003e\nFinished-by: Stephane Eranian \u003ceranian@google.com\u003e\nReviewed-by: Stephane Eranian \u003ceranian@google.com\u003e\nSigned-off-by: Peter Zijlstra \u003ca.p.zijlstra@chello.nl\u003e\nLink: http://lkml.kernel.org/r/20111126014731.GA7030@quad\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n" }, { "commit": "9251f904f95175b4a1d8cbc0449e748f9edd7629", "tree": "585b73028c0c6b955d1759faf8544870eae213f5", "parents": [ "efc96737bd82b508794d2b28061a12af4a3f7766" ], "author": { "name": "Borislav Petkov", "email": "borislav.petkov@amd.com", "time": "Sun Oct 16 17:15:04 2011 +0200" }, "committer": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Mon Nov 14 13:31:26 2011 +0100" }, "message": "perf: Carve out callchain functionality\n\nSplit the callchain code from the perf events core into\na new kernel/events/callchain.c file.\n\nThis simplifies a bit the big core.c\n\nSigned-off-by: Borislav Petkov \u003cborislav.petkov@amd.com\u003e\nCc: Arnaldo Carvalho de Melo \u003cacme@redhat.com\u003e\nCc: Stephane Eranian \u003ceranian@google.com\u003e\n[keep ctx recursion handling inline and use internal headers]\nSigned-off-by: Frederic Weisbecker \u003cfweisbec@gmail.com\u003e\nSigned-off-by: Peter Zijlstra \u003ca.p.zijlstra@chello.nl\u003e\nLink: http://lkml.kernel.org/r/1318778104-17152-1-git-send-email-fweisbec@gmail.com\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n" }, { "commit": "a8b0ca17b80e92faab46ee7179ba9e99ccb61233", "tree": "a4a6282139f26458f80dcbe21c709a9290e84143", "parents": [ "1880c4ae182afb5650c5678949ecfe7ff66a724e" ], "author": { "name": "Peter Zijlstra", "email": "a.p.zijlstra@chello.nl", "time": "Mon Jun 27 14:41:57 2011 +0200" }, "committer": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Fri Jul 01 11:06:35 2011 +0200" }, "message": "perf: Remove the nmi parameter from the swevent and overflow interface\n\nThe nmi parameter indicated if we could do wakeups from the current\ncontext, if not, we would set some state and self-IPI and let the\nresulting interrupt do the wakeup.\n\nFor the various event classes:\n\n - hardware: nmi\u003d0; PMI is in fact an NMI or we run irq_work_run from\n the PMI-tail (ARM etc.)\n - tracepoint: nmi\u003d0; since tracepoint could be from NMI context.\n - software: nmi\u003d[0,1]; some, like the schedule thing cannot\n perform wakeups, and hence need 0.\n\nAs one can see, there is very little nmi\u003d1 usage, and the down-side of\nnot using it is that on some platforms some software events can have a\njiffy delay in wakeup (when arch_irq_work_raise isn\u0027t implemented).\n\nThe up-side however is that we can remove the nmi parameter and save a\nbunch of conditionals in fast paths.\n\nSigned-off-by: Peter Zijlstra \u003ca.p.zijlstra@chello.nl\u003e\nCc: Michael Cree \u003cmcree@orcon.net.nz\u003e\nCc: Will Deacon \u003cwill.deacon@arm.com\u003e\nCc: Deng-Cheng Zhu \u003cdengcheng.zhu@gmail.com\u003e\nCc: Anton Blanchard \u003canton@samba.org\u003e\nCc: Eric B Munson \u003cemunson@mgebm.net\u003e\nCc: Heiko Carstens \u003cheiko.carstens@de.ibm.com\u003e\nCc: Paul Mundt \u003clethal@linux-sh.org\u003e\nCc: David S. Miller \u003cdavem@davemloft.net\u003e\nCc: Frederic Weisbecker \u003cfweisbec@gmail.com\u003e\nCc: Jason Wessel \u003cjason.wessel@windriver.com\u003e\nCc: Don Zickus \u003cdzickus@redhat.com\u003e\nLink: http://lkml.kernel.org/n/tip-agjev8eu666tvknpb3iaj0fg@git.kernel.org\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n" }, { "commit": "76369139ceb955deefc509e6e12ce9d6ce50ccab", "tree": "5ba2732188d2fecd4fd49b93f73e51e17fb69082", "parents": [ "b58f6b0dd3d677338b9065388cc2cc942b86338e" ], "author": { "name": "Frederic Weisbecker", "email": "fweisbec@gmail.com", "time": "Thu May 19 19:55:04 2011 +0200" }, "committer": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Thu Jun 09 12:57:54 2011 +0200" }, "message": "perf: Split up buffer handling from core code\n\nAnd create the internal perf events header.\n\nv2: Keep an internal inlined perf_output_copy()\n\nSigned-off-by: Frederic Weisbecker \u003cfweisbec@gmail.com\u003e\nAcked-by: Peter Zijlstra \u003ca.p.zijlstra@chello.nl\u003e\nCc: Borislav Petkov \u003cbp@alien8.de\u003e\nCc: Stephane Eranian \u003ceranian@google.com\u003e\nCc: Arnaldo Carvalho de Melo \u003cacme@redhat.com\u003e\nCc: Steven Rostedt \u003crostedt@goodmis.org\u003e\nLink: http://lkml.kernel.org/r/1305827704-5607-1-git-send-email-fweisbec@gmail.com\n[ v3: use clearer \u0027ring_buffer\u0027 and \u0027rb\u0027 naming ]\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n" } ] }