2 .\" Copyright (c) 2008, Sun Microsystems, Inc. All Rights Reserved.
3 .\" The contents of this file are subject to the terms of the Common Development and Distribution License (the "License"). You may not use this file except in compliance with the License.
4 .\" You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. See the License for the specific language governing permissions and limitations under the License.
5 .\" When distributing Covered Code, include this CDDL HEADER in each file and include the License file at usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your own identifying information: Portions Copyright [yyyy] [name of copyright owner]
6 .TH LOCKSTAT 1M "Dec 10, 2017"
8 lockstat \- report kernel lock and profiling statistics
12 \fBlockstat\fR [\fB-ACEHI\fR] [\fB-e\fR \fIevent_list\fR] [\fB-i\fR \fIrate\fR]
13 [\fB-b\fR | \fB-t\fR | \fB-h\fR | \fB-s\fR \fIdepth\fR] [\fB-n\fR \fInrecords\fR]
14 [\fB-l\fR \fIlock\fR [, \fIsize\fR]] [\fB-d\fR \fIduration\fR]
15 [\fB-f\fR \fIfunction\fR [, \fIsize\fR]] [\fB-T\fR] [\fB-ckgwWRpP\fR] [\fB-D\fR \fIcount\fR]
16 [\fB-o\fR \fIfilename\fR] [\fB-x\fR \fIopt\fR [=val]] \fIcommand\fR [\fIargs\fR]
21 The \fBlockstat\fR utility gathers and displays kernel locking and profiling
22 statistics. \fBlockstat\fR allows you to specify which events to watch (for
23 example, spin on adaptive mutex, block on read access to rwlock due to waiting
24 writers, and so forth) how much data to gather for each event, and how to
25 display the data. By default, \fBlockstat\fR monitors all lock contention
26 events, gathers frequency and timing data about those events, and displays the
27 data in decreasing frequency order, so that the most common events appear
31 \fBlockstat\fR gathers data until the specified command completes. For example,
32 to gather statistics for a fixed-time interval, use \fBsleep\fR(1) as the
36 \fBexample#\fR \fBlockstat\fR \fBsleep\fR \fB5\fR
39 When the \fB-I\fR option is specified, \fBlockstat\fR establishes a
40 per-processor high-level periodic interrupt source to gather profiling data.
41 The interrupt handler simply generates a \fBlockstat\fR event whose caller is
42 the interrupted PC (program counter). The profiling event is just like any
43 other \fBlockstat\fR event, so all of the normal \fBlockstat\fR options are
47 \fBlockstat\fR relies on DTrace to modify the running kernel's text to
48 intercept events of interest. This imposes a small but measurable overhead on
49 all system activity, so access to \fBlockstat\fR is restricted to super-user by
50 default. The system administrator can permit other users to use \fBlockstat\fR
51 by granting them additional DTrace privileges. Refer to the \fIDynamic
52 Tracing Guide\fR for more information about DTrace security features.
55 The following options are supported:
58 If no event selection options are specified, the default is \fB-C\fR.
66 Watch all lock events. \fB-A\fR is equivalent to \fB-CH\fR.
76 Watch contention events.
92 \fB\fB\fR\fB-e\fR \fIevent_list\fR\fR
96 Only watch the specified events. \fIevent\fR \fIlist\fR is a comma-separated
97 list of events or ranges of events such as 1,4-7,35. Run \fBlockstat\fR with no
98 arguments to get a brief description of all events.
118 Watch profiling interrupt events.
124 \fB\fB\fR\fB-i\fR \fIrate\fR\fR
128 Interrupt rate (per second) for \fB-I\fR. The default is 97 Hz, so that
129 profiling doesn't run in lockstep with the clock interrupt (which runs at 100
136 \fB\fB-x\fR \fIarg\fR[=\fIval\fR]\fR
140 Enable or modify a DTrace runtime option or D compiler option. The list of
141 options is found in the \fI\fR. Boolean options are enabled by specifying their
142 name. Options with values are set by separating the option name and value with
146 .SS "Data Gathering (Mutually Exclusive)"
153 Basic statistics: lock, caller, number of events.
163 Histogram: Timing plus time-distribution histograms.
169 \fB\fB\fR\fB-s\fR \fIdepth\fR\fR
173 Stack trace: Histogram plus stack traces up to \fIdepth\fR frames deep.
183 Timing: Basic plus timing for all events [default].
189 \fB\fB\fR\fB-d\fR \fIduration\fR\fR
193 Only watch events longer than \fIduration\fR.
199 \fB\fB\fR\fB-f\fR \fIfunc[,size]\fR\fR
203 Only watch events generated by \fIfunc\fR, which can be specified as a symbolic
204 name or hex address. \fIsize\fR defaults to the \fBELF\fR symbol size if
205 available, or \fB1\fR if not.
211 \fB\fB\fR\fB-l\fR \fIlock[,size]\fR\fR
215 Only watch \fIlock\fR, which can be specified as a symbolic name or hex
216 address. \fBsize\fR defaults to the \fBELF\fR symbol size or \fB1\fR if the
217 symbol size is not available.
223 \fB\fB\fR\fB-n\fR \fInrecords\fR\fR
227 Maximum number of data records.
237 Trace (rather than sample) events [off by default].
247 Coalesce lock data for lock arrays (for example, \fBpse_mutex[]\fR).
253 \fB\fB\fR\fB-D\fR \fIcount\fR\fR
257 Only display the top \fIcount\fR events of each type.
267 Show total events generated by function. For example, if \fBfoo()\fR calls
268 \fBbar()\fR in a loop, the work done by \fBbar()\fR counts as work generated by
269 \fBfoo()\fR (along with any work done by \fBfoo()\fR itself). The \fB-g\fR
270 option works by counting the total number of stack frames in which each
271 function appears. This implies two things: (1) the data reported by \fB-g\fR
272 can be misleading if the stack traces are not deep enough, and (2) functions
273 that are called recursively might show greater than 100% activity. In light of
274 issue (1), the default data gathering mode when using \fB-g\fR is \fB-s\fR
285 Coalesce PCs within functions.
291 \fB\fB\fR\fB-o\fR \fIfilename\fR\fR
295 Direct output to \fIfilename\fR.
305 Sort data by (\fIcount * time\fR) product.
315 Parsable output format.
325 Display rates (events per second) rather than counts.
335 Whichever: distinguish events only by caller, not by lock.
345 Wherever: distinguish events only by lock, not by caller.
350 The following headers appear over various columns of data.
354 \fB\fBCount\fR or \fBops/s\fR\fR
358 Number of times this event occurred, or the rate (times per second) if \fB-R\fR
369 Percentage of all events represented by this individual event.
379 Percentage of all events generated by this function.
389 Cumulative percentage; a running total of the individuals.
399 Average reference count. This will always be \fB1\fR for exclusive locks
400 (mutexes, spin locks, rwlocks held as writer) but can be greater than \fB1\fR
401 for shared locks (rwlocks held as reader).
411 Average duration of the events in nanoseconds, as appropriate for the event.
412 For the profiling event, duration means interrupt latency.
422 Address of the lock; displayed symbolically if possible.
432 \fBCPU\fR plus processor interrupt level (\fBPIL\fR). For example, if \fBCPU\fR
433 4 is interrupted while at \fBPIL\fR 6, this will be reported as \fBcpu[4]+6\fR.
443 Address of the caller; displayed symbolically if possible.
448 \fBExample 1 \fRMeasuring Kernel Lock Contention
452 example# \fBlockstat sleep 5\fR
453 Adaptive mutex spin: 2210 events in 5.055 seconds (437 events/sec)
461 Count indv cuml rcnt nsec Lock Caller
462 ------------------------------------------------------------------------
463 269 12% 12% 1.00 2160 service_queue background+0xdc
464 249 11% 23% 1.00 86 service_queue qenable_locked+0x64
465 228 10% 34% 1.00 131 service_queue background+0x15c
466 68 3% 37% 1.00 79 0x30000024070 untimeout+0x1c
467 59 3% 40% 1.00 384 0x300066fa8e0 background+0xb0
468 43 2% 41% 1.00 30 rqcred_lock svc_getreq+0x3c
469 42 2% 43% 1.00 341 0x30006834eb8 background+0xb0
470 41 2% 45% 1.00 135 0x30000021058 untimeout+0x1c
471 40 2% 47% 1.00 39 rqcred_lock svc_getreq+0x260
472 37 2% 49% 1.00 2372 0x300068e83d0 hmestart+0x1c4
473 36 2% 50% 1.00 77 0x30000021058 timeout_common+0x4
474 36 2% 52% 1.00 354 0x300066fa120 background+0xb0
475 32 1% 53% 1.00 97 0x30000024070 timeout_common+0x4
476 31 1% 55% 1.00 2923 0x300069883d0 hmestart+0x1c4
477 29 1% 56% 1.00 366 0x300066fb290 background+0xb0
478 28 1% 57% 1.00 117 0x3000001e040 untimeout+0x1c
479 25 1% 59% 1.00 93 0x3000001e040 timeout_common+0x4
480 22 1% 60% 1.00 25 0x30005161110 sync_stream_buf+0xdc
481 21 1% 60% 1.00 291 0x30006834eb8 putq+0xa4
482 19 1% 61% 1.00 43 0x3000515dcb0 mdf_alloc+0xc
483 18 1% 62% 1.00 456 0x30006834eb8 qenable+0x8
484 18 1% 63% 1.00 61 service_queue queuerun+0x168
485 17 1% 64% 1.00 268 0x30005418ee8 vmem_free+0x3c
488 R/W reader blocked by writer: 76 events in 5.055 seconds (15 events/sec)
490 Count indv cuml rcnt nsec Lock Caller
491 ------------------------------------------------------------------------
492 23 30% 30% 1.00 22590137 0x300098ba358 ufs_dirlook+0xd0
493 17 22% 53% 1.00 5820995 0x3000ad815e8 find_bp+0x10
494 13 17% 70% 1.00 2639918 0x300098ba360 ufs_iget+0x198
495 4 5% 75% 1.00 3193015 0x300098ba360 ufs_getattr+0x54
496 3 4% 79% 1.00 7953418 0x3000ad817c0 find_bp+0x10
497 3 4% 83% 1.00 935211 0x3000ad815e8 find_read_lof+0x14
498 2 3% 86% 1.00 16357310 0x300073a4720 find_bp+0x10
499 2 3% 88% 1.00 2072433 0x300073a4720 find_read_lof+0x14
500 2 3% 91% 1.00 1606153 0x300073a4370 find_bp+0x10
501 1 1% 92% 1.00 2656909 0x300107e7400 ufs_iget+0x198
508 \fBExample 2 \fRMeasuring Hold Times
512 example# \fBlockstat -H -D 10 sleep 1\fR
513 Adaptive mutex spin: 513 events
521 Count indv cuml rcnt nsec Lock Caller
522 -------------------------------------------------------------------------
523 480 5% 5% 1.00 1136 0x300007718e8 putnext+0x40
524 286 3% 9% 1.00 666 0x3000077b430 getf+0xd8
525 271 3% 12% 1.00 537 0x3000077b430 msgio32+0x2fc
526 270 3% 15% 1.00 3670 0x300007718e8 strgetmsg+0x3d4
527 270 3% 18% 1.00 1016 0x300007c38b0 getq_noenab+0x200
528 264 3% 20% 1.00 1649 0x300007718e8 strgetmsg+0xa70
529 216 2% 23% 1.00 6251 tcp_mi_lock tcp_snmp_get+0xfc
530 206 2% 25% 1.00 602 thread_free_lock clock+0x250
531 138 2% 27% 1.00 485 0x300007c3998 putnext+0xb8
532 138 2% 28% 1.00 3706 0x300007718e8 strrput+0x5b8
533 -------------------------------------------------------------------------
540 \fBExample 3 \fRMeasuring Hold Times for Stack Traces Containing a Specific
545 example# \fBlockstat -H -f tcp_rput_data -s 50 -D 10 sleep 1\fR
546 Adaptive mutex spin: 11 events in 1.023 seconds (11
555 -------------------------------------------------------------------------
556 Count indv cuml rcnt nsec Lock Caller
557 9 82% 82% 1.00 2540 0x30000031380 tcp_rput_data+0x2b90
559 nsec ------ Time Distribution ------ count Stack
560 256 |@@@@@@@@@@@@@@@@ 5 tcp_rput_data+0x2b90
561 512 |@@@@@@ 2 putnext+0x78
562 1024 |@@@ 1 ip_rput+0xec4
563 2048 | 0 _c_putnext+0x148
564 4096 | 0 hmeread+0x31c
565 8192 | 0 hmeintr+0x36c
567 sbus_intr_wrapper+0x30
570 Count indv cuml rcnt nsec Lock Caller
571 1 9% 91% 1.00 1036 0x30000055380 freemsg+0x44
573 nsec ------ Time Distribution ------ count Stack
574 1024 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ 1 freemsg+0x44
582 sbus_intr_wrapper+0x30
583 -------------------------------------------------------------------------
590 \fBExample 4 \fRBasic Kernel Profiling
593 For basic profiling, we don't care whether the profiling interrupt sampled
594 \fBfoo()\fR\fB+0x4c\fR or \fBfoo()\fR\fB+0x78\fR; we care only that it sampled
595 somewhere in \fBfoo()\fR, so we use \fB-k\fR. The \fBCPU\fR and \fBPIL\fR
596 aren't relevant to basic profiling because we are measuring the system as a
597 whole, not a particular \fBCPU\fR or interrupt level, so we use \fB-W\fR.
602 example# \fBlockstat -kIW -D 20 ./polltest\fR
603 Profiling interrupt: 82 events in 0.424 seconds (194
612 Count indv cuml rcnt nsec Hottest CPU+PIL Caller
613 -----------------------------------------------------------------------
614 8 10% 10% 1.00 698 cpu[1] utl0
615 6 7% 17% 1.00 299 cpu[0] read
616 5 6% 23% 1.00 124 cpu[1] getf
617 4 5% 28% 1.00 327 cpu[0] fifo_read
618 4 5% 33% 1.00 112 cpu[1] poll
619 4 5% 38% 1.00 212 cpu[1] uiomove
620 4 5% 43% 1.00 361 cpu[1] mutex_tryenter
621 3 4% 46% 1.00 682 cpu[0] write
622 3 4% 50% 1.00 89 cpu[0] pcache_poll
623 3 4% 54% 1.00 118 cpu[1] set_active_fd
624 3 4% 57% 1.00 105 cpu[0] syscall_trap32
625 3 4% 61% 1.00 640 cpu[1] (usermode)
626 2 2% 63% 1.00 127 cpu[1] fifo_poll
627 2 2% 66% 1.00 300 cpu[1] fifo_write
628 2 2% 68% 1.00 669 cpu[0] releasef
629 2 2% 71% 1.00 112 cpu[1] bt_getlowbit
630 2 2% 73% 1.00 247 cpu[1] splx
631 2 2% 76% 1.00 503 cpu[0] mutex_enter
632 2 2% 78% 1.00 467 cpu[0]+10 disp_lock_enter
633 2 2% 80% 1.00 139 cpu[1] default_copyin
634 -----------------------------------------------------------------------
641 \fBExample 5 \fRGenerated-load Profiling
644 In the example above, 5% of the samples were in \fBpoll()\fR. This tells us how
645 much time was spent inside \fBpoll()\fR itself, but tells us nothing about how
646 much work was \fBgenerated\fR by \fBpoll()\fR; that is, how much time we spent
647 in functions called by \fBpoll()\fR. To determine that, we use the \fB-g\fR
648 option. The example below shows that although \fBpolltest\fR spends only 5% of
649 its time in \fBpoll()\fR itself, \fBpoll()\fR-induced work accounts for 34% of
654 Note that the functions that generate the profiling interrupt
655 (\fBlockstat_intr()\fR, \fBcyclic_fire()\fR, and so forth) appear in every
656 stack trace, and therefore are considered to have generated 100% of the load.
657 This illustrates an important point: the generated load percentages do
658 \fBnot\fR add up to 100% because they are not independent. If 72% of all stack
659 traces contain both \fBfoo()\fR and \fBbar()\fR, then both \fBfoo()\fR and
660 \fBbar()\fR are 72% load generators.
665 example# \fBlockstat -kgIW -D 20 ./polltest\fR
666 Profiling interrupt: 80 events in 0.412 seconds (194 events/sec)
674 Count genr cuml rcnt nsec Hottest CPU+PIL Caller
675 -------------------------------------------------------------------------
676 80 100% ---- 1.00 310 cpu[1] lockstat_intr
677 80 100% ---- 1.00 310 cpu[1] cyclic_fire
678 80 100% ---- 1.00 310 cpu[1] cbe_level14
679 80 100% ---- 1.00 310 cpu[1] current_thread
680 27 34% ---- 1.00 176 cpu[1] poll
681 20 25% ---- 1.00 221 cpu[0] write
682 19 24% ---- 1.00 249 cpu[1] read
683 17 21% ---- 1.00 232 cpu[0] write32
684 17 21% ---- 1.00 207 cpu[1] pcache_poll
685 14 18% ---- 1.00 319 cpu[0] fifo_write
686 13 16% ---- 1.00 214 cpu[1] read32
687 10 12% ---- 1.00 208 cpu[1] fifo_read
688 10 12% ---- 1.00 787 cpu[1] utl0
689 9 11% ---- 1.00 178 cpu[0] pcacheset_resolve
690 9 11% ---- 1.00 262 cpu[0] uiomove
691 7 9% ---- 1.00 506 cpu[1] (usermode)
692 5 6% ---- 1.00 195 cpu[1] fifo_poll
693 5 6% ---- 1.00 136 cpu[1] syscall_trap32
694 4 5% ---- 1.00 139 cpu[0] releasef
695 3 4% ---- 1.00 277 cpu[1] polllock
696 -------------------------------------------------------------------------
703 \fBExample 6 \fRGathering Lock Contention and Profiling Data for a Specific
707 In this example we use the \fB-f\fR option not to specify a single function,
708 but rather to specify the entire text space of the \fBsbus\fR module. We gather
709 both lock contention and profiling statistics so that contention can be
710 correlated with overall load on the module.
715 example# \fBmodinfo | grep sbus\fR
716 24 102a8b6f b8b4 59 1 sbus (SBus (sysio) nexus driver)
724 example# \fBlockstat -kICE -f 0x102a8b6f,0xb8b4 sleep 10\fR
725 Adaptive mutex spin: 39 events in 10.042 seconds (4 events/sec)
733 Count indv cuml rcnt nsec Lock Caller
734 -------------------------------------------------------------------------
735 15 38% 38% 1.00 206 0x30005160528 sync_stream_buf
736 7 18% 56% 1.00 14 0x30005160d18 sync_stream_buf
737 6 15% 72% 1.00 27 0x300060c3118 sync_stream_buf
738 5 13% 85% 1.00 24 0x300060c3510 sync_stream_buf
739 2 5% 90% 1.00 29 0x300060c2d20 sync_stream_buf
740 2 5% 95% 1.00 24 0x30005161cf8 sync_stream_buf
741 1 3% 97% 1.00 21 0x30005161110 sync_stream_buf
742 1 3% 100% 1.00 23 0x30005160130 sync_stream_buf
745 Adaptive mutex block: 9 events in 10.042 seconds (1 events/sec)
747 Count indv cuml rcnt nsec Lock Caller
748 -------------------------------------------------------------------------
749 4 44% 44% 1.00 156539 0x30005160528 sync_stream_buf
750 2 22% 67% 1.00 763516 0x30005160d18 sync_stream_buf
751 1 11% 78% 1.00 462130 0x300060c3510 sync_stream_buf
752 1 11% 89% 1.00 288749 0x30005161110 sync_stream_buf
753 1 11% 100% 1.00 1015374 0x30005160130 sync_stream_buf
756 Profiling interrupt: 229 events in 10.042 seconds (23 events/sec)
758 Count indv cuml rcnt nsec Hottest CPU+PIL Caller
760 -------------------------------------------------------------------------
761 89 39% 39% 1.00 426 cpu[0]+6 sync_stream_buf
762 64 28% 67% 1.00 398 cpu[0]+6 sbus_intr_wrapper
763 23 10% 77% 1.00 324 cpu[0]+6 iommu_dvma_kaddr_load
764 21 9% 86% 1.00 512 cpu[0]+6 iommu_tlb_flush
765 14 6% 92% 1.00 342 cpu[0]+6 iommu_dvma_unload
766 13 6% 98% 1.00 306 cpu[1] iommu_dvma_sync
767 5 2% 100% 1.00 389 cpu[1] iommu_dma_bindhdl
768 -------------------------------------------------------------------------
775 \fBExample 7 \fRDetermining the Average PIL (processor interrupt level) for a
780 example# \fBlockstat -Iw -l cpu[3] ./testprog\fR
782 Profiling interrupt: 14791 events in 152.463 seconds (97 events/sec)
784 Count indv cuml rcnt nsec CPU+PIL Hottest Caller
786 -----------------------------------------------------------------------
787 13641 92% 92% 1.00 253 cpu[3] (usermode)
788 579 4% 96% 1.00 325 cpu[3]+6 ip_ocsum+0xe8
789 375 3% 99% 1.00 411 cpu[3]+10 splx
790 154 1% 100% 1.00 527 cpu[3]+4 fas_intr_svc+0x80
791 41 0% 100% 1.00 293 cpu[3]+13 send_mondo+0x18
792 1 0% 100% 1.00 266 cpu[3]+12 zsa_rxint+0x400
793 -----------------------------------------------------------------------
800 \fBExample 8 \fRDetermining which Subsystem is Causing the System to be Busy
804 example# \fBlockstat -s 10 -I sleep 20\fR
806 Profiling interrupt: 4863 events in 47.375 seconds (103 events/sec)
808 Count indv cuml rcnt nsec CPU+PIL Caller
810 -----------------------------------------------------------------------
811 1929 40% 40% 0.00 3215 cpu[0] usec_delay+0x78
812 nsec ------ Time Distribution ------ count Stack
813 4096 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ 1872 ata_wait+0x90
814 8192 | 27 acersb_get_intr_status+0x34
815 16384 | 29 ata_set_feature+0x124
816 32768 | 1 ata_disk_start+0x15c
818 ghd_waitq_process_and \e
820 ghd_waitq_process_and \e
823 ata_disk_tran_start+0x108
824 -----------------------------------------------------------------------
832 \fBdtrace\fR(1M), \fBplockstat\fR(1M), \fBattributes\fR(5), \fBlockstat\fR(7D),
833 \fBmutex\fR(9F), \fBrwlock\fR(9F)
836 \fIDynamic Tracing Guide\fR:
839 https://illumos.org/books/dtrace/
843 Tail-call elimination can affect call sites. For example, if
844 \fBfoo()\fR\fB+0x50\fR calls \fBbar()\fR and the last thing \fBbar()\fR does is
845 call \fBmutex_exit()\fR, the compiler can arrange for \fBbar()\fR to branch to
846 \fBmutex_exit()\fR with a return address of \fBfoo()\fR\fB+0x58\fR. Thus, the
847 \fBmutex_exit()\fR in \fBbar()\fR will appear as though it occurred at
848 \fBfoo()\fR\fB+0x58\fR.
851 The \fBPC\fR in the stack frame in which an interrupt occurs can be bogus
852 because, between function calls, the compiler is free to use the return address
853 register for local storage.
856 When using the \fB-I\fR and \fB-s\fR options together, the interrupted PC will
857 usually not appear anywhere in the stack since the interrupt handler is entered
858 asynchronously, not by a function call from that \fBPC\fR.
861 The \fBlockstat\fR technology is provided on an as-is basis. The format and
862 content of \fBlockstat\fR output reflect the current kernel
863 implementation and are therefore subject to change in future releases.