4 .\" The contents of this file are subject to the terms of the
5 .\" Common Development and Distribution License (the "License").
6 .\" You may not use this file except in compliance with the License.
8 .\" You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 .\" or http://www.opensolaris.org/os/licensing.
10 .\" See the License for the specific language governing permissions
11 .\" and limitations under the License.
13 .\" When distributing Covered Code, include this CDDL HEADER in each
14 .\" file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 .\" If applicable, add the following below this CDDL HEADER, with the
16 .\" fields enclosed by brackets "[]" replaced with your own identifying
17 .\" information: Portions Copyright [yyyy] [name of copyright owner]
20 .\" Copyright (c) 2008, Sun Microsystems, Inc. All Rights Reserved.
29 .Nd report kernel lock and profiling statistics
33 .Op Fl e Ar event-list
35 .Op Fl b | t | h | s Ar depth
36 .Op Fl n Ar num-records
37 .Op Fl l Ar lock Oo Ns , Ns Ar size Oc
39 .Op Fl f Ar function Oo Ns , Ns Ar size Oc
44 .Op Fl x Ar opt Oo Ns = Ns Ar val Oc
50 utility gathers and displays kernel locking and profiling statistics.
52 allows you to specify which events to watch (for example, spin on adaptive
53 mutex, block on read access to rwlock due to waiting writers, and so forth), how
54 much data to gather for each event, and how to display the data.
57 monitors all lock contention events, gathers frequency and timing data about
58 those events, and displays the data in decreasing frequency order, so that the
59 most common events appear first.
62 gathers data until the specified command completes.
63 For example, to gather statistics for a fixed-time interval, use
65 as the command, as follows:
67 .Dl # lockstat sleep 5
73 establishes a per-processor high-level periodic interrupt source to gather
75 The interrupt handler simply generates a
77 event whose caller is the interrupted PC (program counter).
78 The profiling event is just like any other
80 event, so all of the normal
82 options are applicable.
85 relies on DTrace to modify the running kernel's text to intercept events of
87 This imposes a small but measurable overhead on all system activity, so access
90 is restricted to super-user by default.
92 The following options are supported:
93 .Bl -tag -width indent
95 Print the D program used to gather the requested data.
98 If no event selection options are specified, the default is
100 .Bl -tag -width indent
102 Watch all lock events.
107 Watch contention events.
110 .It Fl e Ar event-list
111 Only watch the specified events.
113 is a comma-separated list of events or ranges of events such as 1,4-7,35.
116 with no arguments to get a brief description of all events.
120 Watch profiling interrupt events.
122 Interrupt rate (per second) for
124 The default is 97 Hz, so that profiling doesn't run in lockstep with the clock
125 interrupt (which runs at 100 Hz).
128 .Bl -tag -width indent
129 .It Fl x Ar arg Oo Ns = Ns Ar val Oc
132 runtime option or D compiler option.
133 Boolean options are enabled by specifying their name.
134 Options with values are set by separating the option name and value with an
137 .Ss "Data Gathering (Mutually Exclusive)"
138 .Bl -tag -width indent
140 Basic statistics: lock, caller, number of events.
142 Histogram: timing plus time-distribution histograms.
144 Stack trace: histogram plus stack traces up to
148 Timing: Basic plus timing for all events (default).
151 .Bl -tag -width indent
153 Only watch events longer than
155 .It Fl f Ar func Ns Oo Ns , Ns Ar size Oc Ns
156 Only watch events generated by
158 which can be specified as a symbolic name or hex address.
160 defaults to the ELF symbol size if available, or 1 if not.
161 .It Fl l Ar lock Ns Oo Ns , Ns Ar size Oc Ns
164 which can be specified as a symbolic name or hex address.
166 defaults to the ELF symbol size or 1 if the symbol size is not available.
167 .It Fl n Ar num-records
168 Maximum number of data records.
170 Trace (rather than sample) events.
171 This is off by default.
174 .Bl -tag -width indent
180 Show total events generated by function.
185 in a loop, the work done by
187 counts as work generated by
189 (along with any work done by
194 option works by counting the total number of stack frames in which each function
196 This implies two things: (1) the data reported by
198 can be misleading if the stack traces are not deep enough, and (2) functions
199 that are called recursively might show greater than 100% activity.
200 In light of issue (1), the default data gathering mode when using
205 Coalesce PCs within functions.
210 Sort data by (\fIcount * time\fR) product.
212 Parsable output format.
214 Display rates (events per second) rather than counts.
216 Whichever: distinguish events only by caller, not by lock.
218 Wherever: distinguish events only by lock, not by caller.
221 The following headers appear over various columns of data.
222 .Bl -tag -width indent
224 Number of times this event occurred, or the rate (times per second) if
228 Percentage of all events represented by this individual event.
230 Percentage of all events generated by this function.
232 Cumulative percentage; a running total of the individuals.
234 Average reference count.
235 This will always be 1 for exclusive locks (mutexes,
236 spin locks, rwlocks held as writer) but can be greater than 1 for shared locks
237 (rwlocks held as reader).
239 Average duration of the events in nanoseconds, as appropriate for the event.
240 For the profiling event, duration means interrupt latency.
242 Address of the lock; displayed symbolically if possible.
244 CPU plus the priority class of the interrupted thread.
245 For example, if CPU 4 is interrupted while running a timeshare thread, this
249 Address of the caller; displayed symbolically if possible.
253 .It Example 1 Measuring Kernel Lock Contention
255 .Li # lockstat sleep 5
257 Adaptive mutex spin: 41411 events in 5.011 seconds (8263 events/sec)
259 Count indv cuml rcnt nsec Lock Caller
260 -------------------------------------------------------------------------------
261 13750 33% 33% 0.00 72 vm_page_queue_free_mtx vm_page_free_toq+0x12e
262 13648 33% 66% 0.00 66 vm_page_queue_free_mtx vm_page_alloc+0x138
263 4023 10% 76% 0.00 51 vm_dom+0x80 vm_page_dequeue+0x68
264 2672 6% 82% 0.00 186 vm_dom+0x80 vm_page_enqueue+0x63
265 618 1% 84% 0.00 31 0xfffff8000cd83a88 qsyncvp+0x37
266 506 1% 85% 0.00 164 0xfffff8000cb3f098 vputx+0x5a
267 477 1% 86% 0.00 69 0xfffff8000c7eb180 uma_dbg_getslab+0x5b
268 288 1% 87% 0.00 77 0xfffff8000cd8b000 vn_finished_write+0x29
269 263 1% 88% 0.00 103 0xfffff8000cbad448 vinactive+0xdc
270 259 1% 88% 0.00 53 0xfffff8000cd8b000 vfs_ref+0x24
271 237 1% 89% 0.00 20 0xfffff8000cbad448 vfs_hash_get+0xcc
272 233 1% 89% 0.00 22 0xfffff8000bfd9480 uma_dbg_getslab+0x5b
273 223 1% 90% 0.00 20 0xfffff8000cb3f098 cache_lookup+0x561
274 193 0% 90% 0.00 16 0xfffff8000cb40ba8 vref+0x27
275 175 0% 91% 0.00 34 0xfffff8000cbad448 vputx+0x5a
276 169 0% 91% 0.00 51 0xfffff8000cd8b000 vfs_unbusy+0x27
277 164 0% 92% 0.00 31 0xfffff8000cb40ba8 vputx+0x5a
280 Adaptive mutex block: 10 events in 5.011 seconds (2 events/sec)
282 Count indv cuml rcnt nsec Lock Caller
283 -------------------------------------------------------------------------------
284 3 30% 30% 0.00 17592 vm_page_queue_free_mtx vm_page_alloc+0x138
285 2 20% 50% 0.00 20528 vm_dom+0x80 vm_page_enqueue+0x63
286 2 20% 70% 0.00 55502 0xfffff8000cb40ba8 vputx+0x5a
287 1 10% 80% 0.00 12007 vm_page_queue_free_mtx vm_page_free_toq+0x12e
288 1 10% 90% 0.00 9125 0xfffff8000cbad448 vfs_hash_get+0xcc
289 1 10% 100% 0.00 7864 0xfffff8000cd83a88 qsyncvp+0x37
290 -------------------------------------------------------------------------------
293 .It Example 2 Measuring Hold Times
295 .Li # lockstat -H -D 10 sleep 1
297 Adaptive mutex hold: 109589 events in 1.039 seconds (105526 events/sec)
299 Count indv cuml rcnt nsec Lock Caller
300 -------------------------------------------------------------------------------
301 8998 8% 8% 0.00 617 0xfffff8000c7eb180 uma_dbg_getslab+0xd4
302 5901 5% 14% 0.00 917 vm_page_queue_free_mtx vm_object_terminate+0x16a
303 5040 5% 18% 0.00 902 vm_dom+0x80 vm_page_free_toq+0x88
304 4884 4% 23% 0.00 1056 vm_page_queue_free_mtx vm_page_alloc+0x44e
305 4664 4% 27% 0.00 759 vm_dom+0x80 vm_fault_hold+0x1a13
306 4011 4% 31% 0.00 888 vm_dom vm_page_advise+0x11b
307 4010 4% 34% 0.00 957 vm_dom+0x80 _vm_page_deactivate+0x5c
308 3743 3% 38% 0.00 582 0xfffff8000cf04838 pmap_is_prefaultable+0x158
309 2254 2% 40% 0.00 952 vm_dom vm_page_free_toq+0x88
310 1639 1% 41% 0.00 591 0xfffff800d60065b8 trap_pfault+0x1f7
311 -------------------------------------------------------------------------------
314 R/W writer hold: 64314 events in 1.039 seconds (61929 events/sec)
316 Count indv cuml rcnt nsec Lock Caller
317 -------------------------------------------------------------------------------
318 7421 12% 12% 0.00 2994 pvh_global_lock pmap_page_is_mapped+0xb6
319 4668 7% 19% 0.00 3313 pvh_global_lock pmap_enter+0x9ae
320 1639 3% 21% 0.00 733 0xfffff80168d10200 vm_object_deallocate+0x683
321 1639 3% 24% 0.00 3061 0xfffff80168d10200 unlock_and_deallocate+0x2b
322 1639 3% 26% 0.00 2966 0xfffff80168d10200 vm_fault_hold+0x16ee
323 1567 2% 29% 0.00 733 0xfffff80168d10200 vm_fault_hold+0x19bc
324 821 1% 30% 0.00 786 0xfffff801eb0cc000 vm_object_madvise+0x32d
325 649 1% 31% 0.00 4918 0xfffff80191105300 vm_fault_hold+0x16ee
326 648 1% 32% 0.00 8112 0xfffff80191105300 unlock_and_deallocate+0x2b
327 647 1% 33% 0.00 1261 0xfffff80191105300 vm_object_deallocate+0x683
328 -------------------------------------------------------------------------------
330 .It Example 3 Measuring Hold Times for Stack Traces Containing a Specific Function
332 .Li # lockstat -H -f tcp_input -s 50 -D 10 sleep 1
334 Adaptive mutex hold: 68 events in 1.026 seconds (66 events/sec)
336 -------------------------------------------------------------------------------
337 Count indv cuml rcnt nsec Lock Caller
338 32 47% 47% 0.00 1631 0xfffff800686f50d8 tcp_do_segment+0x284b
340 nsec ------ Time Distribution ------ count Stack
341 1024 |@@@@@@@@@@ 11 tcp_input+0xf54
342 2048 |@@@@@@@@@@@@@ 14 ip_input+0xc8
343 4096 |@@@@@ 6 swi_net+0x192
344 8192 | 1 intr_event_execute_handlers+0x93
348 -------------------------------------------------------------------------------
349 Count indv cuml rcnt nsec Lock Caller
350 29 43% 90% 0.00 4851 0xfffff800686f50d8 sowakeup+0xf8
352 nsec ------ Time Distribution ------ count Stack
353 4096 |@@@@@@@@@@@@@@@ 15 tcp_do_segment+0x2423
354 8192 |@@@@@@@@@@@@ 12 tcp_input+0xf54
355 16384 |@@ 2 ip_input+0xc8
357 intr_event_execute_handlers+0x93
361 -------------------------------------------------------------------------------
372 utility first appeared in
375 Tail-call elimination can affect call sites.
384 the compiler can arrange for
388 with a return address of
394 will appear as though it occurred at
397 The PC in the stack frame in which an interrupt occurs can be bogus because,
398 between function calls, the compiler is free to use the return address register
405 options together, the interrupted PC will usually not appear anywhere in the
406 stack since the interrupt handler is entered asynchronously, not by a function