Merge tag 'powerpc-5.1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[linux-2.6-block.git] / Documentation / trace / histogram.rst
CommitLineData
ea272257
MCC
1================
2Event Histograms
3================
b8df4a36 4
ea272257 5Documentation written by Tom Zanussi
b8df4a36
TZ
6
71. Introduction
8===============
9
10 Histogram triggers are special event triggers that can be used to
11 aggregate trace event data into histograms. For information on
5fb94e9c 12 trace events and event triggers, see Documentation/trace/events.rst.
b8df4a36
TZ
13
14
152. Histogram Trigger Command
16============================
17
18 A histogram trigger command is an event trigger command that
19 aggregates event hits into a hash table keyed on one or more trace
20 event format fields (or stacktrace) and a set of running totals
21 derived from one or more trace event format fields and/or event
22 counts (hitcount).
23
ea272257 24 The format of a hist trigger is as follows::
b8df4a36
TZ
25
26 hist:keys=<field1[,field2,...]>[:values=<field1[,field2,...]>]
27 [:sort=<field1[,field2,...]>][:size=#entries][:pause][:continue]
5032b381 28 [:clear][:name=histname1][:<handler>.<action>] [if <filter>]
b8df4a36
TZ
29
30 When a matching event is hit, an entry is added to a hash table
31 using the key(s) and value(s) named. Keys and values correspond to
32 fields in the event's format description. Values must correspond to
33 numeric fields - on an event hit, the value(s) will be added to a
34 sum kept for that field. The special string 'hitcount' can be used
35 in place of an explicit value field - this is simply a count of
36 event hits. If 'values' isn't specified, an implicit 'hitcount'
37 value will be automatically created and used as the only value.
38 Keys can be any field, or the special string 'stacktrace', which
39 will use the event's kernel stacktrace as the key. The keywords
40 'keys' or 'key' can be used to specify keys, and the keywords
41 'values', 'vals', or 'val' can be used to specify values. Compound
42 keys consisting of up to two fields can be specified by the 'keys'
43 keyword. Hashing a compound key produces a unique entry in the
44 table for each unique combination of component keys, and can be
45 useful for providing more fine-grained summaries of event data.
46 Additionally, sort keys consisting of up to two fields can be
47 specified by the 'sort' keyword. If more than one field is
48 specified, the result will be a 'sort within a sort': the first key
49 is taken to be the primary sort key and the second the secondary
50 key. If a hist trigger is given a name using the 'name' parameter,
51 its histogram data will be shared with other triggers of the same
52 name, and trigger hits will update this common data. Only triggers
53 with 'compatible' fields can be combined in this way; triggers are
54 'compatible' if the fields named in the trigger share the same
55 number and type of fields and those fields also have the same names.
56 Note that any two events always share the compatible 'hitcount' and
57 'stacktrace' fields and can therefore be combined using those
58 fields, however pointless that may be.
59
60 'hist' triggers add a 'hist' file to each event's subdirectory.
61 Reading the 'hist' file for the event will dump the hash table in
62 its entirety to stdout. If there are multiple hist triggers
63 attached to an event, there will be a table for each trigger in the
64 output. The table displayed for a named trigger will be the same as
65 any other instance having the same name. Each printed hash table
66 entry is a simple list of the keys and values comprising the entry;
67 keys are printed first and are delineated by curly braces, and are
68 followed by the set of value fields for the entry. By default,
69 numeric fields are displayed as base-10 integers. This can be
70 modified by appending any of the following modifiers to the field
71 name:
72
ea272257 73 =========== ==========================================
b8df4a36
TZ
74 .hex display a number as a hex value
75 .sym display an address as a symbol
76 .sym-offset display an address as a symbol and offset
77 .syscall display a syscall id as a system call name
78 .execname display a common_pid as a program name
442c9484 79 .log2 display log2 value rather than raw number
860f9f6b 80 .usecs display a common_timestamp in microseconds
ea272257 81 =========== ==========================================
b8df4a36
TZ
82
83 Note that in general the semantics of a given field aren't
84 interpreted when applying a modifier to it, but there are some
85 restrictions to be aware of in this regard:
86
87 - only the 'hex' modifier can be used for values (because values
88 are essentially sums, and the other modifiers don't make sense
89 in that context).
90 - the 'execname' modifier can only be used on a 'common_pid'. The
91 reason for this is that the execname is simply the 'comm' value
92 saved for the 'current' process when an event was triggered,
93 which is the same as the common_pid value saved by the event
94 tracing code. Trying to apply that comm value to other pid
95 values wouldn't be correct, and typically events that care save
96 pid-specific comm fields in the event itself.
97
98 A typical usage scenario would be the following to enable a hist
ea272257 99 trigger, read its current contents, and then turn it off::
b8df4a36 100
ea272257
MCC
101 # echo 'hist:keys=skbaddr.hex:vals=len' > \
102 /sys/kernel/debug/tracing/events/net/netif_rx/trigger
b8df4a36 103
ea272257 104 # cat /sys/kernel/debug/tracing/events/net/netif_rx/hist
b8df4a36 105
ea272257
MCC
106 # echo '!hist:keys=skbaddr.hex:vals=len' > \
107 /sys/kernel/debug/tracing/events/net/netif_rx/trigger
b8df4a36
TZ
108
109 The trigger file itself can be read to show the details of the
110 currently attached hist trigger. This information is also displayed
111 at the top of the 'hist' file when read.
112
113 By default, the size of the hash table is 2048 entries. The 'size'
114 parameter can be used to specify more or fewer than that. The units
115 are in terms of hashtable entries - if a run uses more entries than
116 specified, the results will show the number of 'drops', the number
117 of hits that were ignored. The size should be a power of 2 between
118 128 and 131072 (any non- power-of-2 number specified will be rounded
119 up).
120
121 The 'sort' parameter can be used to specify a value field to sort
122 on. The default if unspecified is 'hitcount' and the default sort
123 order is 'ascending'. To sort in the opposite direction, append
124 .descending' to the sort key.
125
126 The 'pause' parameter can be used to pause an existing hist trigger
127 or to start a hist trigger but not log any events until told to do
128 so. 'continue' or 'cont' can be used to start or restart a paused
129 hist trigger.
130
131 The 'clear' parameter will clear the contents of a running hist
132 trigger and leave its current paused/active state.
133
134 Note that the 'pause', 'cont', and 'clear' parameters should be
135 applied using 'append' shell operator ('>>') if applied to an
136 existing trigger, rather than via the '>' operator, which will cause
137 the trigger to be removed through truncation.
138
139- enable_hist/disable_hist
140
141 The enable_hist and disable_hist triggers can be used to have one
142 event conditionally start and stop another event's already-attached
143 hist trigger. Any number of enable_hist and disable_hist triggers
144 can be attached to a given event, allowing that event to kick off
145 and stop aggregations on a host of other events.
146
ea272257 147 The format is very similar to the enable/disable_event triggers::
b8df4a36
TZ
148
149 enable_hist:<system>:<event>[:count]
150 disable_hist:<system>:<event>[:count]
151
152 Instead of enabling or disabling the tracing of the target event
153 into the trace buffer as the enable/disable_event triggers do, the
154 enable/disable_hist triggers enable or disable the aggregation of
155 the target event into a hash table.
156
157 A typical usage scenario for the enable_hist/disable_hist triggers
158 would be to first set up a paused hist trigger on some event,
159 followed by an enable_hist/disable_hist pair that turns the hist
ea272257 160 aggregation on and off when conditions of interest are hit::
b8df4a36 161
ea272257
MCC
162 # echo 'hist:keys=skbaddr.hex:vals=len:pause' > \
163 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
b8df4a36 164
ea272257
MCC
165 # echo 'enable_hist:net:netif_receive_skb if filename==/usr/bin/wget' > \
166 /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger
b8df4a36 167
ea272257
MCC
168 # echo 'disable_hist:net:netif_receive_skb if comm==wget' > \
169 /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger
b8df4a36
TZ
170
171 The above sets up an initially paused hist trigger which is unpaused
172 and starts aggregating events when a given program is executed, and
173 which stops aggregating when the process exits and the hist trigger
174 is paused again.
175
176 The examples below provide a more concrete illustration of the
177 concepts and typical usage patterns discussed above.
178
ea272257
MCC
179'special' event fields
180------------------------
8b7622bf
TZ
181
182 There are a number of 'special event fields' available for use as
183 keys or values in a hist trigger. These look like and behave as if
184 they were actual event fields, but aren't really part of the event's
185 field definition or format file. They are however available for any
186 event, and can be used anywhere an actual event field could be.
187 They are:
188
ea272257
MCC
189 ====================== ==== =======================================
190 common_timestamp u64 timestamp (from ring buffer) associated
191 with the event, in nanoseconds. May be
192 modified by .usecs to have timestamps
193 interpreted as microseconds.
194 cpu int the cpu on which the event occurred.
195 ====================== ==== =======================================
b8df4a36 196
ea272257
MCC
197Extended error information
198--------------------------
f404da6e
TZ
199
200 For some error conditions encountered when invoking a hist trigger
201 command, extended error information is available via the
202 corresponding event's 'hist' file. Reading the hist file after an
203 error will display more detailed information about what went wrong,
204 if information is available. This extended error information will
205 be available until the next hist trigger command for that event.
206
207 If available for a given error condition, the extended error
ea272257 208 information and usage takes the following form::
f404da6e
TZ
209
210 # echo xxx > /sys/kernel/debug/tracing/events/sched/sched_wakeup/trigger
211 echo: write error: Invalid argument
212
213 # cat /sys/kernel/debug/tracing/events/sched/sched_wakeup/hist
214 ERROR: Couldn't yyy: zzz
215 Last command: xxx
216
b8df4a36
TZ
2176.2 'hist' trigger examples
218---------------------------
219
220 The first set of examples creates aggregations using the kmalloc
221 event. The fields that can be used for the hist trigger are listed
ea272257 222 in the kmalloc event's format file::
b8df4a36
TZ
223
224 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/format
225 name: kmalloc
226 ID: 374
227 format:
228 field:unsigned short common_type; offset:0; size:2; signed:0;
229 field:unsigned char common_flags; offset:2; size:1; signed:0;
230 field:unsigned char common_preempt_count; offset:3; size:1; signed:0;
231 field:int common_pid; offset:4; size:4; signed:1;
232
233 field:unsigned long call_site; offset:8; size:8; signed:0;
234 field:const void * ptr; offset:16; size:8; signed:0;
235 field:size_t bytes_req; offset:24; size:8; signed:0;
236 field:size_t bytes_alloc; offset:32; size:8; signed:0;
237 field:gfp_t gfp_flags; offset:40; size:4; signed:0;
238
239 We'll start by creating a hist trigger that generates a simple table
240 that lists the total number of bytes requested for each function in
ea272257 241 the kernel that made one or more calls to kmalloc::
b8df4a36
TZ
242
243 # echo 'hist:key=call_site:val=bytes_req' > \
244 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
245
246 This tells the tracing system to create a 'hist' trigger using the
247 call_site field of the kmalloc event as the key for the table, which
248 just means that each unique call_site address will have an entry
249 created for it in the table. The 'val=bytes_req' parameter tells
250 the hist trigger that for each unique entry (call_site) in the
251 table, it should keep a running total of the number of bytes
252 requested by that call_site.
253
254 We'll let it run for awhile and then dump the contents of the 'hist'
255 file in the kmalloc event's subdirectory (for readability, a number
ea272257 256 of entries have been omitted)::
b8df4a36
TZ
257
258 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist
259 # trigger info: hist:keys=call_site:vals=bytes_req:sort=hitcount:size=2048 [active]
260
261 { call_site: 18446744072106379007 } hitcount: 1 bytes_req: 176
262 { call_site: 18446744071579557049 } hitcount: 1 bytes_req: 1024
263 { call_site: 18446744071580608289 } hitcount: 1 bytes_req: 16384
264 { call_site: 18446744071581827654 } hitcount: 1 bytes_req: 24
265 { call_site: 18446744071580700980 } hitcount: 1 bytes_req: 8
266 { call_site: 18446744071579359876 } hitcount: 1 bytes_req: 152
267 { call_site: 18446744071580795365 } hitcount: 3 bytes_req: 144
268 { call_site: 18446744071581303129 } hitcount: 3 bytes_req: 144
269 { call_site: 18446744071580713234 } hitcount: 4 bytes_req: 2560
270 { call_site: 18446744071580933750 } hitcount: 4 bytes_req: 736
271 .
272 .
273 .
274 { call_site: 18446744072106047046 } hitcount: 69 bytes_req: 5576
275 { call_site: 18446744071582116407 } hitcount: 73 bytes_req: 2336
276 { call_site: 18446744072106054684 } hitcount: 136 bytes_req: 140504
277 { call_site: 18446744072106224230 } hitcount: 136 bytes_req: 19584
278 { call_site: 18446744072106078074 } hitcount: 153 bytes_req: 2448
279 { call_site: 18446744072106062406 } hitcount: 153 bytes_req: 36720
280 { call_site: 18446744071582507929 } hitcount: 153 bytes_req: 37088
281 { call_site: 18446744072102520590 } hitcount: 273 bytes_req: 10920
282 { call_site: 18446744071582143559 } hitcount: 358 bytes_req: 716
283 { call_site: 18446744072106465852 } hitcount: 417 bytes_req: 56712
284 { call_site: 18446744072102523378 } hitcount: 485 bytes_req: 27160
285 { call_site: 18446744072099568646 } hitcount: 1676 bytes_req: 33520
286
287 Totals:
288 Hits: 4610
289 Entries: 45
290 Dropped: 0
291
292 The output displays a line for each entry, beginning with the key
293 specified in the trigger, followed by the value(s) also specified in
294 the trigger. At the beginning of the output is a line that displays
295 the trigger info, which can also be displayed by reading the
ea272257 296 'trigger' file::
b8df4a36
TZ
297
298 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
299 hist:keys=call_site:vals=bytes_req:sort=hitcount:size=2048 [active]
300
301 At the end of the output are a few lines that display the overall
302 totals for the run. The 'Hits' field shows the total number of
303 times the event trigger was hit, the 'Entries' field shows the total
304 number of used entries in the hash table, and the 'Dropped' field
305 shows the number of hits that were dropped because the number of
306 used entries for the run exceeded the maximum number of entries
307 allowed for the table (normally 0, but if not a hint that you may
308 want to increase the size of the table using the 'size' parameter).
309
310 Notice in the above output that there's an extra field, 'hitcount',
311 which wasn't specified in the trigger. Also notice that in the
312 trigger info output, there's a parameter, 'sort=hitcount', which
313 wasn't specified in the trigger either. The reason for that is that
314 every trigger implicitly keeps a count of the total number of hits
315 attributed to a given entry, called the 'hitcount'. That hitcount
316 information is explicitly displayed in the output, and in the
317 absence of a user-specified sort parameter, is used as the default
318 sort field.
319
320 The value 'hitcount' can be used in place of an explicit value in
321 the 'values' parameter if you don't really need to have any
322 particular field summed and are mainly interested in hit
323 frequencies.
324
325 To turn the hist trigger off, simply call up the trigger in the
ea272257 326 command history and re-execute it with a '!' prepended::
b8df4a36
TZ
327
328 # echo '!hist:key=call_site:val=bytes_req' > \
329 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
330
331 Finally, notice that the call_site as displayed in the output above
332 isn't really very useful. It's an address, but normally addresses
333 are displayed in hex. To have a numeric field displayed as a hex
ea272257 334 value, simply append '.hex' to the field name in the trigger::
b8df4a36
TZ
335
336 # echo 'hist:key=call_site.hex:val=bytes_req' > \
337 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
338
339 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist
340 # trigger info: hist:keys=call_site.hex:vals=bytes_req:sort=hitcount:size=2048 [active]
341
342 { call_site: ffffffffa026b291 } hitcount: 1 bytes_req: 433
343 { call_site: ffffffffa07186ff } hitcount: 1 bytes_req: 176
344 { call_site: ffffffff811ae721 } hitcount: 1 bytes_req: 16384
345 { call_site: ffffffff811c5134 } hitcount: 1 bytes_req: 8
346 { call_site: ffffffffa04a9ebb } hitcount: 1 bytes_req: 511
347 { call_site: ffffffff8122e0a6 } hitcount: 1 bytes_req: 12
348 { call_site: ffffffff8107da84 } hitcount: 1 bytes_req: 152
349 { call_site: ffffffff812d8246 } hitcount: 1 bytes_req: 24
350 { call_site: ffffffff811dc1e5 } hitcount: 3 bytes_req: 144
351 { call_site: ffffffffa02515e8 } hitcount: 3 bytes_req: 648
352 { call_site: ffffffff81258159 } hitcount: 3 bytes_req: 144
353 { call_site: ffffffff811c80f4 } hitcount: 4 bytes_req: 544
354 .
355 .
356 .
357 { call_site: ffffffffa06c7646 } hitcount: 106 bytes_req: 8024
358 { call_site: ffffffffa06cb246 } hitcount: 132 bytes_req: 31680
359 { call_site: ffffffffa06cef7a } hitcount: 132 bytes_req: 2112
360 { call_site: ffffffff8137e399 } hitcount: 132 bytes_req: 23232
361 { call_site: ffffffffa06c941c } hitcount: 185 bytes_req: 171360
362 { call_site: ffffffffa06f2a66 } hitcount: 185 bytes_req: 26640
363 { call_site: ffffffffa036a70e } hitcount: 265 bytes_req: 10600
364 { call_site: ffffffff81325447 } hitcount: 292 bytes_req: 584
365 { call_site: ffffffffa072da3c } hitcount: 446 bytes_req: 60656
366 { call_site: ffffffffa036b1f2 } hitcount: 526 bytes_req: 29456
367 { call_site: ffffffffa0099c06 } hitcount: 1780 bytes_req: 35600
368
369 Totals:
370 Hits: 4775
371 Entries: 46
372 Dropped: 0
373
374 Even that's only marginally more useful - while hex values do look
375 more like addresses, what users are typically more interested in
376 when looking at text addresses are the corresponding symbols
377 instead. To have an address displayed as symbolic value instead,
378 simply append '.sym' or '.sym-offset' to the field name in the
ea272257 379 trigger::
b8df4a36
TZ
380
381 # echo 'hist:key=call_site.sym:val=bytes_req' > \
382 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
383
384 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist
385 # trigger info: hist:keys=call_site.sym:vals=bytes_req:sort=hitcount:size=2048 [active]
386
387 { call_site: [ffffffff810adcb9] syslog_print_all } hitcount: 1 bytes_req: 1024
388 { call_site: [ffffffff8154bc62] usb_control_msg } hitcount: 1 bytes_req: 8
389 { call_site: [ffffffffa00bf6fe] hidraw_send_report [hid] } hitcount: 1 bytes_req: 7
390 { call_site: [ffffffff8154acbe] usb_alloc_urb } hitcount: 1 bytes_req: 192
391 { call_site: [ffffffffa00bf1ca] hidraw_report_event [hid] } hitcount: 1 bytes_req: 7
392 { call_site: [ffffffff811e3a25] __seq_open_private } hitcount: 1 bytes_req: 40
393 { call_site: [ffffffff8109524a] alloc_fair_sched_group } hitcount: 2 bytes_req: 128
394 { call_site: [ffffffff811febd5] fsnotify_alloc_group } hitcount: 2 bytes_req: 528
395 { call_site: [ffffffff81440f58] __tty_buffer_request_room } hitcount: 2 bytes_req: 2624
396 { call_site: [ffffffff81200ba6] inotify_new_group } hitcount: 2 bytes_req: 96
397 { call_site: [ffffffffa05e19af] ieee80211_start_tx_ba_session [mac80211] } hitcount: 2 bytes_req: 464
398 { call_site: [ffffffff81672406] tcp_get_metrics } hitcount: 2 bytes_req: 304
399 { call_site: [ffffffff81097ec2] alloc_rt_sched_group } hitcount: 2 bytes_req: 128
400 { call_site: [ffffffff81089b05] sched_create_group } hitcount: 2 bytes_req: 1424
401 .
402 .
403 .
404 { call_site: [ffffffffa04a580c] intel_crtc_page_flip [i915] } hitcount: 1185 bytes_req: 123240
405 { call_site: [ffffffffa0287592] drm_mode_page_flip_ioctl [drm] } hitcount: 1185 bytes_req: 104280
406 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state [i915] } hitcount: 1402 bytes_req: 190672
407 { call_site: [ffffffff812891ca] ext4_find_extent } hitcount: 1518 bytes_req: 146208
408 { call_site: [ffffffffa029070e] drm_vma_node_allow [drm] } hitcount: 1746 bytes_req: 69840
409 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 2021 bytes_req: 792312
410 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc [drm] } hitcount: 2592 bytes_req: 145152
411 { call_site: [ffffffffa0489a66] intel_ring_begin [i915] } hitcount: 2629 bytes_req: 378576
412 { call_site: [ffffffffa046041c] i915_gem_execbuffer2 [i915] } hitcount: 2629 bytes_req: 3783248
413 { call_site: [ffffffff81325607] apparmor_file_alloc_security } hitcount: 5192 bytes_req: 10384
414 { call_site: [ffffffffa00b7c06] hid_report_raw_event [hid] } hitcount: 5529 bytes_req: 110584
415 { call_site: [ffffffff8131ebf7] aa_alloc_task_context } hitcount: 21943 bytes_req: 702176
416 { call_site: [ffffffff8125847d] ext4_htree_store_dirent } hitcount: 55759 bytes_req: 5074265
417
418 Totals:
419 Hits: 109928
420 Entries: 71
421 Dropped: 0
422
423 Because the default sort key above is 'hitcount', the above shows a
424 the list of call_sites by increasing hitcount, so that at the bottom
425 we see the functions that made the most kmalloc calls during the
426 run. If instead we we wanted to see the top kmalloc callers in
427 terms of the number of bytes requested rather than the number of
428 calls, and we wanted the top caller to appear at the top, we can use
ea272257 429 the 'sort' parameter, along with the 'descending' modifier::
b8df4a36
TZ
430
431 # echo 'hist:key=call_site.sym:val=bytes_req:sort=bytes_req.descending' > \
432 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
433
434 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist
435 # trigger info: hist:keys=call_site.sym:vals=bytes_req:sort=bytes_req.descending:size=2048 [active]
436
437 { call_site: [ffffffffa046041c] i915_gem_execbuffer2 [i915] } hitcount: 2186 bytes_req: 3397464
438 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 1790 bytes_req: 712176
439 { call_site: [ffffffff8125847d] ext4_htree_store_dirent } hitcount: 8132 bytes_req: 513135
440 { call_site: [ffffffff811e2a1b] seq_buf_alloc } hitcount: 106 bytes_req: 440128
441 { call_site: [ffffffffa0489a66] intel_ring_begin [i915] } hitcount: 2186 bytes_req: 314784
442 { call_site: [ffffffff812891ca] ext4_find_extent } hitcount: 2174 bytes_req: 208992
443 { call_site: [ffffffff811ae8e1] __kmalloc } hitcount: 8 bytes_req: 131072
444 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state [i915] } hitcount: 859 bytes_req: 116824
445 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc [drm] } hitcount: 1834 bytes_req: 102704
446 { call_site: [ffffffffa04a580c] intel_crtc_page_flip [i915] } hitcount: 972 bytes_req: 101088
447 { call_site: [ffffffffa0287592] drm_mode_page_flip_ioctl [drm] } hitcount: 972 bytes_req: 85536
448 { call_site: [ffffffffa00b7c06] hid_report_raw_event [hid] } hitcount: 3333 bytes_req: 66664
449 { call_site: [ffffffff8137e559] sg_kmalloc } hitcount: 209 bytes_req: 61632
450 .
451 .
452 .
453 { call_site: [ffffffff81095225] alloc_fair_sched_group } hitcount: 2 bytes_req: 128
454 { call_site: [ffffffff81097ec2] alloc_rt_sched_group } hitcount: 2 bytes_req: 128
455 { call_site: [ffffffff812d8406] copy_semundo } hitcount: 2 bytes_req: 48
456 { call_site: [ffffffff81200ba6] inotify_new_group } hitcount: 1 bytes_req: 48
457 { call_site: [ffffffffa027121a] drm_getmagic [drm] } hitcount: 1 bytes_req: 48
458 { call_site: [ffffffff811e3a25] __seq_open_private } hitcount: 1 bytes_req: 40
459 { call_site: [ffffffff811c52f4] bprm_change_interp } hitcount: 2 bytes_req: 16
460 { call_site: [ffffffff8154bc62] usb_control_msg } hitcount: 1 bytes_req: 8
461 { call_site: [ffffffffa00bf1ca] hidraw_report_event [hid] } hitcount: 1 bytes_req: 7
462 { call_site: [ffffffffa00bf6fe] hidraw_send_report [hid] } hitcount: 1 bytes_req: 7
463
464 Totals:
465 Hits: 32133
466 Entries: 81
467 Dropped: 0
468
469 To display the offset and size information in addition to the symbol
ea272257 470 name, just use 'sym-offset' instead::
b8df4a36
TZ
471
472 # echo 'hist:key=call_site.sym-offset:val=bytes_req:sort=bytes_req.descending' > \
473 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
474
475 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist
476 # trigger info: hist:keys=call_site.sym-offset:vals=bytes_req:sort=bytes_req.descending:size=2048 [active]
477
478 { call_site: [ffffffffa046041c] i915_gem_execbuffer2+0x6c/0x2c0 [i915] } hitcount: 4569 bytes_req: 3163720
479 { call_site: [ffffffffa0489a66] intel_ring_begin+0xc6/0x1f0 [i915] } hitcount: 4569 bytes_req: 657936
480 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23+0x694/0x1020 [i915] } hitcount: 1519 bytes_req: 472936
481 { call_site: [ffffffffa045e646] i915_gem_do_execbuffer.isra.23+0x516/0x1020 [i915] } hitcount: 3050 bytes_req: 211832
482 { call_site: [ffffffff811e2a1b] seq_buf_alloc+0x1b/0x50 } hitcount: 34 bytes_req: 148384
483 { call_site: [ffffffffa04a580c] intel_crtc_page_flip+0xbc/0x870 [i915] } hitcount: 1385 bytes_req: 144040
484 { call_site: [ffffffff811ae8e1] __kmalloc+0x191/0x1b0 } hitcount: 8 bytes_req: 131072
485 { call_site: [ffffffffa0287592] drm_mode_page_flip_ioctl+0x282/0x360 [drm] } hitcount: 1385 bytes_req: 121880
486 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc+0x32/0x100 [drm] } hitcount: 1848 bytes_req: 103488
487 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state+0x2c/0xa0 [i915] } hitcount: 461 bytes_req: 62696
488 { call_site: [ffffffffa029070e] drm_vma_node_allow+0x2e/0xd0 [drm] } hitcount: 1541 bytes_req: 61640
489 { call_site: [ffffffff815f8d7b] sk_prot_alloc+0xcb/0x1b0 } hitcount: 57 bytes_req: 57456
490 .
491 .
492 .
493 { call_site: [ffffffff8109524a] alloc_fair_sched_group+0x5a/0x1a0 } hitcount: 2 bytes_req: 128
494 { call_site: [ffffffffa027b921] drm_vm_open_locked+0x31/0xa0 [drm] } hitcount: 3 bytes_req: 96
495 { call_site: [ffffffff8122e266] proc_self_follow_link+0x76/0xb0 } hitcount: 8 bytes_req: 96
496 { call_site: [ffffffff81213e80] load_elf_binary+0x240/0x1650 } hitcount: 3 bytes_req: 84
497 { call_site: [ffffffff8154bc62] usb_control_msg+0x42/0x110 } hitcount: 1 bytes_req: 8
498 { call_site: [ffffffffa00bf6fe] hidraw_send_report+0x7e/0x1a0 [hid] } hitcount: 1 bytes_req: 7
499 { call_site: [ffffffffa00bf1ca] hidraw_report_event+0x8a/0x120 [hid] } hitcount: 1 bytes_req: 7
500
501 Totals:
502 Hits: 26098
503 Entries: 64
504 Dropped: 0
505
506 We can also add multiple fields to the 'values' parameter. For
507 example, we might want to see the total number of bytes allocated
508 alongside bytes requested, and display the result sorted by bytes
ea272257 509 allocated in a descending order::
b8df4a36
TZ
510
511 # echo 'hist:keys=call_site.sym:values=bytes_req,bytes_alloc:sort=bytes_alloc.descending' > \
512 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
513
514 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist
515 # trigger info: hist:keys=call_site.sym:vals=bytes_req,bytes_alloc:sort=bytes_alloc.descending:size=2048 [active]
516
517 { call_site: [ffffffffa046041c] i915_gem_execbuffer2 [i915] } hitcount: 7403 bytes_req: 4084360 bytes_alloc: 5958016
518 { call_site: [ffffffff811e2a1b] seq_buf_alloc } hitcount: 541 bytes_req: 2213968 bytes_alloc: 2228224
519 { call_site: [ffffffffa0489a66] intel_ring_begin [i915] } hitcount: 7404 bytes_req: 1066176 bytes_alloc: 1421568
520 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 1565 bytes_req: 557368 bytes_alloc: 1037760
521 { call_site: [ffffffff8125847d] ext4_htree_store_dirent } hitcount: 9557 bytes_req: 595778 bytes_alloc: 695744
522 { call_site: [ffffffffa045e646] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 5839 bytes_req: 430680 bytes_alloc: 470400
523 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state [i915] } hitcount: 2388 bytes_req: 324768 bytes_alloc: 458496
524 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc [drm] } hitcount: 3911 bytes_req: 219016 bytes_alloc: 250304
525 { call_site: [ffffffff815f8d7b] sk_prot_alloc } hitcount: 235 bytes_req: 236880 bytes_alloc: 240640
526 { call_site: [ffffffff8137e559] sg_kmalloc } hitcount: 557 bytes_req: 169024 bytes_alloc: 221760
527 { call_site: [ffffffffa00b7c06] hid_report_raw_event [hid] } hitcount: 9378 bytes_req: 187548 bytes_alloc: 206312
528 { call_site: [ffffffffa04a580c] intel_crtc_page_flip [i915] } hitcount: 1519 bytes_req: 157976 bytes_alloc: 194432
529 .
530 .
531 .
532 { call_site: [ffffffff8109bd3b] sched_autogroup_create_attach } hitcount: 2 bytes_req: 144 bytes_alloc: 192
533 { call_site: [ffffffff81097ee8] alloc_rt_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128
534 { call_site: [ffffffff8109524a] alloc_fair_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128
535 { call_site: [ffffffff81095225] alloc_fair_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128
536 { call_site: [ffffffff81097ec2] alloc_rt_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128
537 { call_site: [ffffffff81213e80] load_elf_binary } hitcount: 3 bytes_req: 84 bytes_alloc: 96
538 { call_site: [ffffffff81079a2e] kthread_create_on_node } hitcount: 1 bytes_req: 56 bytes_alloc: 64
539 { call_site: [ffffffffa00bf6fe] hidraw_send_report [hid] } hitcount: 1 bytes_req: 7 bytes_alloc: 8
540 { call_site: [ffffffff8154bc62] usb_control_msg } hitcount: 1 bytes_req: 8 bytes_alloc: 8
541 { call_site: [ffffffffa00bf1ca] hidraw_report_event [hid] } hitcount: 1 bytes_req: 7 bytes_alloc: 8
542
543 Totals:
544 Hits: 66598
545 Entries: 65
546 Dropped: 0
547
548 Finally, to finish off our kmalloc example, instead of simply having
549 the hist trigger display symbolic call_sites, we can have the hist
550 trigger additionally display the complete set of kernel stack traces
551 that led to each call_site. To do that, we simply use the special
ea272257 552 value 'stacktrace' for the key parameter::
b8df4a36
TZ
553
554 # echo 'hist:keys=stacktrace:values=bytes_req,bytes_alloc:sort=bytes_alloc' > \
555 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger
556
557 The above trigger will use the kernel stack trace in effect when an
558 event is triggered as the key for the hash table. This allows the
559 enumeration of every kernel callpath that led up to a particular
560 event, along with a running total of any of the event fields for
561 that event. Here we tally bytes requested and bytes allocated for
562 every callpath in the system that led up to a kmalloc (in this case
ea272257 563 every callpath to a kmalloc for a kernel compile)::
b8df4a36
TZ
564
565 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist
566 # trigger info: hist:keys=stacktrace:vals=bytes_req,bytes_alloc:sort=bytes_alloc:size=2048 [active]
567
568 { stacktrace:
569 __kmalloc_track_caller+0x10b/0x1a0
570 kmemdup+0x20/0x50
571 hidraw_report_event+0x8a/0x120 [hid]
572 hid_report_raw_event+0x3ea/0x440 [hid]
573 hid_input_report+0x112/0x190 [hid]
574 hid_irq_in+0xc2/0x260 [usbhid]
575 __usb_hcd_giveback_urb+0x72/0x120
576 usb_giveback_urb_bh+0x9e/0xe0
577 tasklet_hi_action+0xf8/0x100
578 __do_softirq+0x114/0x2c0
579 irq_exit+0xa5/0xb0
580 do_IRQ+0x5a/0xf0
581 ret_from_intr+0x0/0x30
582 cpuidle_enter+0x17/0x20
583 cpu_startup_entry+0x315/0x3e0
584 rest_init+0x7c/0x80
585 } hitcount: 3 bytes_req: 21 bytes_alloc: 24
586 { stacktrace:
587 __kmalloc_track_caller+0x10b/0x1a0
588 kmemdup+0x20/0x50
589 hidraw_report_event+0x8a/0x120 [hid]
590 hid_report_raw_event+0x3ea/0x440 [hid]
591 hid_input_report+0x112/0x190 [hid]
592 hid_irq_in+0xc2/0x260 [usbhid]
593 __usb_hcd_giveback_urb+0x72/0x120
594 usb_giveback_urb_bh+0x9e/0xe0
595 tasklet_hi_action+0xf8/0x100
596 __do_softirq+0x114/0x2c0
597 irq_exit+0xa5/0xb0
598 do_IRQ+0x5a/0xf0
599 ret_from_intr+0x0/0x30
600 } hitcount: 3 bytes_req: 21 bytes_alloc: 24
601 { stacktrace:
602 kmem_cache_alloc_trace+0xeb/0x150
603 aa_alloc_task_context+0x27/0x40
604 apparmor_cred_prepare+0x1f/0x50
605 security_prepare_creds+0x16/0x20
606 prepare_creds+0xdf/0x1a0
607 SyS_capset+0xb5/0x200
608 system_call_fastpath+0x12/0x6a
609 } hitcount: 1 bytes_req: 32 bytes_alloc: 32
610 .
611 .
612 .
613 { stacktrace:
614 __kmalloc+0x11b/0x1b0
615 i915_gem_execbuffer2+0x6c/0x2c0 [i915]
616 drm_ioctl+0x349/0x670 [drm]
617 do_vfs_ioctl+0x2f0/0x4f0
618 SyS_ioctl+0x81/0xa0
619 system_call_fastpath+0x12/0x6a
620 } hitcount: 17726 bytes_req: 13944120 bytes_alloc: 19593808
621 { stacktrace:
622 __kmalloc+0x11b/0x1b0
623 load_elf_phdrs+0x76/0xa0
624 load_elf_binary+0x102/0x1650
625 search_binary_handler+0x97/0x1d0
626 do_execveat_common.isra.34+0x551/0x6e0
627 SyS_execve+0x3a/0x50
628 return_from_execve+0x0/0x23
629 } hitcount: 33348 bytes_req: 17152128 bytes_alloc: 20226048
630 { stacktrace:
631 kmem_cache_alloc_trace+0xeb/0x150
632 apparmor_file_alloc_security+0x27/0x40
633 security_file_alloc+0x16/0x20
634 get_empty_filp+0x93/0x1c0
635 path_openat+0x31/0x5f0
636 do_filp_open+0x3a/0x90
637 do_sys_open+0x128/0x220
638 SyS_open+0x1e/0x20
639 system_call_fastpath+0x12/0x6a
640 } hitcount: 4766422 bytes_req: 9532844 bytes_alloc: 38131376
641 { stacktrace:
642 __kmalloc+0x11b/0x1b0
643 seq_buf_alloc+0x1b/0x50
644 seq_read+0x2cc/0x370
645 proc_reg_read+0x3d/0x80
646 __vfs_read+0x28/0xe0
647 vfs_read+0x86/0x140
648 SyS_read+0x46/0xb0
649 system_call_fastpath+0x12/0x6a
650 } hitcount: 19133 bytes_req: 78368768 bytes_alloc: 78368768
651
652 Totals:
653 Hits: 6085872
654 Entries: 253
655 Dropped: 0
656
657 If you key a hist trigger on common_pid, in order for example to
658 gather and display sorted totals for each process, you can use the
659 special .execname modifier to display the executable names for the
660 processes in the table rather than raw pids. The example below
ea272257 661 keeps a per-process sum of total bytes read::
b8df4a36
TZ
662
663 # echo 'hist:key=common_pid.execname:val=count:sort=count.descending' > \
664 /sys/kernel/debug/tracing/events/syscalls/sys_enter_read/trigger
665
666 # cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_read/hist
667 # trigger info: hist:keys=common_pid.execname:vals=count:sort=count.descending:size=2048 [active]
668
669 { common_pid: gnome-terminal [ 3196] } hitcount: 280 count: 1093512
670 { common_pid: Xorg [ 1309] } hitcount: 525 count: 256640
671 { common_pid: compiz [ 2889] } hitcount: 59 count: 254400
672 { common_pid: bash [ 8710] } hitcount: 3 count: 66369
673 { common_pid: dbus-daemon-lau [ 8703] } hitcount: 49 count: 47739
674 { common_pid: irqbalance [ 1252] } hitcount: 27 count: 27648
675 { common_pid: 01ifupdown [ 8705] } hitcount: 3 count: 17216
676 { common_pid: dbus-daemon [ 772] } hitcount: 10 count: 12396
677 { common_pid: Socket Thread [ 8342] } hitcount: 11 count: 11264
678 { common_pid: nm-dhcp-client. [ 8701] } hitcount: 6 count: 7424
679 { common_pid: gmain [ 1315] } hitcount: 18 count: 6336
680 .
681 .
682 .
683 { common_pid: postgres [ 1892] } hitcount: 2 count: 32
684 { common_pid: postgres [ 1891] } hitcount: 2 count: 32
685 { common_pid: gmain [ 8704] } hitcount: 2 count: 32
686 { common_pid: upstart-dbus-br [ 2740] } hitcount: 21 count: 21
687 { common_pid: nm-dispatcher.a [ 8696] } hitcount: 1 count: 16
688 { common_pid: indicator-datet [ 2904] } hitcount: 1 count: 16
689 { common_pid: gdbus [ 2998] } hitcount: 1 count: 16
690 { common_pid: rtkit-daemon [ 2052] } hitcount: 1 count: 8
691 { common_pid: init [ 1] } hitcount: 2 count: 2
692
693 Totals:
694 Hits: 2116
695 Entries: 51
696 Dropped: 0
697
698 Similarly, if you key a hist trigger on syscall id, for example to
699 gather and display a list of systemwide syscall hits, you can use
700 the special .syscall modifier to display the syscall names rather
701 than raw ids. The example below keeps a running total of syscall
ea272257 702 counts for the system during the run::
b8df4a36
TZ
703
704 # echo 'hist:key=id.syscall:val=hitcount' > \
705 /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/trigger
706
707 # cat /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/hist
708 # trigger info: hist:keys=id.syscall:vals=hitcount:sort=hitcount:size=2048 [active]
709
710 { id: sys_fsync [ 74] } hitcount: 1
711 { id: sys_newuname [ 63] } hitcount: 1
712 { id: sys_prctl [157] } hitcount: 1
713 { id: sys_statfs [137] } hitcount: 1
714 { id: sys_symlink [ 88] } hitcount: 1
715 { id: sys_sendmmsg [307] } hitcount: 1
716 { id: sys_semctl [ 66] } hitcount: 1
717 { id: sys_readlink [ 89] } hitcount: 3
718 { id: sys_bind [ 49] } hitcount: 3
719 { id: sys_getsockname [ 51] } hitcount: 3
720 { id: sys_unlink [ 87] } hitcount: 3
721 { id: sys_rename [ 82] } hitcount: 4
722 { id: unknown_syscall [ 58] } hitcount: 4
723 { id: sys_connect [ 42] } hitcount: 4
724 { id: sys_getpid [ 39] } hitcount: 4
725 .
726 .
727 .
728 { id: sys_rt_sigprocmask [ 14] } hitcount: 952
729 { id: sys_futex [202] } hitcount: 1534
730 { id: sys_write [ 1] } hitcount: 2689
731 { id: sys_setitimer [ 38] } hitcount: 2797
732 { id: sys_read [ 0] } hitcount: 3202
733 { id: sys_select [ 23] } hitcount: 3773
734 { id: sys_writev [ 20] } hitcount: 4531
735 { id: sys_poll [ 7] } hitcount: 8314
736 { id: sys_recvmsg [ 47] } hitcount: 13738
737 { id: sys_ioctl [ 16] } hitcount: 21843
738
739 Totals:
740 Hits: 67612
741 Entries: 72
742 Dropped: 0
743
ea272257
MCC
744 The syscall counts above provide a rough overall picture of system
745 call activity on the system; we can see for example that the most
746 popular system call on this system was the 'sys_ioctl' system call.
b8df4a36 747
ea272257
MCC
748 We can use 'compound' keys to refine that number and provide some
749 further insight as to which processes exactly contribute to the
750 overall ioctl count.
b8df4a36 751
ea272257
MCC
752 The command below keeps a hitcount for every unique combination of
753 system call id and pid - the end result is essentially a table
754 that keeps a per-pid sum of system call hits. The results are
755 sorted using the system call id as the primary key, and the
756 hitcount sum as the secondary key::
b8df4a36
TZ
757
758 # echo 'hist:key=id.syscall,common_pid.execname:val=hitcount:sort=id,hitcount' > \
759 /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/trigger
760
761 # cat /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/hist
762 # trigger info: hist:keys=id.syscall,common_pid.execname:vals=hitcount:sort=id.syscall,hitcount:size=2048 [active]
763
764 { id: sys_read [ 0], common_pid: rtkit-daemon [ 1877] } hitcount: 1
765 { id: sys_read [ 0], common_pid: gdbus [ 2976] } hitcount: 1
766 { id: sys_read [ 0], common_pid: console-kit-dae [ 3400] } hitcount: 1
767 { id: sys_read [ 0], common_pid: postgres [ 1865] } hitcount: 1
768 { id: sys_read [ 0], common_pid: deja-dup-monito [ 3543] } hitcount: 2
769 { id: sys_read [ 0], common_pid: NetworkManager [ 890] } hitcount: 2
770 { id: sys_read [ 0], common_pid: evolution-calen [ 3048] } hitcount: 2
771 { id: sys_read [ 0], common_pid: postgres [ 1864] } hitcount: 2
772 { id: sys_read [ 0], common_pid: nm-applet [ 3022] } hitcount: 2
773 { id: sys_read [ 0], common_pid: whoopsie [ 1212] } hitcount: 2
774 .
775 .
776 .
777 { id: sys_ioctl [ 16], common_pid: bash [ 8479] } hitcount: 1
778 { id: sys_ioctl [ 16], common_pid: bash [ 3472] } hitcount: 12
779 { id: sys_ioctl [ 16], common_pid: gnome-terminal [ 3199] } hitcount: 16
780 { id: sys_ioctl [ 16], common_pid: Xorg [ 1267] } hitcount: 1808
781 { id: sys_ioctl [ 16], common_pid: compiz [ 2994] } hitcount: 5580
782 .
783 .
784 .
785 { id: sys_waitid [247], common_pid: upstart-dbus-br [ 2690] } hitcount: 3
786 { id: sys_waitid [247], common_pid: upstart-dbus-br [ 2688] } hitcount: 16
787 { id: sys_inotify_add_watch [254], common_pid: gmain [ 975] } hitcount: 2
788 { id: sys_inotify_add_watch [254], common_pid: gmain [ 3204] } hitcount: 4
789 { id: sys_inotify_add_watch [254], common_pid: gmain [ 2888] } hitcount: 4
790 { id: sys_inotify_add_watch [254], common_pid: gmain [ 3003] } hitcount: 4
791 { id: sys_inotify_add_watch [254], common_pid: gmain [ 2873] } hitcount: 4
792 { id: sys_inotify_add_watch [254], common_pid: gmain [ 3196] } hitcount: 6
793 { id: sys_openat [257], common_pid: java [ 2623] } hitcount: 2
794 { id: sys_eventfd2 [290], common_pid: ibus-ui-gtk3 [ 2760] } hitcount: 4
795 { id: sys_eventfd2 [290], common_pid: compiz [ 2994] } hitcount: 6
796
797 Totals:
798 Hits: 31536
799 Entries: 323
800 Dropped: 0
801
ea272257
MCC
802 The above list does give us a breakdown of the ioctl syscall by
803 pid, but it also gives us quite a bit more than that, which we
804 don't really care about at the moment. Since we know the syscall
805 id for sys_ioctl (16, displayed next to the sys_ioctl name), we
806 can use that to filter out all the other syscalls::
b8df4a36
TZ
807
808 # echo 'hist:key=id.syscall,common_pid.execname:val=hitcount:sort=id,hitcount if id == 16' > \
809 /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/trigger
810
811 # cat /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/hist
812 # trigger info: hist:keys=id.syscall,common_pid.execname:vals=hitcount:sort=id.syscall,hitcount:size=2048 if id == 16 [active]
813
814 { id: sys_ioctl [ 16], common_pid: gmain [ 2769] } hitcount: 1
815 { id: sys_ioctl [ 16], common_pid: evolution-addre [ 8571] } hitcount: 1
816 { id: sys_ioctl [ 16], common_pid: gmain [ 3003] } hitcount: 1
817 { id: sys_ioctl [ 16], common_pid: gmain [ 2781] } hitcount: 1
818 { id: sys_ioctl [ 16], common_pid: gmain [ 2829] } hitcount: 1
819 { id: sys_ioctl [ 16], common_pid: bash [ 8726] } hitcount: 1
820 { id: sys_ioctl [ 16], common_pid: bash [ 8508] } hitcount: 1
821 { id: sys_ioctl [ 16], common_pid: gmain [ 2970] } hitcount: 1
822 { id: sys_ioctl [ 16], common_pid: gmain [ 2768] } hitcount: 1
823 .
824 .
825 .
826 { id: sys_ioctl [ 16], common_pid: pool [ 8559] } hitcount: 45
827 { id: sys_ioctl [ 16], common_pid: pool [ 8555] } hitcount: 48
828 { id: sys_ioctl [ 16], common_pid: pool [ 8551] } hitcount: 48
829 { id: sys_ioctl [ 16], common_pid: avahi-daemon [ 896] } hitcount: 66
830 { id: sys_ioctl [ 16], common_pid: Xorg [ 1267] } hitcount: 26674
831 { id: sys_ioctl [ 16], common_pid: compiz [ 2994] } hitcount: 73443
832
833 Totals:
834 Hits: 101162
835 Entries: 103
836 Dropped: 0
837
ea272257
MCC
838 The above output shows that 'compiz' and 'Xorg' are far and away
839 the heaviest ioctl callers (which might lead to questions about
840 whether they really need to be making all those calls and to
841 possible avenues for further investigation.)
b8df4a36 842
ea272257
MCC
843 The compound key examples used a key and a sum value (hitcount) to
844 sort the output, but we can just as easily use two keys instead.
845 Here's an example where we use a compound key composed of the the
846 common_pid and size event fields. Sorting with pid as the primary
847 key and 'size' as the secondary key allows us to display an
848 ordered summary of the recvfrom sizes, with counts, received by
849 each process::
b8df4a36
TZ
850
851 # echo 'hist:key=common_pid.execname,size:val=hitcount:sort=common_pid,size' > \
852 /sys/kernel/debug/tracing/events/syscalls/sys_enter_recvfrom/trigger
853
854 # cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_recvfrom/hist
855 # trigger info: hist:keys=common_pid.execname,size:vals=hitcount:sort=common_pid.execname,size:size=2048 [active]
856
857 { common_pid: smbd [ 784], size: 4 } hitcount: 1
858 { common_pid: dnsmasq [ 1412], size: 4096 } hitcount: 672
859 { common_pid: postgres [ 1796], size: 1000 } hitcount: 6
860 { common_pid: postgres [ 1867], size: 1000 } hitcount: 10
861 { common_pid: bamfdaemon [ 2787], size: 28 } hitcount: 2
862 { common_pid: bamfdaemon [ 2787], size: 14360 } hitcount: 1
863 { common_pid: compiz [ 2994], size: 8 } hitcount: 1
864 { common_pid: compiz [ 2994], size: 20 } hitcount: 11
865 { common_pid: gnome-terminal [ 3199], size: 4 } hitcount: 2
866 { common_pid: firefox [ 8817], size: 4 } hitcount: 1
867 { common_pid: firefox [ 8817], size: 8 } hitcount: 5
868 { common_pid: firefox [ 8817], size: 588 } hitcount: 2
869 { common_pid: firefox [ 8817], size: 628 } hitcount: 1
870 { common_pid: firefox [ 8817], size: 6944 } hitcount: 1
871 { common_pid: firefox [ 8817], size: 408880 } hitcount: 2
872 { common_pid: firefox [ 8822], size: 8 } hitcount: 2
873 { common_pid: firefox [ 8822], size: 160 } hitcount: 2
874 { common_pid: firefox [ 8822], size: 320 } hitcount: 2
875 { common_pid: firefox [ 8822], size: 352 } hitcount: 1
876 .
877 .
878 .
879 { common_pid: pool [ 8923], size: 1960 } hitcount: 10
880 { common_pid: pool [ 8923], size: 2048 } hitcount: 10
881 { common_pid: pool [ 8924], size: 1960 } hitcount: 10
882 { common_pid: pool [ 8924], size: 2048 } hitcount: 10
883 { common_pid: pool [ 8928], size: 1964 } hitcount: 4
884 { common_pid: pool [ 8928], size: 1965 } hitcount: 2
885 { common_pid: pool [ 8928], size: 2048 } hitcount: 6
886 { common_pid: pool [ 8929], size: 1982 } hitcount: 1
887 { common_pid: pool [ 8929], size: 2048 } hitcount: 1
888
889 Totals:
890 Hits: 2016
891 Entries: 224
892 Dropped: 0
893
894 The above example also illustrates the fact that although a compound
895 key is treated as a single entity for hashing purposes, the sub-keys
896 it's composed of can be accessed independently.
897
898 The next example uses a string field as the hash key and
899 demonstrates how you can manually pause and continue a hist trigger.
900 In this example, we'll aggregate fork counts and don't expect a
901 large number of entries in the hash table, so we'll drop it to a
ea272257 902 much smaller number, say 256::
b8df4a36
TZ
903
904 # echo 'hist:key=child_comm:val=hitcount:size=256' > \
905 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger
906
907 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist
908 # trigger info: hist:keys=child_comm:vals=hitcount:sort=hitcount:size=256 [active]
909
910 { child_comm: dconf worker } hitcount: 1
911 { child_comm: ibus-daemon } hitcount: 1
912 { child_comm: whoopsie } hitcount: 1
913 { child_comm: smbd } hitcount: 1
914 { child_comm: gdbus } hitcount: 1
915 { child_comm: kthreadd } hitcount: 1
916 { child_comm: dconf worker } hitcount: 1
917 { child_comm: evolution-alarm } hitcount: 2
918 { child_comm: Socket Thread } hitcount: 2
919 { child_comm: postgres } hitcount: 2
920 { child_comm: bash } hitcount: 3
921 { child_comm: compiz } hitcount: 3
922 { child_comm: evolution-sourc } hitcount: 4
923 { child_comm: dhclient } hitcount: 4
924 { child_comm: pool } hitcount: 5
925 { child_comm: nm-dispatcher.a } hitcount: 8
926 { child_comm: firefox } hitcount: 8
927 { child_comm: dbus-daemon } hitcount: 8
928 { child_comm: glib-pacrunner } hitcount: 10
929 { child_comm: evolution } hitcount: 23
930
931 Totals:
932 Hits: 89
933 Entries: 20
934 Dropped: 0
935
936 If we want to pause the hist trigger, we can simply append :pause to
937 the command that started the trigger. Notice that the trigger info
ea272257 938 displays as [paused]::
b8df4a36
TZ
939
940 # echo 'hist:key=child_comm:val=hitcount:size=256:pause' >> \
941 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger
942
943 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist
944 # trigger info: hist:keys=child_comm:vals=hitcount:sort=hitcount:size=256 [paused]
945
946 { child_comm: dconf worker } hitcount: 1
947 { child_comm: kthreadd } hitcount: 1
948 { child_comm: dconf worker } hitcount: 1
949 { child_comm: gdbus } hitcount: 1
950 { child_comm: ibus-daemon } hitcount: 1
951 { child_comm: Socket Thread } hitcount: 2
952 { child_comm: evolution-alarm } hitcount: 2
953 { child_comm: smbd } hitcount: 2
954 { child_comm: bash } hitcount: 3
955 { child_comm: whoopsie } hitcount: 3
956 { child_comm: compiz } hitcount: 3
957 { child_comm: evolution-sourc } hitcount: 4
958 { child_comm: pool } hitcount: 5
959 { child_comm: postgres } hitcount: 6
960 { child_comm: firefox } hitcount: 8
961 { child_comm: dhclient } hitcount: 10
962 { child_comm: emacs } hitcount: 12
963 { child_comm: dbus-daemon } hitcount: 20
964 { child_comm: nm-dispatcher.a } hitcount: 20
965 { child_comm: evolution } hitcount: 35
966 { child_comm: glib-pacrunner } hitcount: 59
967
968 Totals:
969 Hits: 199
970 Entries: 21
971 Dropped: 0
972
973 To manually continue having the trigger aggregate events, append
974 :cont instead. Notice that the trigger info displays as [active]
ea272257 975 again, and the data has changed::
b8df4a36
TZ
976
977 # echo 'hist:key=child_comm:val=hitcount:size=256:cont' >> \
978 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger
979
980 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist
981 # trigger info: hist:keys=child_comm:vals=hitcount:sort=hitcount:size=256 [active]
982
983 { child_comm: dconf worker } hitcount: 1
984 { child_comm: dconf worker } hitcount: 1
985 { child_comm: kthreadd } hitcount: 1
986 { child_comm: gdbus } hitcount: 1
987 { child_comm: ibus-daemon } hitcount: 1
988 { child_comm: Socket Thread } hitcount: 2
989 { child_comm: evolution-alarm } hitcount: 2
990 { child_comm: smbd } hitcount: 2
991 { child_comm: whoopsie } hitcount: 3
992 { child_comm: compiz } hitcount: 3
993 { child_comm: evolution-sourc } hitcount: 4
994 { child_comm: bash } hitcount: 5
995 { child_comm: pool } hitcount: 5
996 { child_comm: postgres } hitcount: 6
997 { child_comm: firefox } hitcount: 8
998 { child_comm: dhclient } hitcount: 11
999 { child_comm: emacs } hitcount: 12
1000 { child_comm: dbus-daemon } hitcount: 22
1001 { child_comm: nm-dispatcher.a } hitcount: 22
1002 { child_comm: evolution } hitcount: 35
1003 { child_comm: glib-pacrunner } hitcount: 59
1004
1005 Totals:
1006 Hits: 206
1007 Entries: 21
1008 Dropped: 0
1009
1010 The previous example showed how to start and stop a hist trigger by
1011 appending 'pause' and 'continue' to the hist trigger command. A
1012 hist trigger can also be started in a paused state by initially
1013 starting the trigger with ':pause' appended. This allows you to
1014 start the trigger only when you're ready to start collecting data
1015 and not before. For example, you could start the trigger in a
1016 paused state, then unpause it and do something you want to measure,
1017 then pause the trigger again when done.
1018
1019 Of course, doing this manually can be difficult and error-prone, but
1020 it is possible to automatically start and stop a hist trigger based
1021 on some condition, via the enable_hist and disable_hist triggers.
1022
1023 For example, suppose we wanted to take a look at the relative
1024 weights in terms of skb length for each callpath that leads to a
1025 netif_receieve_skb event when downloading a decent-sized file using
1026 wget.
1027
1028 First we set up an initially paused stacktrace trigger on the
ea272257 1029 netif_receive_skb event::
b8df4a36
TZ
1030
1031 # echo 'hist:key=stacktrace:vals=len:pause' > \
1032 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1033
1034 Next, we set up an 'enable_hist' trigger on the sched_process_exec
1035 event, with an 'if filename==/usr/bin/wget' filter. The effect of
1036 this new trigger is that it will 'unpause' the hist trigger we just
1037 set up on netif_receive_skb if and only if it sees a
1038 sched_process_exec event with a filename of '/usr/bin/wget'. When
1039 that happens, all netif_receive_skb events are aggregated into a
ea272257 1040 hash table keyed on stacktrace::
b8df4a36
TZ
1041
1042 # echo 'enable_hist:net:netif_receive_skb if filename==/usr/bin/wget' > \
1043 /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger
1044
1045 The aggregation continues until the netif_receive_skb is paused
1046 again, which is what the following disable_hist event does by
1047 creating a similar setup on the sched_process_exit event, using the
ea272257 1048 filter 'comm==wget'::
b8df4a36
TZ
1049
1050 # echo 'disable_hist:net:netif_receive_skb if comm==wget' > \
1051 /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger
1052
1053 Whenever a process exits and the comm field of the disable_hist
1054 trigger filter matches 'comm==wget', the netif_receive_skb hist
1055 trigger is disabled.
1056
1057 The overall effect is that netif_receive_skb events are aggregated
1058 into the hash table for only the duration of the wget. Executing a
1059 wget command and then listing the 'hist' file will display the
ea272257 1060 output generated by the wget command::
b8df4a36
TZ
1061
1062 $ wget https://www.kernel.org/pub/linux/kernel/v3.x/patch-3.19.xz
1063
1064 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist
1065 # trigger info: hist:keys=stacktrace:vals=len:sort=hitcount:size=2048 [paused]
1066
1067 { stacktrace:
1068 __netif_receive_skb_core+0x46d/0x990
1069 __netif_receive_skb+0x18/0x60
1070 netif_receive_skb_internal+0x23/0x90
1071 napi_gro_receive+0xc8/0x100
1072 ieee80211_deliver_skb+0xd6/0x270 [mac80211]
1073 ieee80211_rx_handlers+0xccf/0x22f0 [mac80211]
1074 ieee80211_prepare_and_rx_handle+0x4e7/0xc40 [mac80211]
1075 ieee80211_rx+0x31d/0x900 [mac80211]
1076 iwlagn_rx_reply_rx+0x3db/0x6f0 [iwldvm]
1077 iwl_rx_dispatch+0x8e/0xf0 [iwldvm]
1078 iwl_pcie_irq_handler+0xe3c/0x12f0 [iwlwifi]
1079 irq_thread_fn+0x20/0x50
1080 irq_thread+0x11f/0x150
1081 kthread+0xd2/0xf0
1082 ret_from_fork+0x42/0x70
1083 } hitcount: 85 len: 28884
1084 { stacktrace:
1085 __netif_receive_skb_core+0x46d/0x990
1086 __netif_receive_skb+0x18/0x60
1087 netif_receive_skb_internal+0x23/0x90
1088 napi_gro_complete+0xa4/0xe0
1089 dev_gro_receive+0x23a/0x360
1090 napi_gro_receive+0x30/0x100
1091 ieee80211_deliver_skb+0xd6/0x270 [mac80211]
1092 ieee80211_rx_handlers+0xccf/0x22f0 [mac80211]
1093 ieee80211_prepare_and_rx_handle+0x4e7/0xc40 [mac80211]
1094 ieee80211_rx+0x31d/0x900 [mac80211]
1095 iwlagn_rx_reply_rx+0x3db/0x6f0 [iwldvm]
1096 iwl_rx_dispatch+0x8e/0xf0 [iwldvm]
1097 iwl_pcie_irq_handler+0xe3c/0x12f0 [iwlwifi]
1098 irq_thread_fn+0x20/0x50
1099 irq_thread+0x11f/0x150
1100 kthread+0xd2/0xf0
1101 } hitcount: 98 len: 664329
1102 { stacktrace:
1103 __netif_receive_skb_core+0x46d/0x990
1104 __netif_receive_skb+0x18/0x60
1105 process_backlog+0xa8/0x150
1106 net_rx_action+0x15d/0x340
1107 __do_softirq+0x114/0x2c0
1108 do_softirq_own_stack+0x1c/0x30
1109 do_softirq+0x65/0x70
1110 __local_bh_enable_ip+0xb5/0xc0
1111 ip_finish_output+0x1f4/0x840
1112 ip_output+0x6b/0xc0
1113 ip_local_out_sk+0x31/0x40
1114 ip_send_skb+0x1a/0x50
1115 udp_send_skb+0x173/0x2a0
1116 udp_sendmsg+0x2bf/0x9f0
1117 inet_sendmsg+0x64/0xa0
1118 sock_sendmsg+0x3d/0x50
1119 } hitcount: 115 len: 13030
1120 { stacktrace:
1121 __netif_receive_skb_core+0x46d/0x990
1122 __netif_receive_skb+0x18/0x60
1123 netif_receive_skb_internal+0x23/0x90
1124 napi_gro_complete+0xa4/0xe0
1125 napi_gro_flush+0x6d/0x90
1126 iwl_pcie_irq_handler+0x92a/0x12f0 [iwlwifi]
1127 irq_thread_fn+0x20/0x50
1128 irq_thread+0x11f/0x150
1129 kthread+0xd2/0xf0
1130 ret_from_fork+0x42/0x70
1131 } hitcount: 934 len: 5512212
1132
1133 Totals:
1134 Hits: 1232
1135 Entries: 4
1136 Dropped: 0
1137
1138 The above shows all the netif_receive_skb callpaths and their total
1139 lengths for the duration of the wget command.
1140
1141 The 'clear' hist trigger param can be used to clear the hash table.
1142 Suppose we wanted to try another run of the previous example but
1143 this time also wanted to see the complete list of events that went
1144 into the histogram. In order to avoid having to set everything up
ea272257 1145 again, we can just clear the histogram first::
b8df4a36
TZ
1146
1147 # echo 'hist:key=stacktrace:vals=len:clear' >> \
1148 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1149
1150 Just to verify that it is in fact cleared, here's what we now see in
ea272257 1151 the hist file::
b8df4a36
TZ
1152
1153 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist
1154 # trigger info: hist:keys=stacktrace:vals=len:sort=hitcount:size=2048 [paused]
1155
1156 Totals:
1157 Hits: 0
1158 Entries: 0
1159 Dropped: 0
1160
1161 Since we want to see the detailed list of every netif_receive_skb
1162 event occurring during the new run, which are in fact the same
1163 events being aggregated into the hash table, we add some additional
1164 'enable_event' events to the triggering sched_process_exec and
ea272257 1165 sched_process_exit events as such::
b8df4a36
TZ
1166
1167 # echo 'enable_event:net:netif_receive_skb if filename==/usr/bin/wget' > \
1168 /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger
1169
1170 # echo 'disable_event:net:netif_receive_skb if comm==wget' > \
1171 /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger
1172
1173 If you read the trigger files for the sched_process_exec and
1174 sched_process_exit triggers, you should see two triggers for each:
1175 one enabling/disabling the hist aggregation and the other
ea272257 1176 enabling/disabling the logging of events::
b8df4a36
TZ
1177
1178 # cat /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger
1179 enable_event:net:netif_receive_skb:unlimited if filename==/usr/bin/wget
1180 enable_hist:net:netif_receive_skb:unlimited if filename==/usr/bin/wget
1181
1182 # cat /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger
1183 enable_event:net:netif_receive_skb:unlimited if comm==wget
1184 disable_hist:net:netif_receive_skb:unlimited if comm==wget
1185
1186 In other words, whenever either of the sched_process_exec or
1187 sched_process_exit events is hit and matches 'wget', it enables or
1188 disables both the histogram and the event log, and what you end up
1189 with is a hash table and set of events just covering the specified
ea272257 1190 duration. Run the wget command again::
b8df4a36
TZ
1191
1192 $ wget https://www.kernel.org/pub/linux/kernel/v3.x/patch-3.19.xz
1193
1194 Displaying the 'hist' file should show something similar to what you
1195 saw in the last run, but this time you should also see the
ea272257 1196 individual events in the trace file::
b8df4a36
TZ
1197
1198 # cat /sys/kernel/debug/tracing/trace
1199
1200 # tracer: nop
1201 #
1202 # entries-in-buffer/entries-written: 183/1426 #P:4
1203 #
1204 # _-----=> irqs-off
1205 # / _----=> need-resched
1206 # | / _---=> hardirq/softirq
1207 # || / _--=> preempt-depth
1208 # ||| / delay
1209 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1210 # | | | |||| | |
1211 wget-15108 [000] ..s1 31769.606929: netif_receive_skb: dev=lo skbaddr=ffff88009c353100 len=60
1212 wget-15108 [000] ..s1 31769.606999: netif_receive_skb: dev=lo skbaddr=ffff88009c353200 len=60
1213 dnsmasq-1382 [000] ..s1 31769.677652: netif_receive_skb: dev=lo skbaddr=ffff88009c352b00 len=130
1214 dnsmasq-1382 [000] ..s1 31769.685917: netif_receive_skb: dev=lo skbaddr=ffff88009c352200 len=138
1215 ##### CPU 2 buffer started ####
1216 irq/29-iwlwifi-559 [002] ..s. 31772.031529: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d433d00 len=2948
1217 irq/29-iwlwifi-559 [002] ..s. 31772.031572: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d432200 len=1500
1218 irq/29-iwlwifi-559 [002] ..s. 31772.032196: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d433100 len=2948
1219 irq/29-iwlwifi-559 [002] ..s. 31772.032761: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d433000 len=2948
1220 irq/29-iwlwifi-559 [002] ..s. 31772.033220: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d432e00 len=1500
1221 .
1222 .
1223 .
1224
1225 The following example demonstrates how multiple hist triggers can be
1226 attached to a given event. This capability can be useful for
1227 creating a set of different summaries derived from the same set of
1228 events, or for comparing the effects of different filters, among
ea272257 1229 other things::
b8df4a36
TZ
1230
1231 # echo 'hist:keys=skbaddr.hex:vals=len if len < 0' >> \
1232 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1233 # echo 'hist:keys=skbaddr.hex:vals=len if len > 4096' >> \
1234 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1235 # echo 'hist:keys=skbaddr.hex:vals=len if len == 256' >> \
1236 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1237 # echo 'hist:keys=skbaddr.hex:vals=len' >> \
1238 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1239 # echo 'hist:keys=len:vals=common_preempt_count' >> \
1240 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1241
1242 The above set of commands create four triggers differing only in
1243 their filters, along with a completely different though fairly
1244 nonsensical trigger. Note that in order to append multiple hist
1245 triggers to the same file, you should use the '>>' operator to
1246 append them ('>' will also add the new hist trigger, but will remove
1247 any existing hist triggers beforehand).
1248
1249 Displaying the contents of the 'hist' file for the event shows the
ea272257 1250 contents of all five histograms::
b8df4a36
TZ
1251
1252 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist
1253
1254 # event histogram
1255 #
1256 # trigger info: hist:keys=len:vals=hitcount,common_preempt_count:sort=hitcount:size=2048 [active]
1257 #
1258
1259 { len: 176 } hitcount: 1 common_preempt_count: 0
1260 { len: 223 } hitcount: 1 common_preempt_count: 0
1261 { len: 4854 } hitcount: 1 common_preempt_count: 0
1262 { len: 395 } hitcount: 1 common_preempt_count: 0
1263 { len: 177 } hitcount: 1 common_preempt_count: 0
1264 { len: 446 } hitcount: 1 common_preempt_count: 0
1265 { len: 1601 } hitcount: 1 common_preempt_count: 0
1266 .
1267 .
1268 .
1269 { len: 1280 } hitcount: 66 common_preempt_count: 0
1270 { len: 116 } hitcount: 81 common_preempt_count: 40
1271 { len: 708 } hitcount: 112 common_preempt_count: 0
1272 { len: 46 } hitcount: 221 common_preempt_count: 0
1273 { len: 1264 } hitcount: 458 common_preempt_count: 0
1274
1275 Totals:
1276 Hits: 1428
1277 Entries: 147
1278 Dropped: 0
1279
1280
1281 # event histogram
1282 #
1283 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 [active]
1284 #
1285
1286 { skbaddr: ffff8800baee5e00 } hitcount: 1 len: 130
1287 { skbaddr: ffff88005f3d5600 } hitcount: 1 len: 1280
1288 { skbaddr: ffff88005f3d4900 } hitcount: 1 len: 1280
1289 { skbaddr: ffff88009fed6300 } hitcount: 1 len: 115
1290 { skbaddr: ffff88009fe0ad00 } hitcount: 1 len: 115
1291 { skbaddr: ffff88008cdb1900 } hitcount: 1 len: 46
1292 { skbaddr: ffff880064b5ef00 } hitcount: 1 len: 118
1293 { skbaddr: ffff880044e3c700 } hitcount: 1 len: 60
1294 { skbaddr: ffff880100065900 } hitcount: 1 len: 46
1295 { skbaddr: ffff8800d46bd500 } hitcount: 1 len: 116
1296 { skbaddr: ffff88005f3d5f00 } hitcount: 1 len: 1280
1297 { skbaddr: ffff880100064700 } hitcount: 1 len: 365
1298 { skbaddr: ffff8800badb6f00 } hitcount: 1 len: 60
1299 .
1300 .
1301 .
1302 { skbaddr: ffff88009fe0be00 } hitcount: 27 len: 24677
1303 { skbaddr: ffff88009fe0a400 } hitcount: 27 len: 23052
1304 { skbaddr: ffff88009fe0b700 } hitcount: 31 len: 25589
1305 { skbaddr: ffff88009fe0b600 } hitcount: 32 len: 27326
1306 { skbaddr: ffff88006a462800 } hitcount: 68 len: 71678
1307 { skbaddr: ffff88006a463700 } hitcount: 70 len: 72678
1308 { skbaddr: ffff88006a462b00 } hitcount: 71 len: 77589
1309 { skbaddr: ffff88006a463600 } hitcount: 73 len: 71307
1310 { skbaddr: ffff88006a462200 } hitcount: 81 len: 81032
1311
1312 Totals:
1313 Hits: 1451
1314 Entries: 318
1315 Dropped: 0
1316
1317
1318 # event histogram
1319 #
1320 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 if len == 256 [active]
1321 #
1322
1323
1324 Totals:
1325 Hits: 0
1326 Entries: 0
1327 Dropped: 0
1328
1329
1330 # event histogram
1331 #
1332 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 if len > 4096 [active]
1333 #
1334
1335 { skbaddr: ffff88009fd2c300 } hitcount: 1 len: 7212
1336 { skbaddr: ffff8800d2bcce00 } hitcount: 1 len: 7212
1337 { skbaddr: ffff8800d2bcd700 } hitcount: 1 len: 7212
1338 { skbaddr: ffff8800d2bcda00 } hitcount: 1 len: 21492
1339 { skbaddr: ffff8800ae2e2d00 } hitcount: 1 len: 7212
1340 { skbaddr: ffff8800d2bcdb00 } hitcount: 1 len: 7212
1341 { skbaddr: ffff88006a4df500 } hitcount: 1 len: 4854
1342 { skbaddr: ffff88008ce47b00 } hitcount: 1 len: 18636
1343 { skbaddr: ffff8800ae2e2200 } hitcount: 1 len: 12924
1344 { skbaddr: ffff88005f3e1000 } hitcount: 1 len: 4356
1345 { skbaddr: ffff8800d2bcdc00 } hitcount: 2 len: 24420
1346 { skbaddr: ffff8800d2bcc200 } hitcount: 2 len: 12996
1347
1348 Totals:
1349 Hits: 14
1350 Entries: 12
1351 Dropped: 0
1352
1353
1354 # event histogram
1355 #
1356 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 if len < 0 [active]
1357 #
1358
1359
1360 Totals:
1361 Hits: 0
1362 Entries: 0
1363 Dropped: 0
1364
1365 Named triggers can be used to have triggers share a common set of
1366 histogram data. This capability is mostly useful for combining the
1367 output of events generated by tracepoints contained inside inline
1368 functions, but names can be used in a hist trigger on any event.
1369 For example, these two triggers when hit will update the same 'len'
ea272257 1370 field in the shared 'foo' histogram data::
b8df4a36
TZ
1371
1372 # echo 'hist:name=foo:keys=skbaddr.hex:vals=len' > \
1373 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger
1374 # echo 'hist:name=foo:keys=skbaddr.hex:vals=len' > \
1375 /sys/kernel/debug/tracing/events/net/netif_rx/trigger
1376
1377 You can see that they're updating common histogram data by reading
ea272257 1378 each event's hist files at the same time::
b8df4a36
TZ
1379
1380 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist;
1381 cat /sys/kernel/debug/tracing/events/net/netif_rx/hist
1382
1383 # event histogram
1384 #
1385 # trigger info: hist:name=foo:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 [active]
1386 #
1387
1388 { skbaddr: ffff88000ad53500 } hitcount: 1 len: 46
1389 { skbaddr: ffff8800af5a1500 } hitcount: 1 len: 76
1390 { skbaddr: ffff8800d62a1900 } hitcount: 1 len: 46
1391 { skbaddr: ffff8800d2bccb00 } hitcount: 1 len: 468
1392 { skbaddr: ffff8800d3c69900 } hitcount: 1 len: 46
1393 { skbaddr: ffff88009ff09100 } hitcount: 1 len: 52
1394 { skbaddr: ffff88010f13ab00 } hitcount: 1 len: 168
1395 { skbaddr: ffff88006a54f400 } hitcount: 1 len: 46
1396 { skbaddr: ffff8800d2bcc500 } hitcount: 1 len: 260
1397 { skbaddr: ffff880064505000 } hitcount: 1 len: 46
1398 { skbaddr: ffff8800baf24e00 } hitcount: 1 len: 32
1399 { skbaddr: ffff88009fe0ad00 } hitcount: 1 len: 46
1400 { skbaddr: ffff8800d3edff00 } hitcount: 1 len: 44
1401 { skbaddr: ffff88009fe0b400 } hitcount: 1 len: 168
1402 { skbaddr: ffff8800a1c55a00 } hitcount: 1 len: 40
1403 { skbaddr: ffff8800d2bcd100 } hitcount: 1 len: 40
1404 { skbaddr: ffff880064505f00 } hitcount: 1 len: 174
1405 { skbaddr: ffff8800a8bff200 } hitcount: 1 len: 160
1406 { skbaddr: ffff880044e3cc00 } hitcount: 1 len: 76
1407 { skbaddr: ffff8800a8bfe700 } hitcount: 1 len: 46
1408 { skbaddr: ffff8800d2bcdc00 } hitcount: 1 len: 32
1409 { skbaddr: ffff8800a1f64800 } hitcount: 1 len: 46
1410 { skbaddr: ffff8800d2bcde00 } hitcount: 1 len: 988
1411 { skbaddr: ffff88006a5dea00 } hitcount: 1 len: 46
1412 { skbaddr: ffff88002e37a200 } hitcount: 1 len: 44
1413 { skbaddr: ffff8800a1f32c00 } hitcount: 2 len: 676
1414 { skbaddr: ffff88000ad52600 } hitcount: 2 len: 107
1415 { skbaddr: ffff8800a1f91e00 } hitcount: 2 len: 92
1416 { skbaddr: ffff8800af5a0200 } hitcount: 2 len: 142
1417 { skbaddr: ffff8800d2bcc600 } hitcount: 2 len: 220
1418 { skbaddr: ffff8800ba36f500 } hitcount: 2 len: 92
1419 { skbaddr: ffff8800d021f800 } hitcount: 2 len: 92
1420 { skbaddr: ffff8800a1f33600 } hitcount: 2 len: 675
1421 { skbaddr: ffff8800a8bfff00 } hitcount: 3 len: 138
1422 { skbaddr: ffff8800d62a1300 } hitcount: 3 len: 138
1423 { skbaddr: ffff88002e37a100 } hitcount: 4 len: 184
1424 { skbaddr: ffff880064504400 } hitcount: 4 len: 184
1425 { skbaddr: ffff8800a8bfec00 } hitcount: 4 len: 184
1426 { skbaddr: ffff88000ad53700 } hitcount: 5 len: 230
1427 { skbaddr: ffff8800d2bcdb00 } hitcount: 5 len: 196
1428 { skbaddr: ffff8800a1f90000 } hitcount: 6 len: 276
1429 { skbaddr: ffff88006a54f900 } hitcount: 6 len: 276
1430
1431 Totals:
1432 Hits: 81
1433 Entries: 42
1434 Dropped: 0
1435 # event histogram
1436 #
1437 # trigger info: hist:name=foo:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 [active]
1438 #
1439
1440 { skbaddr: ffff88000ad53500 } hitcount: 1 len: 46
1441 { skbaddr: ffff8800af5a1500 } hitcount: 1 len: 76
1442 { skbaddr: ffff8800d62a1900 } hitcount: 1 len: 46
1443 { skbaddr: ffff8800d2bccb00 } hitcount: 1 len: 468
1444 { skbaddr: ffff8800d3c69900 } hitcount: 1 len: 46
1445 { skbaddr: ffff88009ff09100 } hitcount: 1 len: 52
1446 { skbaddr: ffff88010f13ab00 } hitcount: 1 len: 168
1447 { skbaddr: ffff88006a54f400 } hitcount: 1 len: 46
1448 { skbaddr: ffff8800d2bcc500 } hitcount: 1 len: 260
1449 { skbaddr: ffff880064505000 } hitcount: 1 len: 46
1450 { skbaddr: ffff8800baf24e00 } hitcount: 1 len: 32
1451 { skbaddr: ffff88009fe0ad00 } hitcount: 1 len: 46
1452 { skbaddr: ffff8800d3edff00 } hitcount: 1 len: 44
1453 { skbaddr: ffff88009fe0b400 } hitcount: 1 len: 168
1454 { skbaddr: ffff8800a1c55a00 } hitcount: 1 len: 40
1455 { skbaddr: ffff8800d2bcd100 } hitcount: 1 len: 40
1456 { skbaddr: ffff880064505f00 } hitcount: 1 len: 174
1457 { skbaddr: ffff8800a8bff200 } hitcount: 1 len: 160
1458 { skbaddr: ffff880044e3cc00 } hitcount: 1 len: 76
1459 { skbaddr: ffff8800a8bfe700 } hitcount: 1 len: 46
1460 { skbaddr: ffff8800d2bcdc00 } hitcount: 1 len: 32
1461 { skbaddr: ffff8800a1f64800 } hitcount: 1 len: 46
1462 { skbaddr: ffff8800d2bcde00 } hitcount: 1 len: 988
1463 { skbaddr: ffff88006a5dea00 } hitcount: 1 len: 46
1464 { skbaddr: ffff88002e37a200 } hitcount: 1 len: 44
1465 { skbaddr: ffff8800a1f32c00 } hitcount: 2 len: 676
1466 { skbaddr: ffff88000ad52600 } hitcount: 2 len: 107
1467 { skbaddr: ffff8800a1f91e00 } hitcount: 2 len: 92
1468 { skbaddr: ffff8800af5a0200 } hitcount: 2 len: 142
1469 { skbaddr: ffff8800d2bcc600 } hitcount: 2 len: 220
1470 { skbaddr: ffff8800ba36f500 } hitcount: 2 len: 92
1471 { skbaddr: ffff8800d021f800 } hitcount: 2 len: 92
1472 { skbaddr: ffff8800a1f33600 } hitcount: 2 len: 675
1473 { skbaddr: ffff8800a8bfff00 } hitcount: 3 len: 138
1474 { skbaddr: ffff8800d62a1300 } hitcount: 3 len: 138
1475 { skbaddr: ffff88002e37a100 } hitcount: 4 len: 184
1476 { skbaddr: ffff880064504400 } hitcount: 4 len: 184
1477 { skbaddr: ffff8800a8bfec00 } hitcount: 4 len: 184
1478 { skbaddr: ffff88000ad53700 } hitcount: 5 len: 230
1479 { skbaddr: ffff8800d2bcdb00 } hitcount: 5 len: 196
1480 { skbaddr: ffff8800a1f90000 } hitcount: 6 len: 276
1481 { skbaddr: ffff88006a54f900 } hitcount: 6 len: 276
1482
1483 Totals:
1484 Hits: 81
1485 Entries: 42
1486 Dropped: 0
1487
1488 And here's an example that shows how to combine histogram data from
1489 any two events even if they don't share any 'compatible' fields
1490 other than 'hitcount' and 'stacktrace'. These commands create a
ea272257 1491 couple of triggers named 'bar' using those fields::
b8df4a36
TZ
1492
1493 # echo 'hist:name=bar:key=stacktrace:val=hitcount' > \
1494 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger
1495 # echo 'hist:name=bar:key=stacktrace:val=hitcount' > \
1496 /sys/kernel/debug/tracing/events/net/netif_rx/trigger
1497
1498 And displaying the output of either shows some interesting if
ea272257 1499 somewhat confusing output::
b8df4a36
TZ
1500
1501 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist
1502 # cat /sys/kernel/debug/tracing/events/net/netif_rx/hist
1503
1504 # event histogram
1505 #
1506 # trigger info: hist:name=bar:keys=stacktrace:vals=hitcount:sort=hitcount:size=2048 [active]
1507 #
1508
1509 { stacktrace:
1510 _do_fork+0x18e/0x330
1511 kernel_thread+0x29/0x30
1512 kthreadd+0x154/0x1b0
1513 ret_from_fork+0x3f/0x70
1514 } hitcount: 1
1515 { stacktrace:
1516 netif_rx_internal+0xb2/0xd0
1517 netif_rx_ni+0x20/0x70
1518 dev_loopback_xmit+0xaa/0xd0
1519 ip_mc_output+0x126/0x240
1520 ip_local_out_sk+0x31/0x40
1521 igmp_send_report+0x1e9/0x230
1522 igmp_timer_expire+0xe9/0x120
1523 call_timer_fn+0x39/0xf0
1524 run_timer_softirq+0x1e1/0x290
1525 __do_softirq+0xfd/0x290
1526 irq_exit+0x98/0xb0
1527 smp_apic_timer_interrupt+0x4a/0x60
1528 apic_timer_interrupt+0x6d/0x80
1529 cpuidle_enter+0x17/0x20
1530 call_cpuidle+0x3b/0x60
1531 cpu_startup_entry+0x22d/0x310
1532 } hitcount: 1
1533 { stacktrace:
1534 netif_rx_internal+0xb2/0xd0
1535 netif_rx_ni+0x20/0x70
1536 dev_loopback_xmit+0xaa/0xd0
1537 ip_mc_output+0x17f/0x240
1538 ip_local_out_sk+0x31/0x40
1539 ip_send_skb+0x1a/0x50
1540 udp_send_skb+0x13e/0x270
1541 udp_sendmsg+0x2bf/0x980
1542 inet_sendmsg+0x67/0xa0
1543 sock_sendmsg+0x38/0x50
1544 SYSC_sendto+0xef/0x170
1545 SyS_sendto+0xe/0x10
1546 entry_SYSCALL_64_fastpath+0x12/0x6a
1547 } hitcount: 2
1548 { stacktrace:
1549 netif_rx_internal+0xb2/0xd0
1550 netif_rx+0x1c/0x60
1551 loopback_xmit+0x6c/0xb0
1552 dev_hard_start_xmit+0x219/0x3a0
1553 __dev_queue_xmit+0x415/0x4f0
1554 dev_queue_xmit_sk+0x13/0x20
1555 ip_finish_output2+0x237/0x340
1556 ip_finish_output+0x113/0x1d0
1557 ip_output+0x66/0xc0
1558 ip_local_out_sk+0x31/0x40
1559 ip_send_skb+0x1a/0x50
1560 udp_send_skb+0x16d/0x270
1561 udp_sendmsg+0x2bf/0x980
1562 inet_sendmsg+0x67/0xa0
1563 sock_sendmsg+0x38/0x50
1564 ___sys_sendmsg+0x14e/0x270
1565 } hitcount: 76
1566 { stacktrace:
1567 netif_rx_internal+0xb2/0xd0
1568 netif_rx+0x1c/0x60
1569 loopback_xmit+0x6c/0xb0
1570 dev_hard_start_xmit+0x219/0x3a0
1571 __dev_queue_xmit+0x415/0x4f0
1572 dev_queue_xmit_sk+0x13/0x20
1573 ip_finish_output2+0x237/0x340
1574 ip_finish_output+0x113/0x1d0
1575 ip_output+0x66/0xc0
1576 ip_local_out_sk+0x31/0x40
1577 ip_send_skb+0x1a/0x50
1578 udp_send_skb+0x16d/0x270
1579 udp_sendmsg+0x2bf/0x980
1580 inet_sendmsg+0x67/0xa0
1581 sock_sendmsg+0x38/0x50
1582 ___sys_sendmsg+0x269/0x270
1583 } hitcount: 77
1584 { stacktrace:
1585 netif_rx_internal+0xb2/0xd0
1586 netif_rx+0x1c/0x60
1587 loopback_xmit+0x6c/0xb0
1588 dev_hard_start_xmit+0x219/0x3a0
1589 __dev_queue_xmit+0x415/0x4f0
1590 dev_queue_xmit_sk+0x13/0x20
1591 ip_finish_output2+0x237/0x340
1592 ip_finish_output+0x113/0x1d0
1593 ip_output+0x66/0xc0
1594 ip_local_out_sk+0x31/0x40
1595 ip_send_skb+0x1a/0x50
1596 udp_send_skb+0x16d/0x270
1597 udp_sendmsg+0x2bf/0x980
1598 inet_sendmsg+0x67/0xa0
1599 sock_sendmsg+0x38/0x50
1600 SYSC_sendto+0xef/0x170
1601 } hitcount: 88
1602 { stacktrace:
1603 _do_fork+0x18e/0x330
1604 SyS_clone+0x19/0x20
1605 entry_SYSCALL_64_fastpath+0x12/0x6a
1606 } hitcount: 244
1607
1608 Totals:
1609 Hits: 489
1610 Entries: 7
1611 Dropped: 0
033cbcee 1612
033cbcee
TZ
16132.2 Inter-event hist triggers
1614-----------------------------
1615
1616Inter-event hist triggers are hist triggers that combine values from
1617one or more other events and create a histogram using that data. Data
1618from an inter-event histogram can in turn become the source for
1619further combined histograms, thus providing a chain of related
1620histograms, which is important for some applications.
1621
1622The most important example of an inter-event quantity that can be used
1623in this manner is latency, which is simply a difference in timestamps
1624between two events. Although latency is the most important
1625inter-event quantity, note that because the support is completely
1626general across the trace event subsystem, any event field can be used
1627in an inter-event quantity.
1628
1629An example of a histogram that combines data from other histograms
1630into a useful chain would be a 'wakeupswitch latency' histogram that
1631combines a 'wakeup latency' histogram and a 'switch latency'
1632histogram.
1633
1634Normally, a hist trigger specification consists of a (possibly
1635compound) key along with one or more numeric values, which are
1636continually updated sums associated with that key. A histogram
1637specification in this case consists of individual key and value
1638specifications that refer to trace event fields associated with a
1639single event type.
1640
1641The inter-event hist trigger extension allows fields from multiple
1642events to be referenced and combined into a multi-event histogram
1643specification. In support of this overall goal, a few enabling
1644features have been added to the hist trigger support:
1645
1646 - In order to compute an inter-event quantity, a value from one
1647 event needs to saved and then referenced from another event. This
1648 requires the introduction of support for histogram 'variables'.
1649
1650 - The computation of inter-event quantities and their combination
1651 require some minimal amount of support for applying simple
1652 expressions to variables (+ and -).
1653
1654 - A histogram consisting of inter-event quantities isn't logically a
1655 histogram on either event (so having the 'hist' file for either
1656 event host the histogram output doesn't really make sense). To
1657 address the idea that the histogram is associated with a
1658 combination of events, support is added allowing the creation of
1659 'synthetic' events that are events derived from other events.
1660 These synthetic events are full-fledged events just like any other
1661 and can be used as such, as for instance to create the
1662 'combination' histograms mentioned previously.
1663
1664 - A set of 'actions' can be associated with histogram entries -
1665 these can be used to generate the previously mentioned synthetic
1666 events, but can also be used for other purposes, such as for
1667 example saving context when a 'max' latency has been hit.
1668
1669 - Trace events don't have a 'timestamp' associated with them, but
1670 there is an implicit timestamp saved along with an event in the
1671 underlying ftrace ring buffer. This timestamp is now exposed as a
1672 a synthetic field named 'common_timestamp' which can be used in
1673 histograms as if it were any other event field; it isn't an actual
1674 field in the trace format but rather is a synthesized value that
1675 nonetheless can be used as if it were an actual field. By default
1676 it is in units of nanoseconds; appending '.usecs' to a
1677 common_timestamp field changes the units to microseconds.
1678
a4072fe8
TZ
1679A note on inter-event timestamps: If common_timestamp is used in a
1680histogram, the trace buffer is automatically switched over to using
1681absolute timestamps and the "global" trace clock, in order to avoid
1682bogus timestamp differences with other clocks that aren't coherent
1683across CPUs. This can be overridden by specifying one of the other
1684trace clocks instead, using the "clock=XXX" hist trigger attribute,
1685where XXX is any of the clocks listed in the tracing/trace_clock
1686pseudo-file.
1687
1688These features are described in more detail in the following sections.
033cbcee
TZ
1689
16902.2.1 Histogram Variables
1691-------------------------
1692
1693Variables are simply named locations used for saving and retrieving
1694values between matching events. A 'matching' event is defined as an
1695event that has a matching key - if a variable is saved for a histogram
1696entry corresponding to that key, any subsequent event with a matching
1697key can access that variable.
1698
1699A variable's value is normally available to any subsequent event until
1700it is set to something else by a subsequent event. The one exception
1701to that rule is that any variable used in an expression is essentially
1702'read-once' - once it's used by an expression in a subsequent event,
1703it's reset to its 'unset' state, which means it can't be used again
1704unless it's set again. This ensures not only that an event doesn't
1705use an uninitialized variable in a calculation, but that that variable
1706is used only once and not for any unrelated subsequent match.
1707
1708The basic syntax for saving a variable is to simply prefix a unique
1709variable name not corresponding to any keyword along with an '=' sign
1710to any event field.
1711
1712Either keys or values can be saved and retrieved in this way. This
1713creates a variable named 'ts0' for a histogram entry with the key
ea272257 1714'next_pid'::
033cbcee
TZ
1715
1716 # echo 'hist:keys=next_pid:vals=$ts0:ts0=common_timestamp ... >> \
1717 event/trigger
1718
1719The ts0 variable can be accessed by any subsequent event having the
1720same pid as 'next_pid'.
1721
1722Variable references are formed by prepending the variable name with
1723the '$' sign. Thus for example, the ts0 variable above would be
1724referenced as '$ts0' in expressions.
1725
1726Because 'vals=' is used, the common_timestamp variable value above
1727will also be summed as a normal histogram value would (though for a
1728timestamp it makes little sense).
1729
ea272257 1730The below shows that a key value can also be saved in the same way::
033cbcee
TZ
1731
1732 # echo 'hist:timer_pid=common_pid:key=timer_pid ...' >> event/trigger
1733
1734If a variable isn't a key variable or prefixed with 'vals=', the
1735associated event field will be saved in a variable but won't be summed
ea272257 1736as a value::
033cbcee 1737
064f35a9 1738 # echo 'hist:keys=next_pid:ts1=common_timestamp ...' >> event/trigger
033cbcee
TZ
1739
1740Multiple variables can be assigned at the same time. The below would
1741result in both ts0 and b being created as variables, with both
ea272257 1742common_timestamp and field1 additionally being summed as values::
033cbcee 1743
064f35a9 1744 # echo 'hist:keys=pid:vals=$ts0,$b:ts0=common_timestamp,b=field1 ...' >> \
033cbcee
TZ
1745 event/trigger
1746
1747Note that variable assignments can appear either preceding or
1748following their use. The command below behaves identically to the
ea272257 1749command above::
033cbcee 1750
064f35a9 1751 # echo 'hist:keys=pid:ts0=common_timestamp,b=field1:vals=$ts0,$b ...' >> \
033cbcee
TZ
1752 event/trigger
1753
1754Any number of variables not bound to a 'vals=' prefix can also be
1755assigned by simply separating them with colons. Below is the same
ea272257 1756thing but without the values being summed in the histogram::
033cbcee 1757
064f35a9 1758 # echo 'hist:keys=pid:ts0=common_timestamp:b=field1 ...' >> event/trigger
033cbcee
TZ
1759
1760Variables set as above can be referenced and used in expressions on
1761another event.
1762
ea272257 1763For example, here's how a latency can be calculated::
033cbcee 1764
064f35a9
JFG
1765 # echo 'hist:keys=pid,prio:ts0=common_timestamp ...' >> event1/trigger
1766 # echo 'hist:keys=next_pid:wakeup_lat=common_timestamp-$ts0 ...' >> event2/trigger
033cbcee 1767
aea74de4 1768In the first line above, the event's timestamp is saved into the
033cbcee
TZ
1769variable ts0. In the next line, ts0 is subtracted from the second
1770event's timestamp to produce the latency, which is then assigned into
1771yet another variable, 'wakeup_lat'. The hist trigger below in turn
1772makes use of the wakeup_lat variable to compute a combined latency
ea272257 1773using the same key and variable from yet another event::
033cbcee 1774
064f35a9 1775 # echo 'hist:key=pid:wakeupswitch_lat=$wakeup_lat+$switchtime_lat ...' >> event3/trigger
033cbcee
TZ
1776
17772.2.2 Synthetic Events
1778----------------------
1779
1780Synthetic events are user-defined events generated from hist trigger
1781variables or fields associated with one or more other events. Their
1782purpose is to provide a mechanism for displaying data spanning
1783multiple events consistent with the existing and already familiar
1784usage for normal events.
1785
1786To define a synthetic event, the user writes a simple specification
1787consisting of the name of the new event along with one or more
1788variables and their types, which can be any valid field type,
1789separated by semicolons, to the tracing/synthetic_events file.
1790
1791For instance, the following creates a new event named 'wakeup_latency'
1792with 3 fields: lat, pid, and prio. Each of those fields is simply a
ea272257 1793variable reference to a variable on another event::
033cbcee
TZ
1794
1795 # echo 'wakeup_latency \
1796 u64 lat; \
1797 pid_t pid; \
1798 int prio' >> \
1799 /sys/kernel/debug/tracing/synthetic_events
1800
1801Reading the tracing/synthetic_events file lists all the currently
ea272257 1802defined synthetic events, in this case the event defined above::
033cbcee
TZ
1803
1804 # cat /sys/kernel/debug/tracing/synthetic_events
1805 wakeup_latency u64 lat; pid_t pid; int prio
1806
1807An existing synthetic event definition can be removed by prepending
ea272257 1808the command that defined it with a '!'::
033cbcee
TZ
1809
1810 # echo '!wakeup_latency u64 lat pid_t pid int prio' >> \
1811 /sys/kernel/debug/tracing/synthetic_events
1812
1813At this point, there isn't yet an actual 'wakeup_latency' event
aea74de4 1814instantiated in the event subsystem - for this to happen, a 'hist
033cbcee 1815trigger action' needs to be instantiated and bound to actual fields
064f35a9
JFG
1816and variables defined on other events (see Section 2.2.3 below on
1817how that is done using hist trigger 'onmatch' action). Once that is
1818done, the 'wakeup_latency' synthetic event instance is created.
033cbcee 1819
ea272257 1820A histogram can now be defined for the new synthetic event::
033cbcee
TZ
1821
1822 # echo 'hist:keys=pid,prio,lat.log2:sort=pid,lat' >> \
1823 /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/trigger
1824
1825The new event is created under the tracing/events/synthetic/ directory
ea272257 1826and looks and behaves just like any other event::
033cbcee
TZ
1827
1828 # ls /sys/kernel/debug/tracing/events/synthetic/wakeup_latency
1829 enable filter format hist id trigger
1830
1831Like any other event, once a histogram is enabled for the event, the
1832output can be displayed by reading the event's 'hist' file.
1833
5032b381
TZ
18342.2.3 Hist trigger 'handlers' and 'actions'
1835-------------------------------------------
033cbcee 1836
5032b381
TZ
1837A hist trigger 'action' is a function that's executed (in most cases
1838conditionally) whenever a histogram entry is added or updated.
033cbcee 1839
5032b381
TZ
1840When a histogram entry is added or updated, a hist trigger 'handler'
1841is what decides whether the corresponding action is actually invoked
1842or not.
033cbcee 1843
5032b381
TZ
1844Hist trigger handlers and actions are paired together in the general
1845form:
1846
1847 <handler>.<action>
1848
1849To specify a handler.action pair for a given event, simply specify
1850that handler.action pair between colons in the hist trigger
1851specification.
1852
1853In theory, any handler can be combined with any action, but in
1854practice, not every handler.action combination is currently supported;
1855if a given handler.action combination isn't supported, the hist
1856trigger will fail with -EINVAL;
1857
1858The default 'handler.action' if none is explicity specified is as it
1859always has been, to simply update the set of values associated with an
1860entry. Some applications, however, may want to perform additional
1861actions at that point, such as generate another event, or compare and
1862save a maximum.
1863
1864The supported handlers and actions are listed below, and each is
1865described in more detail in the following paragraphs, in the context
1866of descriptions of some common and useful handler.action combinations.
1867
1868The available handlers are:
1869
1870 - onmatch(matching.event) - invoke action on any addition or update
1871 - onmax(var) - invoke action if var exceeds current max
ff0d35e2 1872 - onchange(var) - invoke action if var changes
5032b381
TZ
1873
1874The available actions are:
1875
e91eefd7 1876 - trace(<synthetic_event_name>,param list) - generate synthetic event
5032b381 1877 - save(field,...) - save current event fields
fd451a3d 1878 - snapshot() - snapshot the trace buffer
5032b381
TZ
1879
1880The following commonly-used handler.action pairs are available:
033cbcee 1881
e91eefd7 1882 - onmatch(matching.event).trace(<synthetic_event_name>,param list)
033cbcee 1883
e91eefd7
TZ
1884 The 'onmatch(matching.event).trace(<synthetic_event_name>,param
1885 list)' hist trigger action is invoked whenever an event matches
1886 and the histogram entry would be added or updated. It causes the
1887 named synthetic event to be generated with the values given in the
033cbcee
TZ
1888 'param list'. The result is the generation of a synthetic event
1889 that consists of the values contained in those variables at the
e91eefd7
TZ
1890 time the invoking event was hit. For example, if the synthetic
1891 event name is 'wakeup_latency', a wakeup_latency event is
1892 generated using onmatch(event).trace(wakeup_latency,arg1,arg2).
1893
1894 There is also an equivalent alternative form available for
1895 generating synthetic events. In this form, the synthetic event
1896 name is used as if it were a function name. For example, using
1897 the 'wakeup_latency' synthetic event name again, the
1898 wakeup_latency event would be generated by invoking it as if it
1899 were a function call, with the event field values passed in as
1900 arguments: onmatch(event).wakeup_latency(arg1,arg2). The syntax
1901 for this form is:
1902
1903 onmatch(matching.event).<synthetic_event_name>(param list)
1904
1905 In either case, the 'param list' consists of one or more
1906 parameters which may be either variables or fields defined on
1907 either the 'matching.event' or the target event. The variables or
1908 fields specified in the param list may be either fully-qualified
1909 or unqualified. If a variable is specified as unqualified, it
1910 must be unique between the two events. A field name used as a
1911 param can be unqualified if it refers to the target event, but
1912 must be fully qualified if it refers to the matching event. A
1913 fully-qualified name is of the form 'system.event_name.$var_name'
1914 or 'system.event_name.field'.
033cbcee
TZ
1915
1916 The 'matching.event' specification is simply the fully qualified
1917 event name of the event that matches the target event for the
1918 onmatch() functionality, in the form 'system.event_name'.
1919
1920 Finally, the number and type of variables/fields in the 'param
1921 list' must match the number and types of the fields in the
1922 synthetic event being generated.
1923
1924 As an example the below defines a simple synthetic event and uses
1925 a variable defined on the sched_wakeup_new event as a parameter
1926 when invoking the synthetic event. Here we define the synthetic
ea272257 1927 event::
033cbcee 1928
ea272257
MCC
1929 # echo 'wakeup_new_test pid_t pid' >> \
1930 /sys/kernel/debug/tracing/synthetic_events
033cbcee 1931
ea272257
MCC
1932 # cat /sys/kernel/debug/tracing/synthetic_events
1933 wakeup_new_test pid_t pid
033cbcee
TZ
1934
1935 The following hist trigger both defines the missing testpid
1936 variable and specifies an onmatch() action that generates a
1937 wakeup_new_test synthetic event whenever a sched_wakeup_new event
1938 occurs, which because of the 'if comm == "cyclictest"' filter only
ea272257 1939 happens when the executable is cyclictest::
033cbcee 1940
ea272257
MCC
1941 # echo 'hist:keys=$testpid:testpid=pid:onmatch(sched.sched_wakeup_new).\
1942 wakeup_new_test($testpid) if comm=="cyclictest"' >> \
1943 /sys/kernel/debug/tracing/events/sched/sched_wakeup_new/trigger
033cbcee 1944
e91eefd7
TZ
1945 Or, equivalently, using the 'trace' keyword syntax:
1946
1947 # echo 'hist:keys=$testpid:testpid=pid:onmatch(sched.sched_wakeup_new).\
1948 trace(wakeup_new_test,$testpid) if comm=="cyclictest"' >> \
1949 /sys/kernel/debug/tracing/events/sched/sched_wakeup_new/trigger
1950
033cbcee
TZ
1951 Creating and displaying a histogram based on those events is now
1952 just a matter of using the fields and new synthetic event in the
ea272257 1953 tracing/events/synthetic directory, as usual::
033cbcee 1954
ea272257
MCC
1955 # echo 'hist:keys=pid:sort=pid' >> \
1956 /sys/kernel/debug/tracing/events/synthetic/wakeup_new_test/trigger
033cbcee
TZ
1957
1958 Running 'cyclictest' should cause wakeup_new events to generate
1959 wakeup_new_test synthetic events which should result in histogram
ea272257 1960 output in the wakeup_new_test event's hist file::
033cbcee 1961
ea272257 1962 # cat /sys/kernel/debug/tracing/events/synthetic/wakeup_new_test/hist
033cbcee
TZ
1963
1964 A more typical usage would be to use two events to calculate a
1965 latency. The following example uses a set of hist triggers to
ea272257 1966 produce a 'wakeup_latency' histogram.
033cbcee 1967
ea272257 1968 First, we define a 'wakeup_latency' synthetic event::
033cbcee 1969
ea272257
MCC
1970 # echo 'wakeup_latency u64 lat; pid_t pid; int prio' >> \
1971 /sys/kernel/debug/tracing/synthetic_events
033cbcee
TZ
1972
1973 Next, we specify that whenever we see a sched_waking event for a
ea272257 1974 cyclictest thread, save the timestamp in a 'ts0' variable::
033cbcee 1975
ea272257
MCC
1976 # echo 'hist:keys=$saved_pid:saved_pid=pid:ts0=common_timestamp.usecs \
1977 if comm=="cyclictest"' >> \
1978 /sys/kernel/debug/tracing/events/sched/sched_waking/trigger
033cbcee
TZ
1979
1980 Then, when the corresponding thread is actually scheduled onto the
1981 CPU by a sched_switch event, calculate the latency and use that
1982 along with another variable and an event field to generate a
ea272257 1983 wakeup_latency synthetic event::
033cbcee 1984
ea272257
MCC
1985 # echo 'hist:keys=next_pid:wakeup_lat=common_timestamp.usecs-$ts0:\
1986 onmatch(sched.sched_waking).wakeup_latency($wakeup_lat,\
1987 $saved_pid,next_prio) if next_comm=="cyclictest"' >> \
1988 /sys/kernel/debug/tracing/events/sched/sched_switch/trigger
033cbcee
TZ
1989
1990 We also need to create a histogram on the wakeup_latency synthetic
ea272257 1991 event in order to aggregate the generated synthetic event data::
033cbcee 1992
ea272257
MCC
1993 # echo 'hist:keys=pid,prio,lat:sort=pid,lat' >> \
1994 /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/trigger
033cbcee
TZ
1995
1996 Finally, once we've run cyclictest to actually generate some
1997 events, we can see the output by looking at the wakeup_latency
ea272257 1998 synthetic event's hist file::
033cbcee 1999
ea272257 2000 # cat /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/hist
033cbcee
TZ
2001
2002 - onmax(var).save(field,.. .)
2003
2004 The 'onmax(var).save(field,...)' hist trigger action is invoked
2005 whenever the value of 'var' associated with a histogram entry
2006 exceeds the current maximum contained in that variable.
2007
2008 The end result is that the trace event fields specified as the
2009 onmax.save() params will be saved if 'var' exceeds the current
2010 maximum for that hist trigger entry. This allows context from the
2011 event that exhibited the new maximum to be saved for later
2012 reference. When the histogram is displayed, additional fields
2013 displaying the saved values will be printed.
2014
2015 As an example the below defines a couple of hist triggers, one for
2016 sched_waking and another for sched_switch, keyed on pid. Whenever
2017 a sched_waking occurs, the timestamp is saved in the entry
2018 corresponding to the current pid, and when the scheduler switches
2019 back to that pid, the timestamp difference is calculated. If the
2020 resulting latency, stored in wakeup_lat, exceeds the current
2021 maximum latency, the values specified in the save() fields are
ea272257 2022 recorded::
033cbcee 2023
ea272257
MCC
2024 # echo 'hist:keys=pid:ts0=common_timestamp.usecs \
2025 if comm=="cyclictest"' >> \
2026 /sys/kernel/debug/tracing/events/sched/sched_waking/trigger
033cbcee 2027
ea272257
MCC
2028 # echo 'hist:keys=next_pid:\
2029 wakeup_lat=common_timestamp.usecs-$ts0:\
2030 onmax($wakeup_lat).save(next_comm,prev_pid,prev_prio,prev_comm) \
2031 if next_comm=="cyclictest"' >> \
2032 /sys/kernel/debug/tracing/events/sched/sched_switch/trigger
033cbcee
TZ
2033
2034 When the histogram is displayed, the max value and the saved
2035 values corresponding to the max are displayed following the rest
ea272257 2036 of the fields::
033cbcee 2037
ea272257
MCC
2038 # cat /sys/kernel/debug/tracing/events/sched/sched_switch/hist
2039 { next_pid: 2255 } hitcount: 239
2040 common_timestamp-ts0: 0
2041 max: 27
2042 next_comm: cyclictest
2043 prev_pid: 0 prev_prio: 120 prev_comm: swapper/1
033cbcee 2044
ea272257
MCC
2045 { next_pid: 2256 } hitcount: 2355
2046 common_timestamp-ts0: 0
2047 max: 49 next_comm: cyclictest
2048 prev_pid: 0 prev_prio: 120 prev_comm: swapper/0
033cbcee 2049
ea272257
MCC
2050 Totals:
2051 Hits: 12970
2052 Entries: 2
2053 Dropped: 0
d3439f9d 2054
fd451a3d
TZ
2055 - onmax(var).snapshot()
2056
2057 The 'onmax(var).snapshot()' hist trigger action is invoked
2058 whenever the value of 'var' associated with a histogram entry
2059 exceeds the current maximum contained in that variable.
2060
2061 The end result is that a global snapshot of the trace buffer will
2062 be saved in the tracing/snapshot file if 'var' exceeds the current
2063 maximum for any hist trigger entry.
2064
2065 Note that in this case the maximum is a global maximum for the
2066 current trace instance, which is the maximum across all buckets of
2067 the histogram. The key of the specific trace event that caused
2068 the global maximum and the global maximum itself are displayed,
2069 along with a message stating that a snapshot has been taken and
2070 where to find it. The user can use the key information displayed
2071 to locate the corresponding bucket in the histogram for even more
2072 detail.
2073
2074 As an example the below defines a couple of hist triggers, one for
2075 sched_waking and another for sched_switch, keyed on pid. Whenever
2076 a sched_waking event occurs, the timestamp is saved in the entry
2077 corresponding to the current pid, and when the scheduler switches
2078 back to that pid, the timestamp difference is calculated. If the
2079 resulting latency, stored in wakeup_lat, exceeds the current
2080 maximum latency, a snapshot is taken. As part of the setup, all
2081 the scheduler events are also enabled, which are the events that
2082 will show up in the snapshot when it is taken at some point:
2083
2084 # echo 1 > /sys/kernel/debug/tracing/events/sched/enable
2085
2086 # echo 'hist:keys=pid:ts0=common_timestamp.usecs \
2087 if comm=="cyclictest"' >> \
2088 /sys/kernel/debug/tracing/events/sched/sched_waking/trigger
2089
2090 # echo 'hist:keys=next_pid:wakeup_lat=common_timestamp.usecs-$ts0: \
2091 onmax($wakeup_lat).save(next_prio,next_comm,prev_pid,prev_prio, \
2092 prev_comm):onmax($wakeup_lat).snapshot() \
2093 if next_comm=="cyclictest"' >> \
2094 /sys/kernel/debug/tracing/events/sched/sched_switch/trigger
2095
2096 When the histogram is displayed, for each bucket the max value
2097 and the saved values corresponding to the max are displayed
2098 following the rest of the fields.
2099
2100 If a snaphot was taken, there is also a message indicating that,
2101 along with the value and event that triggered the global maximum:
2102
2103 # cat /sys/kernel/debug/tracing/events/sched/sched_switch/hist
2104 { next_pid: 2101 } hitcount: 200
2105 max: 52 next_prio: 120 next_comm: cyclictest \
2106 prev_pid: 0 prev_prio: 120 prev_comm: swapper/6
2107
2108 { next_pid: 2103 } hitcount: 1326
2109 max: 572 next_prio: 19 next_comm: cyclictest \
2110 prev_pid: 0 prev_prio: 120 prev_comm: swapper/1
2111
2112 { next_pid: 2102 } hitcount: 1982 \
2113 max: 74 next_prio: 19 next_comm: cyclictest \
2114 prev_pid: 0 prev_prio: 120 prev_comm: swapper/5
2115
2116 Snapshot taken (see tracing/snapshot). Details:
2117 triggering value { onmax($wakeup_lat) }: 572 \
2118 triggered by event with key: { next_pid: 2103 }
2119
2120 Totals:
2121 Hits: 3508
2122 Entries: 3
2123 Dropped: 0
2124
2125 In the above case, the event that triggered the global maximum has
2126 the key with next_pid == 2103. If you look at the bucket that has
2127 2103 as the key, you'll find the additional values save()'d along
2128 with the local maximum for that bucket, which should be the same
2129 as the global maximum (since that was the same value that
2130 triggered the global snapshot).
2131
2132 And finally, looking at the snapshot data should show at or near
2133 the end the event that triggered the snapshot (in this case you
2134 can verify the timestamps between the sched_waking and
2135 sched_switch events, which should match the time displayed in the
2136 global maximum):
2137
2138 # cat /sys/kernel/debug/tracing/snapshot
2139
2140 <...>-2103 [005] d..3 309.873125: sched_switch: prev_comm=cyclictest prev_pid=2103 prev_prio=19 prev_state=D ==> next_comm=swapper/5 next_pid=0 next_prio=120
2141 <idle>-0 [005] d.h3 309.873611: sched_waking: comm=cyclictest pid=2102 prio=19 target_cpu=005
2142 <idle>-0 [005] dNh4 309.873613: sched_wakeup: comm=cyclictest pid=2102 prio=19 target_cpu=005
2143 <idle>-0 [005] d..3 309.873616: sched_switch: prev_comm=swapper/5 prev_pid=0 prev_prio=120 prev_state=S ==> next_comm=cyclictest next_pid=2102 next_prio=19
2144 <...>-2102 [005] d..3 309.873625: sched_switch: prev_comm=cyclictest prev_pid=2102 prev_prio=19 prev_state=D ==> next_comm=swapper/5 next_pid=0 next_prio=120
2145 <idle>-0 [005] d.h3 309.874624: sched_waking: comm=cyclictest pid=2102 prio=19 target_cpu=005
2146 <idle>-0 [005] dNh4 309.874626: sched_wakeup: comm=cyclictest pid=2102 prio=19 target_cpu=005
2147 <idle>-0 [005] dNh3 309.874628: sched_waking: comm=cyclictest pid=2103 prio=19 target_cpu=005
2148 <idle>-0 [005] dNh4 309.874630: sched_wakeup: comm=cyclictest pid=2103 prio=19 target_cpu=005
2149 <idle>-0 [005] d..3 309.874633: sched_switch: prev_comm=swapper/5 prev_pid=0 prev_prio=120 prev_state=S ==> next_comm=cyclictest next_pid=2102 next_prio=19
2150 <idle>-0 [004] d.h3 309.874757: sched_waking: comm=gnome-terminal- pid=1699 prio=120 target_cpu=004
2151 <idle>-0 [004] dNh4 309.874762: sched_wakeup: comm=gnome-terminal- pid=1699 prio=120 target_cpu=004
2152 <idle>-0 [004] d..3 309.874766: sched_switch: prev_comm=swapper/4 prev_pid=0 prev_prio=120 prev_state=S ==> next_comm=gnome-terminal- next_pid=1699 next_prio=120
2153 gnome-terminal--1699 [004] d.h2 309.874941: sched_stat_runtime: comm=gnome-terminal- pid=1699 runtime=180706 [ns] vruntime=1126870572 [ns]
2154 <idle>-0 [003] d.s4 309.874956: sched_waking: comm=rcu_sched pid=9 prio=120 target_cpu=007
2155 <idle>-0 [003] d.s5 309.874960: sched_wake_idle_without_ipi: cpu=7
2156 <idle>-0 [003] d.s5 309.874961: sched_wakeup: comm=rcu_sched pid=9 prio=120 target_cpu=007
2157 <idle>-0 [007] d..3 309.874963: sched_switch: prev_comm=swapper/7 prev_pid=0 prev_prio=120 prev_state=S ==> next_comm=rcu_sched next_pid=9 next_prio=120
2158 rcu_sched-9 [007] d..3 309.874973: sched_stat_runtime: comm=rcu_sched pid=9 runtime=13646 [ns] vruntime=22531430286 [ns]
2159 rcu_sched-9 [007] d..3 309.874978: sched_switch: prev_comm=rcu_sched prev_pid=9 prev_prio=120 prev_state=R+ ==> next_comm=swapper/7 next_pid=0 next_prio=120
2160 <...>-2102 [005] d..4 309.874994: sched_migrate_task: comm=cyclictest pid=2103 prio=19 orig_cpu=5 dest_cpu=1
2161 <...>-2102 [005] d..4 309.875185: sched_wake_idle_without_ipi: cpu=1
2162 <idle>-0 [001] d..3 309.875200: sched_switch: prev_comm=swapper/1 prev_pid=0 prev_prio=120 prev_state=S ==> next_comm=cyclictest next_pid=2103 next_prio=19
2163
ff0d35e2
TZ
2164 - onchange(var).save(field,.. .)
2165
2166 The 'onchange(var).save(field,...)' hist trigger action is invoked
2167 whenever the value of 'var' associated with a histogram entry
2168 changes.
2169
2170 The end result is that the trace event fields specified as the
2171 onchange.save() params will be saved if 'var' changes for that
2172 hist trigger entry. This allows context from the event that
2173 changed the value to be saved for later reference. When the
2174 histogram is displayed, additional fields displaying the saved
2175 values will be printed.
2176
2177 - onchange(var).snapshot()
2178
2179 The 'onchange(var).snapshot()' hist trigger action is invoked
2180 whenever the value of 'var' associated with a histogram entry
2181 changes.
2182
2183 The end result is that a global snapshot of the trace buffer will
2184 be saved in the tracing/snapshot file if 'var' changes for any
2185 hist trigger entry.
2186
2187 Note that in this case the changed value is a global variable
2188 associated withe current trace instance. The key of the specific
2189 trace event that caused the value to change and the global value
2190 itself are displayed, along with a message stating that a snapshot
2191 has been taken and where to find it. The user can use the key
2192 information displayed to locate the corresponding bucket in the
2193 histogram for even more detail.
2194
2195 As an example the below defines a hist trigger on the tcp_probe
2196 event, keyed on dport. Whenever a tcp_probe event occurs, the
2197 cwnd field is checked against the current value stored in the
2198 $cwnd variable. If the value has changed, a snapshot is taken.
2199 As part of the setup, all the scheduler and tcp events are also
2200 enabled, which are the events that will show up in the snapshot
2201 when it is taken at some point:
2202
2203 # echo 1 > /sys/kernel/debug/tracing/events/sched/enable
2204 # echo 1 > /sys/kernel/debug/tracing/events/tcp/enable
2205
2206 # echo 'hist:keys=dport:cwnd=snd_cwnd: \
2207 onchange($cwnd).save(snd_wnd,srtt,rcv_wnd): \
2208 onchange($cwnd).snapshot()' >> \
2209 /sys/kernel/debug/tracing/events/tcp/tcp_probe/trigger
2210
2211 When the histogram is displayed, for each bucket the tracked value
2212 and the saved values corresponding to that value are displayed
2213 following the rest of the fields.
2214
2215 If a snaphot was taken, there is also a message indicating that,
2216 along with the value and event that triggered the snapshot:
2217
2218 # cat /sys/kernel/debug/tracing/events/tcp/tcp_probe/hist
2219 { dport: 1521 } hitcount: 8
2220 changed: 10 snd_wnd: 35456 srtt: 154262 rcv_wnd: 42112
2221
2222 { dport: 80 } hitcount: 23
2223 changed: 10 snd_wnd: 28960 srtt: 19604 rcv_wnd: 29312
2224
2225 { dport: 9001 } hitcount: 172
2226 changed: 10 snd_wnd: 48384 srtt: 260444 rcv_wnd: 55168
2227
2228 { dport: 443 } hitcount: 211
2229 changed: 10 snd_wnd: 26960 srtt: 17379 rcv_wnd: 28800
2230
2231 Snapshot taken (see tracing/snapshot). Details:
2232 triggering value { onchange($cwnd) }: 10
2233 triggered by event with key: { dport: 80 }
2234
2235 Totals:
2236 Hits: 414
2237 Entries: 4
2238 Dropped: 0
2239
2240 In the above case, the event that triggered the snapshot has the
2241 key with dport == 80. If you look at the bucket that has 80 as
2242 the key, you'll find the additional values save()'d along with the
2243 changed value for that bucket, which should be the same as the
2244 global changed value (since that was the same value that triggered
2245 the global snapshot).
2246
2247 And finally, looking at the snapshot data should show at or near
2248 the end the event that triggered the snapshot:
2249
2250 # cat /sys/kernel/debug/tracing/snapshot
2251
2252 gnome-shell-1261 [006] dN.3 49.823113: sched_stat_runtime: comm=gnome-shell pid=1261 runtime=49347 [ns] vruntime=1835730389 [ns]
2253 kworker/u16:4-773 [003] d..3 49.823114: sched_switch: prev_comm=kworker/u16:4 prev_pid=773 prev_prio=120 prev_state=R+ ==> next_comm=kworker/3:2 next_pid=135 next_prio=120
2254 gnome-shell-1261 [006] d..3 49.823114: sched_switch: prev_comm=gnome-shell prev_pid=1261 prev_prio=120 prev_state=R+ ==> next_comm=kworker/6:2 next_pid=387 next_prio=120
2255 kworker/3:2-135 [003] d..3 49.823118: sched_stat_runtime: comm=kworker/3:2 pid=135 runtime=5339 [ns] vruntime=17815800388 [ns]
2256 kworker/6:2-387 [006] d..3 49.823120: sched_stat_runtime: comm=kworker/6:2 pid=387 runtime=9594 [ns] vruntime=14589605367 [ns]
2257 kworker/6:2-387 [006] d..3 49.823122: sched_switch: prev_comm=kworker/6:2 prev_pid=387 prev_prio=120 prev_state=R+ ==> next_comm=gnome-shell next_pid=1261 next_prio=120
2258 kworker/3:2-135 [003] d..3 49.823123: sched_switch: prev_comm=kworker/3:2 prev_pid=135 prev_prio=120 prev_state=T ==> next_comm=swapper/3 next_pid=0 next_prio=120
2259 <idle>-0 [004] ..s7 49.823798: tcp_probe: src=10.0.0.10:54326 dest=23.215.104.193:80 mark=0x0 length=32 snd_nxt=0xe3ae2ff5 snd_una=0xe3ae2ecd snd_cwnd=10 ssthresh=2147483647 snd_wnd=28960 srtt=19604 rcv_wnd=29312
2260
d3439f9d
SRV
22613. User space creating a trigger
2262--------------------------------
2263
2264Writing into /sys/kernel/tracing/trace_marker writes into the ftrace
2265ring buffer. This can also act like an event, by writing into the trigger
2266file located in /sys/kernel/tracing/events/ftrace/print/
2267
2268Modifying cyclictest to write into the trace_marker file before it sleeps
ea272257 2269and after it wakes up, something like this::
d3439f9d 2270
ea272257
MCC
2271 static void traceputs(char *str)
2272 {
d3439f9d
SRV
2273 /* tracemark_fd is the trace_marker file descriptor */
2274 if (tracemark_fd < 0)
2275 return;
2276 /* write the tracemark message */
2277 write(tracemark_fd, str, strlen(str));
ea272257 2278 }
d3439f9d 2279
ea272257 2280And later add something like::
d3439f9d
SRV
2281
2282 traceputs("start");
2283 clock_nanosleep(...);
2284 traceputs("end");
2285
ea272257 2286We can make a histogram from this::
d3439f9d
SRV
2287
2288 # cd /sys/kernel/tracing
2289 # echo 'latency u64 lat' > synthetic_events
2290 # echo 'hist:keys=common_pid:ts0=common_timestamp.usecs if buf == "start"' > events/ftrace/print/trigger
2291 # echo 'hist:keys=common_pid:lat=common_timestamp.usecs-$ts0:onmatch(ftrace.print).latency($lat) if buf == "end"' >> events/ftrace/print/trigger
2292 # echo 'hist:keys=lat,common_pid:sort=lat' > events/synthetic/latency/trigger
2293
2294The above created a synthetic event called "latency" and two histograms
2295against the trace_marker, one gets triggered when "start" is written into the
2296trace_marker file and the other when "end" is written. If the pids match, then
2297it will call the "latency" synthetic event with the calculated latency as its
2298parameter. Finally, a histogram is added to the latency synthetic event to
2299record the calculated latency along with the pid.
2300
ea272257 2301Now running cyclictest with::
d3439f9d
SRV
2302
2303 # ./cyclictest -p80 -d0 -i250 -n -a -t --tracemark -b 1000
2304
2305 -p80 : run threads at priority 80
2306 -d0 : have all threads run at the same interval
2307 -i250 : start the interval at 250 microseconds (all threads will do this)
2308 -n : sleep with nanosleep
2309 -a : affine all threads to a separate CPU
2310 -t : one thread per available CPU
2311 --tracemark : enable trace mark writing
2312 -b 1000 : stop if any latency is greater than 1000 microseconds
2313
2314Note, the -b 1000 is used just to make --tracemark available.
2315
ea272257 2316Then we can see the histogram created by this with::
d3439f9d
SRV
2317
2318 # cat events/synthetic/latency/hist
ea272257
MCC
2319 # event histogram
2320 #
2321 # trigger info: hist:keys=lat,common_pid:vals=hitcount:sort=lat:size=2048 [active]
2322 #
2323
2324 { lat: 107, common_pid: 2039 } hitcount: 1
2325 { lat: 122, common_pid: 2041 } hitcount: 1
2326 { lat: 166, common_pid: 2039 } hitcount: 1
2327 { lat: 174, common_pid: 2039 } hitcount: 1
2328 { lat: 194, common_pid: 2041 } hitcount: 1
2329 { lat: 196, common_pid: 2036 } hitcount: 1
2330 { lat: 197, common_pid: 2038 } hitcount: 1
2331 { lat: 198, common_pid: 2039 } hitcount: 1
2332 { lat: 199, common_pid: 2039 } hitcount: 1
2333 { lat: 200, common_pid: 2041 } hitcount: 1
2334 { lat: 201, common_pid: 2039 } hitcount: 2
2335 { lat: 202, common_pid: 2038 } hitcount: 1
2336 { lat: 202, common_pid: 2043 } hitcount: 1
2337 { lat: 203, common_pid: 2039 } hitcount: 1
2338 { lat: 203, common_pid: 2036 } hitcount: 1
2339 { lat: 203, common_pid: 2041 } hitcount: 1
2340 { lat: 206, common_pid: 2038 } hitcount: 2
2341 { lat: 207, common_pid: 2039 } hitcount: 1
2342 { lat: 207, common_pid: 2036 } hitcount: 1
2343 { lat: 208, common_pid: 2040 } hitcount: 1
2344 { lat: 209, common_pid: 2043 } hitcount: 1
2345 { lat: 210, common_pid: 2039 } hitcount: 1
2346 { lat: 211, common_pid: 2039 } hitcount: 4
2347 { lat: 212, common_pid: 2043 } hitcount: 1
2348 { lat: 212, common_pid: 2039 } hitcount: 2
2349 { lat: 213, common_pid: 2039 } hitcount: 1
2350 { lat: 214, common_pid: 2038 } hitcount: 1
2351 { lat: 214, common_pid: 2039 } hitcount: 2
2352 { lat: 214, common_pid: 2042 } hitcount: 1
2353 { lat: 215, common_pid: 2039 } hitcount: 1
2354 { lat: 217, common_pid: 2036 } hitcount: 1
2355 { lat: 217, common_pid: 2040 } hitcount: 1
2356 { lat: 217, common_pid: 2039 } hitcount: 1
2357 { lat: 218, common_pid: 2039 } hitcount: 6
2358 { lat: 219, common_pid: 2039 } hitcount: 9
2359 { lat: 220, common_pid: 2039 } hitcount: 11
2360 { lat: 221, common_pid: 2039 } hitcount: 5
2361 { lat: 221, common_pid: 2042 } hitcount: 1
2362 { lat: 222, common_pid: 2039 } hitcount: 7
2363 { lat: 223, common_pid: 2036 } hitcount: 1
2364 { lat: 223, common_pid: 2039 } hitcount: 3
2365 { lat: 224, common_pid: 2039 } hitcount: 4
2366 { lat: 224, common_pid: 2037 } hitcount: 1
2367 { lat: 224, common_pid: 2036 } hitcount: 2
2368 { lat: 225, common_pid: 2039 } hitcount: 5
2369 { lat: 225, common_pid: 2042 } hitcount: 1
2370 { lat: 226, common_pid: 2039 } hitcount: 7
2371 { lat: 226, common_pid: 2036 } hitcount: 4
2372 { lat: 227, common_pid: 2039 } hitcount: 6
2373 { lat: 227, common_pid: 2036 } hitcount: 12
2374 { lat: 227, common_pid: 2043 } hitcount: 1
2375 { lat: 228, common_pid: 2039 } hitcount: 7
2376 { lat: 228, common_pid: 2036 } hitcount: 14
2377 { lat: 229, common_pid: 2039 } hitcount: 9
2378 { lat: 229, common_pid: 2036 } hitcount: 8
2379 { lat: 229, common_pid: 2038 } hitcount: 1
2380 { lat: 230, common_pid: 2039 } hitcount: 11
2381 { lat: 230, common_pid: 2036 } hitcount: 6
2382 { lat: 230, common_pid: 2043 } hitcount: 1
2383 { lat: 230, common_pid: 2042 } hitcount: 2
2384 { lat: 231, common_pid: 2041 } hitcount: 1
2385 { lat: 231, common_pid: 2036 } hitcount: 6
2386 { lat: 231, common_pid: 2043 } hitcount: 1
2387 { lat: 231, common_pid: 2039 } hitcount: 8
2388 { lat: 232, common_pid: 2037 } hitcount: 1
2389 { lat: 232, common_pid: 2039 } hitcount: 6
2390 { lat: 232, common_pid: 2040 } hitcount: 2
2391 { lat: 232, common_pid: 2036 } hitcount: 5
2392 { lat: 232, common_pid: 2043 } hitcount: 1
2393 { lat: 233, common_pid: 2036 } hitcount: 5
2394 { lat: 233, common_pid: 2039 } hitcount: 11
2395 { lat: 234, common_pid: 2039 } hitcount: 4
2396 { lat: 234, common_pid: 2038 } hitcount: 2
2397 { lat: 234, common_pid: 2043 } hitcount: 2
2398 { lat: 234, common_pid: 2036 } hitcount: 11
2399 { lat: 234, common_pid: 2040 } hitcount: 1
2400 { lat: 235, common_pid: 2037 } hitcount: 2
2401 { lat: 235, common_pid: 2036 } hitcount: 8
2402 { lat: 235, common_pid: 2043 } hitcount: 2
2403 { lat: 235, common_pid: 2039 } hitcount: 5
2404 { lat: 235, common_pid: 2042 } hitcount: 2
2405 { lat: 235, common_pid: 2040 } hitcount: 4
2406 { lat: 235, common_pid: 2041 } hitcount: 1
2407 { lat: 236, common_pid: 2036 } hitcount: 7
2408 { lat: 236, common_pid: 2037 } hitcount: 1
2409 { lat: 236, common_pid: 2041 } hitcount: 5
2410 { lat: 236, common_pid: 2039 } hitcount: 3
2411 { lat: 236, common_pid: 2043 } hitcount: 9
2412 { lat: 236, common_pid: 2040 } hitcount: 7
2413 { lat: 237, common_pid: 2037 } hitcount: 1
2414 { lat: 237, common_pid: 2040 } hitcount: 1
2415 { lat: 237, common_pid: 2036 } hitcount: 9
2416 { lat: 237, common_pid: 2039 } hitcount: 3
2417 { lat: 237, common_pid: 2043 } hitcount: 8
2418 { lat: 237, common_pid: 2042 } hitcount: 2
2419 { lat: 237, common_pid: 2041 } hitcount: 2
2420 { lat: 238, common_pid: 2043 } hitcount: 10
2421 { lat: 238, common_pid: 2040 } hitcount: 1
2422 { lat: 238, common_pid: 2037 } hitcount: 9
2423 { lat: 238, common_pid: 2038 } hitcount: 1
2424 { lat: 238, common_pid: 2039 } hitcount: 1
2425 { lat: 238, common_pid: 2042 } hitcount: 3
2426 { lat: 238, common_pid: 2036 } hitcount: 7
2427 { lat: 239, common_pid: 2041 } hitcount: 1
2428 { lat: 239, common_pid: 2043 } hitcount: 11
2429 { lat: 239, common_pid: 2037 } hitcount: 11
2430 { lat: 239, common_pid: 2038 } hitcount: 6
2431 { lat: 239, common_pid: 2036 } hitcount: 7
2432 { lat: 239, common_pid: 2040 } hitcount: 1
2433 { lat: 239, common_pid: 2042 } hitcount: 9
2434 { lat: 240, common_pid: 2037 } hitcount: 29
2435 { lat: 240, common_pid: 2043 } hitcount: 15
2436 { lat: 240, common_pid: 2040 } hitcount: 44
2437 { lat: 240, common_pid: 2039 } hitcount: 1
2438 { lat: 240, common_pid: 2041 } hitcount: 2
2439 { lat: 240, common_pid: 2038 } hitcount: 1
2440 { lat: 240, common_pid: 2036 } hitcount: 10
2441 { lat: 240, common_pid: 2042 } hitcount: 13
2442 { lat: 241, common_pid: 2036 } hitcount: 21
2443 { lat: 241, common_pid: 2041 } hitcount: 36
2444 { lat: 241, common_pid: 2037 } hitcount: 34
2445 { lat: 241, common_pid: 2042 } hitcount: 14
2446 { lat: 241, common_pid: 2040 } hitcount: 94
2447 { lat: 241, common_pid: 2039 } hitcount: 12
2448 { lat: 241, common_pid: 2038 } hitcount: 2
2449 { lat: 241, common_pid: 2043 } hitcount: 28
2450 { lat: 242, common_pid: 2040 } hitcount: 109
2451 { lat: 242, common_pid: 2041 } hitcount: 506
2452 { lat: 242, common_pid: 2039 } hitcount: 155
2453 { lat: 242, common_pid: 2042 } hitcount: 21
2454 { lat: 242, common_pid: 2037 } hitcount: 52
2455 { lat: 242, common_pid: 2043 } hitcount: 21
2456 { lat: 242, common_pid: 2036 } hitcount: 16
2457 { lat: 242, common_pid: 2038 } hitcount: 156
2458 { lat: 243, common_pid: 2037 } hitcount: 46
2459 { lat: 243, common_pid: 2039 } hitcount: 40
2460 { lat: 243, common_pid: 2042 } hitcount: 119
2461 { lat: 243, common_pid: 2041 } hitcount: 611
2462 { lat: 243, common_pid: 2036 } hitcount: 69
2463 { lat: 243, common_pid: 2038 } hitcount: 784
2464 { lat: 243, common_pid: 2040 } hitcount: 323
2465 { lat: 243, common_pid: 2043 } hitcount: 14
2466 { lat: 244, common_pid: 2043 } hitcount: 35
2467 { lat: 244, common_pid: 2042 } hitcount: 305
2468 { lat: 244, common_pid: 2039 } hitcount: 8
2469 { lat: 244, common_pid: 2040 } hitcount: 4515
2470 { lat: 244, common_pid: 2038 } hitcount: 371
2471 { lat: 244, common_pid: 2037 } hitcount: 31
2472 { lat: 244, common_pid: 2036 } hitcount: 114
2473 { lat: 244, common_pid: 2041 } hitcount: 3396
2474 { lat: 245, common_pid: 2036 } hitcount: 700
2475 { lat: 245, common_pid: 2041 } hitcount: 2772
2476 { lat: 245, common_pid: 2037 } hitcount: 268
2477 { lat: 245, common_pid: 2039 } hitcount: 472
2478 { lat: 245, common_pid: 2038 } hitcount: 2758
2479 { lat: 245, common_pid: 2042 } hitcount: 3833
2480 { lat: 245, common_pid: 2040 } hitcount: 3105
2481 { lat: 245, common_pid: 2043 } hitcount: 645
2482 { lat: 246, common_pid: 2038 } hitcount: 3451
2483 { lat: 246, common_pid: 2041 } hitcount: 142
2484 { lat: 246, common_pid: 2037 } hitcount: 5101
2485 { lat: 246, common_pid: 2040 } hitcount: 68
2486 { lat: 246, common_pid: 2043 } hitcount: 5099
2487 { lat: 246, common_pid: 2039 } hitcount: 5608
2488 { lat: 246, common_pid: 2042 } hitcount: 3723
2489 { lat: 246, common_pid: 2036 } hitcount: 4738
2490 { lat: 247, common_pid: 2042 } hitcount: 312
2491 { lat: 247, common_pid: 2043 } hitcount: 2385
2492 { lat: 247, common_pid: 2041 } hitcount: 452
2493 { lat: 247, common_pid: 2038 } hitcount: 792
2494 { lat: 247, common_pid: 2040 } hitcount: 78
2495 { lat: 247, common_pid: 2036 } hitcount: 2375
2496 { lat: 247, common_pid: 2039 } hitcount: 1834
2497 { lat: 247, common_pid: 2037 } hitcount: 2655
2498 { lat: 248, common_pid: 2037 } hitcount: 36
2499 { lat: 248, common_pid: 2042 } hitcount: 11
2500 { lat: 248, common_pid: 2038 } hitcount: 122
2501 { lat: 248, common_pid: 2036 } hitcount: 135
2502 { lat: 248, common_pid: 2039 } hitcount: 26
2503 { lat: 248, common_pid: 2041 } hitcount: 503
2504 { lat: 248, common_pid: 2043 } hitcount: 66
2505 { lat: 248, common_pid: 2040 } hitcount: 46
2506 { lat: 249, common_pid: 2037 } hitcount: 29
2507 { lat: 249, common_pid: 2038 } hitcount: 1
2508 { lat: 249, common_pid: 2043 } hitcount: 29
2509 { lat: 249, common_pid: 2039 } hitcount: 8
2510 { lat: 249, common_pid: 2042 } hitcount: 56
2511 { lat: 249, common_pid: 2040 } hitcount: 27
2512 { lat: 249, common_pid: 2041 } hitcount: 11
2513 { lat: 249, common_pid: 2036 } hitcount: 27
2514 { lat: 250, common_pid: 2038 } hitcount: 1
2515 { lat: 250, common_pid: 2036 } hitcount: 30
2516 { lat: 250, common_pid: 2040 } hitcount: 19
2517 { lat: 250, common_pid: 2043 } hitcount: 22
2518 { lat: 250, common_pid: 2042 } hitcount: 20
2519 { lat: 250, common_pid: 2041 } hitcount: 1
2520 { lat: 250, common_pid: 2039 } hitcount: 6
2521 { lat: 250, common_pid: 2037 } hitcount: 48
2522 { lat: 251, common_pid: 2037 } hitcount: 43
2523 { lat: 251, common_pid: 2039 } hitcount: 1
2524 { lat: 251, common_pid: 2036 } hitcount: 12
2525 { lat: 251, common_pid: 2042 } hitcount: 2
2526 { lat: 251, common_pid: 2041 } hitcount: 1
2527 { lat: 251, common_pid: 2043 } hitcount: 15
2528 { lat: 251, common_pid: 2040 } hitcount: 3
2529 { lat: 252, common_pid: 2040 } hitcount: 1
2530 { lat: 252, common_pid: 2036 } hitcount: 12
2531 { lat: 252, common_pid: 2037 } hitcount: 21
2532 { lat: 252, common_pid: 2043 } hitcount: 14
2533 { lat: 253, common_pid: 2037 } hitcount: 21
2534 { lat: 253, common_pid: 2039 } hitcount: 2
2535 { lat: 253, common_pid: 2036 } hitcount: 9
2536 { lat: 253, common_pid: 2043 } hitcount: 6
2537 { lat: 253, common_pid: 2040 } hitcount: 1
2538 { lat: 254, common_pid: 2036 } hitcount: 8
2539 { lat: 254, common_pid: 2043 } hitcount: 3
2540 { lat: 254, common_pid: 2041 } hitcount: 1
2541 { lat: 254, common_pid: 2042 } hitcount: 1
2542 { lat: 254, common_pid: 2039 } hitcount: 1
2543 { lat: 254, common_pid: 2037 } hitcount: 12
2544 { lat: 255, common_pid: 2043 } hitcount: 1
2545 { lat: 255, common_pid: 2037 } hitcount: 2
2546 { lat: 255, common_pid: 2036 } hitcount: 2
2547 { lat: 255, common_pid: 2039 } hitcount: 8
2548 { lat: 256, common_pid: 2043 } hitcount: 1
2549 { lat: 256, common_pid: 2036 } hitcount: 4
2550 { lat: 256, common_pid: 2039 } hitcount: 6
2551 { lat: 257, common_pid: 2039 } hitcount: 5
2552 { lat: 257, common_pid: 2036 } hitcount: 4
2553 { lat: 258, common_pid: 2039 } hitcount: 5
2554 { lat: 258, common_pid: 2036 } hitcount: 2
2555 { lat: 259, common_pid: 2036 } hitcount: 7
2556 { lat: 259, common_pid: 2039 } hitcount: 7
2557 { lat: 260, common_pid: 2036 } hitcount: 8
2558 { lat: 260, common_pid: 2039 } hitcount: 6
2559 { lat: 261, common_pid: 2036 } hitcount: 5
2560 { lat: 261, common_pid: 2039 } hitcount: 7
2561 { lat: 262, common_pid: 2039 } hitcount: 5
2562 { lat: 262, common_pid: 2036 } hitcount: 5
2563 { lat: 263, common_pid: 2039 } hitcount: 7
2564 { lat: 263, common_pid: 2036 } hitcount: 7
2565 { lat: 264, common_pid: 2039 } hitcount: 9
2566 { lat: 264, common_pid: 2036 } hitcount: 9
2567 { lat: 265, common_pid: 2036 } hitcount: 5
2568 { lat: 265, common_pid: 2039 } hitcount: 1
2569 { lat: 266, common_pid: 2036 } hitcount: 1
2570 { lat: 266, common_pid: 2039 } hitcount: 3
2571 { lat: 267, common_pid: 2036 } hitcount: 1
2572 { lat: 267, common_pid: 2039 } hitcount: 3
2573 { lat: 268, common_pid: 2036 } hitcount: 1
2574 { lat: 268, common_pid: 2039 } hitcount: 6
2575 { lat: 269, common_pid: 2036 } hitcount: 1
2576 { lat: 269, common_pid: 2043 } hitcount: 1
2577 { lat: 269, common_pid: 2039 } hitcount: 2
2578 { lat: 270, common_pid: 2040 } hitcount: 1
2579 { lat: 270, common_pid: 2039 } hitcount: 6
2580 { lat: 271, common_pid: 2041 } hitcount: 1
2581 { lat: 271, common_pid: 2039 } hitcount: 5
2582 { lat: 272, common_pid: 2039 } hitcount: 10
2583 { lat: 273, common_pid: 2039 } hitcount: 8
2584 { lat: 274, common_pid: 2039 } hitcount: 2
2585 { lat: 275, common_pid: 2039 } hitcount: 1
2586 { lat: 276, common_pid: 2039 } hitcount: 2
2587 { lat: 276, common_pid: 2037 } hitcount: 1
2588 { lat: 276, common_pid: 2038 } hitcount: 1
2589 { lat: 277, common_pid: 2039 } hitcount: 1
2590 { lat: 277, common_pid: 2042 } hitcount: 1
2591 { lat: 278, common_pid: 2039 } hitcount: 1
2592 { lat: 279, common_pid: 2039 } hitcount: 4
2593 { lat: 279, common_pid: 2043 } hitcount: 1
2594 { lat: 280, common_pid: 2039 } hitcount: 3
2595 { lat: 283, common_pid: 2036 } hitcount: 2
2596 { lat: 284, common_pid: 2039 } hitcount: 1
2597 { lat: 284, common_pid: 2043 } hitcount: 1
2598 { lat: 288, common_pid: 2039 } hitcount: 1
2599 { lat: 289, common_pid: 2039 } hitcount: 1
2600 { lat: 300, common_pid: 2039 } hitcount: 1
2601 { lat: 384, common_pid: 2039 } hitcount: 1
2602
2603 Totals:
2604 Hits: 67625
2605 Entries: 278
2606 Dropped: 0
d3439f9d
SRV
2607
2608Note, the writes are around the sleep, so ideally they will all be of 250
2609microseconds. If you are wondering how there are several that are under
2610250 microseconds, that is because the way cyclictest works, is if one
2611iteration comes in late, the next one will set the timer to wake up less that
2612250. That is, if an iteration came in 50 microseconds late, the next wake up
2613will be at 200 microseconds.
2614
2615But this could easily be done in userspace. To make this even more
2616interesting, we can mix the histogram between events that happened in the
ea272257 2617kernel with trace_marker::
d3439f9d
SRV
2618
2619 # cd /sys/kernel/tracing
2620 # echo 'latency u64 lat' > synthetic_events
2621 # echo 'hist:keys=pid:ts0=common_timestamp.usecs' > events/sched/sched_waking/trigger
2622 # echo 'hist:keys=common_pid:lat=common_timestamp.usecs-$ts0:onmatch(sched.sched_waking).latency($lat) if buf == "end"' > events/ftrace/print/trigger
2623 # echo 'hist:keys=lat,common_pid:sort=lat' > events/synthetic/latency/trigger
2624
2625The difference this time is that instead of using the trace_marker to start
2626the latency, the sched_waking event is used, matching the common_pid for the
2627trace_marker write with the pid that is being woken by sched_waking.
2628
ea272257 2629After running cyclictest again with the same parameters, we now have::
d3439f9d
SRV
2630
2631 # cat events/synthetic/latency/hist
ea272257
MCC
2632 # event histogram
2633 #
2634 # trigger info: hist:keys=lat,common_pid:vals=hitcount:sort=lat:size=2048 [active]
2635 #
2636
2637 { lat: 7, common_pid: 2302 } hitcount: 640
2638 { lat: 7, common_pid: 2299 } hitcount: 42
2639 { lat: 7, common_pid: 2303 } hitcount: 18
2640 { lat: 7, common_pid: 2305 } hitcount: 166
2641 { lat: 7, common_pid: 2306 } hitcount: 1
2642 { lat: 7, common_pid: 2301 } hitcount: 91
2643 { lat: 7, common_pid: 2300 } hitcount: 17
2644 { lat: 8, common_pid: 2303 } hitcount: 8296
2645 { lat: 8, common_pid: 2304 } hitcount: 6864
2646 { lat: 8, common_pid: 2305 } hitcount: 9464
2647 { lat: 8, common_pid: 2301 } hitcount: 9213
2648 { lat: 8, common_pid: 2306 } hitcount: 6246
2649 { lat: 8, common_pid: 2302 } hitcount: 8797
2650 { lat: 8, common_pid: 2299 } hitcount: 8771
2651 { lat: 8, common_pid: 2300 } hitcount: 8119
2652 { lat: 9, common_pid: 2305 } hitcount: 1519
2653 { lat: 9, common_pid: 2299 } hitcount: 2346
2654 { lat: 9, common_pid: 2303 } hitcount: 2841
2655 { lat: 9, common_pid: 2301 } hitcount: 1846
2656 { lat: 9, common_pid: 2304 } hitcount: 3861
2657 { lat: 9, common_pid: 2302 } hitcount: 1210
2658 { lat: 9, common_pid: 2300 } hitcount: 2762
2659 { lat: 9, common_pid: 2306 } hitcount: 4247
2660 { lat: 10, common_pid: 2299 } hitcount: 16
2661 { lat: 10, common_pid: 2306 } hitcount: 333
2662 { lat: 10, common_pid: 2303 } hitcount: 16
2663 { lat: 10, common_pid: 2304 } hitcount: 168
2664 { lat: 10, common_pid: 2302 } hitcount: 240
2665 { lat: 10, common_pid: 2301 } hitcount: 28
2666 { lat: 10, common_pid: 2300 } hitcount: 95
2667 { lat: 10, common_pid: 2305 } hitcount: 18
2668 { lat: 11, common_pid: 2303 } hitcount: 5
2669 { lat: 11, common_pid: 2305 } hitcount: 8
2670 { lat: 11, common_pid: 2306 } hitcount: 221
2671 { lat: 11, common_pid: 2302 } hitcount: 76
2672 { lat: 11, common_pid: 2304 } hitcount: 26
2673 { lat: 11, common_pid: 2300 } hitcount: 125
2674 { lat: 11, common_pid: 2299 } hitcount: 2
2675 { lat: 12, common_pid: 2305 } hitcount: 3
2676 { lat: 12, common_pid: 2300 } hitcount: 6
2677 { lat: 12, common_pid: 2306 } hitcount: 90
2678 { lat: 12, common_pid: 2302 } hitcount: 4
2679 { lat: 12, common_pid: 2303 } hitcount: 1
2680 { lat: 12, common_pid: 2304 } hitcount: 122
2681 { lat: 13, common_pid: 2300 } hitcount: 12
2682 { lat: 13, common_pid: 2301 } hitcount: 1
2683 { lat: 13, common_pid: 2306 } hitcount: 32
2684 { lat: 13, common_pid: 2302 } hitcount: 5
2685 { lat: 13, common_pid: 2305 } hitcount: 1
2686 { lat: 13, common_pid: 2303 } hitcount: 1
2687 { lat: 13, common_pid: 2304 } hitcount: 61
2688 { lat: 14, common_pid: 2303 } hitcount: 4
2689 { lat: 14, common_pid: 2306 } hitcount: 5
2690 { lat: 14, common_pid: 2305 } hitcount: 4
2691 { lat: 14, common_pid: 2304 } hitcount: 62
2692 { lat: 14, common_pid: 2302 } hitcount: 19
2693 { lat: 14, common_pid: 2300 } hitcount: 33
2694 { lat: 14, common_pid: 2299 } hitcount: 1
2695 { lat: 14, common_pid: 2301 } hitcount: 4
2696 { lat: 15, common_pid: 2305 } hitcount: 1
2697 { lat: 15, common_pid: 2302 } hitcount: 25
2698 { lat: 15, common_pid: 2300 } hitcount: 11
2699 { lat: 15, common_pid: 2299 } hitcount: 5
2700 { lat: 15, common_pid: 2301 } hitcount: 1
2701 { lat: 15, common_pid: 2304 } hitcount: 8
2702 { lat: 15, common_pid: 2303 } hitcount: 1
2703 { lat: 15, common_pid: 2306 } hitcount: 6
2704 { lat: 16, common_pid: 2302 } hitcount: 31
2705 { lat: 16, common_pid: 2306 } hitcount: 3
2706 { lat: 16, common_pid: 2300 } hitcount: 5
2707 { lat: 17, common_pid: 2302 } hitcount: 6
2708 { lat: 17, common_pid: 2303 } hitcount: 1
2709 { lat: 18, common_pid: 2304 } hitcount: 1
2710 { lat: 18, common_pid: 2302 } hitcount: 8
2711 { lat: 18, common_pid: 2299 } hitcount: 1
2712 { lat: 18, common_pid: 2301 } hitcount: 1
2713 { lat: 19, common_pid: 2303 } hitcount: 4
2714 { lat: 19, common_pid: 2304 } hitcount: 5
2715 { lat: 19, common_pid: 2302 } hitcount: 4
2716 { lat: 19, common_pid: 2299 } hitcount: 3
2717 { lat: 19, common_pid: 2306 } hitcount: 1
2718 { lat: 19, common_pid: 2300 } hitcount: 4
2719 { lat: 19, common_pid: 2305 } hitcount: 5
2720 { lat: 20, common_pid: 2299 } hitcount: 2
2721 { lat: 20, common_pid: 2302 } hitcount: 3
2722 { lat: 20, common_pid: 2305 } hitcount: 1
2723 { lat: 20, common_pid: 2300 } hitcount: 2
2724 { lat: 20, common_pid: 2301 } hitcount: 2
2725 { lat: 20, common_pid: 2303 } hitcount: 3
2726 { lat: 21, common_pid: 2305 } hitcount: 1
2727 { lat: 21, common_pid: 2299 } hitcount: 5
2728 { lat: 21, common_pid: 2303 } hitcount: 4
2729 { lat: 21, common_pid: 2302 } hitcount: 7
2730 { lat: 21, common_pid: 2300 } hitcount: 1
2731 { lat: 21, common_pid: 2301 } hitcount: 5
2732 { lat: 21, common_pid: 2304 } hitcount: 2
2733 { lat: 22, common_pid: 2302 } hitcount: 5
2734 { lat: 22, common_pid: 2303 } hitcount: 1
2735 { lat: 22, common_pid: 2306 } hitcount: 3
2736 { lat: 22, common_pid: 2301 } hitcount: 2
2737 { lat: 22, common_pid: 2300 } hitcount: 1
2738 { lat: 22, common_pid: 2299 } hitcount: 1
2739 { lat: 22, common_pid: 2305 } hitcount: 1
2740 { lat: 22, common_pid: 2304 } hitcount: 1
2741 { lat: 23, common_pid: 2299 } hitcount: 1
2742 { lat: 23, common_pid: 2306 } hitcount: 2
2743 { lat: 23, common_pid: 2302 } hitcount: 6
2744 { lat: 24, common_pid: 2302 } hitcount: 3
2745 { lat: 24, common_pid: 2300 } hitcount: 1
2746 { lat: 24, common_pid: 2306 } hitcount: 2
2747 { lat: 24, common_pid: 2305 } hitcount: 1
2748 { lat: 24, common_pid: 2299 } hitcount: 1
2749 { lat: 25, common_pid: 2300 } hitcount: 1
2750 { lat: 25, common_pid: 2302 } hitcount: 4
2751 { lat: 26, common_pid: 2302 } hitcount: 2
2752 { lat: 27, common_pid: 2305 } hitcount: 1
2753 { lat: 27, common_pid: 2300 } hitcount: 1
2754 { lat: 27, common_pid: 2302 } hitcount: 3
2755 { lat: 28, common_pid: 2306 } hitcount: 1
2756 { lat: 28, common_pid: 2302 } hitcount: 4
2757 { lat: 29, common_pid: 2302 } hitcount: 1
2758 { lat: 29, common_pid: 2300 } hitcount: 2
2759 { lat: 29, common_pid: 2306 } hitcount: 1
2760 { lat: 29, common_pid: 2304 } hitcount: 1
2761 { lat: 30, common_pid: 2302 } hitcount: 4
2762 { lat: 31, common_pid: 2302 } hitcount: 6
2763 { lat: 32, common_pid: 2302 } hitcount: 1
2764 { lat: 33, common_pid: 2299 } hitcount: 1
2765 { lat: 33, common_pid: 2302 } hitcount: 3
2766 { lat: 34, common_pid: 2302 } hitcount: 2
2767 { lat: 35, common_pid: 2302 } hitcount: 1
2768 { lat: 35, common_pid: 2304 } hitcount: 1
2769 { lat: 36, common_pid: 2302 } hitcount: 4
2770 { lat: 37, common_pid: 2302 } hitcount: 6
2771 { lat: 38, common_pid: 2302 } hitcount: 2
2772 { lat: 39, common_pid: 2302 } hitcount: 2
2773 { lat: 39, common_pid: 2304 } hitcount: 1
2774 { lat: 40, common_pid: 2304 } hitcount: 2
2775 { lat: 40, common_pid: 2302 } hitcount: 5
2776 { lat: 41, common_pid: 2304 } hitcount: 1
2777 { lat: 41, common_pid: 2302 } hitcount: 8
2778 { lat: 42, common_pid: 2302 } hitcount: 6
2779 { lat: 42, common_pid: 2304 } hitcount: 1
2780 { lat: 43, common_pid: 2302 } hitcount: 3
2781 { lat: 43, common_pid: 2304 } hitcount: 4
2782 { lat: 44, common_pid: 2302 } hitcount: 6
2783 { lat: 45, common_pid: 2302 } hitcount: 5
2784 { lat: 46, common_pid: 2302 } hitcount: 5
2785 { lat: 47, common_pid: 2302 } hitcount: 7
2786 { lat: 48, common_pid: 2301 } hitcount: 1
2787 { lat: 48, common_pid: 2302 } hitcount: 9
2788 { lat: 49, common_pid: 2302 } hitcount: 3
2789 { lat: 50, common_pid: 2302 } hitcount: 1
2790 { lat: 50, common_pid: 2301 } hitcount: 1
2791 { lat: 51, common_pid: 2302 } hitcount: 2
2792 { lat: 51, common_pid: 2301 } hitcount: 1
2793 { lat: 61, common_pid: 2302 } hitcount: 1
2794 { lat: 110, common_pid: 2302 } hitcount: 1
2795
2796 Totals:
2797 Hits: 89565
2798 Entries: 158
2799 Dropped: 0
d3439f9d
SRV
2800
2801This doesn't tell us any information about how late cyclictest may have
2802woken up, but it does show us a nice histogram of how long it took from
2803the time that cyclictest was woken to the time it made it into user space.