histogram.rst (157241B)
1================ 2Event Histograms 3================ 4 5Documentation written by Tom Zanussi 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 12 trace events and event triggers, see Documentation/trace/events.rst. 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 24 The format of a hist trigger is as follows:: 25 26 hist:keys=<field1[,field2,...]>[:values=<field1[,field2,...]>] 27 [:sort=<field1[,field2,...]>][:size=#entries][:pause][:continue] 28 [:clear][:name=histname1][:<handler>.<action>] [if <filter>] 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 73 ============= ================================================= 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 79 .log2 display log2 value rather than raw number 80 .buckets=size display grouping of values rather than raw number 81 .usecs display a common_timestamp in microseconds 82 ============= ================================================= 83 84 Note that in general the semantics of a given field aren't 85 interpreted when applying a modifier to it, but there are some 86 restrictions to be aware of in this regard: 87 88 - only the 'hex' modifier can be used for values (because values 89 are essentially sums, and the other modifiers don't make sense 90 in that context). 91 - the 'execname' modifier can only be used on a 'common_pid'. The 92 reason for this is that the execname is simply the 'comm' value 93 saved for the 'current' process when an event was triggered, 94 which is the same as the common_pid value saved by the event 95 tracing code. Trying to apply that comm value to other pid 96 values wouldn't be correct, and typically events that care save 97 pid-specific comm fields in the event itself. 98 99 A typical usage scenario would be the following to enable a hist 100 trigger, read its current contents, and then turn it off:: 101 102 # echo 'hist:keys=skbaddr.hex:vals=len' > \ 103 /sys/kernel/debug/tracing/events/net/netif_rx/trigger 104 105 # cat /sys/kernel/debug/tracing/events/net/netif_rx/hist 106 107 # echo '!hist:keys=skbaddr.hex:vals=len' > \ 108 /sys/kernel/debug/tracing/events/net/netif_rx/trigger 109 110 The trigger file itself can be read to show the details of the 111 currently attached hist trigger. This information is also displayed 112 at the top of the 'hist' file when read. 113 114 By default, the size of the hash table is 2048 entries. The 'size' 115 parameter can be used to specify more or fewer than that. The units 116 are in terms of hashtable entries - if a run uses more entries than 117 specified, the results will show the number of 'drops', the number 118 of hits that were ignored. The size should be a power of 2 between 119 128 and 131072 (any non- power-of-2 number specified will be rounded 120 up). 121 122 The 'sort' parameter can be used to specify a value field to sort 123 on. The default if unspecified is 'hitcount' and the default sort 124 order is 'ascending'. To sort in the opposite direction, append 125 .descending' to the sort key. 126 127 The 'pause' parameter can be used to pause an existing hist trigger 128 or to start a hist trigger but not log any events until told to do 129 so. 'continue' or 'cont' can be used to start or restart a paused 130 hist trigger. 131 132 The 'clear' parameter will clear the contents of a running hist 133 trigger and leave its current paused/active state. 134 135 Note that the 'pause', 'cont', and 'clear' parameters should be 136 applied using 'append' shell operator ('>>') if applied to an 137 existing trigger, rather than via the '>' operator, which will cause 138 the trigger to be removed through truncation. 139 140- enable_hist/disable_hist 141 142 The enable_hist and disable_hist triggers can be used to have one 143 event conditionally start and stop another event's already-attached 144 hist trigger. Any number of enable_hist and disable_hist triggers 145 can be attached to a given event, allowing that event to kick off 146 and stop aggregations on a host of other events. 147 148 The format is very similar to the enable/disable_event triggers:: 149 150 enable_hist:<system>:<event>[:count] 151 disable_hist:<system>:<event>[:count] 152 153 Instead of enabling or disabling the tracing of the target event 154 into the trace buffer as the enable/disable_event triggers do, the 155 enable/disable_hist triggers enable or disable the aggregation of 156 the target event into a hash table. 157 158 A typical usage scenario for the enable_hist/disable_hist triggers 159 would be to first set up a paused hist trigger on some event, 160 followed by an enable_hist/disable_hist pair that turns the hist 161 aggregation on and off when conditions of interest are hit:: 162 163 # echo 'hist:keys=skbaddr.hex:vals=len:pause' > \ 164 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 165 166 # echo 'enable_hist:net:netif_receive_skb if filename==/usr/bin/wget' > \ 167 /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger 168 169 # echo 'disable_hist:net:netif_receive_skb if comm==wget' > \ 170 /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger 171 172 The above sets up an initially paused hist trigger which is unpaused 173 and starts aggregating events when a given program is executed, and 174 which stops aggregating when the process exits and the hist trigger 175 is paused again. 176 177 The examples below provide a more concrete illustration of the 178 concepts and typical usage patterns discussed above. 179 180'special' event fields 181------------------------ 182 183 There are a number of 'special event fields' available for use as 184 keys or values in a hist trigger. These look like and behave as if 185 they were actual event fields, but aren't really part of the event's 186 field definition or format file. They are however available for any 187 event, and can be used anywhere an actual event field could be. 188 They are: 189 190 ====================== ==== ======================================= 191 common_timestamp u64 timestamp (from ring buffer) associated 192 with the event, in nanoseconds. May be 193 modified by .usecs to have timestamps 194 interpreted as microseconds. 195 common_cpu int the cpu on which the event occurred. 196 ====================== ==== ======================================= 197 198Extended error information 199-------------------------- 200 201 For some error conditions encountered when invoking a hist trigger 202 command, extended error information is available via the 203 tracing/error_log file. See Error Conditions in 204 :file:`Documentation/trace/ftrace.rst` for details. 205 2066.2 'hist' trigger examples 207--------------------------- 208 209 The first set of examples creates aggregations using the kmalloc 210 event. The fields that can be used for the hist trigger are listed 211 in the kmalloc event's format file:: 212 213 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/format 214 name: kmalloc 215 ID: 374 216 format: 217 field:unsigned short common_type; offset:0; size:2; signed:0; 218 field:unsigned char common_flags; offset:2; size:1; signed:0; 219 field:unsigned char common_preempt_count; offset:3; size:1; signed:0; 220 field:int common_pid; offset:4; size:4; signed:1; 221 222 field:unsigned long call_site; offset:8; size:8; signed:0; 223 field:const void * ptr; offset:16; size:8; signed:0; 224 field:size_t bytes_req; offset:24; size:8; signed:0; 225 field:size_t bytes_alloc; offset:32; size:8; signed:0; 226 field:gfp_t gfp_flags; offset:40; size:4; signed:0; 227 228 We'll start by creating a hist trigger that generates a simple table 229 that lists the total number of bytes requested for each function in 230 the kernel that made one or more calls to kmalloc:: 231 232 # echo 'hist:key=call_site:val=bytes_req.buckets=32' > \ 233 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 234 235 This tells the tracing system to create a 'hist' trigger using the 236 call_site field of the kmalloc event as the key for the table, which 237 just means that each unique call_site address will have an entry 238 created for it in the table. The 'val=bytes_req' parameter tells 239 the hist trigger that for each unique entry (call_site) in the 240 table, it should keep a running total of the number of bytes 241 requested by that call_site. 242 243 We'll let it run for awhile and then dump the contents of the 'hist' 244 file in the kmalloc event's subdirectory (for readability, a number 245 of entries have been omitted):: 246 247 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist 248 # trigger info: hist:keys=call_site:vals=bytes_req:sort=hitcount:size=2048 [active] 249 250 { call_site: 18446744072106379007 } hitcount: 1 bytes_req: 176 251 { call_site: 18446744071579557049 } hitcount: 1 bytes_req: 1024 252 { call_site: 18446744071580608289 } hitcount: 1 bytes_req: 16384 253 { call_site: 18446744071581827654 } hitcount: 1 bytes_req: 24 254 { call_site: 18446744071580700980 } hitcount: 1 bytes_req: 8 255 { call_site: 18446744071579359876 } hitcount: 1 bytes_req: 152 256 { call_site: 18446744071580795365 } hitcount: 3 bytes_req: 144 257 { call_site: 18446744071581303129 } hitcount: 3 bytes_req: 144 258 { call_site: 18446744071580713234 } hitcount: 4 bytes_req: 2560 259 { call_site: 18446744071580933750 } hitcount: 4 bytes_req: 736 260 . 261 . 262 . 263 { call_site: 18446744072106047046 } hitcount: 69 bytes_req: 5576 264 { call_site: 18446744071582116407 } hitcount: 73 bytes_req: 2336 265 { call_site: 18446744072106054684 } hitcount: 136 bytes_req: 140504 266 { call_site: 18446744072106224230 } hitcount: 136 bytes_req: 19584 267 { call_site: 18446744072106078074 } hitcount: 153 bytes_req: 2448 268 { call_site: 18446744072106062406 } hitcount: 153 bytes_req: 36720 269 { call_site: 18446744071582507929 } hitcount: 153 bytes_req: 37088 270 { call_site: 18446744072102520590 } hitcount: 273 bytes_req: 10920 271 { call_site: 18446744071582143559 } hitcount: 358 bytes_req: 716 272 { call_site: 18446744072106465852 } hitcount: 417 bytes_req: 56712 273 { call_site: 18446744072102523378 } hitcount: 485 bytes_req: 27160 274 { call_site: 18446744072099568646 } hitcount: 1676 bytes_req: 33520 275 276 Totals: 277 Hits: 4610 278 Entries: 45 279 Dropped: 0 280 281 The output displays a line for each entry, beginning with the key 282 specified in the trigger, followed by the value(s) also specified in 283 the trigger. At the beginning of the output is a line that displays 284 the trigger info, which can also be displayed by reading the 285 'trigger' file:: 286 287 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 288 hist:keys=call_site:vals=bytes_req:sort=hitcount:size=2048 [active] 289 290 At the end of the output are a few lines that display the overall 291 totals for the run. The 'Hits' field shows the total number of 292 times the event trigger was hit, the 'Entries' field shows the total 293 number of used entries in the hash table, and the 'Dropped' field 294 shows the number of hits that were dropped because the number of 295 used entries for the run exceeded the maximum number of entries 296 allowed for the table (normally 0, but if not a hint that you may 297 want to increase the size of the table using the 'size' parameter). 298 299 Notice in the above output that there's an extra field, 'hitcount', 300 which wasn't specified in the trigger. Also notice that in the 301 trigger info output, there's a parameter, 'sort=hitcount', which 302 wasn't specified in the trigger either. The reason for that is that 303 every trigger implicitly keeps a count of the total number of hits 304 attributed to a given entry, called the 'hitcount'. That hitcount 305 information is explicitly displayed in the output, and in the 306 absence of a user-specified sort parameter, is used as the default 307 sort field. 308 309 The value 'hitcount' can be used in place of an explicit value in 310 the 'values' parameter if you don't really need to have any 311 particular field summed and are mainly interested in hit 312 frequencies. 313 314 To turn the hist trigger off, simply call up the trigger in the 315 command history and re-execute it with a '!' prepended:: 316 317 # echo '!hist:key=call_site:val=bytes_req' > \ 318 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 319 320 Finally, notice that the call_site as displayed in the output above 321 isn't really very useful. It's an address, but normally addresses 322 are displayed in hex. To have a numeric field displayed as a hex 323 value, simply append '.hex' to the field name in the trigger:: 324 325 # echo 'hist:key=call_site.hex:val=bytes_req' > \ 326 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 327 328 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist 329 # trigger info: hist:keys=call_site.hex:vals=bytes_req:sort=hitcount:size=2048 [active] 330 331 { call_site: ffffffffa026b291 } hitcount: 1 bytes_req: 433 332 { call_site: ffffffffa07186ff } hitcount: 1 bytes_req: 176 333 { call_site: ffffffff811ae721 } hitcount: 1 bytes_req: 16384 334 { call_site: ffffffff811c5134 } hitcount: 1 bytes_req: 8 335 { call_site: ffffffffa04a9ebb } hitcount: 1 bytes_req: 511 336 { call_site: ffffffff8122e0a6 } hitcount: 1 bytes_req: 12 337 { call_site: ffffffff8107da84 } hitcount: 1 bytes_req: 152 338 { call_site: ffffffff812d8246 } hitcount: 1 bytes_req: 24 339 { call_site: ffffffff811dc1e5 } hitcount: 3 bytes_req: 144 340 { call_site: ffffffffa02515e8 } hitcount: 3 bytes_req: 648 341 { call_site: ffffffff81258159 } hitcount: 3 bytes_req: 144 342 { call_site: ffffffff811c80f4 } hitcount: 4 bytes_req: 544 343 . 344 . 345 . 346 { call_site: ffffffffa06c7646 } hitcount: 106 bytes_req: 8024 347 { call_site: ffffffffa06cb246 } hitcount: 132 bytes_req: 31680 348 { call_site: ffffffffa06cef7a } hitcount: 132 bytes_req: 2112 349 { call_site: ffffffff8137e399 } hitcount: 132 bytes_req: 23232 350 { call_site: ffffffffa06c941c } hitcount: 185 bytes_req: 171360 351 { call_site: ffffffffa06f2a66 } hitcount: 185 bytes_req: 26640 352 { call_site: ffffffffa036a70e } hitcount: 265 bytes_req: 10600 353 { call_site: ffffffff81325447 } hitcount: 292 bytes_req: 584 354 { call_site: ffffffffa072da3c } hitcount: 446 bytes_req: 60656 355 { call_site: ffffffffa036b1f2 } hitcount: 526 bytes_req: 29456 356 { call_site: ffffffffa0099c06 } hitcount: 1780 bytes_req: 35600 357 358 Totals: 359 Hits: 4775 360 Entries: 46 361 Dropped: 0 362 363 Even that's only marginally more useful - while hex values do look 364 more like addresses, what users are typically more interested in 365 when looking at text addresses are the corresponding symbols 366 instead. To have an address displayed as symbolic value instead, 367 simply append '.sym' or '.sym-offset' to the field name in the 368 trigger:: 369 370 # echo 'hist:key=call_site.sym:val=bytes_req' > \ 371 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 372 373 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist 374 # trigger info: hist:keys=call_site.sym:vals=bytes_req:sort=hitcount:size=2048 [active] 375 376 { call_site: [ffffffff810adcb9] syslog_print_all } hitcount: 1 bytes_req: 1024 377 { call_site: [ffffffff8154bc62] usb_control_msg } hitcount: 1 bytes_req: 8 378 { call_site: [ffffffffa00bf6fe] hidraw_send_report [hid] } hitcount: 1 bytes_req: 7 379 { call_site: [ffffffff8154acbe] usb_alloc_urb } hitcount: 1 bytes_req: 192 380 { call_site: [ffffffffa00bf1ca] hidraw_report_event [hid] } hitcount: 1 bytes_req: 7 381 { call_site: [ffffffff811e3a25] __seq_open_private } hitcount: 1 bytes_req: 40 382 { call_site: [ffffffff8109524a] alloc_fair_sched_group } hitcount: 2 bytes_req: 128 383 { call_site: [ffffffff811febd5] fsnotify_alloc_group } hitcount: 2 bytes_req: 528 384 { call_site: [ffffffff81440f58] __tty_buffer_request_room } hitcount: 2 bytes_req: 2624 385 { call_site: [ffffffff81200ba6] inotify_new_group } hitcount: 2 bytes_req: 96 386 { call_site: [ffffffffa05e19af] ieee80211_start_tx_ba_session [mac80211] } hitcount: 2 bytes_req: 464 387 { call_site: [ffffffff81672406] tcp_get_metrics } hitcount: 2 bytes_req: 304 388 { call_site: [ffffffff81097ec2] alloc_rt_sched_group } hitcount: 2 bytes_req: 128 389 { call_site: [ffffffff81089b05] sched_create_group } hitcount: 2 bytes_req: 1424 390 . 391 . 392 . 393 { call_site: [ffffffffa04a580c] intel_crtc_page_flip [i915] } hitcount: 1185 bytes_req: 123240 394 { call_site: [ffffffffa0287592] drm_mode_page_flip_ioctl [drm] } hitcount: 1185 bytes_req: 104280 395 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state [i915] } hitcount: 1402 bytes_req: 190672 396 { call_site: [ffffffff812891ca] ext4_find_extent } hitcount: 1518 bytes_req: 146208 397 { call_site: [ffffffffa029070e] drm_vma_node_allow [drm] } hitcount: 1746 bytes_req: 69840 398 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 2021 bytes_req: 792312 399 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc [drm] } hitcount: 2592 bytes_req: 145152 400 { call_site: [ffffffffa0489a66] intel_ring_begin [i915] } hitcount: 2629 bytes_req: 378576 401 { call_site: [ffffffffa046041c] i915_gem_execbuffer2 [i915] } hitcount: 2629 bytes_req: 3783248 402 { call_site: [ffffffff81325607] apparmor_file_alloc_security } hitcount: 5192 bytes_req: 10384 403 { call_site: [ffffffffa00b7c06] hid_report_raw_event [hid] } hitcount: 5529 bytes_req: 110584 404 { call_site: [ffffffff8131ebf7] aa_alloc_task_context } hitcount: 21943 bytes_req: 702176 405 { call_site: [ffffffff8125847d] ext4_htree_store_dirent } hitcount: 55759 bytes_req: 5074265 406 407 Totals: 408 Hits: 109928 409 Entries: 71 410 Dropped: 0 411 412 Because the default sort key above is 'hitcount', the above shows a 413 the list of call_sites by increasing hitcount, so that at the bottom 414 we see the functions that made the most kmalloc calls during the 415 run. If instead we we wanted to see the top kmalloc callers in 416 terms of the number of bytes requested rather than the number of 417 calls, and we wanted the top caller to appear at the top, we can use 418 the 'sort' parameter, along with the 'descending' modifier:: 419 420 # echo 'hist:key=call_site.sym:val=bytes_req:sort=bytes_req.descending' > \ 421 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 422 423 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist 424 # trigger info: hist:keys=call_site.sym:vals=bytes_req:sort=bytes_req.descending:size=2048 [active] 425 426 { call_site: [ffffffffa046041c] i915_gem_execbuffer2 [i915] } hitcount: 2186 bytes_req: 3397464 427 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 1790 bytes_req: 712176 428 { call_site: [ffffffff8125847d] ext4_htree_store_dirent } hitcount: 8132 bytes_req: 513135 429 { call_site: [ffffffff811e2a1b] seq_buf_alloc } hitcount: 106 bytes_req: 440128 430 { call_site: [ffffffffa0489a66] intel_ring_begin [i915] } hitcount: 2186 bytes_req: 314784 431 { call_site: [ffffffff812891ca] ext4_find_extent } hitcount: 2174 bytes_req: 208992 432 { call_site: [ffffffff811ae8e1] __kmalloc } hitcount: 8 bytes_req: 131072 433 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state [i915] } hitcount: 859 bytes_req: 116824 434 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc [drm] } hitcount: 1834 bytes_req: 102704 435 { call_site: [ffffffffa04a580c] intel_crtc_page_flip [i915] } hitcount: 972 bytes_req: 101088 436 { call_site: [ffffffffa0287592] drm_mode_page_flip_ioctl [drm] } hitcount: 972 bytes_req: 85536 437 { call_site: [ffffffffa00b7c06] hid_report_raw_event [hid] } hitcount: 3333 bytes_req: 66664 438 { call_site: [ffffffff8137e559] sg_kmalloc } hitcount: 209 bytes_req: 61632 439 . 440 . 441 . 442 { call_site: [ffffffff81095225] alloc_fair_sched_group } hitcount: 2 bytes_req: 128 443 { call_site: [ffffffff81097ec2] alloc_rt_sched_group } hitcount: 2 bytes_req: 128 444 { call_site: [ffffffff812d8406] copy_semundo } hitcount: 2 bytes_req: 48 445 { call_site: [ffffffff81200ba6] inotify_new_group } hitcount: 1 bytes_req: 48 446 { call_site: [ffffffffa027121a] drm_getmagic [drm] } hitcount: 1 bytes_req: 48 447 { call_site: [ffffffff811e3a25] __seq_open_private } hitcount: 1 bytes_req: 40 448 { call_site: [ffffffff811c52f4] bprm_change_interp } hitcount: 2 bytes_req: 16 449 { call_site: [ffffffff8154bc62] usb_control_msg } hitcount: 1 bytes_req: 8 450 { call_site: [ffffffffa00bf1ca] hidraw_report_event [hid] } hitcount: 1 bytes_req: 7 451 { call_site: [ffffffffa00bf6fe] hidraw_send_report [hid] } hitcount: 1 bytes_req: 7 452 453 Totals: 454 Hits: 32133 455 Entries: 81 456 Dropped: 0 457 458 To display the offset and size information in addition to the symbol 459 name, just use 'sym-offset' instead:: 460 461 # echo 'hist:key=call_site.sym-offset:val=bytes_req:sort=bytes_req.descending' > \ 462 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 463 464 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist 465 # trigger info: hist:keys=call_site.sym-offset:vals=bytes_req:sort=bytes_req.descending:size=2048 [active] 466 467 { call_site: [ffffffffa046041c] i915_gem_execbuffer2+0x6c/0x2c0 [i915] } hitcount: 4569 bytes_req: 3163720 468 { call_site: [ffffffffa0489a66] intel_ring_begin+0xc6/0x1f0 [i915] } hitcount: 4569 bytes_req: 657936 469 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23+0x694/0x1020 [i915] } hitcount: 1519 bytes_req: 472936 470 { call_site: [ffffffffa045e646] i915_gem_do_execbuffer.isra.23+0x516/0x1020 [i915] } hitcount: 3050 bytes_req: 211832 471 { call_site: [ffffffff811e2a1b] seq_buf_alloc+0x1b/0x50 } hitcount: 34 bytes_req: 148384 472 { call_site: [ffffffffa04a580c] intel_crtc_page_flip+0xbc/0x870 [i915] } hitcount: 1385 bytes_req: 144040 473 { call_site: [ffffffff811ae8e1] __kmalloc+0x191/0x1b0 } hitcount: 8 bytes_req: 131072 474 { call_site: [ffffffffa0287592] drm_mode_page_flip_ioctl+0x282/0x360 [drm] } hitcount: 1385 bytes_req: 121880 475 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc+0x32/0x100 [drm] } hitcount: 1848 bytes_req: 103488 476 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state+0x2c/0xa0 [i915] } hitcount: 461 bytes_req: 62696 477 { call_site: [ffffffffa029070e] drm_vma_node_allow+0x2e/0xd0 [drm] } hitcount: 1541 bytes_req: 61640 478 { call_site: [ffffffff815f8d7b] sk_prot_alloc+0xcb/0x1b0 } hitcount: 57 bytes_req: 57456 479 . 480 . 481 . 482 { call_site: [ffffffff8109524a] alloc_fair_sched_group+0x5a/0x1a0 } hitcount: 2 bytes_req: 128 483 { call_site: [ffffffffa027b921] drm_vm_open_locked+0x31/0xa0 [drm] } hitcount: 3 bytes_req: 96 484 { call_site: [ffffffff8122e266] proc_self_follow_link+0x76/0xb0 } hitcount: 8 bytes_req: 96 485 { call_site: [ffffffff81213e80] load_elf_binary+0x240/0x1650 } hitcount: 3 bytes_req: 84 486 { call_site: [ffffffff8154bc62] usb_control_msg+0x42/0x110 } hitcount: 1 bytes_req: 8 487 { call_site: [ffffffffa00bf6fe] hidraw_send_report+0x7e/0x1a0 [hid] } hitcount: 1 bytes_req: 7 488 { call_site: [ffffffffa00bf1ca] hidraw_report_event+0x8a/0x120 [hid] } hitcount: 1 bytes_req: 7 489 490 Totals: 491 Hits: 26098 492 Entries: 64 493 Dropped: 0 494 495 We can also add multiple fields to the 'values' parameter. For 496 example, we might want to see the total number of bytes allocated 497 alongside bytes requested, and display the result sorted by bytes 498 allocated in a descending order:: 499 500 # echo 'hist:keys=call_site.sym:values=bytes_req,bytes_alloc:sort=bytes_alloc.descending' > \ 501 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 502 503 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist 504 # trigger info: hist:keys=call_site.sym:vals=bytes_req,bytes_alloc:sort=bytes_alloc.descending:size=2048 [active] 505 506 { call_site: [ffffffffa046041c] i915_gem_execbuffer2 [i915] } hitcount: 7403 bytes_req: 4084360 bytes_alloc: 5958016 507 { call_site: [ffffffff811e2a1b] seq_buf_alloc } hitcount: 541 bytes_req: 2213968 bytes_alloc: 2228224 508 { call_site: [ffffffffa0489a66] intel_ring_begin [i915] } hitcount: 7404 bytes_req: 1066176 bytes_alloc: 1421568 509 { call_site: [ffffffffa045e7c4] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 1565 bytes_req: 557368 bytes_alloc: 1037760 510 { call_site: [ffffffff8125847d] ext4_htree_store_dirent } hitcount: 9557 bytes_req: 595778 bytes_alloc: 695744 511 { call_site: [ffffffffa045e646] i915_gem_do_execbuffer.isra.23 [i915] } hitcount: 5839 bytes_req: 430680 bytes_alloc: 470400 512 { call_site: [ffffffffa04c4a3c] intel_plane_duplicate_state [i915] } hitcount: 2388 bytes_req: 324768 bytes_alloc: 458496 513 { call_site: [ffffffffa02911f2] drm_modeset_lock_crtc [drm] } hitcount: 3911 bytes_req: 219016 bytes_alloc: 250304 514 { call_site: [ffffffff815f8d7b] sk_prot_alloc } hitcount: 235 bytes_req: 236880 bytes_alloc: 240640 515 { call_site: [ffffffff8137e559] sg_kmalloc } hitcount: 557 bytes_req: 169024 bytes_alloc: 221760 516 { call_site: [ffffffffa00b7c06] hid_report_raw_event [hid] } hitcount: 9378 bytes_req: 187548 bytes_alloc: 206312 517 { call_site: [ffffffffa04a580c] intel_crtc_page_flip [i915] } hitcount: 1519 bytes_req: 157976 bytes_alloc: 194432 518 . 519 . 520 . 521 { call_site: [ffffffff8109bd3b] sched_autogroup_create_attach } hitcount: 2 bytes_req: 144 bytes_alloc: 192 522 { call_site: [ffffffff81097ee8] alloc_rt_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128 523 { call_site: [ffffffff8109524a] alloc_fair_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128 524 { call_site: [ffffffff81095225] alloc_fair_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128 525 { call_site: [ffffffff81097ec2] alloc_rt_sched_group } hitcount: 2 bytes_req: 128 bytes_alloc: 128 526 { call_site: [ffffffff81213e80] load_elf_binary } hitcount: 3 bytes_req: 84 bytes_alloc: 96 527 { call_site: [ffffffff81079a2e] kthread_create_on_node } hitcount: 1 bytes_req: 56 bytes_alloc: 64 528 { call_site: [ffffffffa00bf6fe] hidraw_send_report [hid] } hitcount: 1 bytes_req: 7 bytes_alloc: 8 529 { call_site: [ffffffff8154bc62] usb_control_msg } hitcount: 1 bytes_req: 8 bytes_alloc: 8 530 { call_site: [ffffffffa00bf1ca] hidraw_report_event [hid] } hitcount: 1 bytes_req: 7 bytes_alloc: 8 531 532 Totals: 533 Hits: 66598 534 Entries: 65 535 Dropped: 0 536 537 Finally, to finish off our kmalloc example, instead of simply having 538 the hist trigger display symbolic call_sites, we can have the hist 539 trigger additionally display the complete set of kernel stack traces 540 that led to each call_site. To do that, we simply use the special 541 value 'stacktrace' for the key parameter:: 542 543 # echo 'hist:keys=stacktrace:values=bytes_req,bytes_alloc:sort=bytes_alloc' > \ 544 /sys/kernel/debug/tracing/events/kmem/kmalloc/trigger 545 546 The above trigger will use the kernel stack trace in effect when an 547 event is triggered as the key for the hash table. This allows the 548 enumeration of every kernel callpath that led up to a particular 549 event, along with a running total of any of the event fields for 550 that event. Here we tally bytes requested and bytes allocated for 551 every callpath in the system that led up to a kmalloc (in this case 552 every callpath to a kmalloc for a kernel compile):: 553 554 # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/hist 555 # trigger info: hist:keys=stacktrace:vals=bytes_req,bytes_alloc:sort=bytes_alloc:size=2048 [active] 556 557 { stacktrace: 558 __kmalloc_track_caller+0x10b/0x1a0 559 kmemdup+0x20/0x50 560 hidraw_report_event+0x8a/0x120 [hid] 561 hid_report_raw_event+0x3ea/0x440 [hid] 562 hid_input_report+0x112/0x190 [hid] 563 hid_irq_in+0xc2/0x260 [usbhid] 564 __usb_hcd_giveback_urb+0x72/0x120 565 usb_giveback_urb_bh+0x9e/0xe0 566 tasklet_hi_action+0xf8/0x100 567 __do_softirq+0x114/0x2c0 568 irq_exit+0xa5/0xb0 569 do_IRQ+0x5a/0xf0 570 ret_from_intr+0x0/0x30 571 cpuidle_enter+0x17/0x20 572 cpu_startup_entry+0x315/0x3e0 573 rest_init+0x7c/0x80 574 } hitcount: 3 bytes_req: 21 bytes_alloc: 24 575 { stacktrace: 576 __kmalloc_track_caller+0x10b/0x1a0 577 kmemdup+0x20/0x50 578 hidraw_report_event+0x8a/0x120 [hid] 579 hid_report_raw_event+0x3ea/0x440 [hid] 580 hid_input_report+0x112/0x190 [hid] 581 hid_irq_in+0xc2/0x260 [usbhid] 582 __usb_hcd_giveback_urb+0x72/0x120 583 usb_giveback_urb_bh+0x9e/0xe0 584 tasklet_hi_action+0xf8/0x100 585 __do_softirq+0x114/0x2c0 586 irq_exit+0xa5/0xb0 587 do_IRQ+0x5a/0xf0 588 ret_from_intr+0x0/0x30 589 } hitcount: 3 bytes_req: 21 bytes_alloc: 24 590 { stacktrace: 591 kmem_cache_alloc_trace+0xeb/0x150 592 aa_alloc_task_context+0x27/0x40 593 apparmor_cred_prepare+0x1f/0x50 594 security_prepare_creds+0x16/0x20 595 prepare_creds+0xdf/0x1a0 596 SyS_capset+0xb5/0x200 597 system_call_fastpath+0x12/0x6a 598 } hitcount: 1 bytes_req: 32 bytes_alloc: 32 599 . 600 . 601 . 602 { stacktrace: 603 __kmalloc+0x11b/0x1b0 604 i915_gem_execbuffer2+0x6c/0x2c0 [i915] 605 drm_ioctl+0x349/0x670 [drm] 606 do_vfs_ioctl+0x2f0/0x4f0 607 SyS_ioctl+0x81/0xa0 608 system_call_fastpath+0x12/0x6a 609 } hitcount: 17726 bytes_req: 13944120 bytes_alloc: 19593808 610 { stacktrace: 611 __kmalloc+0x11b/0x1b0 612 load_elf_phdrs+0x76/0xa0 613 load_elf_binary+0x102/0x1650 614 search_binary_handler+0x97/0x1d0 615 do_execveat_common.isra.34+0x551/0x6e0 616 SyS_execve+0x3a/0x50 617 return_from_execve+0x0/0x23 618 } hitcount: 33348 bytes_req: 17152128 bytes_alloc: 20226048 619 { stacktrace: 620 kmem_cache_alloc_trace+0xeb/0x150 621 apparmor_file_alloc_security+0x27/0x40 622 security_file_alloc+0x16/0x20 623 get_empty_filp+0x93/0x1c0 624 path_openat+0x31/0x5f0 625 do_filp_open+0x3a/0x90 626 do_sys_open+0x128/0x220 627 SyS_open+0x1e/0x20 628 system_call_fastpath+0x12/0x6a 629 } hitcount: 4766422 bytes_req: 9532844 bytes_alloc: 38131376 630 { stacktrace: 631 __kmalloc+0x11b/0x1b0 632 seq_buf_alloc+0x1b/0x50 633 seq_read+0x2cc/0x370 634 proc_reg_read+0x3d/0x80 635 __vfs_read+0x28/0xe0 636 vfs_read+0x86/0x140 637 SyS_read+0x46/0xb0 638 system_call_fastpath+0x12/0x6a 639 } hitcount: 19133 bytes_req: 78368768 bytes_alloc: 78368768 640 641 Totals: 642 Hits: 6085872 643 Entries: 253 644 Dropped: 0 645 646 If you key a hist trigger on common_pid, in order for example to 647 gather and display sorted totals for each process, you can use the 648 special .execname modifier to display the executable names for the 649 processes in the table rather than raw pids. The example below 650 keeps a per-process sum of total bytes read:: 651 652 # echo 'hist:key=common_pid.execname:val=count:sort=count.descending' > \ 653 /sys/kernel/debug/tracing/events/syscalls/sys_enter_read/trigger 654 655 # cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_read/hist 656 # trigger info: hist:keys=common_pid.execname:vals=count:sort=count.descending:size=2048 [active] 657 658 { common_pid: gnome-terminal [ 3196] } hitcount: 280 count: 1093512 659 { common_pid: Xorg [ 1309] } hitcount: 525 count: 256640 660 { common_pid: compiz [ 2889] } hitcount: 59 count: 254400 661 { common_pid: bash [ 8710] } hitcount: 3 count: 66369 662 { common_pid: dbus-daemon-lau [ 8703] } hitcount: 49 count: 47739 663 { common_pid: irqbalance [ 1252] } hitcount: 27 count: 27648 664 { common_pid: 01ifupdown [ 8705] } hitcount: 3 count: 17216 665 { common_pid: dbus-daemon [ 772] } hitcount: 10 count: 12396 666 { common_pid: Socket Thread [ 8342] } hitcount: 11 count: 11264 667 { common_pid: nm-dhcp-client. [ 8701] } hitcount: 6 count: 7424 668 { common_pid: gmain [ 1315] } hitcount: 18 count: 6336 669 . 670 . 671 . 672 { common_pid: postgres [ 1892] } hitcount: 2 count: 32 673 { common_pid: postgres [ 1891] } hitcount: 2 count: 32 674 { common_pid: gmain [ 8704] } hitcount: 2 count: 32 675 { common_pid: upstart-dbus-br [ 2740] } hitcount: 21 count: 21 676 { common_pid: nm-dispatcher.a [ 8696] } hitcount: 1 count: 16 677 { common_pid: indicator-datet [ 2904] } hitcount: 1 count: 16 678 { common_pid: gdbus [ 2998] } hitcount: 1 count: 16 679 { common_pid: rtkit-daemon [ 2052] } hitcount: 1 count: 8 680 { common_pid: init [ 1] } hitcount: 2 count: 2 681 682 Totals: 683 Hits: 2116 684 Entries: 51 685 Dropped: 0 686 687 Similarly, if you key a hist trigger on syscall id, for example to 688 gather and display a list of systemwide syscall hits, you can use 689 the special .syscall modifier to display the syscall names rather 690 than raw ids. The example below keeps a running total of syscall 691 counts for the system during the run:: 692 693 # echo 'hist:key=id.syscall:val=hitcount' > \ 694 /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/trigger 695 696 # cat /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/hist 697 # trigger info: hist:keys=id.syscall:vals=hitcount:sort=hitcount:size=2048 [active] 698 699 { id: sys_fsync [ 74] } hitcount: 1 700 { id: sys_newuname [ 63] } hitcount: 1 701 { id: sys_prctl [157] } hitcount: 1 702 { id: sys_statfs [137] } hitcount: 1 703 { id: sys_symlink [ 88] } hitcount: 1 704 { id: sys_sendmmsg [307] } hitcount: 1 705 { id: sys_semctl [ 66] } hitcount: 1 706 { id: sys_readlink [ 89] } hitcount: 3 707 { id: sys_bind [ 49] } hitcount: 3 708 { id: sys_getsockname [ 51] } hitcount: 3 709 { id: sys_unlink [ 87] } hitcount: 3 710 { id: sys_rename [ 82] } hitcount: 4 711 { id: unknown_syscall [ 58] } hitcount: 4 712 { id: sys_connect [ 42] } hitcount: 4 713 { id: sys_getpid [ 39] } hitcount: 4 714 . 715 . 716 . 717 { id: sys_rt_sigprocmask [ 14] } hitcount: 952 718 { id: sys_futex [202] } hitcount: 1534 719 { id: sys_write [ 1] } hitcount: 2689 720 { id: sys_setitimer [ 38] } hitcount: 2797 721 { id: sys_read [ 0] } hitcount: 3202 722 { id: sys_select [ 23] } hitcount: 3773 723 { id: sys_writev [ 20] } hitcount: 4531 724 { id: sys_poll [ 7] } hitcount: 8314 725 { id: sys_recvmsg [ 47] } hitcount: 13738 726 { id: sys_ioctl [ 16] } hitcount: 21843 727 728 Totals: 729 Hits: 67612 730 Entries: 72 731 Dropped: 0 732 733 The syscall counts above provide a rough overall picture of system 734 call activity on the system; we can see for example that the most 735 popular system call on this system was the 'sys_ioctl' system call. 736 737 We can use 'compound' keys to refine that number and provide some 738 further insight as to which processes exactly contribute to the 739 overall ioctl count. 740 741 The command below keeps a hitcount for every unique combination of 742 system call id and pid - the end result is essentially a table 743 that keeps a per-pid sum of system call hits. The results are 744 sorted using the system call id as the primary key, and the 745 hitcount sum as the secondary key:: 746 747 # echo 'hist:key=id.syscall,common_pid.execname:val=hitcount:sort=id,hitcount' > \ 748 /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/trigger 749 750 # cat /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/hist 751 # trigger info: hist:keys=id.syscall,common_pid.execname:vals=hitcount:sort=id.syscall,hitcount:size=2048 [active] 752 753 { id: sys_read [ 0], common_pid: rtkit-daemon [ 1877] } hitcount: 1 754 { id: sys_read [ 0], common_pid: gdbus [ 2976] } hitcount: 1 755 { id: sys_read [ 0], common_pid: console-kit-dae [ 3400] } hitcount: 1 756 { id: sys_read [ 0], common_pid: postgres [ 1865] } hitcount: 1 757 { id: sys_read [ 0], common_pid: deja-dup-monito [ 3543] } hitcount: 2 758 { id: sys_read [ 0], common_pid: NetworkManager [ 890] } hitcount: 2 759 { id: sys_read [ 0], common_pid: evolution-calen [ 3048] } hitcount: 2 760 { id: sys_read [ 0], common_pid: postgres [ 1864] } hitcount: 2 761 { id: sys_read [ 0], common_pid: nm-applet [ 3022] } hitcount: 2 762 { id: sys_read [ 0], common_pid: whoopsie [ 1212] } hitcount: 2 763 . 764 . 765 . 766 { id: sys_ioctl [ 16], common_pid: bash [ 8479] } hitcount: 1 767 { id: sys_ioctl [ 16], common_pid: bash [ 3472] } hitcount: 12 768 { id: sys_ioctl [ 16], common_pid: gnome-terminal [ 3199] } hitcount: 16 769 { id: sys_ioctl [ 16], common_pid: Xorg [ 1267] } hitcount: 1808 770 { id: sys_ioctl [ 16], common_pid: compiz [ 2994] } hitcount: 5580 771 . 772 . 773 . 774 { id: sys_waitid [247], common_pid: upstart-dbus-br [ 2690] } hitcount: 3 775 { id: sys_waitid [247], common_pid: upstart-dbus-br [ 2688] } hitcount: 16 776 { id: sys_inotify_add_watch [254], common_pid: gmain [ 975] } hitcount: 2 777 { id: sys_inotify_add_watch [254], common_pid: gmain [ 3204] } hitcount: 4 778 { id: sys_inotify_add_watch [254], common_pid: gmain [ 2888] } hitcount: 4 779 { id: sys_inotify_add_watch [254], common_pid: gmain [ 3003] } hitcount: 4 780 { id: sys_inotify_add_watch [254], common_pid: gmain [ 2873] } hitcount: 4 781 { id: sys_inotify_add_watch [254], common_pid: gmain [ 3196] } hitcount: 6 782 { id: sys_openat [257], common_pid: java [ 2623] } hitcount: 2 783 { id: sys_eventfd2 [290], common_pid: ibus-ui-gtk3 [ 2760] } hitcount: 4 784 { id: sys_eventfd2 [290], common_pid: compiz [ 2994] } hitcount: 6 785 786 Totals: 787 Hits: 31536 788 Entries: 323 789 Dropped: 0 790 791 The above list does give us a breakdown of the ioctl syscall by 792 pid, but it also gives us quite a bit more than that, which we 793 don't really care about at the moment. Since we know the syscall 794 id for sys_ioctl (16, displayed next to the sys_ioctl name), we 795 can use that to filter out all the other syscalls:: 796 797 # echo 'hist:key=id.syscall,common_pid.execname:val=hitcount:sort=id,hitcount if id == 16' > \ 798 /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/trigger 799 800 # cat /sys/kernel/debug/tracing/events/raw_syscalls/sys_enter/hist 801 # trigger info: hist:keys=id.syscall,common_pid.execname:vals=hitcount:sort=id.syscall,hitcount:size=2048 if id == 16 [active] 802 803 { id: sys_ioctl [ 16], common_pid: gmain [ 2769] } hitcount: 1 804 { id: sys_ioctl [ 16], common_pid: evolution-addre [ 8571] } hitcount: 1 805 { id: sys_ioctl [ 16], common_pid: gmain [ 3003] } hitcount: 1 806 { id: sys_ioctl [ 16], common_pid: gmain [ 2781] } hitcount: 1 807 { id: sys_ioctl [ 16], common_pid: gmain [ 2829] } hitcount: 1 808 { id: sys_ioctl [ 16], common_pid: bash [ 8726] } hitcount: 1 809 { id: sys_ioctl [ 16], common_pid: bash [ 8508] } hitcount: 1 810 { id: sys_ioctl [ 16], common_pid: gmain [ 2970] } hitcount: 1 811 { id: sys_ioctl [ 16], common_pid: gmain [ 2768] } hitcount: 1 812 . 813 . 814 . 815 { id: sys_ioctl [ 16], common_pid: pool [ 8559] } hitcount: 45 816 { id: sys_ioctl [ 16], common_pid: pool [ 8555] } hitcount: 48 817 { id: sys_ioctl [ 16], common_pid: pool [ 8551] } hitcount: 48 818 { id: sys_ioctl [ 16], common_pid: avahi-daemon [ 896] } hitcount: 66 819 { id: sys_ioctl [ 16], common_pid: Xorg [ 1267] } hitcount: 26674 820 { id: sys_ioctl [ 16], common_pid: compiz [ 2994] } hitcount: 73443 821 822 Totals: 823 Hits: 101162 824 Entries: 103 825 Dropped: 0 826 827 The above output shows that 'compiz' and 'Xorg' are far and away 828 the heaviest ioctl callers (which might lead to questions about 829 whether they really need to be making all those calls and to 830 possible avenues for further investigation.) 831 832 The compound key examples used a key and a sum value (hitcount) to 833 sort the output, but we can just as easily use two keys instead. 834 Here's an example where we use a compound key composed of the the 835 common_pid and size event fields. Sorting with pid as the primary 836 key and 'size' as the secondary key allows us to display an 837 ordered summary of the recvfrom sizes, with counts, received by 838 each process:: 839 840 # echo 'hist:key=common_pid.execname,size:val=hitcount:sort=common_pid,size' > \ 841 /sys/kernel/debug/tracing/events/syscalls/sys_enter_recvfrom/trigger 842 843 # cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_recvfrom/hist 844 # trigger info: hist:keys=common_pid.execname,size:vals=hitcount:sort=common_pid.execname,size:size=2048 [active] 845 846 { common_pid: smbd [ 784], size: 4 } hitcount: 1 847 { common_pid: dnsmasq [ 1412], size: 4096 } hitcount: 672 848 { common_pid: postgres [ 1796], size: 1000 } hitcount: 6 849 { common_pid: postgres [ 1867], size: 1000 } hitcount: 10 850 { common_pid: bamfdaemon [ 2787], size: 28 } hitcount: 2 851 { common_pid: bamfdaemon [ 2787], size: 14360 } hitcount: 1 852 { common_pid: compiz [ 2994], size: 8 } hitcount: 1 853 { common_pid: compiz [ 2994], size: 20 } hitcount: 11 854 { common_pid: gnome-terminal [ 3199], size: 4 } hitcount: 2 855 { common_pid: firefox [ 8817], size: 4 } hitcount: 1 856 { common_pid: firefox [ 8817], size: 8 } hitcount: 5 857 { common_pid: firefox [ 8817], size: 588 } hitcount: 2 858 { common_pid: firefox [ 8817], size: 628 } hitcount: 1 859 { common_pid: firefox [ 8817], size: 6944 } hitcount: 1 860 { common_pid: firefox [ 8817], size: 408880 } hitcount: 2 861 { common_pid: firefox [ 8822], size: 8 } hitcount: 2 862 { common_pid: firefox [ 8822], size: 160 } hitcount: 2 863 { common_pid: firefox [ 8822], size: 320 } hitcount: 2 864 { common_pid: firefox [ 8822], size: 352 } hitcount: 1 865 . 866 . 867 . 868 { common_pid: pool [ 8923], size: 1960 } hitcount: 10 869 { common_pid: pool [ 8923], size: 2048 } hitcount: 10 870 { common_pid: pool [ 8924], size: 1960 } hitcount: 10 871 { common_pid: pool [ 8924], size: 2048 } hitcount: 10 872 { common_pid: pool [ 8928], size: 1964 } hitcount: 4 873 { common_pid: pool [ 8928], size: 1965 } hitcount: 2 874 { common_pid: pool [ 8928], size: 2048 } hitcount: 6 875 { common_pid: pool [ 8929], size: 1982 } hitcount: 1 876 { common_pid: pool [ 8929], size: 2048 } hitcount: 1 877 878 Totals: 879 Hits: 2016 880 Entries: 224 881 Dropped: 0 882 883 The above example also illustrates the fact that although a compound 884 key is treated as a single entity for hashing purposes, the sub-keys 885 it's composed of can be accessed independently. 886 887 The next example uses a string field as the hash key and 888 demonstrates how you can manually pause and continue a hist trigger. 889 In this example, we'll aggregate fork counts and don't expect a 890 large number of entries in the hash table, so we'll drop it to a 891 much smaller number, say 256:: 892 893 # echo 'hist:key=child_comm:val=hitcount:size=256' > \ 894 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger 895 896 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist 897 # trigger info: hist:keys=child_comm:vals=hitcount:sort=hitcount:size=256 [active] 898 899 { child_comm: dconf worker } hitcount: 1 900 { child_comm: ibus-daemon } hitcount: 1 901 { child_comm: whoopsie } hitcount: 1 902 { child_comm: smbd } hitcount: 1 903 { child_comm: gdbus } hitcount: 1 904 { child_comm: kthreadd } hitcount: 1 905 { child_comm: dconf worker } hitcount: 1 906 { child_comm: evolution-alarm } hitcount: 2 907 { child_comm: Socket Thread } hitcount: 2 908 { child_comm: postgres } hitcount: 2 909 { child_comm: bash } hitcount: 3 910 { child_comm: compiz } hitcount: 3 911 { child_comm: evolution-sourc } hitcount: 4 912 { child_comm: dhclient } hitcount: 4 913 { child_comm: pool } hitcount: 5 914 { child_comm: nm-dispatcher.a } hitcount: 8 915 { child_comm: firefox } hitcount: 8 916 { child_comm: dbus-daemon } hitcount: 8 917 { child_comm: glib-pacrunner } hitcount: 10 918 { child_comm: evolution } hitcount: 23 919 920 Totals: 921 Hits: 89 922 Entries: 20 923 Dropped: 0 924 925 If we want to pause the hist trigger, we can simply append :pause to 926 the command that started the trigger. Notice that the trigger info 927 displays as [paused]:: 928 929 # echo 'hist:key=child_comm:val=hitcount:size=256:pause' >> \ 930 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger 931 932 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist 933 # trigger info: hist:keys=child_comm:vals=hitcount:sort=hitcount:size=256 [paused] 934 935 { child_comm: dconf worker } hitcount: 1 936 { child_comm: kthreadd } hitcount: 1 937 { child_comm: dconf worker } hitcount: 1 938 { child_comm: gdbus } hitcount: 1 939 { child_comm: ibus-daemon } hitcount: 1 940 { child_comm: Socket Thread } hitcount: 2 941 { child_comm: evolution-alarm } hitcount: 2 942 { child_comm: smbd } hitcount: 2 943 { child_comm: bash } hitcount: 3 944 { child_comm: whoopsie } hitcount: 3 945 { child_comm: compiz } hitcount: 3 946 { child_comm: evolution-sourc } hitcount: 4 947 { child_comm: pool } hitcount: 5 948 { child_comm: postgres } hitcount: 6 949 { child_comm: firefox } hitcount: 8 950 { child_comm: dhclient } hitcount: 10 951 { child_comm: emacs } hitcount: 12 952 { child_comm: dbus-daemon } hitcount: 20 953 { child_comm: nm-dispatcher.a } hitcount: 20 954 { child_comm: evolution } hitcount: 35 955 { child_comm: glib-pacrunner } hitcount: 59 956 957 Totals: 958 Hits: 199 959 Entries: 21 960 Dropped: 0 961 962 To manually continue having the trigger aggregate events, append 963 :cont instead. Notice that the trigger info displays as [active] 964 again, and the data has changed:: 965 966 # echo 'hist:key=child_comm:val=hitcount:size=256:cont' >> \ 967 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger 968 969 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist 970 # trigger info: hist:keys=child_comm:vals=hitcount:sort=hitcount:size=256 [active] 971 972 { child_comm: dconf worker } hitcount: 1 973 { child_comm: dconf worker } hitcount: 1 974 { child_comm: kthreadd } hitcount: 1 975 { child_comm: gdbus } hitcount: 1 976 { child_comm: ibus-daemon } hitcount: 1 977 { child_comm: Socket Thread } hitcount: 2 978 { child_comm: evolution-alarm } hitcount: 2 979 { child_comm: smbd } hitcount: 2 980 { child_comm: whoopsie } hitcount: 3 981 { child_comm: compiz } hitcount: 3 982 { child_comm: evolution-sourc } hitcount: 4 983 { child_comm: bash } hitcount: 5 984 { child_comm: pool } hitcount: 5 985 { child_comm: postgres } hitcount: 6 986 { child_comm: firefox } hitcount: 8 987 { child_comm: dhclient } hitcount: 11 988 { child_comm: emacs } hitcount: 12 989 { child_comm: dbus-daemon } hitcount: 22 990 { child_comm: nm-dispatcher.a } hitcount: 22 991 { child_comm: evolution } hitcount: 35 992 { child_comm: glib-pacrunner } hitcount: 59 993 994 Totals: 995 Hits: 206 996 Entries: 21 997 Dropped: 0 998 999 The previous example showed how to start and stop a hist trigger by 1000 appending 'pause' and 'continue' to the hist trigger command. A 1001 hist trigger can also be started in a paused state by initially 1002 starting the trigger with ':pause' appended. This allows you to 1003 start the trigger only when you're ready to start collecting data 1004 and not before. For example, you could start the trigger in a 1005 paused state, then unpause it and do something you want to measure, 1006 then pause the trigger again when done. 1007 1008 Of course, doing this manually can be difficult and error-prone, but 1009 it is possible to automatically start and stop a hist trigger based 1010 on some condition, via the enable_hist and disable_hist triggers. 1011 1012 For example, suppose we wanted to take a look at the relative 1013 weights in terms of skb length for each callpath that leads to a 1014 netif_receive_skb event when downloading a decent-sized file using 1015 wget. 1016 1017 First we set up an initially paused stacktrace trigger on the 1018 netif_receive_skb event:: 1019 1020 # echo 'hist:key=stacktrace:vals=len:pause' > \ 1021 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1022 1023 Next, we set up an 'enable_hist' trigger on the sched_process_exec 1024 event, with an 'if filename==/usr/bin/wget' filter. The effect of 1025 this new trigger is that it will 'unpause' the hist trigger we just 1026 set up on netif_receive_skb if and only if it sees a 1027 sched_process_exec event with a filename of '/usr/bin/wget'. When 1028 that happens, all netif_receive_skb events are aggregated into a 1029 hash table keyed on stacktrace:: 1030 1031 # echo 'enable_hist:net:netif_receive_skb if filename==/usr/bin/wget' > \ 1032 /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger 1033 1034 The aggregation continues until the netif_receive_skb is paused 1035 again, which is what the following disable_hist event does by 1036 creating a similar setup on the sched_process_exit event, using the 1037 filter 'comm==wget':: 1038 1039 # echo 'disable_hist:net:netif_receive_skb if comm==wget' > \ 1040 /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger 1041 1042 Whenever a process exits and the comm field of the disable_hist 1043 trigger filter matches 'comm==wget', the netif_receive_skb hist 1044 trigger is disabled. 1045 1046 The overall effect is that netif_receive_skb events are aggregated 1047 into the hash table for only the duration of the wget. Executing a 1048 wget command and then listing the 'hist' file will display the 1049 output generated by the wget command:: 1050 1051 $ wget https://www.kernel.org/pub/linux/kernel/v3.x/patch-3.19.xz 1052 1053 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist 1054 # trigger info: hist:keys=stacktrace:vals=len:sort=hitcount:size=2048 [paused] 1055 1056 { stacktrace: 1057 __netif_receive_skb_core+0x46d/0x990 1058 __netif_receive_skb+0x18/0x60 1059 netif_receive_skb_internal+0x23/0x90 1060 napi_gro_receive+0xc8/0x100 1061 ieee80211_deliver_skb+0xd6/0x270 [mac80211] 1062 ieee80211_rx_handlers+0xccf/0x22f0 [mac80211] 1063 ieee80211_prepare_and_rx_handle+0x4e7/0xc40 [mac80211] 1064 ieee80211_rx+0x31d/0x900 [mac80211] 1065 iwlagn_rx_reply_rx+0x3db/0x6f0 [iwldvm] 1066 iwl_rx_dispatch+0x8e/0xf0 [iwldvm] 1067 iwl_pcie_irq_handler+0xe3c/0x12f0 [iwlwifi] 1068 irq_thread_fn+0x20/0x50 1069 irq_thread+0x11f/0x150 1070 kthread+0xd2/0xf0 1071 ret_from_fork+0x42/0x70 1072 } hitcount: 85 len: 28884 1073 { stacktrace: 1074 __netif_receive_skb_core+0x46d/0x990 1075 __netif_receive_skb+0x18/0x60 1076 netif_receive_skb_internal+0x23/0x90 1077 napi_gro_complete+0xa4/0xe0 1078 dev_gro_receive+0x23a/0x360 1079 napi_gro_receive+0x30/0x100 1080 ieee80211_deliver_skb+0xd6/0x270 [mac80211] 1081 ieee80211_rx_handlers+0xccf/0x22f0 [mac80211] 1082 ieee80211_prepare_and_rx_handle+0x4e7/0xc40 [mac80211] 1083 ieee80211_rx+0x31d/0x900 [mac80211] 1084 iwlagn_rx_reply_rx+0x3db/0x6f0 [iwldvm] 1085 iwl_rx_dispatch+0x8e/0xf0 [iwldvm] 1086 iwl_pcie_irq_handler+0xe3c/0x12f0 [iwlwifi] 1087 irq_thread_fn+0x20/0x50 1088 irq_thread+0x11f/0x150 1089 kthread+0xd2/0xf0 1090 } hitcount: 98 len: 664329 1091 { stacktrace: 1092 __netif_receive_skb_core+0x46d/0x990 1093 __netif_receive_skb+0x18/0x60 1094 process_backlog+0xa8/0x150 1095 net_rx_action+0x15d/0x340 1096 __do_softirq+0x114/0x2c0 1097 do_softirq_own_stack+0x1c/0x30 1098 do_softirq+0x65/0x70 1099 __local_bh_enable_ip+0xb5/0xc0 1100 ip_finish_output+0x1f4/0x840 1101 ip_output+0x6b/0xc0 1102 ip_local_out_sk+0x31/0x40 1103 ip_send_skb+0x1a/0x50 1104 udp_send_skb+0x173/0x2a0 1105 udp_sendmsg+0x2bf/0x9f0 1106 inet_sendmsg+0x64/0xa0 1107 sock_sendmsg+0x3d/0x50 1108 } hitcount: 115 len: 13030 1109 { stacktrace: 1110 __netif_receive_skb_core+0x46d/0x990 1111 __netif_receive_skb+0x18/0x60 1112 netif_receive_skb_internal+0x23/0x90 1113 napi_gro_complete+0xa4/0xe0 1114 napi_gro_flush+0x6d/0x90 1115 iwl_pcie_irq_handler+0x92a/0x12f0 [iwlwifi] 1116 irq_thread_fn+0x20/0x50 1117 irq_thread+0x11f/0x150 1118 kthread+0xd2/0xf0 1119 ret_from_fork+0x42/0x70 1120 } hitcount: 934 len: 5512212 1121 1122 Totals: 1123 Hits: 1232 1124 Entries: 4 1125 Dropped: 0 1126 1127 The above shows all the netif_receive_skb callpaths and their total 1128 lengths for the duration of the wget command. 1129 1130 The 'clear' hist trigger param can be used to clear the hash table. 1131 Suppose we wanted to try another run of the previous example but 1132 this time also wanted to see the complete list of events that went 1133 into the histogram. In order to avoid having to set everything up 1134 again, we can just clear the histogram first:: 1135 1136 # echo 'hist:key=stacktrace:vals=len:clear' >> \ 1137 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1138 1139 Just to verify that it is in fact cleared, here's what we now see in 1140 the hist file:: 1141 1142 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist 1143 # trigger info: hist:keys=stacktrace:vals=len:sort=hitcount:size=2048 [paused] 1144 1145 Totals: 1146 Hits: 0 1147 Entries: 0 1148 Dropped: 0 1149 1150 Since we want to see the detailed list of every netif_receive_skb 1151 event occurring during the new run, which are in fact the same 1152 events being aggregated into the hash table, we add some additional 1153 'enable_event' events to the triggering sched_process_exec and 1154 sched_process_exit events as such:: 1155 1156 # echo 'enable_event:net:netif_receive_skb if filename==/usr/bin/wget' > \ 1157 /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger 1158 1159 # echo 'disable_event:net:netif_receive_skb if comm==wget' > \ 1160 /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger 1161 1162 If you read the trigger files for the sched_process_exec and 1163 sched_process_exit triggers, you should see two triggers for each: 1164 one enabling/disabling the hist aggregation and the other 1165 enabling/disabling the logging of events:: 1166 1167 # cat /sys/kernel/debug/tracing/events/sched/sched_process_exec/trigger 1168 enable_event:net:netif_receive_skb:unlimited if filename==/usr/bin/wget 1169 enable_hist:net:netif_receive_skb:unlimited if filename==/usr/bin/wget 1170 1171 # cat /sys/kernel/debug/tracing/events/sched/sched_process_exit/trigger 1172 enable_event:net:netif_receive_skb:unlimited if comm==wget 1173 disable_hist:net:netif_receive_skb:unlimited if comm==wget 1174 1175 In other words, whenever either of the sched_process_exec or 1176 sched_process_exit events is hit and matches 'wget', it enables or 1177 disables both the histogram and the event log, and what you end up 1178 with is a hash table and set of events just covering the specified 1179 duration. Run the wget command again:: 1180 1181 $ wget https://www.kernel.org/pub/linux/kernel/v3.x/patch-3.19.xz 1182 1183 Displaying the 'hist' file should show something similar to what you 1184 saw in the last run, but this time you should also see the 1185 individual events in the trace file:: 1186 1187 # cat /sys/kernel/debug/tracing/trace 1188 1189 # tracer: nop 1190 # 1191 # entries-in-buffer/entries-written: 183/1426 #P:4 1192 # 1193 # _-----=> irqs-off 1194 # / _----=> need-resched 1195 # | / _---=> hardirq/softirq 1196 # || / _--=> preempt-depth 1197 # ||| / delay 1198 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 1199 # | | | |||| | | 1200 wget-15108 [000] ..s1 31769.606929: netif_receive_skb: dev=lo skbaddr=ffff88009c353100 len=60 1201 wget-15108 [000] ..s1 31769.606999: netif_receive_skb: dev=lo skbaddr=ffff88009c353200 len=60 1202 dnsmasq-1382 [000] ..s1 31769.677652: netif_receive_skb: dev=lo skbaddr=ffff88009c352b00 len=130 1203 dnsmasq-1382 [000] ..s1 31769.685917: netif_receive_skb: dev=lo skbaddr=ffff88009c352200 len=138 1204 ##### CPU 2 buffer started #### 1205 irq/29-iwlwifi-559 [002] ..s. 31772.031529: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d433d00 len=2948 1206 irq/29-iwlwifi-559 [002] ..s. 31772.031572: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d432200 len=1500 1207 irq/29-iwlwifi-559 [002] ..s. 31772.032196: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d433100 len=2948 1208 irq/29-iwlwifi-559 [002] ..s. 31772.032761: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d433000 len=2948 1209 irq/29-iwlwifi-559 [002] ..s. 31772.033220: netif_receive_skb: dev=wlan0 skbaddr=ffff88009d432e00 len=1500 1210 . 1211 . 1212 . 1213 1214 The following example demonstrates how multiple hist triggers can be 1215 attached to a given event. This capability can be useful for 1216 creating a set of different summaries derived from the same set of 1217 events, or for comparing the effects of different filters, among 1218 other things:: 1219 1220 # echo 'hist:keys=skbaddr.hex:vals=len if len < 0' >> \ 1221 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1222 # echo 'hist:keys=skbaddr.hex:vals=len if len > 4096' >> \ 1223 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1224 # echo 'hist:keys=skbaddr.hex:vals=len if len == 256' >> \ 1225 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1226 # echo 'hist:keys=skbaddr.hex:vals=len' >> \ 1227 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1228 # echo 'hist:keys=len:vals=common_preempt_count' >> \ 1229 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1230 1231 The above set of commands create four triggers differing only in 1232 their filters, along with a completely different though fairly 1233 nonsensical trigger. Note that in order to append multiple hist 1234 triggers to the same file, you should use the '>>' operator to 1235 append them ('>' will also add the new hist trigger, but will remove 1236 any existing hist triggers beforehand). 1237 1238 Displaying the contents of the 'hist' file for the event shows the 1239 contents of all five histograms:: 1240 1241 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist 1242 1243 # event histogram 1244 # 1245 # trigger info: hist:keys=len:vals=hitcount,common_preempt_count:sort=hitcount:size=2048 [active] 1246 # 1247 1248 { len: 176 } hitcount: 1 common_preempt_count: 0 1249 { len: 223 } hitcount: 1 common_preempt_count: 0 1250 { len: 4854 } hitcount: 1 common_preempt_count: 0 1251 { len: 395 } hitcount: 1 common_preempt_count: 0 1252 { len: 177 } hitcount: 1 common_preempt_count: 0 1253 { len: 446 } hitcount: 1 common_preempt_count: 0 1254 { len: 1601 } hitcount: 1 common_preempt_count: 0 1255 . 1256 . 1257 . 1258 { len: 1280 } hitcount: 66 common_preempt_count: 0 1259 { len: 116 } hitcount: 81 common_preempt_count: 40 1260 { len: 708 } hitcount: 112 common_preempt_count: 0 1261 { len: 46 } hitcount: 221 common_preempt_count: 0 1262 { len: 1264 } hitcount: 458 common_preempt_count: 0 1263 1264 Totals: 1265 Hits: 1428 1266 Entries: 147 1267 Dropped: 0 1268 1269 1270 # event histogram 1271 # 1272 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 [active] 1273 # 1274 1275 { skbaddr: ffff8800baee5e00 } hitcount: 1 len: 130 1276 { skbaddr: ffff88005f3d5600 } hitcount: 1 len: 1280 1277 { skbaddr: ffff88005f3d4900 } hitcount: 1 len: 1280 1278 { skbaddr: ffff88009fed6300 } hitcount: 1 len: 115 1279 { skbaddr: ffff88009fe0ad00 } hitcount: 1 len: 115 1280 { skbaddr: ffff88008cdb1900 } hitcount: 1 len: 46 1281 { skbaddr: ffff880064b5ef00 } hitcount: 1 len: 118 1282 { skbaddr: ffff880044e3c700 } hitcount: 1 len: 60 1283 { skbaddr: ffff880100065900 } hitcount: 1 len: 46 1284 { skbaddr: ffff8800d46bd500 } hitcount: 1 len: 116 1285 { skbaddr: ffff88005f3d5f00 } hitcount: 1 len: 1280 1286 { skbaddr: ffff880100064700 } hitcount: 1 len: 365 1287 { skbaddr: ffff8800badb6f00 } hitcount: 1 len: 60 1288 . 1289 . 1290 . 1291 { skbaddr: ffff88009fe0be00 } hitcount: 27 len: 24677 1292 { skbaddr: ffff88009fe0a400 } hitcount: 27 len: 23052 1293 { skbaddr: ffff88009fe0b700 } hitcount: 31 len: 25589 1294 { skbaddr: ffff88009fe0b600 } hitcount: 32 len: 27326 1295 { skbaddr: ffff88006a462800 } hitcount: 68 len: 71678 1296 { skbaddr: ffff88006a463700 } hitcount: 70 len: 72678 1297 { skbaddr: ffff88006a462b00 } hitcount: 71 len: 77589 1298 { skbaddr: ffff88006a463600 } hitcount: 73 len: 71307 1299 { skbaddr: ffff88006a462200 } hitcount: 81 len: 81032 1300 1301 Totals: 1302 Hits: 1451 1303 Entries: 318 1304 Dropped: 0 1305 1306 1307 # event histogram 1308 # 1309 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 if len == 256 [active] 1310 # 1311 1312 1313 Totals: 1314 Hits: 0 1315 Entries: 0 1316 Dropped: 0 1317 1318 1319 # event histogram 1320 # 1321 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 if len > 4096 [active] 1322 # 1323 1324 { skbaddr: ffff88009fd2c300 } hitcount: 1 len: 7212 1325 { skbaddr: ffff8800d2bcce00 } hitcount: 1 len: 7212 1326 { skbaddr: ffff8800d2bcd700 } hitcount: 1 len: 7212 1327 { skbaddr: ffff8800d2bcda00 } hitcount: 1 len: 21492 1328 { skbaddr: ffff8800ae2e2d00 } hitcount: 1 len: 7212 1329 { skbaddr: ffff8800d2bcdb00 } hitcount: 1 len: 7212 1330 { skbaddr: ffff88006a4df500 } hitcount: 1 len: 4854 1331 { skbaddr: ffff88008ce47b00 } hitcount: 1 len: 18636 1332 { skbaddr: ffff8800ae2e2200 } hitcount: 1 len: 12924 1333 { skbaddr: ffff88005f3e1000 } hitcount: 1 len: 4356 1334 { skbaddr: ffff8800d2bcdc00 } hitcount: 2 len: 24420 1335 { skbaddr: ffff8800d2bcc200 } hitcount: 2 len: 12996 1336 1337 Totals: 1338 Hits: 14 1339 Entries: 12 1340 Dropped: 0 1341 1342 1343 # event histogram 1344 # 1345 # trigger info: hist:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 if len < 0 [active] 1346 # 1347 1348 1349 Totals: 1350 Hits: 0 1351 Entries: 0 1352 Dropped: 0 1353 1354 Named triggers can be used to have triggers share a common set of 1355 histogram data. This capability is mostly useful for combining the 1356 output of events generated by tracepoints contained inside inline 1357 functions, but names can be used in a hist trigger on any event. 1358 For example, these two triggers when hit will update the same 'len' 1359 field in the shared 'foo' histogram data:: 1360 1361 # echo 'hist:name=foo:keys=skbaddr.hex:vals=len' > \ 1362 /sys/kernel/debug/tracing/events/net/netif_receive_skb/trigger 1363 # echo 'hist:name=foo:keys=skbaddr.hex:vals=len' > \ 1364 /sys/kernel/debug/tracing/events/net/netif_rx/trigger 1365 1366 You can see that they're updating common histogram data by reading 1367 each event's hist files at the same time:: 1368 1369 # cat /sys/kernel/debug/tracing/events/net/netif_receive_skb/hist; 1370 cat /sys/kernel/debug/tracing/events/net/netif_rx/hist 1371 1372 # event histogram 1373 # 1374 # trigger info: hist:name=foo:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 [active] 1375 # 1376 1377 { skbaddr: ffff88000ad53500 } hitcount: 1 len: 46 1378 { skbaddr: ffff8800af5a1500 } hitcount: 1 len: 76 1379 { skbaddr: ffff8800d62a1900 } hitcount: 1 len: 46 1380 { skbaddr: ffff8800d2bccb00 } hitcount: 1 len: 468 1381 { skbaddr: ffff8800d3c69900 } hitcount: 1 len: 46 1382 { skbaddr: ffff88009ff09100 } hitcount: 1 len: 52 1383 { skbaddr: ffff88010f13ab00 } hitcount: 1 len: 168 1384 { skbaddr: ffff88006a54f400 } hitcount: 1 len: 46 1385 { skbaddr: ffff8800d2bcc500 } hitcount: 1 len: 260 1386 { skbaddr: ffff880064505000 } hitcount: 1 len: 46 1387 { skbaddr: ffff8800baf24e00 } hitcount: 1 len: 32 1388 { skbaddr: ffff88009fe0ad00 } hitcount: 1 len: 46 1389 { skbaddr: ffff8800d3edff00 } hitcount: 1 len: 44 1390 { skbaddr: ffff88009fe0b400 } hitcount: 1 len: 168 1391 { skbaddr: ffff8800a1c55a00 } hitcount: 1 len: 40 1392 { skbaddr: ffff8800d2bcd100 } hitcount: 1 len: 40 1393 { skbaddr: ffff880064505f00 } hitcount: 1 len: 174 1394 { skbaddr: ffff8800a8bff200 } hitcount: 1 len: 160 1395 { skbaddr: ffff880044e3cc00 } hitcount: 1 len: 76 1396 { skbaddr: ffff8800a8bfe700 } hitcount: 1 len: 46 1397 { skbaddr: ffff8800d2bcdc00 } hitcount: 1 len: 32 1398 { skbaddr: ffff8800a1f64800 } hitcount: 1 len: 46 1399 { skbaddr: ffff8800d2bcde00 } hitcount: 1 len: 988 1400 { skbaddr: ffff88006a5dea00 } hitcount: 1 len: 46 1401 { skbaddr: ffff88002e37a200 } hitcount: 1 len: 44 1402 { skbaddr: ffff8800a1f32c00 } hitcount: 2 len: 676 1403 { skbaddr: ffff88000ad52600 } hitcount: 2 len: 107 1404 { skbaddr: ffff8800a1f91e00 } hitcount: 2 len: 92 1405 { skbaddr: ffff8800af5a0200 } hitcount: 2 len: 142 1406 { skbaddr: ffff8800d2bcc600 } hitcount: 2 len: 220 1407 { skbaddr: ffff8800ba36f500 } hitcount: 2 len: 92 1408 { skbaddr: ffff8800d021f800 } hitcount: 2 len: 92 1409 { skbaddr: ffff8800a1f33600 } hitcount: 2 len: 675 1410 { skbaddr: ffff8800a8bfff00 } hitcount: 3 len: 138 1411 { skbaddr: ffff8800d62a1300 } hitcount: 3 len: 138 1412 { skbaddr: ffff88002e37a100 } hitcount: 4 len: 184 1413 { skbaddr: ffff880064504400 } hitcount: 4 len: 184 1414 { skbaddr: ffff8800a8bfec00 } hitcount: 4 len: 184 1415 { skbaddr: ffff88000ad53700 } hitcount: 5 len: 230 1416 { skbaddr: ffff8800d2bcdb00 } hitcount: 5 len: 196 1417 { skbaddr: ffff8800a1f90000 } hitcount: 6 len: 276 1418 { skbaddr: ffff88006a54f900 } hitcount: 6 len: 276 1419 1420 Totals: 1421 Hits: 81 1422 Entries: 42 1423 Dropped: 0 1424 # event histogram 1425 # 1426 # trigger info: hist:name=foo:keys=skbaddr.hex:vals=hitcount,len:sort=hitcount:size=2048 [active] 1427 # 1428 1429 { skbaddr: ffff88000ad53500 } hitcount: 1 len: 46 1430 { skbaddr: ffff8800af5a1500 } hitcount: 1 len: 76 1431 { skbaddr: ffff8800d62a1900 } hitcount: 1 len: 46 1432 { skbaddr: ffff8800d2bccb00 } hitcount: 1 len: 468 1433 { skbaddr: ffff8800d3c69900 } hitcount: 1 len: 46 1434 { skbaddr: ffff88009ff09100 } hitcount: 1 len: 52 1435 { skbaddr: ffff88010f13ab00 } hitcount: 1 len: 168 1436 { skbaddr: ffff88006a54f400 } hitcount: 1 len: 46 1437 { skbaddr: ffff8800d2bcc500 } hitcount: 1 len: 260 1438 { skbaddr: ffff880064505000 } hitcount: 1 len: 46 1439 { skbaddr: ffff8800baf24e00 } hitcount: 1 len: 32 1440 { skbaddr: ffff88009fe0ad00 } hitcount: 1 len: 46 1441 { skbaddr: ffff8800d3edff00 } hitcount: 1 len: 44 1442 { skbaddr: ffff88009fe0b400 } hitcount: 1 len: 168 1443 { skbaddr: ffff8800a1c55a00 } hitcount: 1 len: 40 1444 { skbaddr: ffff8800d2bcd100 } hitcount: 1 len: 40 1445 { skbaddr: ffff880064505f00 } hitcount: 1 len: 174 1446 { skbaddr: ffff8800a8bff200 } hitcount: 1 len: 160 1447 { skbaddr: ffff880044e3cc00 } hitcount: 1 len: 76 1448 { skbaddr: ffff8800a8bfe700 } hitcount: 1 len: 46 1449 { skbaddr: ffff8800d2bcdc00 } hitcount: 1 len: 32 1450 { skbaddr: ffff8800a1f64800 } hitcount: 1 len: 46 1451 { skbaddr: ffff8800d2bcde00 } hitcount: 1 len: 988 1452 { skbaddr: ffff88006a5dea00 } hitcount: 1 len: 46 1453 { skbaddr: ffff88002e37a200 } hitcount: 1 len: 44 1454 { skbaddr: ffff8800a1f32c00 } hitcount: 2 len: 676 1455 { skbaddr: ffff88000ad52600 } hitcount: 2 len: 107 1456 { skbaddr: ffff8800a1f91e00 } hitcount: 2 len: 92 1457 { skbaddr: ffff8800af5a0200 } hitcount: 2 len: 142 1458 { skbaddr: ffff8800d2bcc600 } hitcount: 2 len: 220 1459 { skbaddr: ffff8800ba36f500 } hitcount: 2 len: 92 1460 { skbaddr: ffff8800d021f800 } hitcount: 2 len: 92 1461 { skbaddr: ffff8800a1f33600 } hitcount: 2 len: 675 1462 { skbaddr: ffff8800a8bfff00 } hitcount: 3 len: 138 1463 { skbaddr: ffff8800d62a1300 } hitcount: 3 len: 138 1464 { skbaddr: ffff88002e37a100 } hitcount: 4 len: 184 1465 { skbaddr: ffff880064504400 } hitcount: 4 len: 184 1466 { skbaddr: ffff8800a8bfec00 } hitcount: 4 len: 184 1467 { skbaddr: ffff88000ad53700 } hitcount: 5 len: 230 1468 { skbaddr: ffff8800d2bcdb00 } hitcount: 5 len: 196 1469 { skbaddr: ffff8800a1f90000 } hitcount: 6 len: 276 1470 { skbaddr: ffff88006a54f900 } hitcount: 6 len: 276 1471 1472 Totals: 1473 Hits: 81 1474 Entries: 42 1475 Dropped: 0 1476 1477 And here's an example that shows how to combine histogram data from 1478 any two events even if they don't share any 'compatible' fields 1479 other than 'hitcount' and 'stacktrace'. These commands create a 1480 couple of triggers named 'bar' using those fields:: 1481 1482 # echo 'hist:name=bar:key=stacktrace:val=hitcount' > \ 1483 /sys/kernel/debug/tracing/events/sched/sched_process_fork/trigger 1484 # echo 'hist:name=bar:key=stacktrace:val=hitcount' > \ 1485 /sys/kernel/debug/tracing/events/net/netif_rx/trigger 1486 1487 And displaying the output of either shows some interesting if 1488 somewhat confusing output:: 1489 1490 # cat /sys/kernel/debug/tracing/events/sched/sched_process_fork/hist 1491 # cat /sys/kernel/debug/tracing/events/net/netif_rx/hist 1492 1493 # event histogram 1494 # 1495 # trigger info: hist:name=bar:keys=stacktrace:vals=hitcount:sort=hitcount:size=2048 [active] 1496 # 1497 1498 { stacktrace: 1499 kernel_clone+0x18e/0x330 1500 kernel_thread+0x29/0x30 1501 kthreadd+0x154/0x1b0 1502 ret_from_fork+0x3f/0x70 1503 } hitcount: 1 1504 { stacktrace: 1505 netif_rx_internal+0xb2/0xd0 1506 netif_rx_ni+0x20/0x70 1507 dev_loopback_xmit+0xaa/0xd0 1508 ip_mc_output+0x126/0x240 1509 ip_local_out_sk+0x31/0x40 1510 igmp_send_report+0x1e9/0x230 1511 igmp_timer_expire+0xe9/0x120 1512 call_timer_fn+0x39/0xf0 1513 run_timer_softirq+0x1e1/0x290 1514 __do_softirq+0xfd/0x290 1515 irq_exit+0x98/0xb0 1516 smp_apic_timer_interrupt+0x4a/0x60 1517 apic_timer_interrupt+0x6d/0x80 1518 cpuidle_enter+0x17/0x20 1519 call_cpuidle+0x3b/0x60 1520 cpu_startup_entry+0x22d/0x310 1521 } hitcount: 1 1522 { stacktrace: 1523 netif_rx_internal+0xb2/0xd0 1524 netif_rx_ni+0x20/0x70 1525 dev_loopback_xmit+0xaa/0xd0 1526 ip_mc_output+0x17f/0x240 1527 ip_local_out_sk+0x31/0x40 1528 ip_send_skb+0x1a/0x50 1529 udp_send_skb+0x13e/0x270 1530 udp_sendmsg+0x2bf/0x980 1531 inet_sendmsg+0x67/0xa0 1532 sock_sendmsg+0x38/0x50 1533 SYSC_sendto+0xef/0x170 1534 SyS_sendto+0xe/0x10 1535 entry_SYSCALL_64_fastpath+0x12/0x6a 1536 } hitcount: 2 1537 { stacktrace: 1538 netif_rx_internal+0xb2/0xd0 1539 netif_rx+0x1c/0x60 1540 loopback_xmit+0x6c/0xb0 1541 dev_hard_start_xmit+0x219/0x3a0 1542 __dev_queue_xmit+0x415/0x4f0 1543 dev_queue_xmit_sk+0x13/0x20 1544 ip_finish_output2+0x237/0x340 1545 ip_finish_output+0x113/0x1d0 1546 ip_output+0x66/0xc0 1547 ip_local_out_sk+0x31/0x40 1548 ip_send_skb+0x1a/0x50 1549 udp_send_skb+0x16d/0x270 1550 udp_sendmsg+0x2bf/0x980 1551 inet_sendmsg+0x67/0xa0 1552 sock_sendmsg+0x38/0x50 1553 ___sys_sendmsg+0x14e/0x270 1554 } hitcount: 76 1555 { stacktrace: 1556 netif_rx_internal+0xb2/0xd0 1557 netif_rx+0x1c/0x60 1558 loopback_xmit+0x6c/0xb0 1559 dev_hard_start_xmit+0x219/0x3a0 1560 __dev_queue_xmit+0x415/0x4f0 1561 dev_queue_xmit_sk+0x13/0x20 1562 ip_finish_output2+0x237/0x340 1563 ip_finish_output+0x113/0x1d0 1564 ip_output+0x66/0xc0 1565 ip_local_out_sk+0x31/0x40 1566 ip_send_skb+0x1a/0x50 1567 udp_send_skb+0x16d/0x270 1568 udp_sendmsg+0x2bf/0x980 1569 inet_sendmsg+0x67/0xa0 1570 sock_sendmsg+0x38/0x50 1571 ___sys_sendmsg+0x269/0x270 1572 } hitcount: 77 1573 { stacktrace: 1574 netif_rx_internal+0xb2/0xd0 1575 netif_rx+0x1c/0x60 1576 loopback_xmit+0x6c/0xb0 1577 dev_hard_start_xmit+0x219/0x3a0 1578 __dev_queue_xmit+0x415/0x4f0 1579 dev_queue_xmit_sk+0x13/0x20 1580 ip_finish_output2+0x237/0x340 1581 ip_finish_output+0x113/0x1d0 1582 ip_output+0x66/0xc0 1583 ip_local_out_sk+0x31/0x40 1584 ip_send_skb+0x1a/0x50 1585 udp_send_skb+0x16d/0x270 1586 udp_sendmsg+0x2bf/0x980 1587 inet_sendmsg+0x67/0xa0 1588 sock_sendmsg+0x38/0x50 1589 SYSC_sendto+0xef/0x170 1590 } hitcount: 88 1591 { stacktrace: 1592 kernel_clone+0x18e/0x330 1593 SyS_clone+0x19/0x20 1594 entry_SYSCALL_64_fastpath+0x12/0x6a 1595 } hitcount: 244 1596 1597 Totals: 1598 Hits: 489 1599 Entries: 7 1600 Dropped: 0 1601 16022.2 Inter-event hist triggers 1603----------------------------- 1604 1605Inter-event hist triggers are hist triggers that combine values from 1606one or more other events and create a histogram using that data. Data 1607from an inter-event histogram can in turn become the source for 1608further combined histograms, thus providing a chain of related 1609histograms, which is important for some applications. 1610 1611The most important example of an inter-event quantity that can be used 1612in this manner is latency, which is simply a difference in timestamps 1613between two events. Although latency is the most important 1614inter-event quantity, note that because the support is completely 1615general across the trace event subsystem, any event field can be used 1616in an inter-event quantity. 1617 1618An example of a histogram that combines data from other histograms 1619into a useful chain would be a 'wakeupswitch latency' histogram that 1620combines a 'wakeup latency' histogram and a 'switch latency' 1621histogram. 1622 1623Normally, a hist trigger specification consists of a (possibly 1624compound) key along with one or more numeric values, which are 1625continually updated sums associated with that key. A histogram 1626specification in this case consists of individual key and value 1627specifications that refer to trace event fields associated with a 1628single event type. 1629 1630The inter-event hist trigger extension allows fields from multiple 1631events to be referenced and combined into a multi-event histogram 1632specification. In support of this overall goal, a few enabling 1633features have been added to the hist trigger support: 1634 1635 - In order to compute an inter-event quantity, a value from one 1636 event needs to saved and then referenced from another event. This 1637 requires the introduction of support for histogram 'variables'. 1638 1639 - The computation of inter-event quantities and their combination 1640 require some minimal amount of support for applying simple 1641 expressions to variables (+ and -). 1642 1643 - A histogram consisting of inter-event quantities isn't logically a 1644 histogram on either event (so having the 'hist' file for either 1645 event host the histogram output doesn't really make sense). To 1646 address the idea that the histogram is associated with a 1647 combination of events, support is added allowing the creation of 1648 'synthetic' events that are events derived from other events. 1649 These synthetic events are full-fledged events just like any other 1650 and can be used as such, as for instance to create the 1651 'combination' histograms mentioned previously. 1652 1653 - A set of 'actions' can be associated with histogram entries - 1654 these can be used to generate the previously mentioned synthetic 1655 events, but can also be used for other purposes, such as for 1656 example saving context when a 'max' latency has been hit. 1657 1658 - Trace events don't have a 'timestamp' associated with them, but 1659 there is an implicit timestamp saved along with an event in the 1660 underlying ftrace ring buffer. This timestamp is now exposed as a 1661 a synthetic field named 'common_timestamp' which can be used in 1662 histograms as if it were any other event field; it isn't an actual 1663 field in the trace format but rather is a synthesized value that 1664 nonetheless can be used as if it were an actual field. By default 1665 it is in units of nanoseconds; appending '.usecs' to a 1666 common_timestamp field changes the units to microseconds. 1667 1668A note on inter-event timestamps: If common_timestamp is used in a 1669histogram, the trace buffer is automatically switched over to using 1670absolute timestamps and the "global" trace clock, in order to avoid 1671bogus timestamp differences with other clocks that aren't coherent 1672across CPUs. This can be overridden by specifying one of the other 1673trace clocks instead, using the "clock=XXX" hist trigger attribute, 1674where XXX is any of the clocks listed in the tracing/trace_clock 1675pseudo-file. 1676 1677These features are described in more detail in the following sections. 1678 16792.2.1 Histogram Variables 1680------------------------- 1681 1682Variables are simply named locations used for saving and retrieving 1683values between matching events. A 'matching' event is defined as an 1684event that has a matching key - if a variable is saved for a histogram 1685entry corresponding to that key, any subsequent event with a matching 1686key can access that variable. 1687 1688A variable's value is normally available to any subsequent event until 1689it is set to something else by a subsequent event. The one exception 1690to that rule is that any variable used in an expression is essentially 1691'read-once' - once it's used by an expression in a subsequent event, 1692it's reset to its 'unset' state, which means it can't be used again 1693unless it's set again. This ensures not only that an event doesn't 1694use an uninitialized variable in a calculation, but that that variable 1695is used only once and not for any unrelated subsequent match. 1696 1697The basic syntax for saving a variable is to simply prefix a unique 1698variable name not corresponding to any keyword along with an '=' sign 1699to any event field. 1700 1701Either keys or values can be saved and retrieved in this way. This 1702creates a variable named 'ts0' for a histogram entry with the key 1703'next_pid':: 1704 1705 # echo 'hist:keys=next_pid:vals=$ts0:ts0=common_timestamp ... >> \ 1706 event/trigger 1707 1708The ts0 variable can be accessed by any subsequent event having the 1709same pid as 'next_pid'. 1710 1711Variable references are formed by prepending the variable name with 1712the '$' sign. Thus for example, the ts0 variable above would be 1713referenced as '$ts0' in expressions. 1714 1715Because 'vals=' is used, the common_timestamp variable value above 1716will also be summed as a normal histogram value would (though for a 1717timestamp it makes little sense). 1718 1719The below shows that a key value can also be saved in the same way:: 1720 1721 # echo 'hist:timer_pid=common_pid:key=timer_pid ...' >> event/trigger 1722 1723If a variable isn't a key variable or prefixed with 'vals=', the 1724associated event field will be saved in a variable but won't be summed 1725as a value:: 1726 1727 # echo 'hist:keys=next_pid:ts1=common_timestamp ...' >> event/trigger 1728 1729Multiple variables can be assigned at the same time. The below would 1730result in both ts0 and b being created as variables, with both 1731common_timestamp and field1 additionally being summed as values:: 1732 1733 # echo 'hist:keys=pid:vals=$ts0,$b:ts0=common_timestamp,b=field1 ...' >> \ 1734 event/trigger 1735 1736Note that variable assignments can appear either preceding or 1737following their use. The command below behaves identically to the 1738command above:: 1739 1740 # echo 'hist:keys=pid:ts0=common_timestamp,b=field1:vals=$ts0,$b ...' >> \ 1741 event/trigger 1742 1743Any number of variables not bound to a 'vals=' prefix can also be 1744assigned by simply separating them with colons. Below is the same 1745thing but without the values being summed in the histogram:: 1746 1747 # echo 'hist:keys=pid:ts0=common_timestamp:b=field1 ...' >> event/trigger 1748 1749Variables set as above can be referenced and used in expressions on 1750another event. 1751 1752For example, here's how a latency can be calculated:: 1753 1754 # echo 'hist:keys=pid,prio:ts0=common_timestamp ...' >> event1/trigger 1755 # echo 'hist:keys=next_pid:wakeup_lat=common_timestamp-$ts0 ...' >> event2/trigger 1756 1757In the first line above, the event's timestamp is saved into the 1758variable ts0. In the next line, ts0 is subtracted from the second 1759event's timestamp to produce the latency, which is then assigned into 1760yet another variable, 'wakeup_lat'. The hist trigger below in turn 1761makes use of the wakeup_lat variable to compute a combined latency 1762using the same key and variable from yet another event:: 1763 1764 # echo 'hist:key=pid:wakeupswitch_lat=$wakeup_lat+$switchtime_lat ...' >> event3/trigger 1765 1766Expressions support the use of addition, subtraction, multiplication and 1767division operators (+-\*/). 1768 1769Note if division by zero cannot be detected at parse time (i.e. the 1770divisor is not a constant), the result will be -1. 1771 1772Numeric constants can also be used directly in an expression:: 1773 1774 # echo 'hist:keys=next_pid:timestamp_secs=common_timestamp/1000000 ...' >> event/trigger 1775 1776or assigned to a variable and referenced in a subsequent expression:: 1777 1778 # echo 'hist:keys=next_pid:us_per_sec=1000000 ...' >> event/trigger 1779 # echo 'hist:keys=next_pid:timestamp_secs=common_timestamp/$us_per_sec ...' >> event/trigger 1780 17812.2.2 Synthetic Events 1782---------------------- 1783 1784Synthetic events are user-defined events generated from hist trigger 1785variables or fields associated with one or more other events. Their 1786purpose is to provide a mechanism for displaying data spanning 1787multiple events consistent with the existing and already familiar 1788usage for normal events. 1789 1790To define a synthetic event, the user writes a simple specification 1791consisting of the name of the new event along with one or more 1792variables and their types, which can be any valid field type, 1793separated by semicolons, to the tracing/synthetic_events file. 1794 1795See synth_field_size() for available types. 1796 1797If field_name contains [n], the field is considered to be a static array. 1798 1799If field_names contains[] (no subscript), the field is considered to 1800be a dynamic array, which will only take as much space in the event as 1801is required to hold the array. 1802 1803A string field can be specified using either the static notation: 1804 1805 char name[32]; 1806 1807Or the dynamic: 1808 1809 char name[]; 1810 1811The size limit for either is 256. 1812 1813For instance, the following creates a new event named 'wakeup_latency' 1814with 3 fields: lat, pid, and prio. Each of those fields is simply a 1815variable reference to a variable on another event:: 1816 1817 # echo 'wakeup_latency \ 1818 u64 lat; \ 1819 pid_t pid; \ 1820 int prio' >> \ 1821 /sys/kernel/debug/tracing/synthetic_events 1822 1823Reading the tracing/synthetic_events file lists all the currently 1824defined synthetic events, in this case the event defined above:: 1825 1826 # cat /sys/kernel/debug/tracing/synthetic_events 1827 wakeup_latency u64 lat; pid_t pid; int prio 1828 1829An existing synthetic event definition can be removed by prepending 1830the command that defined it with a '!':: 1831 1832 # echo '!wakeup_latency u64 lat pid_t pid int prio' >> \ 1833 /sys/kernel/debug/tracing/synthetic_events 1834 1835At this point, there isn't yet an actual 'wakeup_latency' event 1836instantiated in the event subsystem - for this to happen, a 'hist 1837trigger action' needs to be instantiated and bound to actual fields 1838and variables defined on other events (see Section 2.2.3 below on 1839how that is done using hist trigger 'onmatch' action). Once that is 1840done, the 'wakeup_latency' synthetic event instance is created. 1841 1842The new event is created under the tracing/events/synthetic/ directory 1843and looks and behaves just like any other event:: 1844 1845 # ls /sys/kernel/debug/tracing/events/synthetic/wakeup_latency 1846 enable filter format hist id trigger 1847 1848A histogram can now be defined for the new synthetic event:: 1849 1850 # echo 'hist:keys=pid,prio,lat.log2:sort=lat' >> \ 1851 /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/trigger 1852 1853The above shows the latency "lat" in a power of 2 grouping. 1854 1855Like any other event, once a histogram is enabled for the event, the 1856output can be displayed by reading the event's 'hist' file. 1857 1858 # cat /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/hist 1859 1860 # event histogram 1861 # 1862 # trigger info: hist:keys=pid,prio,lat.log2:vals=hitcount:sort=lat.log2:size=2048 [active] 1863 # 1864 1865 { pid: 2035, prio: 9, lat: ~ 2^2 } hitcount: 43 1866 { pid: 2034, prio: 9, lat: ~ 2^2 } hitcount: 60 1867 { pid: 2029, prio: 9, lat: ~ 2^2 } hitcount: 965 1868 { pid: 2034, prio: 120, lat: ~ 2^2 } hitcount: 9 1869 { pid: 2033, prio: 120, lat: ~ 2^2 } hitcount: 5 1870 { pid: 2030, prio: 9, lat: ~ 2^2 } hitcount: 335 1871 { pid: 2030, prio: 120, lat: ~ 2^2 } hitcount: 10 1872 { pid: 2032, prio: 120, lat: ~ 2^2 } hitcount: 1 1873 { pid: 2035, prio: 120, lat: ~ 2^2 } hitcount: 2 1874 { pid: 2031, prio: 9, lat: ~ 2^2 } hitcount: 176 1875 { pid: 2028, prio: 120, lat: ~ 2^2 } hitcount: 15 1876 { pid: 2033, prio: 9, lat: ~ 2^2 } hitcount: 91 1877 { pid: 2032, prio: 9, lat: ~ 2^2 } hitcount: 125 1878 { pid: 2029, prio: 120, lat: ~ 2^2 } hitcount: 4 1879 { pid: 2031, prio: 120, lat: ~ 2^2 } hitcount: 3 1880 { pid: 2029, prio: 120, lat: ~ 2^3 } hitcount: 2 1881 { pid: 2035, prio: 9, lat: ~ 2^3 } hitcount: 41 1882 { pid: 2030, prio: 120, lat: ~ 2^3 } hitcount: 1 1883 { pid: 2032, prio: 9, lat: ~ 2^3 } hitcount: 32 1884 { pid: 2031, prio: 9, lat: ~ 2^3 } hitcount: 44 1885 { pid: 2034, prio: 9, lat: ~ 2^3 } hitcount: 40 1886 { pid: 2030, prio: 9, lat: ~ 2^3 } hitcount: 29 1887 { pid: 2033, prio: 9, lat: ~ 2^3 } hitcount: 31 1888 { pid: 2029, prio: 9, lat: ~ 2^3 } hitcount: 31 1889 { pid: 2028, prio: 120, lat: ~ 2^3 } hitcount: 18 1890 { pid: 2031, prio: 120, lat: ~ 2^3 } hitcount: 2 1891 { pid: 2028, prio: 120, lat: ~ 2^4 } hitcount: 1 1892 { pid: 2029, prio: 9, lat: ~ 2^4 } hitcount: 4 1893 { pid: 2031, prio: 120, lat: ~ 2^7 } hitcount: 1 1894 { pid: 2032, prio: 120, lat: ~ 2^7 } hitcount: 1 1895 1896 Totals: 1897 Hits: 2122 1898 Entries: 30 1899 Dropped: 0 1900 1901 1902The latency values can also be grouped linearly by a given size with 1903the ".buckets" modifier and specify a size (in this case groups of 10). 1904 1905 # echo 'hist:keys=pid,prio,lat.buckets=10:sort=lat' >> \ 1906 /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/trigger 1907 1908 # event histogram 1909 # 1910 # trigger info: hist:keys=pid,prio,lat.buckets=10:vals=hitcount:sort=lat.buckets=10:size=2048 [active] 1911 # 1912 1913 { pid: 2067, prio: 9, lat: ~ 0-9 } hitcount: 220 1914 { pid: 2068, prio: 9, lat: ~ 0-9 } hitcount: 157 1915 { pid: 2070, prio: 9, lat: ~ 0-9 } hitcount: 100 1916 { pid: 2067, prio: 120, lat: ~ 0-9 } hitcount: 6 1917 { pid: 2065, prio: 120, lat: ~ 0-9 } hitcount: 2 1918 { pid: 2066, prio: 120, lat: ~ 0-9 } hitcount: 2 1919 { pid: 2069, prio: 9, lat: ~ 0-9 } hitcount: 122 1920 { pid: 2069, prio: 120, lat: ~ 0-9 } hitcount: 8 1921 { pid: 2070, prio: 120, lat: ~ 0-9 } hitcount: 1 1922 { pid: 2068, prio: 120, lat: ~ 0-9 } hitcount: 7 1923 { pid: 2066, prio: 9, lat: ~ 0-9 } hitcount: 365 1924 { pid: 2064, prio: 120, lat: ~ 0-9 } hitcount: 35 1925 { pid: 2065, prio: 9, lat: ~ 0-9 } hitcount: 998 1926 { pid: 2071, prio: 9, lat: ~ 0-9 } hitcount: 85 1927 { pid: 2065, prio: 9, lat: ~ 10-19 } hitcount: 2 1928 { pid: 2064, prio: 120, lat: ~ 10-19 } hitcount: 2 1929 1930 Totals: 1931 Hits: 2112 1932 Entries: 16 1933 Dropped: 0 1934 19352.2.3 Hist trigger 'handlers' and 'actions' 1936------------------------------------------- 1937 1938A hist trigger 'action' is a function that's executed (in most cases 1939conditionally) whenever a histogram entry is added or updated. 1940 1941When a histogram entry is added or updated, a hist trigger 'handler' 1942is what decides whether the corresponding action is actually invoked 1943or not. 1944 1945Hist trigger handlers and actions are paired together in the general 1946form: 1947 1948 <handler>.<action> 1949 1950To specify a handler.action pair for a given event, simply specify 1951that handler.action pair between colons in the hist trigger 1952specification. 1953 1954In theory, any handler can be combined with any action, but in 1955practice, not every handler.action combination is currently supported; 1956if a given handler.action combination isn't supported, the hist 1957trigger will fail with -EINVAL; 1958 1959The default 'handler.action' if none is explicitly specified is as it 1960always has been, to simply update the set of values associated with an 1961entry. Some applications, however, may want to perform additional 1962actions at that point, such as generate another event, or compare and 1963save a maximum. 1964 1965The supported handlers and actions are listed below, and each is 1966described in more detail in the following paragraphs, in the context 1967of descriptions of some common and useful handler.action combinations. 1968 1969The available handlers are: 1970 1971 - onmatch(matching.event) - invoke action on any addition or update 1972 - onmax(var) - invoke action if var exceeds current max 1973 - onchange(var) - invoke action if var changes 1974 1975The available actions are: 1976 1977 - trace(<synthetic_event_name>,param list) - generate synthetic event 1978 - save(field,...) - save current event fields 1979 - snapshot() - snapshot the trace buffer 1980 1981The following commonly-used handler.action pairs are available: 1982 1983 - onmatch(matching.event).trace(<synthetic_event_name>,param list) 1984 1985 The 'onmatch(matching.event).trace(<synthetic_event_name>,param 1986 list)' hist trigger action is invoked whenever an event matches 1987 and the histogram entry would be added or updated. It causes the 1988 named synthetic event to be generated with the values given in the 1989 'param list'. The result is the generation of a synthetic event 1990 that consists of the values contained in those variables at the 1991 time the invoking event was hit. For example, if the synthetic 1992 event name is 'wakeup_latency', a wakeup_latency event is 1993 generated using onmatch(event).trace(wakeup_latency,arg1,arg2). 1994 1995 There is also an equivalent alternative form available for 1996 generating synthetic events. In this form, the synthetic event 1997 name is used as if it were a function name. For example, using 1998 the 'wakeup_latency' synthetic event name again, the 1999 wakeup_latency event would be generated by invoking it as if it 2000 were a function call, with the event field values passed in as 2001 arguments: onmatch(event).wakeup_latency(arg1,arg2). The syntax 2002 for this form is: 2003 2004 onmatch(matching.event).<synthetic_event_name>(param list) 2005 2006 In either case, the 'param list' consists of one or more 2007 parameters which may be either variables or fields defined on 2008 either the 'matching.event' or the target event. The variables or 2009 fields specified in the param list may be either fully-qualified 2010 or unqualified. If a variable is specified as unqualified, it 2011 must be unique between the two events. A field name used as a 2012 param can be unqualified if it refers to the target event, but 2013 must be fully qualified if it refers to the matching event. A 2014 fully-qualified name is of the form 'system.event_name.$var_name' 2015 or 'system.event_name.field'. 2016 2017 The 'matching.event' specification is simply the fully qualified 2018 event name of the event that matches the target event for the 2019 onmatch() functionality, in the form 'system.event_name'. Histogram 2020 keys of both events are compared to find if events match. In case 2021 multiple histogram keys are used, they all must match in the specified 2022 order. 2023 2024 Finally, the number and type of variables/fields in the 'param 2025 list' must match the number and types of the fields in the 2026 synthetic event being generated. 2027 2028 As an example the below defines a simple synthetic event and uses 2029 a variable defined on the sched_wakeup_new event as a parameter 2030 when invoking the synthetic event. Here we define the synthetic 2031 event:: 2032 2033 # echo 'wakeup_new_test pid_t pid' >> \ 2034 /sys/kernel/debug/tracing/synthetic_events 2035 2036 # cat /sys/kernel/debug/tracing/synthetic_events 2037 wakeup_new_test pid_t pid 2038 2039 The following hist trigger both defines the missing testpid 2040 variable and specifies an onmatch() action that generates a 2041 wakeup_new_test synthetic event whenever a sched_wakeup_new event 2042 occurs, which because of the 'if comm == "cyclictest"' filter only 2043 happens when the executable is cyclictest:: 2044 2045 # echo 'hist:keys=$testpid:testpid=pid:onmatch(sched.sched_wakeup_new).\ 2046 wakeup_new_test($testpid) if comm=="cyclictest"' >> \ 2047 /sys/kernel/debug/tracing/events/sched/sched_wakeup_new/trigger 2048 2049 Or, equivalently, using the 'trace' keyword syntax: 2050 2051 # echo 'hist:keys=$testpid:testpid=pid:onmatch(sched.sched_wakeup_new).\ 2052 trace(wakeup_new_test,$testpid) if comm=="cyclictest"' >> \ 2053 /sys/kernel/debug/tracing/events/sched/sched_wakeup_new/trigger 2054 2055 Creating and displaying a histogram based on those events is now 2056 just a matter of using the fields and new synthetic event in the 2057 tracing/events/synthetic directory, as usual:: 2058 2059 # echo 'hist:keys=pid:sort=pid' >> \ 2060 /sys/kernel/debug/tracing/events/synthetic/wakeup_new_test/trigger 2061 2062 Running 'cyclictest' should cause wakeup_new events to generate 2063 wakeup_new_test synthetic events which should result in histogram 2064 output in the wakeup_new_test event's hist file:: 2065 2066 # cat /sys/kernel/debug/tracing/events/synthetic/wakeup_new_test/hist 2067 2068 A more typical usage would be to use two events to calculate a 2069 latency. The following example uses a set of hist triggers to 2070 produce a 'wakeup_latency' histogram. 2071 2072 First, we define a 'wakeup_latency' synthetic event:: 2073 2074 # echo 'wakeup_latency u64 lat; pid_t pid; int prio' >> \ 2075 /sys/kernel/debug/tracing/synthetic_events 2076 2077 Next, we specify that whenever we see a sched_waking event for a 2078 cyclictest thread, save the timestamp in a 'ts0' variable:: 2079 2080 # echo 'hist:keys=$saved_pid:saved_pid=pid:ts0=common_timestamp.usecs \ 2081 if comm=="cyclictest"' >> \ 2082 /sys/kernel/debug/tracing/events/sched/sched_waking/trigger 2083 2084 Then, when the corresponding thread is actually scheduled onto the 2085 CPU by a sched_switch event (saved_pid matches next_pid), calculate 2086 the latency and use that along with another variable and an event field 2087 to generate a wakeup_latency synthetic event:: 2088 2089 # echo 'hist:keys=next_pid:wakeup_lat=common_timestamp.usecs-$ts0:\ 2090 onmatch(sched.sched_waking).wakeup_latency($wakeup_lat,\ 2091 $saved_pid,next_prio) if next_comm=="cyclictest"' >> \ 2092 /sys/kernel/debug/tracing/events/sched/sched_switch/trigger 2093 2094 We also need to create a histogram on the wakeup_latency synthetic 2095 event in order to aggregate the generated synthetic event data:: 2096 2097 # echo 'hist:keys=pid,prio,lat:sort=pid,lat' >> \ 2098 /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/trigger 2099 2100 Finally, once we've run cyclictest to actually generate some 2101 events, we can see the output by looking at the wakeup_latency 2102 synthetic event's hist file:: 2103 2104 # cat /sys/kernel/debug/tracing/events/synthetic/wakeup_latency/hist 2105 2106 - onmax(var).save(field,.. .) 2107 2108 The 'onmax(var).save(field,...)' hist trigger action is invoked 2109 whenever the value of 'var' associated with a histogram entry 2110 exceeds the current maximum contained in that variable. 2111 2112 The end result is that the trace event fields specified as the 2113 onmax.save() params will be saved if 'var' exceeds the current 2114 maximum for that hist trigger entry. This allows context from the 2115 event that exhibited the new maximum to be saved for later 2116 reference. When the histogram is displayed, additional fields 2117 displaying the saved values will be printed. 2118 2119 As an example the below defines a couple of hist triggers, one for 2120 sched_waking and another for sched_switch, keyed on pid. Whenever 2121 a sched_waking occurs, the timestamp is saved in the entry 2122 corresponding to the current pid, and when the scheduler switches 2123 back to that pid, the timestamp difference is calculated. If the 2124 resulting latency, stored in wakeup_lat, exceeds the current 2125 maximum latency, the values specified in the save() fields are 2126 recorded:: 2127 2128 # echo 'hist:keys=pid:ts0=common_timestamp.usecs \ 2129 if comm=="cyclictest"' >> \ 2130 /sys/kernel/debug/tracing/events/sched/sched_waking/trigger 2131 2132 # echo 'hist:keys=next_pid:\ 2133 wakeup_lat=common_timestamp.usecs-$ts0:\ 2134 onmax($wakeup_lat).save(next_comm,prev_pid,prev_prio,prev_comm) \ 2135 if next_comm=="cyclictest"' >> \ 2136 /sys/kernel/debug/tracing/events/sched/sched_switch/trigger 2137 2138 When the histogram is displayed, the max value and the saved 2139 values corresponding to the max are displayed following the rest 2140 of the fields:: 2141 2142 # cat /sys/kernel/debug/tracing/events/sched/sched_switch/hist 2143 { next_pid: 2255 } hitcount: 239 2144 common_timestamp-ts0: 0 2145 max: 27 2146 next_comm: cyclictest 2147 prev_pid: 0 prev_prio: 120 prev_comm: swapper/1 2148 2149 { next_pid: 2256 } hitcount: 2355 2150 common_timestamp-ts0: 0 2151 max: 49 next_comm: cyclictest 2152 prev_pid: 0 prev_prio: 120 prev_comm: swapper/0 2153 2154 Totals: 2155 Hits: 12970 2156 Entries: 2 2157 Dropped: 0 2158 2159 - onmax(var).snapshot() 2160 2161 The 'onmax(var).snapshot()' hist trigger action is invoked 2162 whenever the value of 'var' associated with a histogram entry 2163 exceeds the current maximum contained in that variable. 2164 2165 The end result is that a global snapshot of the trace buffer will 2166 be saved in the tracing/snapshot file if 'var' exceeds the current 2167 maximum for any hist trigger entry. 2168 2169 Note that in this case the maximum is a global maximum for the 2170 current trace instance, which is the maximum across all buckets of 2171 the histogram. The key of the specific trace event that caused 2172 the global maximum and the global maximum itself are displayed, 2173 along with a message stating that a snapshot has been taken and 2174 where to find it. The user can use the key information displayed 2175 to locate the corresponding bucket in the histogram for even more 2176 detail. 2177 2178 As an example the below defines a couple of hist triggers, one for 2179 sched_waking and another for sched_switch, keyed on pid. Whenever 2180 a sched_waking event occurs, the timestamp is saved in the entry 2181 corresponding to the current pid, and when the scheduler switches 2182 back to that pid, the timestamp difference is calculated. If the 2183 resulting latency, stored in wakeup_lat, exceeds the current 2184 maximum latency, a snapshot is taken. As part of the setup, all 2185 the scheduler events are also enabled, which are the events that 2186 will show up in the snapshot when it is taken at some point: 2187 2188 # echo 1 > /sys/kernel/debug/tracing/events/sched/enable 2189 2190 # echo 'hist:keys=pid:ts0=common_timestamp.usecs \ 2191 if comm=="cyclictest"' >> \ 2192 /sys/kernel/debug/tracing/events/sched/sched_waking/trigger 2193 2194 # echo 'hist:keys=next_pid:wakeup_lat=common_timestamp.usecs-$ts0: \ 2195 onmax($wakeup_lat).save(next_prio,next_comm,prev_pid,prev_prio, \ 2196 prev_comm):onmax($wakeup_lat).snapshot() \ 2197 if next_comm=="cyclictest"' >> \ 2198 /sys/kernel/debug/tracing/events/sched/sched_switch/trigger 2199 2200 When the histogram is displayed, for each bucket the max value 2201 and the saved values corresponding to the max are displayed 2202 following the rest of the fields. 2203 2204 If a snapshot was taken, there is also a message indicating that, 2205 along with the value and event that triggered the global maximum: 2206 2207 # cat /sys/kernel/debug/tracing/events/sched/sched_switch/hist 2208 { next_pid: 2101 } hitcount: 200 2209 max: 52 next_prio: 120 next_comm: cyclictest \ 2210 prev_pid: 0 prev_prio: 120 prev_comm: swapper/6 2211 2212 { next_pid: 2103 } hitcount: 1326 2213 max: 572 next_prio: 19 next_comm: cyclictest \ 2214 prev_pid: 0 prev_prio: 120 prev_comm: swapper/1 2215 2216 { next_pid: 2102 } hitcount: 1982 \ 2217 max: 74 next_prio: 19 next_comm: cyclictest \ 2218 prev_pid: 0 prev_prio: 120 prev_comm: swapper/5 2219 2220 Snapshot taken (see tracing/snapshot). Details: 2221 triggering value { onmax($wakeup_lat) }: 572 \ 2222 triggered by event with key: { next_pid: 2103 } 2223 2224 Totals: 2225 Hits: 3508 2226 Entries: 3 2227 Dropped: 0 2228 2229 In the above case, the event that triggered the global maximum has 2230 the key with next_pid == 2103. If you look at the bucket that has 2231 2103 as the key, you'll find the additional values save()'d along 2232 with the local maximum for that bucket, which should be the same 2233 as the global maximum (since that was the same value that 2234 triggered the global snapshot). 2235 2236 And finally, looking at the snapshot data should show at or near 2237 the end the event that triggered the snapshot (in this case you 2238 can verify the timestamps between the sched_waking and 2239 sched_switch events, which should match the time displayed in the 2240 global maximum):: 2241 2242 # cat /sys/kernel/debug/tracing/snapshot 2243 2244 <...>-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 2245 <idle>-0 [005] d.h3 309.873611: sched_waking: comm=cyclictest pid=2102 prio=19 target_cpu=005 2246 <idle>-0 [005] dNh4 309.873613: sched_wakeup: comm=cyclictest pid=2102 prio=19 target_cpu=005 2247 <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 2248 <...>-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 2249 <idle>-0 [005] d.h3 309.874624: sched_waking: comm=cyclictest pid=2102 prio=19 target_cpu=005 2250 <idle>-0 [005] dNh4 309.874626: sched_wakeup: comm=cyclictest pid=2102 prio=19 target_cpu=005 2251 <idle>-0 [005] dNh3 309.874628: sched_waking: comm=cyclictest pid=2103 prio=19 target_cpu=005 2252 <idle>-0 [005] dNh4 309.874630: sched_wakeup: comm=cyclictest pid=2103 prio=19 target_cpu=005 2253 <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 2254 <idle>-0 [004] d.h3 309.874757: sched_waking: comm=gnome-terminal- pid=1699 prio=120 target_cpu=004 2255 <idle>-0 [004] dNh4 309.874762: sched_wakeup: comm=gnome-terminal- pid=1699 prio=120 target_cpu=004 2256 <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 2257 gnome-terminal--1699 [004] d.h2 309.874941: sched_stat_runtime: comm=gnome-terminal- pid=1699 runtime=180706 [ns] vruntime=1126870572 [ns] 2258 <idle>-0 [003] d.s4 309.874956: sched_waking: comm=rcu_sched pid=9 prio=120 target_cpu=007 2259 <idle>-0 [003] d.s5 309.874960: sched_wake_idle_without_ipi: cpu=7 2260 <idle>-0 [003] d.s5 309.874961: sched_wakeup: comm=rcu_sched pid=9 prio=120 target_cpu=007 2261 <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 2262 rcu_sched-9 [007] d..3 309.874973: sched_stat_runtime: comm=rcu_sched pid=9 runtime=13646 [ns] vruntime=22531430286 [ns] 2263 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 2264 <...>-2102 [005] d..4 309.874994: sched_migrate_task: comm=cyclictest pid=2103 prio=19 orig_cpu=5 dest_cpu=1 2265 <...>-2102 [005] d..4 309.875185: sched_wake_idle_without_ipi: cpu=1 2266 <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 2267 2268 - onchange(var).save(field,.. .) 2269 2270 The 'onchange(var).save(field,...)' hist trigger action is invoked 2271 whenever the value of 'var' associated with a histogram entry 2272 changes. 2273 2274 The end result is that the trace event fields specified as the 2275 onchange.save() params will be saved if 'var' changes for that 2276 hist trigger entry. This allows context from the event that 2277 changed the value to be saved for later reference. When the 2278 histogram is displayed, additional fields displaying the saved 2279 values will be printed. 2280 2281 - onchange(var).snapshot() 2282 2283 The 'onchange(var).snapshot()' hist trigger action is invoked 2284 whenever the value of 'var' associated with a histogram entry 2285 changes. 2286 2287 The end result is that a global snapshot of the trace buffer will 2288 be saved in the tracing/snapshot file if 'var' changes for any 2289 hist trigger entry. 2290 2291 Note that in this case the changed value is a global variable 2292 associated with current trace instance. The key of the specific 2293 trace event that caused the value to change and the global value 2294 itself are displayed, along with a message stating that a snapshot 2295 has been taken and where to find it. The user can use the key 2296 information displayed to locate the corresponding bucket in the 2297 histogram for even more detail. 2298 2299 As an example the below defines a hist trigger on the tcp_probe 2300 event, keyed on dport. Whenever a tcp_probe event occurs, the 2301 cwnd field is checked against the current value stored in the 2302 $cwnd variable. If the value has changed, a snapshot is taken. 2303 As part of the setup, all the scheduler and tcp events are also 2304 enabled, which are the events that will show up in the snapshot 2305 when it is taken at some point: 2306 2307 # echo 1 > /sys/kernel/debug/tracing/events/sched/enable 2308 # echo 1 > /sys/kernel/debug/tracing/events/tcp/enable 2309 2310 # echo 'hist:keys=dport:cwnd=snd_cwnd: \ 2311 onchange($cwnd).save(snd_wnd,srtt,rcv_wnd): \ 2312 onchange($cwnd).snapshot()' >> \ 2313 /sys/kernel/debug/tracing/events/tcp/tcp_probe/trigger 2314 2315 When the histogram is displayed, for each bucket the tracked value 2316 and the saved values corresponding to that value are displayed 2317 following the rest of the fields. 2318 2319 If a snapshot was taken, there is also a message indicating that, 2320 along with the value and event that triggered the snapshot:: 2321 2322 # cat /sys/kernel/debug/tracing/events/tcp/tcp_probe/hist 2323 2324 { dport: 1521 } hitcount: 8 2325 changed: 10 snd_wnd: 35456 srtt: 154262 rcv_wnd: 42112 2326 2327 { dport: 80 } hitcount: 23 2328 changed: 10 snd_wnd: 28960 srtt: 19604 rcv_wnd: 29312 2329 2330 { dport: 9001 } hitcount: 172 2331 changed: 10 snd_wnd: 48384 srtt: 260444 rcv_wnd: 55168 2332 2333 { dport: 443 } hitcount: 211 2334 changed: 10 snd_wnd: 26960 srtt: 17379 rcv_wnd: 28800 2335 2336 Snapshot taken (see tracing/snapshot). Details:: 2337 2338 triggering value { onchange($cwnd) }: 10 2339 triggered by event with key: { dport: 80 } 2340 2341 Totals: 2342 Hits: 414 2343 Entries: 4 2344 Dropped: 0 2345 2346 In the above case, the event that triggered the snapshot has the 2347 key with dport == 80. If you look at the bucket that has 80 as 2348 the key, you'll find the additional values save()'d along with the 2349 changed value for that bucket, which should be the same as the 2350 global changed value (since that was the same value that triggered 2351 the global snapshot). 2352 2353 And finally, looking at the snapshot data should show at or near 2354 the end the event that triggered the snapshot:: 2355 2356 # cat /sys/kernel/debug/tracing/snapshot 2357 2358 gnome-shell-1261 [006] dN.3 49.823113: sched_stat_runtime: comm=gnome-shell pid=1261 runtime=49347 [ns] vruntime=1835730389 [ns] 2359 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 2360 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 2361 kworker/3:2-135 [003] d..3 49.823118: sched_stat_runtime: comm=kworker/3:2 pid=135 runtime=5339 [ns] vruntime=17815800388 [ns] 2362 kworker/6:2-387 [006] d..3 49.823120: sched_stat_runtime: comm=kworker/6:2 pid=387 runtime=9594 [ns] vruntime=14589605367 [ns] 2363 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 2364 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 2365 <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 2366 23673. User space creating a trigger 2368-------------------------------- 2369 2370Writing into /sys/kernel/tracing/trace_marker writes into the ftrace 2371ring buffer. This can also act like an event, by writing into the trigger 2372file located in /sys/kernel/tracing/events/ftrace/print/ 2373 2374Modifying cyclictest to write into the trace_marker file before it sleeps 2375and after it wakes up, something like this:: 2376 2377 static void traceputs(char *str) 2378 { 2379 /* tracemark_fd is the trace_marker file descriptor */ 2380 if (tracemark_fd < 0) 2381 return; 2382 /* write the tracemark message */ 2383 write(tracemark_fd, str, strlen(str)); 2384 } 2385 2386And later add something like:: 2387 2388 traceputs("start"); 2389 clock_nanosleep(...); 2390 traceputs("end"); 2391 2392We can make a histogram from this:: 2393 2394 # cd /sys/kernel/tracing 2395 # echo 'latency u64 lat' > synthetic_events 2396 # echo 'hist:keys=common_pid:ts0=common_timestamp.usecs if buf == "start"' > events/ftrace/print/trigger 2397 # echo 'hist:keys=common_pid:lat=common_timestamp.usecs-$ts0:onmatch(ftrace.print).latency($lat) if buf == "end"' >> events/ftrace/print/trigger 2398 # echo 'hist:keys=lat,common_pid:sort=lat' > events/synthetic/latency/trigger 2399 2400The above created a synthetic event called "latency" and two histograms 2401against the trace_marker, one gets triggered when "start" is written into the 2402trace_marker file and the other when "end" is written. If the pids match, then 2403it will call the "latency" synthetic event with the calculated latency as its 2404parameter. Finally, a histogram is added to the latency synthetic event to 2405record the calculated latency along with the pid. 2406 2407Now running cyclictest with:: 2408 2409 # ./cyclictest -p80 -d0 -i250 -n -a -t --tracemark -b 1000 2410 2411 -p80 : run threads at priority 80 2412 -d0 : have all threads run at the same interval 2413 -i250 : start the interval at 250 microseconds (all threads will do this) 2414 -n : sleep with nanosleep 2415 -a : affine all threads to a separate CPU 2416 -t : one thread per available CPU 2417 --tracemark : enable trace mark writing 2418 -b 1000 : stop if any latency is greater than 1000 microseconds 2419 2420Note, the -b 1000 is used just to make --tracemark available. 2421 2422Then we can see the histogram created by this with:: 2423 2424 # cat events/synthetic/latency/hist 2425 # event histogram 2426 # 2427 # trigger info: hist:keys=lat,common_pid:vals=hitcount:sort=lat:size=2048 [active] 2428 # 2429 2430 { lat: 107, common_pid: 2039 } hitcount: 1 2431 { lat: 122, common_pid: 2041 } hitcount: 1 2432 { lat: 166, common_pid: 2039 } hitcount: 1 2433 { lat: 174, common_pid: 2039 } hitcount: 1 2434 { lat: 194, common_pid: 2041 } hitcount: 1 2435 { lat: 196, common_pid: 2036 } hitcount: 1 2436 { lat: 197, common_pid: 2038 } hitcount: 1 2437 { lat: 198, common_pid: 2039 } hitcount: 1 2438 { lat: 199, common_pid: 2039 } hitcount: 1 2439 { lat: 200, common_pid: 2041 } hitcount: 1 2440 { lat: 201, common_pid: 2039 } hitcount: 2 2441 { lat: 202, common_pid: 2038 } hitcount: 1 2442 { lat: 202, common_pid: 2043 } hitcount: 1 2443 { lat: 203, common_pid: 2039 } hitcount: 1 2444 { lat: 203, common_pid: 2036 } hitcount: 1 2445 { lat: 203, common_pid: 2041 } hitcount: 1 2446 { lat: 206, common_pid: 2038 } hitcount: 2 2447 { lat: 207, common_pid: 2039 } hitcount: 1 2448 { lat: 207, common_pid: 2036 } hitcount: 1 2449 { lat: 208, common_pid: 2040 } hitcount: 1 2450 { lat: 209, common_pid: 2043 } hitcount: 1 2451 { lat: 210, common_pid: 2039 } hitcount: 1 2452 { lat: 211, common_pid: 2039 } hitcount: 4 2453 { lat: 212, common_pid: 2043 } hitcount: 1 2454 { lat: 212, common_pid: 2039 } hitcount: 2 2455 { lat: 213, common_pid: 2039 } hitcount: 1 2456 { lat: 214, common_pid: 2038 } hitcount: 1 2457 { lat: 214, common_pid: 2039 } hitcount: 2 2458 { lat: 214, common_pid: 2042 } hitcount: 1 2459 { lat: 215, common_pid: 2039 } hitcount: 1 2460 { lat: 217, common_pid: 2036 } hitcount: 1 2461 { lat: 217, common_pid: 2040 } hitcount: 1 2462 { lat: 217, common_pid: 2039 } hitcount: 1 2463 { lat: 218, common_pid: 2039 } hitcount: 6 2464 { lat: 219, common_pid: 2039 } hitcount: 9 2465 { lat: 220, common_pid: 2039 } hitcount: 11 2466 { lat: 221, common_pid: 2039 } hitcount: 5 2467 { lat: 221, common_pid: 2042 } hitcount: 1 2468 { lat: 222, common_pid: 2039 } hitcount: 7 2469 { lat: 223, common_pid: 2036 } hitcount: 1 2470 { lat: 223, common_pid: 2039 } hitcount: 3 2471 { lat: 224, common_pid: 2039 } hitcount: 4 2472 { lat: 224, common_pid: 2037 } hitcount: 1 2473 { lat: 224, common_pid: 2036 } hitcount: 2 2474 { lat: 225, common_pid: 2039 } hitcount: 5 2475 { lat: 225, common_pid: 2042 } hitcount: 1 2476 { lat: 226, common_pid: 2039 } hitcount: 7 2477 { lat: 226, common_pid: 2036 } hitcount: 4 2478 { lat: 227, common_pid: 2039 } hitcount: 6 2479 { lat: 227, common_pid: 2036 } hitcount: 12 2480 { lat: 227, common_pid: 2043 } hitcount: 1 2481 { lat: 228, common_pid: 2039 } hitcount: 7 2482 { lat: 228, common_pid: 2036 } hitcount: 14 2483 { lat: 229, common_pid: 2039 } hitcount: 9 2484 { lat: 229, common_pid: 2036 } hitcount: 8 2485 { lat: 229, common_pid: 2038 } hitcount: 1 2486 { lat: 230, common_pid: 2039 } hitcount: 11 2487 { lat: 230, common_pid: 2036 } hitcount: 6 2488 { lat: 230, common_pid: 2043 } hitcount: 1 2489 { lat: 230, common_pid: 2042 } hitcount: 2 2490 { lat: 231, common_pid: 2041 } hitcount: 1 2491 { lat: 231, common_pid: 2036 } hitcount: 6 2492 { lat: 231, common_pid: 2043 } hitcount: 1 2493 { lat: 231, common_pid: 2039 } hitcount: 8 2494 { lat: 232, common_pid: 2037 } hitcount: 1 2495 { lat: 232, common_pid: 2039 } hitcount: 6 2496 { lat: 232, common_pid: 2040 } hitcount: 2 2497 { lat: 232, common_pid: 2036 } hitcount: 5 2498 { lat: 232, common_pid: 2043 } hitcount: 1 2499 { lat: 233, common_pid: 2036 } hitcount: 5 2500 { lat: 233, common_pid: 2039 } hitcount: 11 2501 { lat: 234, common_pid: 2039 } hitcount: 4 2502 { lat: 234, common_pid: 2038 } hitcount: 2 2503 { lat: 234, common_pid: 2043 } hitcount: 2 2504 { lat: 234, common_pid: 2036 } hitcount: 11 2505 { lat: 234, common_pid: 2040 } hitcount: 1 2506 { lat: 235, common_pid: 2037 } hitcount: 2 2507 { lat: 235, common_pid: 2036 } hitcount: 8 2508 { lat: 235, common_pid: 2043 } hitcount: 2 2509 { lat: 235, common_pid: 2039 } hitcount: 5 2510 { lat: 235, common_pid: 2042 } hitcount: 2 2511 { lat: 235, common_pid: 2040 } hitcount: 4 2512 { lat: 235, common_pid: 2041 } hitcount: 1 2513 { lat: 236, common_pid: 2036 } hitcount: 7 2514 { lat: 236, common_pid: 2037 } hitcount: 1 2515 { lat: 236, common_pid: 2041 } hitcount: 5 2516 { lat: 236, common_pid: 2039 } hitcount: 3 2517 { lat: 236, common_pid: 2043 } hitcount: 9 2518 { lat: 236, common_pid: 2040 } hitcount: 7 2519 { lat: 237, common_pid: 2037 } hitcount: 1 2520 { lat: 237, common_pid: 2040 } hitcount: 1 2521 { lat: 237, common_pid: 2036 } hitcount: 9 2522 { lat: 237, common_pid: 2039 } hitcount: 3 2523 { lat: 237, common_pid: 2043 } hitcount: 8 2524 { lat: 237, common_pid: 2042 } hitcount: 2 2525 { lat: 237, common_pid: 2041 } hitcount: 2 2526 { lat: 238, common_pid: 2043 } hitcount: 10 2527 { lat: 238, common_pid: 2040 } hitcount: 1 2528 { lat: 238, common_pid: 2037 } hitcount: 9 2529 { lat: 238, common_pid: 2038 } hitcount: 1 2530 { lat: 238, common_pid: 2039 } hitcount: 1 2531 { lat: 238, common_pid: 2042 } hitcount: 3 2532 { lat: 238, common_pid: 2036 } hitcount: 7 2533 { lat: 239, common_pid: 2041 } hitcount: 1 2534 { lat: 239, common_pid: 2043 } hitcount: 11 2535 { lat: 239, common_pid: 2037 } hitcount: 11 2536 { lat: 239, common_pid: 2038 } hitcount: 6 2537 { lat: 239, common_pid: 2036 } hitcount: 7 2538 { lat: 239, common_pid: 2040 } hitcount: 1 2539 { lat: 239, common_pid: 2042 } hitcount: 9 2540 { lat: 240, common_pid: 2037 } hitcount: 29 2541 { lat: 240, common_pid: 2043 } hitcount: 15 2542 { lat: 240, common_pid: 2040 } hitcount: 44 2543 { lat: 240, common_pid: 2039 } hitcount: 1 2544 { lat: 240, common_pid: 2041 } hitcount: 2 2545 { lat: 240, common_pid: 2038 } hitcount: 1 2546 { lat: 240, common_pid: 2036 } hitcount: 10 2547 { lat: 240, common_pid: 2042 } hitcount: 13 2548 { lat: 241, common_pid: 2036 } hitcount: 21 2549 { lat: 241, common_pid: 2041 } hitcount: 36 2550 { lat: 241, common_pid: 2037 } hitcount: 34 2551 { lat: 241, common_pid: 2042 } hitcount: 14 2552 { lat: 241, common_pid: 2040 } hitcount: 94 2553 { lat: 241, common_pid: 2039 } hitcount: 12 2554 { lat: 241, common_pid: 2038 } hitcount: 2 2555 { lat: 241, common_pid: 2043 } hitcount: 28 2556 { lat: 242, common_pid: 2040 } hitcount: 109 2557 { lat: 242, common_pid: 2041 } hitcount: 506 2558 { lat: 242, common_pid: 2039 } hitcount: 155 2559 { lat: 242, common_pid: 2042 } hitcount: 21 2560 { lat: 242, common_pid: 2037 } hitcount: 52 2561 { lat: 242, common_pid: 2043 } hitcount: 21 2562 { lat: 242, common_pid: 2036 } hitcount: 16 2563 { lat: 242, common_pid: 2038 } hitcount: 156 2564 { lat: 243, common_pid: 2037 } hitcount: 46 2565 { lat: 243, common_pid: 2039 } hitcount: 40 2566 { lat: 243, common_pid: 2042 } hitcount: 119 2567 { lat: 243, common_pid: 2041 } hitcount: 611 2568 { lat: 243, common_pid: 2036 } hitcount: 69 2569 { lat: 243, common_pid: 2038 } hitcount: 784 2570 { lat: 243, common_pid: 2040 } hitcount: 323 2571 { lat: 243, common_pid: 2043 } hitcount: 14 2572 { lat: 244, common_pid: 2043 } hitcount: 35 2573 { lat: 244, common_pid: 2042 } hitcount: 305 2574 { lat: 244, common_pid: 2039 } hitcount: 8 2575 { lat: 244, common_pid: 2040 } hitcount: 4515 2576 { lat: 244, common_pid: 2038 } hitcount: 371 2577 { lat: 244, common_pid: 2037 } hitcount: 31 2578 { lat: 244, common_pid: 2036 } hitcount: 114 2579 { lat: 244, common_pid: 2041 } hitcount: 3396 2580 { lat: 245, common_pid: 2036 } hitcount: 700 2581 { lat: 245, common_pid: 2041 } hitcount: 2772 2582 { lat: 245, common_pid: 2037 } hitcount: 268 2583 { lat: 245, common_pid: 2039 } hitcount: 472 2584 { lat: 245, common_pid: 2038 } hitcount: 2758 2585 { lat: 245, common_pid: 2042 } hitcount: 3833 2586 { lat: 245, common_pid: 2040 } hitcount: 3105 2587 { lat: 245, common_pid: 2043 } hitcount: 645 2588 { lat: 246, common_pid: 2038 } hitcount: 3451 2589 { lat: 246, common_pid: 2041 } hitcount: 142 2590 { lat: 246, common_pid: 2037 } hitcount: 5101 2591 { lat: 246, common_pid: 2040 } hitcount: 68 2592 { lat: 246, common_pid: 2043 } hitcount: 5099 2593 { lat: 246, common_pid: 2039 } hitcount: 5608 2594 { lat: 246, common_pid: 2042 } hitcount: 3723 2595 { lat: 246, common_pid: 2036 } hitcount: 4738 2596 { lat: 247, common_pid: 2042 } hitcount: 312 2597 { lat: 247, common_pid: 2043 } hitcount: 2385 2598 { lat: 247, common_pid: 2041 } hitcount: 452 2599 { lat: 247, common_pid: 2038 } hitcount: 792 2600 { lat: 247, common_pid: 2040 } hitcount: 78 2601 { lat: 247, common_pid: 2036 } hitcount: 2375 2602 { lat: 247, common_pid: 2039 } hitcount: 1834 2603 { lat: 247, common_pid: 2037 } hitcount: 2655 2604 { lat: 248, common_pid: 2037 } hitcount: 36 2605 { lat: 248, common_pid: 2042 } hitcount: 11 2606 { lat: 248, common_pid: 2038 } hitcount: 122 2607 { lat: 248, common_pid: 2036 } hitcount: 135 2608 { lat: 248, common_pid: 2039 } hitcount: 26 2609 { lat: 248, common_pid: 2041 } hitcount: 503 2610 { lat: 248, common_pid: 2043 } hitcount: 66 2611 { lat: 248, common_pid: 2040 } hitcount: 46 2612 { lat: 249, common_pid: 2037 } hitcount: 29 2613 { lat: 249, common_pid: 2038 } hitcount: 1 2614 { lat: 249, common_pid: 2043 } hitcount: 29 2615 { lat: 249, common_pid: 2039 } hitcount: 8 2616 { lat: 249, common_pid: 2042 } hitcount: 56 2617 { lat: 249, common_pid: 2040 } hitcount: 27 2618 { lat: 249, common_pid: 2041 } hitcount: 11 2619 { lat: 249, common_pid: 2036 } hitcount: 27 2620 { lat: 250, common_pid: 2038 } hitcount: 1 2621 { lat: 250, common_pid: 2036 } hitcount: 30 2622 { lat: 250, common_pid: 2040 } hitcount: 19 2623 { lat: 250, common_pid: 2043 } hitcount: 22 2624 { lat: 250, common_pid: 2042 } hitcount: 20 2625 { lat: 250, common_pid: 2041 } hitcount: 1 2626 { lat: 250, common_pid: 2039 } hitcount: 6 2627 { lat: 250, common_pid: 2037 } hitcount: 48 2628 { lat: 251, common_pid: 2037 } hitcount: 43 2629 { lat: 251, common_pid: 2039 } hitcount: 1 2630 { lat: 251, common_pid: 2036 } hitcount: 12 2631 { lat: 251, common_pid: 2042 } hitcount: 2 2632 { lat: 251, common_pid: 2041 } hitcount: 1 2633 { lat: 251, common_pid: 2043 } hitcount: 15 2634 { lat: 251, common_pid: 2040 } hitcount: 3 2635 { lat: 252, common_pid: 2040 } hitcount: 1 2636 { lat: 252, common_pid: 2036 } hitcount: 12 2637 { lat: 252, common_pid: 2037 } hitcount: 21 2638 { lat: 252, common_pid: 2043 } hitcount: 14 2639 { lat: 253, common_pid: 2037 } hitcount: 21 2640 { lat: 253, common_pid: 2039 } hitcount: 2 2641 { lat: 253, common_pid: 2036 } hitcount: 9 2642 { lat: 253, common_pid: 2043 } hitcount: 6 2643 { lat: 253, common_pid: 2040 } hitcount: 1 2644 { lat: 254, common_pid: 2036 } hitcount: 8 2645 { lat: 254, common_pid: 2043 } hitcount: 3 2646 { lat: 254, common_pid: 2041 } hitcount: 1 2647 { lat: 254, common_pid: 2042 } hitcount: 1 2648 { lat: 254, common_pid: 2039 } hitcount: 1 2649 { lat: 254, common_pid: 2037 } hitcount: 12 2650 { lat: 255, common_pid: 2043 } hitcount: 1 2651 { lat: 255, common_pid: 2037 } hitcount: 2 2652 { lat: 255, common_pid: 2036 } hitcount: 2 2653 { lat: 255, common_pid: 2039 } hitcount: 8 2654 { lat: 256, common_pid: 2043 } hitcount: 1 2655 { lat: 256, common_pid: 2036 } hitcount: 4 2656 { lat: 256, common_pid: 2039 } hitcount: 6 2657 { lat: 257, common_pid: 2039 } hitcount: 5 2658 { lat: 257, common_pid: 2036 } hitcount: 4 2659 { lat: 258, common_pid: 2039 } hitcount: 5 2660 { lat: 258, common_pid: 2036 } hitcount: 2 2661 { lat: 259, common_pid: 2036 } hitcount: 7 2662 { lat: 259, common_pid: 2039 } hitcount: 7 2663 { lat: 260, common_pid: 2036 } hitcount: 8 2664 { lat: 260, common_pid: 2039 } hitcount: 6 2665 { lat: 261, common_pid: 2036 } hitcount: 5 2666 { lat: 261, common_pid: 2039 } hitcount: 7 2667 { lat: 262, common_pid: 2039 } hitcount: 5 2668 { lat: 262, common_pid: 2036 } hitcount: 5 2669 { lat: 263, common_pid: 2039 } hitcount: 7 2670 { lat: 263, common_pid: 2036 } hitcount: 7 2671 { lat: 264, common_pid: 2039 } hitcount: 9 2672 { lat: 264, common_pid: 2036 } hitcount: 9 2673 { lat: 265, common_pid: 2036 } hitcount: 5 2674 { lat: 265, common_pid: 2039 } hitcount: 1 2675 { lat: 266, common_pid: 2036 } hitcount: 1 2676 { lat: 266, common_pid: 2039 } hitcount: 3 2677 { lat: 267, common_pid: 2036 } hitcount: 1 2678 { lat: 267, common_pid: 2039 } hitcount: 3 2679 { lat: 268, common_pid: 2036 } hitcount: 1 2680 { lat: 268, common_pid: 2039 } hitcount: 6 2681 { lat: 269, common_pid: 2036 } hitcount: 1 2682 { lat: 269, common_pid: 2043 } hitcount: 1 2683 { lat: 269, common_pid: 2039 } hitcount: 2 2684 { lat: 270, common_pid: 2040 } hitcount: 1 2685 { lat: 270, common_pid: 2039 } hitcount: 6 2686 { lat: 271, common_pid: 2041 } hitcount: 1 2687 { lat: 271, common_pid: 2039 } hitcount: 5 2688 { lat: 272, common_pid: 2039 } hitcount: 10 2689 { lat: 273, common_pid: 2039 } hitcount: 8 2690 { lat: 274, common_pid: 2039 } hitcount: 2 2691 { lat: 275, common_pid: 2039 } hitcount: 1 2692 { lat: 276, common_pid: 2039 } hitcount: 2 2693 { lat: 276, common_pid: 2037 } hitcount: 1 2694 { lat: 276, common_pid: 2038 } hitcount: 1 2695 { lat: 277, common_pid: 2039 } hitcount: 1 2696 { lat: 277, common_pid: 2042 } hitcount: 1 2697 { lat: 278, common_pid: 2039 } hitcount: 1 2698 { lat: 279, common_pid: 2039 } hitcount: 4 2699 { lat: 279, common_pid: 2043 } hitcount: 1 2700 { lat: 280, common_pid: 2039 } hitcount: 3 2701 { lat: 283, common_pid: 2036 } hitcount: 2 2702 { lat: 284, common_pid: 2039 } hitcount: 1 2703 { lat: 284, common_pid: 2043 } hitcount: 1 2704 { lat: 288, common_pid: 2039 } hitcount: 1 2705 { lat: 289, common_pid: 2039 } hitcount: 1 2706 { lat: 300, common_pid: 2039 } hitcount: 1 2707 { lat: 384, common_pid: 2039 } hitcount: 1 2708 2709 Totals: 2710 Hits: 67625 2711 Entries: 278 2712 Dropped: 0 2713 2714Note, the writes are around the sleep, so ideally they will all be of 250 2715microseconds. If you are wondering how there are several that are under 2716250 microseconds, that is because the way cyclictest works, is if one 2717iteration comes in late, the next one will set the timer to wake up less that 2718250. That is, if an iteration came in 50 microseconds late, the next wake up 2719will be at 200 microseconds. 2720 2721But this could easily be done in userspace. To make this even more 2722interesting, we can mix the histogram between events that happened in the 2723kernel with trace_marker:: 2724 2725 # cd /sys/kernel/tracing 2726 # echo 'latency u64 lat' > synthetic_events 2727 # echo 'hist:keys=pid:ts0=common_timestamp.usecs' > events/sched/sched_waking/trigger 2728 # echo 'hist:keys=common_pid:lat=common_timestamp.usecs-$ts0:onmatch(sched.sched_waking).latency($lat) if buf == "end"' > events/ftrace/print/trigger 2729 # echo 'hist:keys=lat,common_pid:sort=lat' > events/synthetic/latency/trigger 2730 2731The difference this time is that instead of using the trace_marker to start 2732the latency, the sched_waking event is used, matching the common_pid for the 2733trace_marker write with the pid that is being woken by sched_waking. 2734 2735After running cyclictest again with the same parameters, we now have:: 2736 2737 # cat events/synthetic/latency/hist 2738 # event histogram 2739 # 2740 # trigger info: hist:keys=lat,common_pid:vals=hitcount:sort=lat:size=2048 [active] 2741 # 2742 2743 { lat: 7, common_pid: 2302 } hitcount: 640 2744 { lat: 7, common_pid: 2299 } hitcount: 42 2745 { lat: 7, common_pid: 2303 } hitcount: 18 2746 { lat: 7, common_pid: 2305 } hitcount: 166 2747 { lat: 7, common_pid: 2306 } hitcount: 1 2748 { lat: 7, common_pid: 2301 } hitcount: 91 2749 { lat: 7, common_pid: 2300 } hitcount: 17 2750 { lat: 8, common_pid: 2303 } hitcount: 8296 2751 { lat: 8, common_pid: 2304 } hitcount: 6864 2752 { lat: 8, common_pid: 2305 } hitcount: 9464 2753 { lat: 8, common_pid: 2301 } hitcount: 9213 2754 { lat: 8, common_pid: 2306 } hitcount: 6246 2755 { lat: 8, common_pid: 2302 } hitcount: 8797 2756 { lat: 8, common_pid: 2299 } hitcount: 8771 2757 { lat: 8, common_pid: 2300 } hitcount: 8119 2758 { lat: 9, common_pid: 2305 } hitcount: 1519 2759 { lat: 9, common_pid: 2299 } hitcount: 2346 2760 { lat: 9, common_pid: 2303 } hitcount: 2841 2761 { lat: 9, common_pid: 2301 } hitcount: 1846 2762 { lat: 9, common_pid: 2304 } hitcount: 3861 2763 { lat: 9, common_pid: 2302 } hitcount: 1210 2764 { lat: 9, common_pid: 2300 } hitcount: 2762 2765 { lat: 9, common_pid: 2306 } hitcount: 4247 2766 { lat: 10, common_pid: 2299 } hitcount: 16 2767 { lat: 10, common_pid: 2306 } hitcount: 333 2768 { lat: 10, common_pid: 2303 } hitcount: 16 2769 { lat: 10, common_pid: 2304 } hitcount: 168 2770 { lat: 10, common_pid: 2302 } hitcount: 240 2771 { lat: 10, common_pid: 2301 } hitcount: 28 2772 { lat: 10, common_pid: 2300 } hitcount: 95 2773 { lat: 10, common_pid: 2305 } hitcount: 18 2774 { lat: 11, common_pid: 2303 } hitcount: 5 2775 { lat: 11, common_pid: 2305 } hitcount: 8 2776 { lat: 11, common_pid: 2306 } hitcount: 221 2777 { lat: 11, common_pid: 2302 } hitcount: 76 2778 { lat: 11, common_pid: 2304 } hitcount: 26 2779 { lat: 11, common_pid: 2300 } hitcount: 125 2780 { lat: 11, common_pid: 2299 } hitcount: 2 2781 { lat: 12, common_pid: 2305 } hitcount: 3 2782 { lat: 12, common_pid: 2300 } hitcount: 6 2783 { lat: 12, common_pid: 2306 } hitcount: 90 2784 { lat: 12, common_pid: 2302 } hitcount: 4 2785 { lat: 12, common_pid: 2303 } hitcount: 1 2786 { lat: 12, common_pid: 2304 } hitcount: 122 2787 { lat: 13, common_pid: 2300 } hitcount: 12 2788 { lat: 13, common_pid: 2301 } hitcount: 1 2789 { lat: 13, common_pid: 2306 } hitcount: 32 2790 { lat: 13, common_pid: 2302 } hitcount: 5 2791 { lat: 13, common_pid: 2305 } hitcount: 1 2792 { lat: 13, common_pid: 2303 } hitcount: 1 2793 { lat: 13, common_pid: 2304 } hitcount: 61 2794 { lat: 14, common_pid: 2303 } hitcount: 4 2795 { lat: 14, common_pid: 2306 } hitcount: 5 2796 { lat: 14, common_pid: 2305 } hitcount: 4 2797 { lat: 14, common_pid: 2304 } hitcount: 62 2798 { lat: 14, common_pid: 2302 } hitcount: 19 2799 { lat: 14, common_pid: 2300 } hitcount: 33 2800 { lat: 14, common_pid: 2299 } hitcount: 1 2801 { lat: 14, common_pid: 2301 } hitcount: 4 2802 { lat: 15, common_pid: 2305 } hitcount: 1 2803 { lat: 15, common_pid: 2302 } hitcount: 25 2804 { lat: 15, common_pid: 2300 } hitcount: 11 2805 { lat: 15, common_pid: 2299 } hitcount: 5 2806 { lat: 15, common_pid: 2301 } hitcount: 1 2807 { lat: 15, common_pid: 2304 } hitcount: 8 2808 { lat: 15, common_pid: 2303 } hitcount: 1 2809 { lat: 15, common_pid: 2306 } hitcount: 6 2810 { lat: 16, common_pid: 2302 } hitcount: 31 2811 { lat: 16, common_pid: 2306 } hitcount: 3 2812 { lat: 16, common_pid: 2300 } hitcount: 5 2813 { lat: 17, common_pid: 2302 } hitcount: 6 2814 { lat: 17, common_pid: 2303 } hitcount: 1 2815 { lat: 18, common_pid: 2304 } hitcount: 1 2816 { lat: 18, common_pid: 2302 } hitcount: 8 2817 { lat: 18, common_pid: 2299 } hitcount: 1 2818 { lat: 18, common_pid: 2301 } hitcount: 1 2819 { lat: 19, common_pid: 2303 } hitcount: 4 2820 { lat: 19, common_pid: 2304 } hitcount: 5 2821 { lat: 19, common_pid: 2302 } hitcount: 4 2822 { lat: 19, common_pid: 2299 } hitcount: 3 2823 { lat: 19, common_pid: 2306 } hitcount: 1 2824 { lat: 19, common_pid: 2300 } hitcount: 4 2825 { lat: 19, common_pid: 2305 } hitcount: 5 2826 { lat: 20, common_pid: 2299 } hitcount: 2 2827 { lat: 20, common_pid: 2302 } hitcount: 3 2828 { lat: 20, common_pid: 2305 } hitcount: 1 2829 { lat: 20, common_pid: 2300 } hitcount: 2 2830 { lat: 20, common_pid: 2301 } hitcount: 2 2831 { lat: 20, common_pid: 2303 } hitcount: 3 2832 { lat: 21, common_pid: 2305 } hitcount: 1 2833 { lat: 21, common_pid: 2299 } hitcount: 5 2834 { lat: 21, common_pid: 2303 } hitcount: 4 2835 { lat: 21, common_pid: 2302 } hitcount: 7 2836 { lat: 21, common_pid: 2300 } hitcount: 1 2837 { lat: 21, common_pid: 2301 } hitcount: 5 2838 { lat: 21, common_pid: 2304 } hitcount: 2 2839 { lat: 22, common_pid: 2302 } hitcount: 5 2840 { lat: 22, common_pid: 2303 } hitcount: 1 2841 { lat: 22, common_pid: 2306 } hitcount: 3 2842 { lat: 22, common_pid: 2301 } hitcount: 2 2843 { lat: 22, common_pid: 2300 } hitcount: 1 2844 { lat: 22, common_pid: 2299 } hitcount: 1 2845 { lat: 22, common_pid: 2305 } hitcount: 1 2846 { lat: 22, common_pid: 2304 } hitcount: 1 2847 { lat: 23, common_pid: 2299 } hitcount: 1 2848 { lat: 23, common_pid: 2306 } hitcount: 2 2849 { lat: 23, common_pid: 2302 } hitcount: 6 2850 { lat: 24, common_pid: 2302 } hitcount: 3 2851 { lat: 24, common_pid: 2300 } hitcount: 1 2852 { lat: 24, common_pid: 2306 } hitcount: 2 2853 { lat: 24, common_pid: 2305 } hitcount: 1 2854 { lat: 24, common_pid: 2299 } hitcount: 1 2855 { lat: 25, common_pid: 2300 } hitcount: 1 2856 { lat: 25, common_pid: 2302 } hitcount: 4 2857 { lat: 26, common_pid: 2302 } hitcount: 2 2858 { lat: 27, common_pid: 2305 } hitcount: 1 2859 { lat: 27, common_pid: 2300 } hitcount: 1 2860 { lat: 27, common_pid: 2302 } hitcount: 3 2861 { lat: 28, common_pid: 2306 } hitcount: 1 2862 { lat: 28, common_pid: 2302 } hitcount: 4 2863 { lat: 29, common_pid: 2302 } hitcount: 1 2864 { lat: 29, common_pid: 2300 } hitcount: 2 2865 { lat: 29, common_pid: 2306 } hitcount: 1 2866 { lat: 29, common_pid: 2304 } hitcount: 1 2867 { lat: 30, common_pid: 2302 } hitcount: 4 2868 { lat: 31, common_pid: 2302 } hitcount: 6 2869 { lat: 32, common_pid: 2302 } hitcount: 1 2870 { lat: 33, common_pid: 2299 } hitcount: 1 2871 { lat: 33, common_pid: 2302 } hitcount: 3 2872 { lat: 34, common_pid: 2302 } hitcount: 2 2873 { lat: 35, common_pid: 2302 } hitcount: 1 2874 { lat: 35, common_pid: 2304 } hitcount: 1 2875 { lat: 36, common_pid: 2302 } hitcount: 4 2876 { lat: 37, common_pid: 2302 } hitcount: 6 2877 { lat: 38, common_pid: 2302 } hitcount: 2 2878 { lat: 39, common_pid: 2302 } hitcount: 2 2879 { lat: 39, common_pid: 2304 } hitcount: 1 2880 { lat: 40, common_pid: 2304 } hitcount: 2 2881 { lat: 40, common_pid: 2302 } hitcount: 5 2882 { lat: 41, common_pid: 2304 } hitcount: 1 2883 { lat: 41, common_pid: 2302 } hitcount: 8 2884 { lat: 42, common_pid: 2302 } hitcount: 6 2885 { lat: 42, common_pid: 2304 } hitcount: 1 2886 { lat: 43, common_pid: 2302 } hitcount: 3 2887 { lat: 43, common_pid: 2304 } hitcount: 4 2888 { lat: 44, common_pid: 2302 } hitcount: 6 2889 { lat: 45, common_pid: 2302 } hitcount: 5 2890 { lat: 46, common_pid: 2302 } hitcount: 5 2891 { lat: 47, common_pid: 2302 } hitcount: 7 2892 { lat: 48, common_pid: 2301 } hitcount: 1 2893 { lat: 48, common_pid: 2302 } hitcount: 9 2894 { lat: 49, common_pid: 2302 } hitcount: 3 2895 { lat: 50, common_pid: 2302 } hitcount: 1 2896 { lat: 50, common_pid: 2301 } hitcount: 1 2897 { lat: 51, common_pid: 2302 } hitcount: 2 2898 { lat: 51, common_pid: 2301 } hitcount: 1 2899 { lat: 61, common_pid: 2302 } hitcount: 1 2900 { lat: 110, common_pid: 2302 } hitcount: 1 2901 2902 Totals: 2903 Hits: 89565 2904 Entries: 158 2905 Dropped: 0 2906 2907This doesn't tell us any information about how late cyclictest may have 2908woken up, but it does show us a nice histogram of how long it took from 2909the time that cyclictest was woken to the time it made it into user space.