1 ============================================== 1 =========================================================
2 Notes on Analysing Behaviour Using Events and 2 Notes on Analysing Behaviour Using Events and Tracepoints
3 ============================================== 3 =========================================================
4 :Author: Mel Gorman (PCL information heavily b 4 :Author: Mel Gorman (PCL information heavily based on email from Ingo Molnar)
5 5
6 1. Introduction 6 1. Introduction
7 =============== 7 ===============
8 8
9 Tracepoints (see Documentation/trace/tracepoin 9 Tracepoints (see Documentation/trace/tracepoints.rst) can be used without
10 creating custom kernel modules to register pro 10 creating custom kernel modules to register probe functions using the event
11 tracing infrastructure. 11 tracing infrastructure.
12 12
13 Simplistically, tracepoints represent importan 13 Simplistically, tracepoints represent important events that can be
14 taken in conjunction with other tracepoints to 14 taken in conjunction with other tracepoints to build a "Big Picture" of
15 what is going on within the system. There are 15 what is going on within the system. There are a large number of methods for
16 gathering and interpreting these events. Lacki 16 gathering and interpreting these events. Lacking any current Best Practises,
17 this document describes some of the methods th 17 this document describes some of the methods that can be used.
18 18
19 This document assumes that debugfs is mounted 19 This document assumes that debugfs is mounted on /sys/kernel/debug and that
20 the appropriate tracing options have been conf 20 the appropriate tracing options have been configured into the kernel. It is
21 assumed that the PCL tool tools/perf has been 21 assumed that the PCL tool tools/perf has been installed and is in your path.
22 22
23 2. Listing Available Events 23 2. Listing Available Events
24 =========================== 24 ===========================
25 25
26 2.1 Standard Utilities 26 2.1 Standard Utilities
27 ---------------------- 27 ----------------------
28 28
29 All possible events are visible from /sys/kern !! 29 All possible events are visible from /sys/kernel/debug/tracing/events. Simply
30 calling:: 30 calling::
31 31
32 $ find /sys/kernel/tracing/events -type d !! 32 $ find /sys/kernel/debug/tracing/events -type d
33 33
34 will give a fair indication of the number of e 34 will give a fair indication of the number of events available.
35 35
36 2.2 PCL (Performance Counters for Linux) 36 2.2 PCL (Performance Counters for Linux)
37 ---------------------------------------- 37 ----------------------------------------
38 38
39 Discovery and enumeration of all counters and 39 Discovery and enumeration of all counters and events, including tracepoints,
40 are available with the perf tool. Getting a li 40 are available with the perf tool. Getting a list of available events is a
41 simple case of:: 41 simple case of::
42 42
43 $ perf list 2>&1 | grep Tracepoint 43 $ perf list 2>&1 | grep Tracepoint
44 ext4:ext4_free_inode [Tr 44 ext4:ext4_free_inode [Tracepoint event]
45 ext4:ext4_request_inode [Tr 45 ext4:ext4_request_inode [Tracepoint event]
46 ext4:ext4_allocate_inode [Tr 46 ext4:ext4_allocate_inode [Tracepoint event]
47 ext4:ext4_write_begin [Tr 47 ext4:ext4_write_begin [Tracepoint event]
48 ext4:ext4_ordered_write_end [Tr 48 ext4:ext4_ordered_write_end [Tracepoint event]
49 [ .... remaining output snipped .... ] 49 [ .... remaining output snipped .... ]
50 50
51 51
52 3. Enabling Events 52 3. Enabling Events
53 ================== 53 ==================
54 54
55 3.1 System-Wide Event Enabling 55 3.1 System-Wide Event Enabling
56 ------------------------------ 56 ------------------------------
57 57
58 See Documentation/trace/events.rst for a prope 58 See Documentation/trace/events.rst for a proper description on how events
59 can be enabled system-wide. A short example of 59 can be enabled system-wide. A short example of enabling all events related
60 to page allocation would look something like:: 60 to page allocation would look something like::
61 61
62 $ for i in `find /sys/kernel/tracing/events !! 62 $ for i in `find /sys/kernel/debug/tracing/events -name "enable" | grep mm_`; do echo 1 > $i; done
63 63
64 3.2 System-Wide Event Enabling with SystemTap 64 3.2 System-Wide Event Enabling with SystemTap
65 --------------------------------------------- 65 ---------------------------------------------
66 66
67 In SystemTap, tracepoints are accessible using 67 In SystemTap, tracepoints are accessible using the kernel.trace() function
68 call. The following is an example that reports 68 call. The following is an example that reports every 5 seconds what processes
69 were allocating the pages. 69 were allocating the pages.
70 :: 70 ::
71 71
72 global page_allocs 72 global page_allocs
73 73
74 probe kernel.trace("mm_page_alloc") { 74 probe kernel.trace("mm_page_alloc") {
75 page_allocs[execname()]++ 75 page_allocs[execname()]++
76 } 76 }
77 77
78 function print_count() { 78 function print_count() {
79 printf ("%-25s %-s\n", "#Pages Allocat 79 printf ("%-25s %-s\n", "#Pages Allocated", "Process Name")
80 foreach (proc in page_allocs-) 80 foreach (proc in page_allocs-)
81 printf("%-25d %s\n", page_allo 81 printf("%-25d %s\n", page_allocs[proc], proc)
82 printf ("\n") 82 printf ("\n")
83 delete page_allocs 83 delete page_allocs
84 } 84 }
85 85
86 probe timer.s(5) { 86 probe timer.s(5) {
87 print_count() 87 print_count()
88 } 88 }
89 89
90 3.3 System-Wide Event Enabling with PCL 90 3.3 System-Wide Event Enabling with PCL
91 --------------------------------------- 91 ---------------------------------------
92 92
93 By specifying the -a switch and analysing slee 93 By specifying the -a switch and analysing sleep, the system-wide events
94 for a duration of time can be examined. 94 for a duration of time can be examined.
95 :: 95 ::
96 96
97 $ perf stat -a \ 97 $ perf stat -a \
98 -e kmem:mm_page_alloc -e kmem:mm_page_ 98 -e kmem:mm_page_alloc -e kmem:mm_page_free \
99 -e kmem:mm_page_free_batched \ 99 -e kmem:mm_page_free_batched \
100 sleep 10 100 sleep 10
101 Performance counter stats for 'sleep 10': 101 Performance counter stats for 'sleep 10':
102 102
103 9630 kmem:mm_page_alloc 103 9630 kmem:mm_page_alloc
104 2143 kmem:mm_page_free 104 2143 kmem:mm_page_free
105 7424 kmem:mm_page_free_batched 105 7424 kmem:mm_page_free_batched
106 106
107 10.002577764 seconds time elapsed 107 10.002577764 seconds time elapsed
108 108
109 Similarly, one could execute a shell and exit 109 Similarly, one could execute a shell and exit it as desired to get a report
110 at that point. 110 at that point.
111 111
112 3.4 Local Event Enabling 112 3.4 Local Event Enabling
113 ------------------------ 113 ------------------------
114 114
115 Documentation/trace/ftrace.rst describes how t 115 Documentation/trace/ftrace.rst describes how to enable events on a per-thread
116 basis using set_ftrace_pid. 116 basis using set_ftrace_pid.
117 117
118 3.5 Local Event Enablement with PCL 118 3.5 Local Event Enablement with PCL
119 ----------------------------------- 119 -----------------------------------
120 120
121 Events can be activated and tracked for the du 121 Events can be activated and tracked for the duration of a process on a local
122 basis using PCL such as follows. 122 basis using PCL such as follows.
123 :: 123 ::
124 124
125 $ perf stat -e kmem:mm_page_alloc -e kmem:mm 125 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free \
126 -e kmem:mm_page_free_batched 126 -e kmem:mm_page_free_batched ./hackbench 10
127 Time: 0.909 127 Time: 0.909
128 128
129 Performance counter stats for './hackbench 129 Performance counter stats for './hackbench 10':
130 130
131 17803 kmem:mm_page_alloc 131 17803 kmem:mm_page_alloc
132 12398 kmem:mm_page_free 132 12398 kmem:mm_page_free
133 4827 kmem:mm_page_free_batched 133 4827 kmem:mm_page_free_batched
134 134
135 0.973913387 seconds time elapsed 135 0.973913387 seconds time elapsed
136 136
137 4. Event Filtering 137 4. Event Filtering
138 ================== 138 ==================
139 139
140 Documentation/trace/ftrace.rst covers in-depth 140 Documentation/trace/ftrace.rst covers in-depth how to filter events in
141 ftrace. Obviously using grep and awk of trace 141 ftrace. Obviously using grep and awk of trace_pipe is an option as well
142 as any script reading trace_pipe. 142 as any script reading trace_pipe.
143 143
144 5. Analysing Event Variances with PCL 144 5. Analysing Event Variances with PCL
145 ===================================== 145 =====================================
146 146
147 Any workload can exhibit variances between run 147 Any workload can exhibit variances between runs and it can be important
148 to know what the standard deviation is. By and 148 to know what the standard deviation is. By and large, this is left to the
149 performance analyst to do it by hand. In the e 149 performance analyst to do it by hand. In the event that the discrete event
150 occurrences are useful to the performance anal 150 occurrences are useful to the performance analyst, then perf can be used.
151 :: 151 ::
152 152
153 $ perf stat --repeat 5 -e kmem:mm_page_alloc 153 $ perf stat --repeat 5 -e kmem:mm_page_alloc -e kmem:mm_page_free
154 -e kmem:mm_page_free_b 154 -e kmem:mm_page_free_batched ./hackbench 10
155 Time: 0.890 155 Time: 0.890
156 Time: 0.895 156 Time: 0.895
157 Time: 0.915 157 Time: 0.915
158 Time: 1.001 158 Time: 1.001
159 Time: 0.899 159 Time: 0.899
160 160
161 Performance counter stats for './hackbench 161 Performance counter stats for './hackbench 10' (5 runs):
162 162
163 16630 kmem:mm_page_alloc ( 163 16630 kmem:mm_page_alloc ( +- 3.542% )
164 11486 kmem:mm_page_free ( 164 11486 kmem:mm_page_free ( +- 4.771% )
165 4730 kmem:mm_page_free_batched ( 165 4730 kmem:mm_page_free_batched ( +- 2.325% )
166 166
167 0.982653002 seconds time elapsed ( +- 167 0.982653002 seconds time elapsed ( +- 1.448% )
168 168
169 In the event that some higher-level event is r 169 In the event that some higher-level event is required that depends on some
170 aggregation of discrete events, then a script 170 aggregation of discrete events, then a script would need to be developed.
171 171
172 Using --repeat, it is also possible to view ho 172 Using --repeat, it is also possible to view how events are fluctuating over
173 time on a system-wide basis using -a and sleep 173 time on a system-wide basis using -a and sleep.
174 :: 174 ::
175 175
176 $ perf stat -e kmem:mm_page_alloc -e kmem:mm 176 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free \
177 -e kmem:mm_page_free_batched \ 177 -e kmem:mm_page_free_batched \
178 -a --repeat 10 \ 178 -a --repeat 10 \
179 sleep 1 179 sleep 1
180 Performance counter stats for 'sleep 1' (10 180 Performance counter stats for 'sleep 1' (10 runs):
181 181
182 1066 kmem:mm_page_alloc ( 182 1066 kmem:mm_page_alloc ( +- 26.148% )
183 182 kmem:mm_page_free ( 183 182 kmem:mm_page_free ( +- 5.464% )
184 890 kmem:mm_page_free_batched ( 184 890 kmem:mm_page_free_batched ( +- 30.079% )
185 185
186 1.002251757 seconds time elapsed ( +- 186 1.002251757 seconds time elapsed ( +- 0.005% )
187 187
188 6. Higher-Level Analysis with Helper Scripts 188 6. Higher-Level Analysis with Helper Scripts
189 ============================================ 189 ============================================
190 190
191 When events are enabled the events that are tr 191 When events are enabled the events that are triggering can be read from
192 /sys/kernel/tracing/trace_pipe in human-readab !! 192 /sys/kernel/debug/tracing/trace_pipe in human-readable format although binary
193 options exist as well. By post-processing the 193 options exist as well. By post-processing the output, further information can
194 be gathered on-line as appropriate. Examples o 194 be gathered on-line as appropriate. Examples of post-processing might include
195 195
196 - Reading information from /proc for the PID 196 - Reading information from /proc for the PID that triggered the event
197 - Deriving a higher-level event from a serie 197 - Deriving a higher-level event from a series of lower-level events.
198 - Calculating latencies between two events 198 - Calculating latencies between two events
199 199
200 Documentation/trace/postprocess/trace-pageallo 200 Documentation/trace/postprocess/trace-pagealloc-postprocess.pl is an example
201 script that can read trace_pipe from STDIN or 201 script that can read trace_pipe from STDIN or a copy of a trace. When used
202 on-line, it can be interrupted once to generat 202 on-line, it can be interrupted once to generate a report without exiting
203 and twice to exit. 203 and twice to exit.
204 204
205 Simplistically, the script just reads STDIN an 205 Simplistically, the script just reads STDIN and counts up events but it
206 also can do more such as 206 also can do more such as
207 207
208 - Derive high-level events from many low-lev 208 - Derive high-level events from many low-level events. If a number of pages
209 are freed to the main allocator from the p 209 are freed to the main allocator from the per-CPU lists, it recognises
210 that as one per-CPU drain even though ther 210 that as one per-CPU drain even though there is no specific tracepoint
211 for that event 211 for that event
212 - It can aggregate based on PID or individua 212 - It can aggregate based on PID or individual process number
213 - In the event memory is getting externally 213 - In the event memory is getting externally fragmented, it reports
214 on whether the fragmentation event was sev 214 on whether the fragmentation event was severe or moderate.
215 - When receiving an event about a PID, it ca 215 - When receiving an event about a PID, it can record who the parent was so
216 that if large numbers of events are coming 216 that if large numbers of events are coming from very short-lived
217 processes, the parent process responsible 217 processes, the parent process responsible for creating all the helpers
218 can be identified 218 can be identified
219 219
220 7. Lower-Level Analysis with PCL 220 7. Lower-Level Analysis with PCL
221 ================================ 221 ================================
222 222
223 There may also be a requirement to identify wh 223 There may also be a requirement to identify what functions within a program
224 were generating events within the kernel. To b 224 were generating events within the kernel. To begin this sort of analysis, the
225 data must be recorded. At the time of writing, 225 data must be recorded. At the time of writing, this required root:
226 :: 226 ::
227 227
228 $ perf record -c 1 \ 228 $ perf record -c 1 \
229 -e kmem:mm_page_alloc -e kmem:mm_page_ 229 -e kmem:mm_page_alloc -e kmem:mm_page_free \
230 -e kmem:mm_page_free_batched \ 230 -e kmem:mm_page_free_batched \
231 ./hackbench 10 231 ./hackbench 10
232 Time: 0.894 232 Time: 0.894
233 [ perf record: Captured and wrote 0.733 MB p 233 [ perf record: Captured and wrote 0.733 MB perf.data (~32010 samples) ]
234 234
235 Note the use of '-c 1' to set the event period 235 Note the use of '-c 1' to set the event period to sample. The default sample
236 period is quite high to minimise overhead but 236 period is quite high to minimise overhead but the information collected can be
237 very coarse as a result. 237 very coarse as a result.
238 238
239 This record outputted a file called perf.data 239 This record outputted a file called perf.data which can be analysed using
240 perf report. 240 perf report.
241 :: 241 ::
242 242
243 $ perf report 243 $ perf report
244 # Samples: 30922 244 # Samples: 30922
245 # 245 #
246 # Overhead Command Sh 246 # Overhead Command Shared Object
247 # ........ ......... ..................... 247 # ........ ......... ................................
248 # 248 #
249 87.27% hackbench [vdso] 249 87.27% hackbench [vdso]
250 6.85% hackbench /lib/i686/cmov/libc-2 250 6.85% hackbench /lib/i686/cmov/libc-2.9.so
251 2.62% hackbench /lib/ld-2.9.so 251 2.62% hackbench /lib/ld-2.9.so
252 1.52% perf [vdso] 252 1.52% perf [vdso]
253 1.22% hackbench ./hackbench 253 1.22% hackbench ./hackbench
254 0.48% hackbench [kernel] 254 0.48% hackbench [kernel]
255 0.02% perf /lib/i686/cmov/libc-2 255 0.02% perf /lib/i686/cmov/libc-2.9.so
256 0.01% perf /usr/bin/perf 256 0.01% perf /usr/bin/perf
257 0.01% perf /lib/ld-2.9.so 257 0.01% perf /lib/ld-2.9.so
258 0.00% hackbench /lib/i686/cmov/libpth 258 0.00% hackbench /lib/i686/cmov/libpthread-2.9.so
259 # 259 #
260 # (For more details, try: perf report --sort 260 # (For more details, try: perf report --sort comm,dso,symbol)
261 # 261 #
262 262
263 According to this, the vast majority of events 263 According to this, the vast majority of events triggered on events
264 within the VDSO. With simple binaries, this wi 264 within the VDSO. With simple binaries, this will often be the case so let's
265 take a slightly different example. In the cour 265 take a slightly different example. In the course of writing this, it was
266 noticed that X was generating an insane amount 266 noticed that X was generating an insane amount of page allocations so let's look
267 at it: 267 at it:
268 :: 268 ::
269 269
270 $ perf record -c 1 -f \ 270 $ perf record -c 1 -f \
271 -e kmem:mm_page_alloc -e kmem: 271 -e kmem:mm_page_alloc -e kmem:mm_page_free \
272 -e kmem:mm_page_free_batched \ 272 -e kmem:mm_page_free_batched \
273 -p `pidof X` 273 -p `pidof X`
274 274
275 This was interrupted after a few seconds and 275 This was interrupted after a few seconds and
276 :: 276 ::
277 277
278 $ perf report 278 $ perf report
279 # Samples: 27666 279 # Samples: 27666
280 # 280 #
281 # Overhead Command 281 # Overhead Command Shared Object
282 # ........ ....... ....................... 282 # ........ ....... .......................................
283 # 283 #
284 51.95% Xorg [vdso] 284 51.95% Xorg [vdso]
285 47.95% Xorg /opt/gfx-test/lib/libpi 285 47.95% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1
286 0.09% Xorg /lib/i686/cmov/libc-2.9 286 0.09% Xorg /lib/i686/cmov/libc-2.9.so
287 0.01% Xorg [kernel] 287 0.01% Xorg [kernel]
288 # 288 #
289 # (For more details, try: perf report --sort 289 # (For more details, try: perf report --sort comm,dso,symbol)
290 # 290 #
291 291
292 So, almost half of the events are occurring in 292 So, almost half of the events are occurring in a library. To get an idea which
293 symbol: 293 symbol:
294 :: 294 ::
295 295
296 $ perf report --sort comm,dso,symbol 296 $ perf report --sort comm,dso,symbol
297 # Samples: 27666 297 # Samples: 27666
298 # 298 #
299 # Overhead Command 299 # Overhead Command Shared Object Symbol
300 # ........ ....... ....................... 300 # ........ ....... ....................................... ......
301 # 301 #
302 51.95% Xorg [vdso] 302 51.95% Xorg [vdso] [.] 0x000000ffffe424
303 47.93% Xorg /opt/gfx-test/lib/libpi 303 47.93% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] pixmanFillsse2
304 0.09% Xorg /lib/i686/cmov/libc-2.9 304 0.09% Xorg /lib/i686/cmov/libc-2.9.so [.] _int_malloc
305 0.01% Xorg /opt/gfx-test/lib/libpi 305 0.01% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] pixman_region32_copy_f
306 0.01% Xorg [kernel] 306 0.01% Xorg [kernel] [k] read_hpet
307 0.01% Xorg /opt/gfx-test/lib/libpi 307 0.01% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] get_fast_path
308 0.00% Xorg [kernel] 308 0.00% Xorg [kernel] [k] ftrace_trace_userstack
309 309
310 To see where within the function pixmanFillsse 310 To see where within the function pixmanFillsse2 things are going wrong:
311 :: 311 ::
312 312
313 $ perf annotate pixmanFillsse2 313 $ perf annotate pixmanFillsse2
314 [ ... ] 314 [ ... ]
315 0.00 : 34eeb: 0f 18 08 315 0.00 : 34eeb: 0f 18 08 prefetcht0 (%eax)
316 : } 316 : }
317 : 317 :
318 : extern __inline void __attribu 318 : extern __inline void __attribute__((__gnu_inline__, __always_inline__, _
319 : _mm_store_si128 (__m128i *__P, 319 : _mm_store_si128 (__m128i *__P, __m128i __B) : {
320 : *__P = __B; 320 : *__P = __B;
321 12.40 : 34eee: 66 0f 7f 80 40 321 12.40 : 34eee: 66 0f 7f 80 40 ff ff movdqa %xmm0,-0xc0(%eax)
322 0.00 : 34ef5: ff 322 0.00 : 34ef5: ff
323 12.40 : 34ef6: 66 0f 7f 80 50 323 12.40 : 34ef6: 66 0f 7f 80 50 ff ff movdqa %xmm0,-0xb0(%eax)
324 0.00 : 34efd: ff 324 0.00 : 34efd: ff
325 12.39 : 34efe: 66 0f 7f 80 60 325 12.39 : 34efe: 66 0f 7f 80 60 ff ff movdqa %xmm0,-0xa0(%eax)
326 0.00 : 34f05: ff 326 0.00 : 34f05: ff
327 12.67 : 34f06: 66 0f 7f 80 70 327 12.67 : 34f06: 66 0f 7f 80 70 ff ff movdqa %xmm0,-0x90(%eax)
328 0.00 : 34f0d: ff 328 0.00 : 34f0d: ff
329 12.58 : 34f0e: 66 0f 7f 40 80 329 12.58 : 34f0e: 66 0f 7f 40 80 movdqa %xmm0,-0x80(%eax)
330 12.31 : 34f13: 66 0f 7f 40 90 330 12.31 : 34f13: 66 0f 7f 40 90 movdqa %xmm0,-0x70(%eax)
331 12.40 : 34f18: 66 0f 7f 40 a0 331 12.40 : 34f18: 66 0f 7f 40 a0 movdqa %xmm0,-0x60(%eax)
332 12.31 : 34f1d: 66 0f 7f 40 b0 332 12.31 : 34f1d: 66 0f 7f 40 b0 movdqa %xmm0,-0x50(%eax)
333 333
334 At a glance, it looks like the time is being s 334 At a glance, it looks like the time is being spent copying pixmaps to
335 the card. Further investigation would be need 335 the card. Further investigation would be needed to determine why pixmaps
336 are being copied around so much but a starting 336 are being copied around so much but a starting point would be to take an
337 ancient build of libpixmap out of the library 337 ancient build of libpixmap out of the library path where it was totally
338 forgotten about from months ago! 338 forgotten about from months ago!
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.