339 lines · plain
1=========================================================2Notes on Analysing Behaviour Using Events and Tracepoints3=========================================================4:Author: Mel Gorman (PCL information heavily based on email from Ingo Molnar)5 61. Introduction7===============8 9Tracepoints (see Documentation/trace/tracepoints.rst) can be used without10creating custom kernel modules to register probe functions using the event11tracing infrastructure.12 13Simplistically, tracepoints represent important events that can be14taken in conjunction with other tracepoints to build a "Big Picture" of15what is going on within the system. There are a large number of methods for16gathering and interpreting these events. Lacking any current Best Practises,17this document describes some of the methods that can be used.18 19This document assumes that debugfs is mounted on /sys/kernel/debug and that20the appropriate tracing options have been configured into the kernel. It is21assumed that the PCL tool tools/perf has been installed and is in your path.22 232. Listing Available Events24===========================25 262.1 Standard Utilities27----------------------28 29All possible events are visible from /sys/kernel/tracing/events. Simply30calling::31 32 $ find /sys/kernel/tracing/events -type d33 34will give a fair indication of the number of events available.35 362.2 PCL (Performance Counters for Linux)37----------------------------------------38 39Discovery and enumeration of all counters and events, including tracepoints,40are available with the perf tool. Getting a list of available events is a41simple case of::42 43 $ perf list 2>&1 | grep Tracepoint44 ext4:ext4_free_inode [Tracepoint event]45 ext4:ext4_request_inode [Tracepoint event]46 ext4:ext4_allocate_inode [Tracepoint event]47 ext4:ext4_write_begin [Tracepoint event]48 ext4:ext4_ordered_write_end [Tracepoint event]49 [ .... remaining output snipped .... ]50 51 523. Enabling Events53==================54 553.1 System-Wide Event Enabling56------------------------------57 58See Documentation/trace/events.rst for a proper description on how events59can be enabled system-wide. A short example of enabling all events related60to page allocation would look something like::61 62 $ for i in `find /sys/kernel/tracing/events -name "enable" | grep mm_`; do echo 1 > $i; done63 643.2 System-Wide Event Enabling with SystemTap65---------------------------------------------66 67In SystemTap, tracepoints are accessible using the kernel.trace() function68call. The following is an example that reports every 5 seconds what processes69were allocating the pages.70::71 72 global page_allocs73 74 probe kernel.trace("mm_page_alloc") {75 page_allocs[execname()]++76 }77 78 function print_count() {79 printf ("%-25s %-s\n", "#Pages Allocated", "Process Name")80 foreach (proc in page_allocs-)81 printf("%-25d %s\n", page_allocs[proc], proc)82 printf ("\n")83 delete page_allocs84 }85 86 probe timer.s(5) {87 print_count()88 }89 903.3 System-Wide Event Enabling with PCL91---------------------------------------92 93By specifying the -a switch and analysing sleep, the system-wide events94for a duration of time can be examined.95::96 97 $ perf stat -a \98 -e kmem:mm_page_alloc -e kmem:mm_page_free \99 -e kmem:mm_page_free_batched \100 sleep 10101 Performance counter stats for 'sleep 10':102 103 9630 kmem:mm_page_alloc104 2143 kmem:mm_page_free105 7424 kmem:mm_page_free_batched106 107 10.002577764 seconds time elapsed108 109Similarly, one could execute a shell and exit it as desired to get a report110at that point.111 1123.4 Local Event Enabling113------------------------114 115Documentation/trace/ftrace.rst describes how to enable events on a per-thread116basis using set_ftrace_pid.117 1183.5 Local Event Enablement with PCL119-----------------------------------120 121Events can be activated and tracked for the duration of a process on a local122basis using PCL such as follows.123::124 125 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free \126 -e kmem:mm_page_free_batched ./hackbench 10127 Time: 0.909128 129 Performance counter stats for './hackbench 10':130 131 17803 kmem:mm_page_alloc132 12398 kmem:mm_page_free133 4827 kmem:mm_page_free_batched134 135 0.973913387 seconds time elapsed136 1374. Event Filtering138==================139 140Documentation/trace/ftrace.rst covers in-depth how to filter events in141ftrace. Obviously using grep and awk of trace_pipe is an option as well142as any script reading trace_pipe.143 1445. Analysing Event Variances with PCL145=====================================146 147Any workload can exhibit variances between runs and it can be important148to know what the standard deviation is. By and large, this is left to the149performance analyst to do it by hand. In the event that the discrete event150occurrences are useful to the performance analyst, then perf can be used.151::152 153 $ perf stat --repeat 5 -e kmem:mm_page_alloc -e kmem:mm_page_free154 -e kmem:mm_page_free_batched ./hackbench 10155 Time: 0.890156 Time: 0.895157 Time: 0.915158 Time: 1.001159 Time: 0.899160 161 Performance counter stats for './hackbench 10' (5 runs):162 163 16630 kmem:mm_page_alloc ( +- 3.542% )164 11486 kmem:mm_page_free ( +- 4.771% )165 4730 kmem:mm_page_free_batched ( +- 2.325% )166 167 0.982653002 seconds time elapsed ( +- 1.448% )168 169In the event that some higher-level event is required that depends on some170aggregation of discrete events, then a script would need to be developed.171 172Using --repeat, it is also possible to view how events are fluctuating over173time on a system-wide basis using -a and sleep.174::175 176 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free \177 -e kmem:mm_page_free_batched \178 -a --repeat 10 \179 sleep 1180 Performance counter stats for 'sleep 1' (10 runs):181 182 1066 kmem:mm_page_alloc ( +- 26.148% )183 182 kmem:mm_page_free ( +- 5.464% )184 890 kmem:mm_page_free_batched ( +- 30.079% )185 186 1.002251757 seconds time elapsed ( +- 0.005% )187 1886. Higher-Level Analysis with Helper Scripts189============================================190 191When events are enabled the events that are triggering can be read from192/sys/kernel/tracing/trace_pipe in human-readable format although binary193options exist as well. By post-processing the output, further information can194be gathered on-line as appropriate. Examples of post-processing might include195 196 - Reading information from /proc for the PID that triggered the event197 - Deriving a higher-level event from a series of lower-level events.198 - Calculating latencies between two events199 200Documentation/trace/postprocess/trace-pagealloc-postprocess.pl is an example201script that can read trace_pipe from STDIN or a copy of a trace. When used202on-line, it can be interrupted once to generate a report without exiting203and twice to exit.204 205Simplistically, the script just reads STDIN and counts up events but it206also can do more such as207 208 - Derive high-level events from many low-level events. If a number of pages209 are freed to the main allocator from the per-CPU lists, it recognises210 that as one per-CPU drain even though there is no specific tracepoint211 for that event212 - It can aggregate based on PID or individual process number213 - In the event memory is getting externally fragmented, it reports214 on whether the fragmentation event was severe or moderate.215 - When receiving an event about a PID, it can record who the parent was so216 that if large numbers of events are coming from very short-lived217 processes, the parent process responsible for creating all the helpers218 can be identified219 2207. Lower-Level Analysis with PCL221================================222 223There may also be a requirement to identify what functions within a program224were generating events within the kernel. To begin this sort of analysis, the225data must be recorded. At the time of writing, this required root:226::227 228 $ perf record -c 1 \229 -e kmem:mm_page_alloc -e kmem:mm_page_free \230 -e kmem:mm_page_free_batched \231 ./hackbench 10232 Time: 0.894233 [ perf record: Captured and wrote 0.733 MB perf.data (~32010 samples) ]234 235Note the use of '-c 1' to set the event period to sample. The default sample236period is quite high to minimise overhead but the information collected can be237very coarse as a result.238 239This record outputted a file called perf.data which can be analysed using240perf report.241::242 243 $ perf report244 # Samples: 30922245 #246 # Overhead Command Shared Object247 # ........ ......... ................................248 #249 87.27% hackbench [vdso]250 6.85% hackbench /lib/i686/cmov/libc-2.9.so251 2.62% hackbench /lib/ld-2.9.so252 1.52% perf [vdso]253 1.22% hackbench ./hackbench254 0.48% hackbench [kernel]255 0.02% perf /lib/i686/cmov/libc-2.9.so256 0.01% perf /usr/bin/perf257 0.01% perf /lib/ld-2.9.so258 0.00% hackbench /lib/i686/cmov/libpthread-2.9.so259 #260 # (For more details, try: perf report --sort comm,dso,symbol)261 #262 263According to this, the vast majority of events triggered on events264within the VDSO. With simple binaries, this will often be the case so let's265take a slightly different example. In the course of writing this, it was266noticed that X was generating an insane amount of page allocations so let's look267at it:268::269 270 $ perf record -c 1 -f \271 -e kmem:mm_page_alloc -e kmem:mm_page_free \272 -e kmem:mm_page_free_batched \273 -p `pidof X`274 275This was interrupted after a few seconds and276::277 278 $ perf report279 # Samples: 27666280 #281 # Overhead Command Shared Object282 # ........ ....... .......................................283 #284 51.95% Xorg [vdso]285 47.95% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1286 0.09% Xorg /lib/i686/cmov/libc-2.9.so287 0.01% Xorg [kernel]288 #289 # (For more details, try: perf report --sort comm,dso,symbol)290 #291 292So, almost half of the events are occurring in a library. To get an idea which293symbol:294::295 296 $ perf report --sort comm,dso,symbol297 # Samples: 27666298 #299 # Overhead Command Shared Object Symbol300 # ........ ....... ....................................... ......301 #302 51.95% Xorg [vdso] [.] 0x000000ffffe424303 47.93% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] pixmanFillsse2304 0.09% Xorg /lib/i686/cmov/libc-2.9.so [.] _int_malloc305 0.01% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] pixman_region32_copy_f306 0.01% Xorg [kernel] [k] read_hpet307 0.01% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] get_fast_path308 0.00% Xorg [kernel] [k] ftrace_trace_userstack309 310To see where within the function pixmanFillsse2 things are going wrong:311::312 313 $ perf annotate pixmanFillsse2314 [ ... ]315 0.00 : 34eeb: 0f 18 08 prefetcht0 (%eax)316 : }317 :318 : extern __inline void __attribute__((__gnu_inline__, __always_inline__, _319 : _mm_store_si128 (__m128i *__P, __m128i __B) : {320 : *__P = __B;321 12.40 : 34eee: 66 0f 7f 80 40 ff ff movdqa %xmm0,-0xc0(%eax)322 0.00 : 34ef5: ff323 12.40 : 34ef6: 66 0f 7f 80 50 ff ff movdqa %xmm0,-0xb0(%eax)324 0.00 : 34efd: ff325 12.39 : 34efe: 66 0f 7f 80 60 ff ff movdqa %xmm0,-0xa0(%eax)326 0.00 : 34f05: ff327 12.67 : 34f06: 66 0f 7f 80 70 ff ff movdqa %xmm0,-0x90(%eax)328 0.00 : 34f0d: ff329 12.58 : 34f0e: 66 0f 7f 40 80 movdqa %xmm0,-0x80(%eax)330 12.31 : 34f13: 66 0f 7f 40 90 movdqa %xmm0,-0x70(%eax)331 12.40 : 34f18: 66 0f 7f 40 a0 movdqa %xmm0,-0x60(%eax)332 12.31 : 34f1d: 66 0f 7f 40 b0 movdqa %xmm0,-0x50(%eax)333 334At a glance, it looks like the time is being spent copying pixmaps to335the card. Further investigation would be needed to determine why pixmaps336are being copied around so much but a starting point would be to take an337ancient build of libpixmap out of the library path where it was totally338forgotten about from months ago!339