aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation
diff options
context:
space:
mode:
authorRandy Dunlap <randy.dunlap@oracle.com>2009-12-18 18:17:04 -0500
committerIngo Molnar <mingo@elte.hu>2009-12-20 00:18:25 -0500
commitb41df645c829d961068aecd30909c2675acbaaea (patch)
treedd2faee24e1c4ce487fa88e66f9f8cd6f0de2da0 /Documentation
parent360b6e5cab1cea1d838b0100956ce0d3dbccbb6f (diff)
Documentation: Update tracepoint-analysis.txt
Fix grammar, spelling, punctuation, hyphenation, section numbering. Tell what PCL means. Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Steven Rostedt <rostedt@goodmis.org> LKML-Reference: <4B2C0D70.4030707@oracle.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'Documentation')
-rw-r--r--Documentation/trace/tracepoint-analysis.txt60
1 files changed, 30 insertions, 30 deletions
diff --git a/Documentation/trace/tracepoint-analysis.txt b/Documentation/trace/tracepoint-analysis.txt
index 5eb4e487e667..87bee3c129ba 100644
--- a/Documentation/trace/tracepoint-analysis.txt
+++ b/Documentation/trace/tracepoint-analysis.txt
@@ -10,8 +10,8 @@ Tracepoints (see Documentation/trace/tracepoints.txt) can be used without
10creating custom kernel modules to register probe functions using the event 10creating custom kernel modules to register probe functions using the event
11tracing infrastructure. 11tracing infrastructure.
12 12
13Simplistically, tracepoints will represent an important event that when can 13Simplistically, tracepoints represent important events that can be
14be taken in conjunction with other tracepoints to build a "Big Picture" of 14taken in conjunction with other tracepoints to build a "Big Picture" of
15what is going on within the system. There are a large number of methods for 15what is going on within the system. There are a large number of methods for
16gathering and interpreting these events. Lacking any current Best Practises, 16gathering and interpreting these events. Lacking any current Best Practises,
17this document describes some of the methods that can be used. 17this document describes some of the methods that can be used.
@@ -33,12 +33,12 @@ calling
33 33
34will give a fair indication of the number of events available. 34will give a fair indication of the number of events available.
35 35
362.2 PCL 362.2 PCL (Performance Counters for Linux)
37------- 37-------
38 38
39Discovery and enumeration of all counters and events, including tracepoints 39Discovery and enumeration of all counters and events, including tracepoints,
40are available with the perf tool. Getting a list of available events is a 40are available with the perf tool. Getting a list of available events is a
41simple case of 41simple case of:
42 42
43 $ perf list 2>&1 | grep Tracepoint 43 $ perf list 2>&1 | grep Tracepoint
44 ext4:ext4_free_inode [Tracepoint event] 44 ext4:ext4_free_inode [Tracepoint event]
@@ -49,19 +49,19 @@ simple case of
49 [ .... remaining output snipped .... ] 49 [ .... remaining output snipped .... ]
50 50
51 51
522. Enabling Events 523. Enabling Events
53================== 53==================
54 54
552.1 System-Wide Event Enabling 553.1 System-Wide Event Enabling
56------------------------------ 56------------------------------
57 57
58See Documentation/trace/events.txt for a proper description on how events 58See Documentation/trace/events.txt for a proper description on how events
59can be enabled system-wide. A short example of enabling all events related 59can be enabled system-wide. A short example of enabling all events related
60to page allocation would look something like 60to page allocation would look something like:
61 61
62 $ for i in `find /sys/kernel/debug/tracing/events -name "enable" | grep mm_`; do echo 1 > $i; done 62 $ for i in `find /sys/kernel/debug/tracing/events -name "enable" | grep mm_`; do echo 1 > $i; done
63 63
642.2 System-Wide Event Enabling with SystemTap 643.2 System-Wide Event Enabling with SystemTap
65--------------------------------------------- 65---------------------------------------------
66 66
67In SystemTap, tracepoints are accessible using the kernel.trace() function 67In SystemTap, tracepoints are accessible using the kernel.trace() function
@@ -86,7 +86,7 @@ were allocating the pages.
86 print_count() 86 print_count()
87 } 87 }
88 88
892.3 System-Wide Event Enabling with PCL 893.3 System-Wide Event Enabling with PCL
90--------------------------------------- 90---------------------------------------
91 91
92By specifying the -a switch and analysing sleep, the system-wide events 92By specifying the -a switch and analysing sleep, the system-wide events
@@ -107,16 +107,16 @@ for a duration of time can be examined.
107Similarly, one could execute a shell and exit it as desired to get a report 107Similarly, one could execute a shell and exit it as desired to get a report
108at that point. 108at that point.
109 109
1102.4 Local Event Enabling 1103.4 Local Event Enabling
111------------------------ 111------------------------
112 112
113Documentation/trace/ftrace.txt describes how to enable events on a per-thread 113Documentation/trace/ftrace.txt describes how to enable events on a per-thread
114basis using set_ftrace_pid. 114basis using set_ftrace_pid.
115 115
1162.5 Local Event Enablement with PCL 1163.5 Local Event Enablement with PCL
117----------------------------------- 117-----------------------------------
118 118
119Events can be activate and tracked for the duration of a process on a local 119Events can be activated and tracked for the duration of a process on a local
120basis using PCL such as follows. 120basis using PCL such as follows.
121 121
122 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \ 122 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \
@@ -131,18 +131,18 @@ basis using PCL such as follows.
131 131
132 0.973913387 seconds time elapsed 132 0.973913387 seconds time elapsed
133 133
1343. Event Filtering 1344. Event Filtering
135================== 135==================
136 136
137Documentation/trace/ftrace.txt covers in-depth how to filter events in 137Documentation/trace/ftrace.txt covers in-depth how to filter events in
138ftrace. Obviously using grep and awk of trace_pipe is an option as well 138ftrace. Obviously using grep and awk of trace_pipe is an option as well
139as any script reading trace_pipe. 139as any script reading trace_pipe.
140 140
1414. Analysing Event Variances with PCL 1415. Analysing Event Variances with PCL
142===================================== 142=====================================
143 143
144Any workload can exhibit variances between runs and it can be important 144Any workload can exhibit variances between runs and it can be important
145to know what the standard deviation in. By and large, this is left to the 145to know what the standard deviation is. By and large, this is left to the
146performance analyst to do it by hand. In the event that the discrete event 146performance analyst to do it by hand. In the event that the discrete event
147occurrences are useful to the performance analyst, then perf can be used. 147occurrences are useful to the performance analyst, then perf can be used.
148 148
@@ -166,7 +166,7 @@ In the event that some higher-level event is required that depends on some
166aggregation of discrete events, then a script would need to be developed. 166aggregation of discrete events, then a script would need to be developed.
167 167
168Using --repeat, it is also possible to view how events are fluctuating over 168Using --repeat, it is also possible to view how events are fluctuating over
169time on a system wide basis using -a and sleep. 169time on a system-wide basis using -a and sleep.
170 170
171 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \ 171 $ perf stat -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \
172 -e kmem:mm_pagevec_free \ 172 -e kmem:mm_pagevec_free \
@@ -180,7 +180,7 @@ time on a system wide basis using -a and sleep.
180 180
181 1.002251757 seconds time elapsed ( +- 0.005% ) 181 1.002251757 seconds time elapsed ( +- 0.005% )
182 182
1835. Higher-Level Analysis with Helper Scripts 1836. Higher-Level Analysis with Helper Scripts
184============================================ 184============================================
185 185
186When events are enabled the events that are triggering can be read from 186When events are enabled the events that are triggering can be read from
@@ -190,11 +190,11 @@ be gathered on-line as appropriate. Examples of post-processing might include
190 190
191 o Reading information from /proc for the PID that triggered the event 191 o Reading information from /proc for the PID that triggered the event
192 o Deriving a higher-level event from a series of lower-level events. 192 o Deriving a higher-level event from a series of lower-level events.
193 o Calculate latencies between two events 193 o Calculating latencies between two events
194 194
195Documentation/trace/postprocess/trace-pagealloc-postprocess.pl is an example 195Documentation/trace/postprocess/trace-pagealloc-postprocess.pl is an example
196script that can read trace_pipe from STDIN or a copy of a trace. When used 196script that can read trace_pipe from STDIN or a copy of a trace. When used
197on-line, it can be interrupted once to generate a report without existing 197on-line, it can be interrupted once to generate a report without exiting
198and twice to exit. 198and twice to exit.
199 199
200Simplistically, the script just reads STDIN and counts up events but it 200Simplistically, the script just reads STDIN and counts up events but it
@@ -212,12 +212,12 @@ also can do more such as
212 processes, the parent process responsible for creating all the helpers 212 processes, the parent process responsible for creating all the helpers
213 can be identified 213 can be identified
214 214
2156. Lower-Level Analysis with PCL 2157. Lower-Level Analysis with PCL
216================================ 216================================
217 217
218There may also be a requirement to identify what functions with a program 218There may also be a requirement to identify what functions within a program
219were generating events within the kernel. To begin this sort of analysis, the 219were generating events within the kernel. To begin this sort of analysis, the
220data must be recorded. At the time of writing, this required root 220data must be recorded. At the time of writing, this required root:
221 221
222 $ perf record -c 1 \ 222 $ perf record -c 1 \
223 -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \ 223 -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \
@@ -253,11 +253,11 @@ perf report.
253 # (For more details, try: perf report --sort comm,dso,symbol) 253 # (For more details, try: perf report --sort comm,dso,symbol)
254 # 254 #
255 255
256According to this, the vast majority of events occured triggered on events 256According to this, the vast majority of events triggered on events
257within the VDSO. With simple binaries, this will often be the case so lets 257within the VDSO. With simple binaries, this will often be the case so let's
258take a slightly different example. In the course of writing this, it was 258take a slightly different example. In the course of writing this, it was
259noticed that X was generating an insane amount of page allocations so lets look 259noticed that X was generating an insane amount of page allocations so let's look
260at it 260at it:
261 261
262 $ perf record -c 1 -f \ 262 $ perf record -c 1 -f \
263 -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \ 263 -e kmem:mm_page_alloc -e kmem:mm_page_free_direct \
@@ -280,8 +280,8 @@ This was interrupted after a few seconds and
280 # (For more details, try: perf report --sort comm,dso,symbol) 280 # (For more details, try: perf report --sort comm,dso,symbol)
281 # 281 #
282 282
283So, almost half of the events are occuring in a library. To get an idea which 283So, almost half of the events are occurring in a library. To get an idea which
284symbol. 284symbol:
285 285
286 $ perf report --sort comm,dso,symbol 286 $ perf report --sort comm,dso,symbol
287 # Samples: 27666 287 # Samples: 27666
@@ -297,7 +297,7 @@ symbol.
297 0.01% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] get_fast_path 297 0.01% Xorg /opt/gfx-test/lib/libpixman-1.so.0.13.1 [.] get_fast_path
298 0.00% Xorg [kernel] [k] ftrace_trace_userstack 298 0.00% Xorg [kernel] [k] ftrace_trace_userstack
299 299
300To see where within the function pixmanFillsse2 things are going wrong 300To see where within the function pixmanFillsse2 things are going wrong:
301 301
302 $ perf annotate pixmanFillsse2 302 $ perf annotate pixmanFillsse2
303 [ ... ] 303 [ ... ]