blob: 0fd172ddc1b19d05330ba7bb35a7551949c30c6d (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
|
<?xml version='1.0'?>
<!DOCTYPE section PUBLIC "-//OASIS//DTD DocBook XML V4.5//EN" "http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd" [
]>
<section id="kernelprofsect">
<title>Kernel Profiling</title>
<remark>
WAR STORY: Kernel Profiling http://sourceware.org/systemtap/wiki/WSKernelProfile?highlight=((WarStories))
</remark>
<remark>
http://sourceware.org/systemtap/examples/process/pf2.stp
</remark>
<para>In <xref linkend="countcallssect"/>, you can revise the wildcards used in the probe to target <emphasis>all</emphasis> kernel functions. This can be useful if you are interested in indentifying what the kernel is performing over a specific time period.</para>
<para>However, doing so can cause considerable stress on the machine. In addition, this does not provide any indication of whether a specific function is being called too often during small time increments.</para>
<para>This section describes how to profile the kernel properly. <xref linkend="kernelprof"/> does this by providing a list of the top ten kernel functions called within a specific time period, and how many times each function was called during that time.</para>
<formalpara id="kernelprof">
<title>kernelprof.stp</title>
<para>
<programlisting>
#! /usr/bin/env stap
global profile, pcount
probe timer.profile {
pcount <<< 1
fn = probefunc ()
if (fn != "") profile[fn] <<< 1
}
probe timer.ms(5000) {
printf ("\n--- %d samples recorded:\n", @count(pcount))
foreach (f in profile- limit 10) {
printf ("%-30s\t%6d\n", f, @count(profile[f]))
}
delete profile
delete pcount
}
</programlisting>
</para>
</formalpara>
<para><xref linkend="kernelprof"/> records and outputs kernel functions called every 5 seconds. You can change this setting by editing <command>probe timer.ms(5000)</command> accordingly. <xref linkend="kernelprofoutput"/> contains an excerpt of the output over a 20-second period:</para>
<example id="kernelprofoutput">
<title><xref linkend="kernelprof"/> Sample Output</title>
<screen>
[...]
--- 10002 samples recorded:
sys_recvfrom 1
memmove 1
__copy_from_user_ll 17
__copy_to_user_ll 15
mwait_idle 5868
link_path_walk 1
kfree 1
fget_light 1
audit_syscall_exit 1
__d_lookup 1
--- 10002 samples recorded:
sysfs_read_file 1
free_poll_entry 1
syscall_exit_work 1
profile_hit 1
do_page_fault 2
_read_lock 3
kmap_atomic 4
strncpy_from_user 1
find_vma_prepare 1
__copy_from_user_ll 12
--- 10000 samples recorded:
system_call 4
unix_stream_sendmsg 1
__d_path 1
do_page_fault 2
kmap_atomic 1
find_vma 1
__copy_to_user_ll 16
__copy_from_user_ll 10
mwait_idle 5759
memcpy_fromiovec 1
--- 10004 samples recorded:
syscall_exit 1
__d_path 1
fput 2
do_page_fault 4
strncpy_from_user 1
may_open 1
ide_outb 1
ide_outsw 1
__copy_to_user_ll 11
__copy_from_user_ll 14
[...]
</screen>
</example>
<para><xref linkend="kernelprof"/> allows you to identify interesting kernel-specific performance facts about the system, such as kernel-intensive workloads and idle times (in this case, <command>mwait_idle</command>. Note that <xref linkend="kernelprof"/> does not count user-space functions, although it counts all kernel functions as part of the sample (even if the script cannot identify from which specific kernel function a call originated.</para>
</section>
|
