README
1$NetBSD: README,v 1.8 2008/01/14 12:48:52 yamt Exp $
2
3NOTE:
4 - tprof driver currently only supports pentium4 (netburst) processors.
5 - it samples program counters on every PMIs.
6 - it's currently hardcoded to use global_power_events events.
7 for details, see x86/x86/tprof_pmi.c and intel's processor manuals.
8
9usage:
10
110. set SIZEOF_PTR environment variable, which is used by tpfmt.sh and tpann.sh.
12 if not set, SIZEOF_PTR=4 is assumed.
13
141. add a line to your kernel config.
15
16 pseudo-device tprof
17
182. create a device special file.
19
20 # mknod /dev/tprof c 191 0
21
223. run the tprof command.
23
24 # tprof -o /tmp/foo sleep 1
25
26 tprof statistics:
27 sample 57
28 overflow 0
29 buf 3
30 emptybuf 3
31 dropbuf 0
32 dropbuf_sample 0
33
344. format the result.
35 the first line in the following example means that 11 samples have been
36 taken at 0xc0396c36, whose symbolic name is lapic_gettick+0x6.
37
38 # sh ./tpfmt.sh < /tmp/foo
39 11 c0396c36 lapic_gettick+0x6
40 5 c039b98a x86_pause+0x2
41 4 c010cf9d __cpu_simple_lock+0xd
42 2 c010cfcd __cpu_simple_lock_try+0xd
43 2 c039b571 bus_space_read_4+0x11
44 1 c01005c8 sse2_zero_page+0x18
45 1 c0100624 sse2_copy_page+0x34
46 1 c010ceeb mutex_spin_enter+0x2b
47 1 c010cef5 mutex_spin_enter+0x35
48 1 c010cf32 mutex_spin_exit+0x32
49 1 c0119ed0 in_localaddr+0x30
50 1 c012d0fd tcp_output+0x1fbd
51 1 c02980c2 amap_copy+0x42
52 1 c02a0100 uvm_map_lookup_entry_bytree+0x20
53 1 c02a27fe uvm_tree_RB_REMOVE+0xee
54 1 c02a8914 uvm_pagelookup+0x4
55 1 c02a9d5c uvm_pagefree+0xfc
56 1 c02a9e36 uvm_pagefree+0x1d6
57 1 c02dd9d1 _kernel_unlock+0xa1
58 1 c02e0285 mutex_vector_enter+0x15
59 1 c02eb83a sleepq_wake+0x5a
60 1 c0303467 pool_cache_get_paddr+0x97
61 1 c030368b pool_cache_put_slow+0x6b
62 1 c0321ed3 pffasttimo+0x33
63 1 c034547a VOP_LOCK+0xa
64 1 c0346235 VOP_ACCESS+0x45
65 1 c034a749 genfs_unlock+0x29
66 1 c038f251 cpu_idle+0x31
67 1 c03938da pmap_write_protect+0xaa
68 1 c0394305 pmap_do_remove+0x2e5
69 1 c03944b3 pmap_do_remove+0x493
70 1 c0396cdf lapic_delay+0x5f
71 1 c0396d19 lapic_delay+0x99
72 1 c0396d1d lapic_delay+0x9d
73 1 c0397429 lapic_clockintr+0x19
74 1 c039b984 x86_mwait+0xc
75 1 c042f66a _atomic_swap_32+0xa
76
775. tpann.sh is another formatter. it outputs "objdump -d" with numbers of
78 samples for each addresses.
79
80 # tprof -o /tmp/bar sleep 100
81 # sh ./tpann.sh < /tmp/bar
82
83 :
84 snip
85 :
86
87 c01005e0 <sse2_zero_page>:
88 4 c01005e0: 55 push %ebp
89 11 c01005e1: 89 e5 mov %esp,%ebp
90 1 c01005e3: 8b 54 24 08 mov 0x8(%esp),%edx
91 3 c01005e7: b9 00 10 00 00 mov $0x1000,%ecx
92 1 c01005ec: 31 c0 xor %eax,%eax
93 1 c01005ee: 89 f6 mov %esi,%esi
94 7936 c01005f0: 0f c3 42 00 movnti %eax,0x0(%edx)
95 6371 c01005f4: 0f c3 42 04 movnti %eax,0x4(%edx)
96 1220 c01005f8: 0f c3 42 08 movnti %eax,0x8(%edx)
97 741 c01005fc: 0f c3 42 0c movnti %eax,0xc(%edx)
98 1178 c0100600: 0f c3 42 10 movnti %eax,0x10(%edx)
99 1334 c0100604: 0f c3 42 14 movnti %eax,0x14(%edx)
100 976 c0100608: 0f c3 42 18 movnti %eax,0x18(%edx)
101 1299 c010060c: 0f c3 42 1c movnti %eax,0x1c(%edx)
102 954 c0100610: 83 e9 20 sub $0x20,%ecx
103 45 c0100613: 8d 52 20 lea 0x20(%edx),%edx
104 238 c0100616: 75 d8 jne c01005f0 <sse2_zero_page+0x10>
105 71 c0100618: 0f ae f8 sfence
106 297 c010061b: 5d pop %ebp
107 19 c010061c: c3 ret
108 0 c010061d: 8d 76 00 lea 0x0(%esi),%esi
109
110 :
111 snip
112 :
113