usr.sbin/tprof/README

   1 $NetBSD: README,v 1.8 2008/01/14 12:48:52 yamt Exp $
   2
   3 NOTE:
   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
   9 usage:
  10
  11 0. set SIZEOF_PTR environment variable, which is used by tpfmt.sh and tpann.sh.
  12    if not set, SIZEOF_PTR=4 is assumed.
  13
  14 1. add a line to your kernel config.
  15
  16         pseudo-device   tprof
  17
  18 2. create a device special file.
  19
  20         # mknod /dev/tprof c 191 0
  21
  22 3. 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
  34 4. 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
  77 5. 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                 :