x86, ftrace, hw-branch-tracer: documentation
authorMarkus Metzger <markus.t.metzger@intel.com>
Mon, 19 Jan 2009 09:35:58 +0000 (10:35 +0100)
committerIngo Molnar <mingo@elte.hu>
Tue, 20 Jan 2009 12:04:30 +0000 (13:04 +0100)
Document the hw-branch-tracer in the ftrace documentation.

Signed-off-by: Markus Metzger <markus.t.metzger@intel.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Documentation/ftrace.txt

index 803b1318b13da11242a3558123d28293d3ee4378..758fb42a1b6846971776454727a6d588322a0dcc 100644 (file)
@@ -165,6 +165,8 @@ Here is the list of current tracers that may be configured.
   nop - This is not a tracer. To remove all tracers from tracing
                simply echo "nop" into current_tracer.
 
+  hw-branch-tracer - traces branches on all cpu's in a circular buffer.
+
 
 Examples of using the tracer
 ----------------------------
@@ -1152,6 +1154,78 @@ int main (int argc, char **argv)
         return 0;
 }
 
+
+hw-branch-tracer (x86 only)
+---------------------------
+
+This tracer uses the x86 last branch tracing hardware feature to
+collect a branch trace on all cpus with relatively low overhead.
+
+The tracer uses a fixed-size circular buffer per cpu and only
+traces ring 0 branches. The trace file dumps that buffer in the
+following format:
+
+# tracer: hw-branch-tracer
+#
+# CPU#        TO  <-  FROM
+   0  scheduler_tick+0xb5/0x1bf          <-  task_tick_idle+0x5/0x6
+   2  run_posix_cpu_timers+0x2b/0x72a    <-  run_posix_cpu_timers+0x25/0x72a
+   0  scheduler_tick+0x139/0x1bf         <-  scheduler_tick+0xed/0x1bf
+   0  scheduler_tick+0x17c/0x1bf         <-  scheduler_tick+0x148/0x1bf
+   2  run_posix_cpu_timers+0x9e/0x72a    <-  run_posix_cpu_timers+0x5e/0x72a
+   0  scheduler_tick+0x1b6/0x1bf         <-  scheduler_tick+0x1aa/0x1bf
+
+
+The tracer may be used to dump the trace for the oops'ing cpu on a
+kernel oops into the system log. To enable this, ftrace_dump_on_oops
+must be set. To set ftrace_dump_on_oops, one can either use the sysctl
+function or set it via the proc system interface.
+
+  sysctl kernel.ftrace_dump_on_oops=1
+
+or
+
+  echo 1 > /proc/sys/kernel/ftrace_dump_on_oops
+
+
+Here's an example of such a dump after a null pointer dereference in a
+kernel module:
+
+[57848.105921] BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
+[57848.106019] IP: [<ffffffffa0000006>] open+0x6/0x14 [oops]
+[57848.106019] PGD 2354e9067 PUD 2375e7067 PMD 0
+[57848.106019] Oops: 0002 [#1] SMP
+[57848.106019] last sysfs file: /sys/devices/pci0000:00/0000:00:1e.0/0000:20:05.0/local_cpus
+[57848.106019] Dumping ftrace buffer:
+[57848.106019] ---------------------------------
+[...]
+[57848.106019]    0  chrdev_open+0xe6/0x165      <-  cdev_put+0x23/0x24
+[57848.106019]    0  chrdev_open+0x117/0x165     <-  chrdev_open+0xfa/0x165
+[57848.106019]    0  chrdev_open+0x120/0x165     <-  chrdev_open+0x11c/0x165
+[57848.106019]    0  chrdev_open+0x134/0x165     <-  chrdev_open+0x12b/0x165
+[57848.106019]    0  open+0x0/0x14 [oops]        <-  chrdev_open+0x144/0x165
+[57848.106019]    0  page_fault+0x0/0x30         <-  open+0x6/0x14 [oops]
+[57848.106019]    0  error_entry+0x0/0x5b        <-  page_fault+0x4/0x30
+[57848.106019]    0  error_kernelspace+0x0/0x31          <-  error_entry+0x59/0x5b
+[57848.106019]    0  error_sti+0x0/0x1   <-  error_kernelspace+0x2d/0x31
+[57848.106019]    0  page_fault+0x9/0x30         <-  error_sti+0x0/0x1
+[57848.106019]    0  do_page_fault+0x0/0x881     <-  page_fault+0x1a/0x30
+[...]
+[57848.106019]    0  do_page_fault+0x66b/0x881   <-  is_prefetch+0x1ee/0x1f2
+[57848.106019]    0  do_page_fault+0x6e0/0x881   <-  do_page_fault+0x67a/0x881
+[57848.106019]    0  oops_begin+0x0/0x96         <-  do_page_fault+0x6e0/0x881
+[57848.106019]    0  trace_hw_branch_oops+0x0/0x2d       <-  oops_begin+0x9/0x96
+[...]
+[57848.106019]    0  ds_suspend_bts+0x2a/0xe3    <-  ds_suspend_bts+0x1a/0xe3
+[57848.106019] ---------------------------------
+[57848.106019] CPU 0
+[57848.106019] Modules linked in: oops
+[57848.106019] Pid: 5542, comm: cat Tainted: G        W  2.6.28 #23
+[57848.106019] RIP: 0010:[<ffffffffa0000006>]  [<ffffffffa0000006>] open+0x6/0x14 [oops]
+[57848.106019] RSP: 0018:ffff880235457d48  EFLAGS: 00010246
+[...]
+
+
 dynamic ftrace
 --------------