File: [local] / www / ddb.html (download) (as text)
Revision 1.16, Sat Nov 19 13:15:10 2016 UTC (7 years, 6 months ago) by tb
Branch: MAIN
Changes since 1.15: +1 -1 lines
one more obj/ needed
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<title>OpenBSD: Crash Reports</title>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<meta name="description" content="How to report an OpenBSD kernel crash">
<meta name="copyright" content="This document copyright 1998-2016 by OpenBSD.">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="stylesheet" type="text/css" href="openbsd.css">
<link rel="canonical" href="https://www.openbsd.org/report.html">
<style type="text/css">
h3, h4 { color: #0000e0; }
</style>
</head>
<body bgcolor="#ffffff" text="#000000" link="#23238e">
<h2>
<a href="index.html">
<font color="#0000ff"><i>Open</i></font><font color="#000084">BSD</font></a>
<font color="#e00000">Crash Reports</font>
</h2>
<hr>
<p>
<h3>Minimum information for kernel problems</h3>
Familiarize yourself with
<a href="report.html">the general bug reporting procedures</a>
first.
All of that will apply.
When reporting a kernel panic or crash, please remember:
<ul>
<li><i>We need the console output on the screen</i>.
Capture it and save it.
Serial consoles are best, but if you are on a VGA console you can
<a href="faq/faq7.html">scroll the console back</a>
and take readable pictures with a phone or camera.<br>
<li><i>If the kernel panicked we need the traceback.</i>
It may be displayed on the screen.
If you are at a
<tt><a href="http://man.openbsd.org/ddb.4">ddb</a>></tt>
prompt, type <tt>trace</tt>.
If you are running SMP, use the <tt>mach ddbcpu N</tt> command for each
of the <tt>N</tt> processors you have and repeat the <tt>trace</tt>
command for each processor.<br>
<li><i>We need the process list.</i>
Use the command <tt>ps</tt> to get that.
</ul>
<i>
Reports without the above information are useless.
This is the minimum we need to be able to track down the issue.
</i>
<h3>Additional information you can send</h3>
In some situations more information is desirable.
Below are outlined some additional steps you can take in certain situations:
<ul>
<li><i>If your crash appears to involve filesystems.</i>
The following additional things would be helpful
<ul>
<li>The output of the
<tt><a href="http://man.openbsd.org/ddb.4">ddb</a>></tt> command
<tt>show uvm</tt>
<li>The output of the
<tt><a href="http://man.openbsd.org/ddb.4">ddb</a>></tt>
command <tt>show bcstats</tt>
<li>The output of the <tt>mount</tt> command from your running machine, so
we know what filesystems are mounted and how.
</ul>
<li> ... XXX boot crash? XXX
<li> ... XXX show regs? XXX
</ul>
<h3>Lost the panic message?</h3>
Under some circumstances, you may lose the very first message of a panic,
stating the reason for the panic.
<blockquote><pre>
ddb> <b>show panic</b>
0: kernel: page fault trap, code=0
ddb>
</pre></blockquote>
<h3>Note for SMP systems</h3>
You should get a trace from each processor as part of your report:
<blockquote><pre>
ddb{0}> <b>trace</b>
pool_get(d05e7c20,0,dab19ef8,d0169414,80) at pool_get+0x226
fxp_add_rfabuf(d0a62000,d3c12b00,dab19f10,dab19f10) at fxp_add_rfabuf+0xa5
fxp_intr(d0a62000) at fxp_intr+0x1e7
Xintr_ioapic0() at Xintr_ioapic0+0x6d
--- interrupt ---
idle_loop+0x21:
ddb{0}> <b>machine ddbcpu 1</b>
Stopped at Debugger+0x4: leave
ddb{1}> <b>trace</b>
Debugger(d0319e28,d05ff5a0,dab1bee8,d031cc6e,d0a61800) at Debugger+0x4
i386_ipi_db(d0a61800,d05ff5a0,dab1bef8,d01eb997) at i386_ipi_db+0xb
i386_ipi_handler(b0,d05f0058,dab10010,d01d0010,dab10010) at i386_ipi_handler+0x
4a
Xintripi() at Xintripi+0x47
--- interrupt ---
i386_softintlock(0,58,dab10010,dab10010,d01e0010) at i386_softintlock+0x37
Xintrltimer() at Xintrltimer+0x47
--- interrupt ---
idle_loop+0x21:
ddb{1}>
</pre></blockquote>
Repeat the <tt>machine ddbcpu x</tt> followed by <tt>trace</tt> for each
processor in your machine.
<h3>How do I gather further information from a kernel crash?</h3><p>
A typical kernel crash on OpenBSD might look like this:
(things to watch for are marked with bold font)
<blockquote><pre>
kernel: page fault trap, code=0
Stopped at <b>_pf_route+0x263</b>: mov 0x40(%edi),%edx
ddb>
</pre></blockquote>
The first command to run from the <tt>ddb></tt> prompt is <tt>trace</tt>
(see <a href="http://man.openbsd.org/ddb">ddb(4)</a> for details):
<blockquote><pre>
ddb> <b>trace</b>
<b>_pf_route</b>(e28cb7e4,e28bc978,2,1fad,d0b8b120) at <b>_pf_route+0x263</b>
_pf_test(2,1f4ad,e28cb7e4,b4c1) at _pf_test+0x706
_pf_route(e28cbb00,e28bc978,2,d0a65440,d0b8b120) at _pf_route+0x207
_pf_test(2,d0a65440,e28cbb00,d023c282) at _pf_test+0x706
_ip_output(d0b6a200,0,0,0,0) at _ip_output+0xb67
_icmp_send(d0b6a200,0,1,a012) at _icmp_send+0x57
_icmp_reflect(d0b6a200,0,1,0,3) at _icmp_reflect+0x26b
_icmp_input(d0b6a200,14,0,0,d0b6a200) at _icmp_input+0x42c
_ipv4_input(d0b6a200,e289f140,d0a489e0,e289f140) at _ipv4_input+0x6eb
_ipintr(10,10,e289f140,e289f140,e28cbd38) at _ipintr+0x8d
Bad frame pointer: 0xe28cbcac
ddb>
</pre></blockquote>
This tells us what function calls lead to the crash.
<p>
To find out the particular line of C code that caused the crash, you can
do the following:
<p>
Find the source file where the crashing function is defined in.
In this example, that would be <tt>pf_route()</tt> in <tt>sys/net/pf.c</tt>.
Recompile that source file with debug information:
<blockquote><pre>
# <b>cd /usr/src/sys/arch/$(uname -m)/compile/GENERIC</b>
# <b>rm obj/pf.o</b>
# <b>DEBUG=-g make pf.o</b>
</pre></blockquote>
Then use <a href="http://man.openbsd.org/objdump">objdump(1)</a> to get the
disassembly:
<blockquote><pre>
# <b>objdump --line --disassemble --reloc obj/pf.o >pf.dis</b>
</pre></blockquote>
In the output, grep for the function name:
<blockquote><pre>
# <b>grep "<_pf_route>:" pf.dis</b>
0000<b>7d88</b> <_pf_route>:
</pre></blockquote>
Take this first hex number and add the offset from the <tt>Stopped at</tt> line:
<tt>0x7d88 + 0x263 == 0x7feb</tt>.
<p>
Scroll down to that line (the assembler instruction should match the one
quoted in the <tt>Stopped at</tt> line), then up to the nearest C line number:
<blockquote><pre>
# <b>more pf.dis</b>
/usr/src/sys/arch/i386/compile/GENERIC/../../../../net/pf.c:<b>3872</b>
7fe7: 0f b7 43 02 movzwl 0x2(%ebx),%eax
<b>7feb</b>: 8b 57 40 mov 0x40(%edi),%edx
7fee: 39 d0 cmp %edx,%eax
7ff0: 0f 87 92 00 00 00 ja 8088 <_pf_route+0x300>
</pre></blockquote>
So, it's precisely line <tt>3872</tt> of <tt>pf.c</tt> that crashes:
<blockquote><pre>
# <b>cat -n pf.c | head -n 3872 | tail -n 1</b>
3872 if ((u_int16_t)ip->ip_len <= ifp->if_mtu) {
</pre></blockquote>
Note that the kernel that produced the crash output and the object file
for objdump must be compiled from the exact same source file, otherwise
the offsets won't match.
<p>
If you provide both the ddb trace output and the relevant objdump section,
that's very helpful.
<p>
</body>
</html>