Re: [PATCH] stack2core: show stack message and convert it to core file when kernel die

[Hui Zhu <teawater@gmail.com>]

Thanks Russell,

This S2C: message just for program s2c.
s2c can convert it to a core file.  Then gdb can do a clear analyse
with this file.
Then you can get more message than current we can get.

For example:
(gdb) bt
#0  0xc0008470 in kernel_init (unused=<value optimized out>)
   at /home/teawater/kernel/arm_versatile_926ejs.glibc_std.standard/build/linux/init/main.c:916
#1  0xc0042660 in sys_waitid (which=<value optimized out>, upid=<value
optimized out>, infop=0x0, options=0, ru=0x14)
   at /home/teawater/kernel/arm_versatile_926ejs.glibc_std.standard/build/linux/kernel/exit.c:1798
Backtrace stopped: previous frame inner to this frame (corrupt stack?)

a more clear backtrace.

(gdb) frame 1
#1  0xc0042660 in sys_waitid (which=<value optimized out>, upid=<value
optimized out>, infop=0x0, options=0, ru=0x14)
   at /home/teawater/kernel/arm_versatile_926ejs.glibc_std.standard/build/linux/kernel/exit.c:1798
1798                    pid = find_get_pid(upid);
(gdb) p pid
$1 = (struct pid *) 0x0

A value of a val in stack in the frame.


I think it will more helpful for user to deal with the kernel die.

Best regards,
Hui

On Mon, Jan 4, 2010 at 00:03, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
> On Sun, Jan 03, 2010 at 11:05:05PM +0800, Hui Zhu wrote:
>> Hello,
>>
>> For, when the kernel die, the user will get some message like:
>> PC is at kernel_init+0xd4/0x104
>> LR is at _atomic_dec_and_lock+0x48/0x6c
>> pc : [<c0008470>]    lr : [<c01911f8>]    psr: 60000013
>> sp : c7823fd8  ip : c7823f48  fp : c7823ff4
>> Stack: (0xc7823fd8 to 0xc7824000)
>> 3fc0:                                                       00000000 00000001
>> Backtrace:
>> [<c000839c>] (kernel_init+0x0/0x104) from [<c0042660>] (do_exit+0x0/0x880)
>> This backtrace have some wrong message sometime and cannot get any
>> val. Of course, kdump can get more message.  But it need do some a lot
>> of other config.
>
> If you have frame pointers enabled, the backtrace is _never_ wrong.
> It only goes wrong if you disable frame pointers, at which point the
> unwind tables have to be used.
>
>> When kernel die, show some message:
>> S2C:elf_class=1
>> S2C:elf_data=1
>> S2C:elf_arch=40
>> S2C:elf_osabi=0
>> S2C:r0=0x00000000;
>> S2C:r1=0xc7822000;
>> S2C:r2=0xc7823f48;
>> S2C:r3=0x00000003;
>> S2C:r4=0x00000000;
>> S2C:r5=0x00000000;
>> S2C:r6=0x00000000;
>> S2C:r7=0x00000000;
>> S2C:r8=0x00000000;
>> S2C:r9=0x00000000;
>> S2C:r10=0x00000000;
>> S2C:fp=0xc7823ff4;
>> S2C:ip=0xc7823f48;
>> S2C:sp=0xc7823fd8;
>> S2C:lr=0xc01911f8;
>> S2C:pc=0xc0008470;
>> S2C:cpsr=0x60000013;
>> S2C:ORIG_r0=0xffffffff;
>>
>> S2C:stack=0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
>> S2C:stack=0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>> S2C:stack=0x00, 0x00, 0x00, 0x00, 0xf8, 0x3f, 0x82, 0xc7,
>> S2C:stack=0x60, 0x26, 0x04, 0xc0, 0xa8, 0x83, 0x00, 0xc0,
>> S2C:stack=0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
>
> Please don't invent yet another way of dumping stuff out of the kernel.
> What we already have is sufficient for your needs - there's no reason
> what so ever to change it to achieve your goals.  We already dump the
> registers and the stack, which seems to be all that you require.
>