Re: Ideas/Suggestions for 2.2

ralf@uni-koblenz.de
Fri, 12 Dec 1997 01:11:57 +0100


On Thu, Dec 11, 1997 at 09:56:55AM -0800, Ben Woodard wrote:

> Is it concievable to make something like sun's kadb but from the gdb
> sources? What kadb does is you boot it instead of the kernel and then
> it boots the kernel. Then when when an non maskable interrupt comes,
> it drops into kadb rather than rebooting.
>
> I haven't looked at the linux source yet but I know it is possible to
> vector certain keyboard interrupts into the debugger. So I believe it
> is possible to do this from linux, however could someone comment on
> the feasability of making a kadb type of debugger out of gdb.

For my MIPS hacking PMON, a ROM resident debugger running on an Algorithmics
P4032 evaluation board has been a gift sent by higher powers. The
kernel crashes, you press the debug button on the board and are in PMON,
where you now have things like stack backtraces etc. For example:

PMON> boot indy:/disk2/algor/vmlinux
Loading file: indy:/disk2/algor/vmlinux (elf)
0x80020000/1035072 + 0x8011cb40/216840 + 1910 syms\
Entry address is 80020508
PMON> g

Exception Cause=TLBL
bi_TagFind+0x20 8c43fff8 lw v1,-8(v0) # 0xfffffff8
PMON> bt
bi_TagFind+0x20 ()
setup_arch+0x78 ()
start_kernel+0x38 ()
PMON> boot indy:/disk2/algor/vmlinux
Loading file: indy:/disk2/algor/vmlinux (elf)
0x80020000/1034688 + 0x8011c9c0/216840 + 1910 syms-
Entry address is 80020508
PMON> g

Nothing happens ... Let's press the debug interrupt button:

Debug Interrupt
panic+0x110 08011635 j panic+0x110 # 0x800458d4
panic+0x114 00000000 nop

So, we had a kernel panic. Why the hell? Who was calling panic()?
What was the text for panic?

PMON> bt
panic+0x110 (0x800f4040,0x8010ad20,0x8010ad24,0x20004801)
setup_arch+0x34 ()
start_kernel+0x38 ()
PMON> d -s 0x800f4040
Unsupported architecture

That wasn't supposed to happen. Let's take a look at
arch/mips/kernel/setup.c:setup_arch():

[...]
switch(mips_machgroup)
{
#ifdef CONFIG_ALGOR_P0432
case MACH_GROUP_ALGORITHMICS:
p4032_setup();
break;
#endif
[...]

So, it was a stupid typo ...

PMON was originally written by IDT; enhancements have been made by
other parties and the code is available freely. It's a bit MIPS specific
but a lot of it should be portable to other systems for real programmers.
The cool thing is that it is ROM based, I rarely had the problem that a
kernel bug made the debugger unuseable which I've seen relativly often
with kgdb, works over a serial interface and can be interfaced to a full
blown gdb - which I never used since I'm actually a fan of printk() for
debugging. I use PMON though because it has it's strengths were printk()
fails.

Ralf