gdb/kgdb register mismatch

gdb/kgdb register mismatch

[Pete Popov]

I'm having the following problems with gdb/kgdb:

64bit CPU, running a 32 bit kernel. When I break at a certain point, I
can see that the kernel kgdb stub appears to be sending the correct
register information, where each reg value written in the kgdb packet is
32bit. However, the cross gdb client seems to be interpreting, or
expecting, 64bit register values so it 'skips' over every other value.
For example:

reg values sent by kgdb:      reg values shown by gdb client:
reg 0:  0x00000000            reg0:  0x00000001
reg 1:  0x00000001            reg1:  0x00000002
reg 2:  0x00000002            reg2:  0x00000004
reg 3:  0x00000003
reg 4:  0x00000004

Should the kgdb stub be sending 64bit values for the registers, even
though it's a 32bit kernel?  If the stub is supposed to be sending 32bit
register values, any suggestion why cross gdb is not interpreting them
correctly?

Thanks,

Pete

Re: gdb/kgdb register mismatch

[Daniel Jacobowitz]

On Thu, Feb 23, 2006 at 11:40:06AM -0800, Pete Popov wrote:
> 
> I'm having the following problems with gdb/kgdb:
> 
> 64bit CPU, running a 32 bit kernel. When I break at a certain point, I
> can see that the kernel kgdb stub appears to be sending the correct
> register information, where each reg value written in the kgdb packet is
> 32bit. However, the cross gdb client seems to be interpreting, or
> expecting, 64bit register values so it 'skips' over every other value.
> For example:
> 
> reg values sent by kgdb:      reg values shown by gdb client:
> reg 0:  0x00000000            reg0:  0x00000001
> reg 1:  0x00000001            reg1:  0x00000002
> reg 2:  0x00000002            reg2:  0x00000004
> reg 3:  0x00000003
> reg 4:  0x00000004
> 
> Should the kgdb stub be sending 64bit values for the registers, even
> though it's a 32bit kernel?  If the stub is supposed to be sending 32bit
> register values, any suggestion why cross gdb is not interpreting them
> correctly?

This is a FAQ; it's a bug in GDB that will be fixed someday, but for
now use "set architecture mips:isa32".

-- 
Daniel Jacobowitz
CodeSourcery

Re: gdb/kgdb register mismatch

[Pete Popov]

On Thu, 2006-02-23 at 16:31 -0500, Daniel Jacobowitz wrote:
> On Thu, Feb 23, 2006 at 11:40:06AM -0800, Pete Popov wrote:
> > 
> > I'm having the following problems with gdb/kgdb:
> > 
> > 64bit CPU, running a 32 bit kernel. When I break at a certain point, I
> > can see that the kernel kgdb stub appears to be sending the correct
> > register information, where each reg value written in the kgdb packet is
> > 32bit. However, the cross gdb client seems to be interpreting, or
> > expecting, 64bit register values so it 'skips' over every other value.
> > For example:
> > 
> > reg values sent by kgdb:      reg values shown by gdb client:
> > reg 0:  0x00000000            reg0:  0x00000001
> > reg 1:  0x00000001            reg1:  0x00000002
> > reg 2:  0x00000002            reg2:  0x00000004
> > reg 3:  0x00000003
> > reg 4:  0x00000004
> > 
> > Should the kgdb stub be sending 64bit values for the registers, even
> > though it's a 32bit kernel?  If the stub is supposed to be sending 32bit
> > register values, any suggestion why cross gdb is not interpreting them
> > correctly?
> 
> This is a FAQ; it's a bug in GDB that will be fixed someday, but for
> now use "set architecture mips:isa32".

It works, thanks! I'm not sure how well documented this is but perhaps
it's just because I don't use kgdb often.

Pete