No CONFIG_DEBUG_INFO on x86_64 Xen; why not?

No CONFIG_DEBUG_INFO on x86_64 Xen; why not?

[Stephen C. Tweedie]

Hi all,

lib/Kconfig.debug contains:

config DEBUG_INFO
        bool "Compile the kernel with debug info"
        depends on DEBUG_KERNEL && !X86_64_XEN

Does anyone recall why this was disabled for X86_64_XEN?  

hg annotate xen-unstable.hg leads to the cset description

        Update linux sparse tree to subarch layout.

and linux-2.6-merge.hg shows 

        various fixup in Kconfig to handle Xen
        
neither of which are particularly helpful.  I've just completed a build
with this change reverted and -g enabled and it built fine (although I'm
remote from the box right now so can't boot it yet); and gdb on the
vmlinux shows symbolic info is present.

Anyone remember, or object to reenabling CONFIG_DEBUG_INFO?

Thanks,
 Stephen

Re: No CONFIG_DEBUG_INFO on x86_64 Xen; why not?

[Keir Fraser]

On 6 Apr 2006, at 01:17, Stephen C. Tweedie wrote:

> neither of which are particularly helpful.  I've just completed a build
> with this change reverted and -g enabled and it built fine (although 
> I'm
> remote from the box right now so can't boot it yet); and gdb on the
> vmlinux shows symbolic info is present.
>
> Anyone remember, or object to reenabling CONFIG_DEBUG_INFO?

Well, it builds and boots fine, and we leave it disabled by default 
anyway, so I'll remove our patch to Kconfig.debug. I think maybe there 
was a problem with our port of debug code in entry-xen.S, but I see 
that file forcibly disables CONFIG_DEBUG_INFO when it builds, which is 
probably the better 'fix' for the time being.

  -- Keir

Re: No CONFIG_DEBUG_INFO on x86_64 Xen; why not?

[Stephen Tweedie]

Hi,

On Thu, Apr 06, 2006 at 09:51:06AM +0100, Keir Fraser wrote:
 
> >Anyone remember, or object to reenabling CONFIG_DEBUG_INFO?
> 
> Well, it builds and boots fine, and we leave it disabled by default 
> anyway, so I'll remove our patch to Kconfig.debug. I think maybe there 
> was a problem with our port of debug code in entry-xen.S, but I see 
> that file forcibly disables CONFIG_DEBUG_INFO when it builds, which is 
> probably the better 'fix' for the time being.

Right, entry-xen.S has problems --- it's broken even with the
CONFIG_DEBUG_INFO disable, as setting the new CONFIG_UNWIND_INFO
option still breaks it again.  But that's just as easy to disable as a
temporary workaround.  (I'd still like to find out why CFI is breaking
on Xen with those options disabled, though.)

--Stephen