Re: [PATCH] BUG preserve registers

[Andrew Morton <akpm@zip.com.au>]

Denis Vlasenko wrote:
> 
> On 10 February 2002 07:05, Andrew Morton wrote:
> > Andrew Morton wrote:
> > > I think I'm done with this now.
> >
> > Famous last words.  There's a slightly updated version at
> > http://www.zip.com.au/~akpm/linux/2.4/2.4.18-pre9/
> >
> > CONFIG_DEBUG_BUGVEROSE has been retained - it's now used to enable
> > BUG() in a couple of bloaty places.  But BUG() now always
> > reports file-and-line.
> >
> > Also, here's the kill-BUG-in-headers patch.  With this, the
> > kernel image is in fact a few kbytes smaller than stock
> > 2.4.18-pre9 with CONFIG_DEBUG_BUGVERBOSE=n, and we get full
> > file-and-line and the registers aren't trashed, so everyone's
> > happy.
> 
> I downloaded BUG.patch and inline-BUG.patch and compiled 2.4.18-pre9
> bzImage with and without these patches:
> 
> 2.4.18-pre9:         bzImage 996963  vmlinux 2440880
> 2.4.18-pre9+patches: bzImage 997531  vmlinux 2441182
> 
> .config is the same, you may find it below.
> "# CONFIG_DEBUG_BUGVERBOSE is not set"
> Andrew, are these numbers expected?

Pretty much.  In adding the patch, you've added six more bytes
to each BUG() call site and numerous filenames.  For me, the
kernel came out a little smaller, but for you, it grew a bit.  I
wonder why.  From a quick test it seems that gcc-3.0.3 and
recent binutils do fold common strings between compilation
units.  But I need to double-check that.  Maybe it's due to
that somehow.

Of course, for the extra 300 bytes of kernel you've gained
file-and-line info in all the BUG calls.

-