Re: [PATCH 2/2] x86/boot/compressed: Remove unnecessary sections from bzImage

[Arvind Sankar]

On Sat, Feb 22, 2020 at 10:37:47AM -0500, Arvind Sankar wrote:
> On Sat, Feb 22, 2020 at 12:42:42AM -0700, Nathan Chancellor wrote:
> > 
> > Thanks for the clarity. With your suggestion (diff below), I see the
> > following error:
> > 
> > arch/x86/boot/compressed/vmlinux: no symbols
> > ld.lld: error: undefined symbol: ZO_input_data
> > >>> referenced by arch/x86/boot/header.o:(.header+0x59)
> > 
> > ld.lld: error: undefined symbol: ZO_z_input_len
> > >>> referenced by arch/x86/boot/header.o:(.header+0x5D)
> > make[3]: *** [../arch/x86/boot/Makefile:108: arch/x86/boot/setup.elf]
> > 
> > It seems like the section still isn't being added?
> > 
> > Cheers,
> > Nathan
> 
> It seems like lld also doesn't treat .symtab as special and is
> discarding it, but that one is actually essential to be able to build
> the bzImage.
> 
> The sections that GNU ld ends up discarding via that *(*) directive are
> .dynsym, .dynstr, .gnu.hash, .eh_frame, .rela.dyn, .comment and
> .dynamic.
> 
> Out of these, only .eh_frame has any significant size, and that's what
> we discard in the other linker scripts (in kernel/vmlinux.lds.S and
> boot/setup.ld).
> 
> It looks like it would be safest to just do
> 	/DISCARD/ : {
> 		*(.eh_frame)
> 	}
> instead. If you can double-check that that works with lld, I can send
> out a new version.
> 
> Thanks and sorry for the breakage.

Tested with lld and it seems fine with that change.

Boris, should I send the fix as a diff to the current patch in tip, or
as a fresh one that can replace it?

Thanks.