On Tue, Jan 03, 2023 at 11:58:48AM +0100, Ard Biesheuvel wrote:
> On Tue, 3 Jan 2023 at 03:13, Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
> >
> > On Mon, Jan 2, 2023 at 5:45 PM Guenter Roeck <linux@roeck-us.net> wrote:
> > >
> > > ... and reverting commit 99cb0d917ff indeed fixes the problem.
> >
> > Hmm. My gut feel is that this just exposes some bug in binutils.
> >
> > That said, maybe that commit should not have added its own /DISCARDS/
> > thing, and instead just added that "*(.note.GNU-stack)" to the general
> > /DISCARDS/ thing that is defined by the
> >
> >   #define DISCARDS  ..
> >
> > a little bit later, so that we only end up with one single DISCARD
> > list. Something like this (broken patch on purpose):
> >
> >   --- a/include/asm-generic/vmlinux.lds.h
> >   +++ b/include/asm-generic/vmlinux.lds.h
> >   @@ -897,5 +897,4 @@
> >     */
> >    #define NOTES                                        \
> >   -     /DISCARD/ : { *(.note.GNU-stack) }              \
> >         .notes : AT(ADDR(.notes) - LOAD_OFFSET) {       \
> >                 BOUNDED_SECTION_BY(.note.*, _notes)     \
> >   @@ -1016,4 +1015,5 @@
> >    #define DISCARDS                                     \
> >         /DISCARD/ : {                                   \
> >   +     *(.note.GNU-stack)                              \
> >         EXIT_DISCARDS                                   \
> >         EXIT_CALL                                       \
> >
> > But maybe that DISCARDS macrop ends up being used too late?
> >
> 
> Masahiro's v1 did something like this, and it caused an issue on
> RISC-V, which is why we ended up with this approach instead.

FWIW, I gave this one a go & it didn't produce the link failures that
Masahiro's v1 did on RISC-V...

> > It really shouldn't matter, but here we are, with a build problem with
> > some random old binutils on an odd platform..
> >
> 
> AIUI, the way ld.bfd used to combine output sections may also affect
> the /DISCARD/ pseudo-section, and so introducing it much earlier
> results in these discards to be interpreted in a different order.
> 
> The purpose of this change is to prevent .note.GNU-stack from deciding
> the section type of the .notes output section, and so keeping it in
> its own section should be sufficient. E.g.,
> 
> --- a/include/asm-generic/vmlinux.lds.h
> +++ b/include/asm-generic/vmlinux.lds.h
> @@ -896,7 +896,7 @@
>   * Otherwise, the type of .notes section would become PROGBITS
> instead of NOTES.
>   */
>  #define NOTES                                                          \
> -       /DISCARD/ : { *(.note.GNU-stack) }                              \
> +       .note.GNU-stack : { *(.note.GNU-stack) }                        \
>         .notes : AT(ADDR(.notes) - LOAD_OFFSET) {                       \
>                 BOUNDED_SECTION_BY(.note.*, _notes)                     \
>         } NOTES_HEADERS                                                 \
> 
> The .note.GNU-stack has zero size, so the result should be the same.

...and this one doesn't move DISCARDS around either, so also shouldn't
hit that issue. Famous last words, so I did run it through a config that
produced the link failures before & it was fine.