[PATCH v3] elf: Set p_align to the minimum page size if possible

[H.J. Lu]

On Fri, Jan 14, 2022 at 5:40 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> On 14.01.2022 14:03, H.J. Lu wrote:
> > On Fri, Jan 14, 2022 at 12:27 AM Jan Beulich <jbeulich@suse.com> wrote:
> >>
> >> On 29.12.2021 20:39, H.J. Lu via Binutils wrote:
> >>> --- a/ld/testsuite/ld-elf/linux-x86.exp
> >>> +++ b/ld/testsuite/ld-elf/linux-x86.exp
> >>> @@ -185,6 +185,42 @@ run_ld_link_exec_tests [list \
> >>>       "" \
> >>>       "tmpdir/indirect-extern-access-2.so" \
> >>>      ] \
> >>> +    [list \
> >>> +     "Run p_align-1a without PIE" \
> >>> +     "$NOPIE_LDFLAGS" \
> >>> +     "" \
> >>> +     { p_align-1.c } \
> >>> +     "p_align-1a" \
> >>> +     "pass.out" \
> >>> +     "$NOPIE_CFLAGS" \
> >>> +    ] \
> >>> +    [list \
> >>> +     "Run p_align-1b with PIE" \
> >>> +     "-pie" \
> >>> +     "" \
> >>> +     { p_align-1.c } \
> >>> +     "p_align-1b" \
> >>> +     "pass.out" \
> >>> +     "-fpie" \
> >>> +    ] \
> >>> +    [list \
> >>> +     "Run p_align-1c with -Wl,-z,max-page-size=0x1000 without PIE" \
> >>> +     "$NOPIE_LDFLAGS -Wl,-z,max-page-size=0x1000" \
> >>> +     "" \
> >>> +     { p_align-1.c } \
> >>> +     "p_align-1c" \
> >>> +     "pass.out" \
> >>> +     "$NOPIE_CFLAGS" \
> >>> +    ] \
> >>> +    [list \
> >>> +     "Run p_align-1d with -Wl,-z,max-page-size=0x1000 with PIE" \
> >>> +     "-pie -Wl,-z,max-page-size=0x1000" \
> >>> +     "" \
> >>> +     { p_align-1.c } \
> >>> +     "p_align-1d" \
> >>> +     "pass.out" \
> >>> +     "-fpie" \
> >>> +    ] \
> >>>  ]
> >>
> >> The two PIE variants of this also fail for me on glibc 2.26. Looks
> >> like LOAD segments' alignment isn't being honored there, at least
> >> not if it's as big as it is here.
> >>
> >
> > The PIE alignment needs the kernel fix:
> >
> > commit ce81bb256a224259ab686742a6284930cbe4f1fa
> > Author: Chris Kennelly <ckennelly@google.com>
> > Date:   Thu Oct 15 20:12:32 2020 -0700
> >
> >     fs/binfmt_elf: use PT_LOAD p_align values for suitable start address
>
> Well, then the test needs to be skipped if that fix is not in place.
> After all you're testing binutils behavior here, not kernel or libc one.
> I'm running a variety of (largely up-to-date) kernels on all of my
> systems. But it looks like our kernel folks decided against backporting
> this particular change. And I don't think you expect people to remember
> to run the testsuite only on top of "certain" kernels?

Care to submit a patch?

-- 
H.J.