On Wed, 13 May 2026, at 10:50, Mike Rapoport wrote:
> On Tue, May 12, 2026 at 02:56:16PM +0200, Ard Biesheuvel wrote:
>> On Mon, 11 May 2026, at 16:40, Jann Horn wrote:
>> > On Mon, May 11, 2026 at 10:59 AM Ard Biesheuvel <[email protected]> wrote:
>> >> I think we should simply do something along the lines of the below,
>> >> considering that the size of a data object tends to correlate with
>> >> its minimum alignment.
>> >>
>> >> I do find it rather puzzling that the compiler emits empty_zero_page
>> >> *after* zero_page_pfn - ideally, we'd combine the below with
>> >> -fdata-sections so that the linker sees all individual objects, but
>> >> I suspect that would create some problems elsewhere.
>> >>
>> >>
>> >> --- a/include/asm-generic/vmlinux.lds.h
>> >> +++ b/include/asm-generic/vmlinux.lds.h
>> >> @@ -452,7 +452,7 @@
>> >> #define RO_AFTER_INIT_DATA \
>> >> . = ALIGN(8); \
>> >> __start_ro_after_init = .; \
>> >> - *(.data..ro_after_init) \
>> >> + *(SORT_BY_ALIGNMENT(.data..ro_after_init)) \
>> >
>> > Oh, neat, I didn't realize that's possible. That seems like a nicer
>> > approach...
>>
>> Neat but rather ineffective, unfortunately. (I don't see a size
>> difference with the arm64 defconfig kernel)
>>
>> Given that empty_zero_page only ever gets its address taken, we
>> might just move it into the linker script if that requires tweaking
>> anyway. We can just place it at the start of .rodata, which is
>> already page aligned on most architectures (and will become page
>> aligned unless EMPTY_ZERO_PAGE is #define'd by the arch linker
>> script to something else)
>>
>>
>> --- a/include/asm-generic/vmlinux.lds.h
>> +++ b/include/asm-generic/vmlinux.lds.h
>> @@ -472,6 +472,17 @@
>> #endif
>> #endif
>>
>> +#ifndef EMPTY_ZERO_PAGE
>> +#ifndef __HAVE_COLOR_ZERO_PAGE
>
> I don't think we want let architectures that don't use colored zero pages
> redefine it.
> If it will be really required we can add the ability to redefine
> EMPTY_ZERO_PAGE later.
>
I was actually intending to add use this for arm64 in the next patch. It
already has a reserved_pg_dir in .rodata which is page-sized (i.e., up
to 64k in size) and guaranteed to remain all zeroes, so empty_zero_page
could actually be an alias for that. This is what I had in a previous
revision, before you turned the empty_zero_page definition into common
code:
https://lore.kernel.org/all/[email protected]/
