String literals in __init functions

[Mathias Krause]

On 26 March 2015 at 22:40, Andrew Morton <akpm at linux-foundation.org> wrote:
> On Thu, 26 Mar 2015 21:49:06 +0100 Mathias Krause <minipli at googlemail.com> wrote:
>
>> Andrew, what's your opinion on such a patch set? Do you too think it's
>> useful? Or do you share Ingo's fear about the additional maintenance
>> burden?
>
> I don't think the burden would be toooo high, although it will mess the
> code up a bit.
>
> The post-build checking for section reference mismatches will help,
> although that seems to have got itself turned off (what happened
> there?).

Seem to be working fine here.

This is make M=lib/, building the test module:
  CC [M]  lib/test_module.o
  Building modules, stage 2.
  MODPOST 1 modules
  LD [M]  lib/test_module.ko

This is the same module with two pi_info() calls in a non-__init
function, therefore generating section mismatches:
  CC [M]  lib/test_module.o
  Building modules, stage 2.
  MODPOST 1 modules
WARNING: modpost: Found 2 section mismatch(es).
To see full details build your kernel with:
'make CONFIG_DEBUG_SECTION_MISMATCH=y'
  LD [M]  lib/test_module.ko

> Did anyone look at writing a postprocessor for the .s files?  It
> doesn't look like it will be too hard from an initial peek.
>
> Did anyone ask the gcc developers?  I'd have thought that a function-wide
>
>         __attribute__((__string_section__(foo))
>
> wouldn't be a ton of work to implement.

The point is you cannot blindly mark all strings referenced from
__init / __exit code to end up in a matching string section because
strings in this code might have to live longer when passed to
functions keeping a pointer on them. For example, the name passed to
class_create() won't be copied. If that one would go into the
.init.rodata section automatically, we would have dangling pointers
after the .init.* memory got freed.
To know if it's safe to automatically put a string into an .init /
.exit section one needs to see the whole code. That's why I'm
reasoning it needs to be an LTO pass, not a .s post-processor or
function wide section attribute. Or, as in my approach, a source level
annotation.

Thanks,
Mathias