Re: [PATCH 6/6] x86: Fix -Os build (BZ #29576)

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 05/10/22 15:06, H.J. Lu wrote:
> On Wed, Oct 5, 2022 at 10:38 AM Adhemerval Zanella Netto
> <adhemerval.zanella@linaro.org> wrote:
>>
>>
>>
>> On 05/10/22 14:21, H.J. Lu wrote:
>>> On Wed, Oct 5, 2022 at 7:10 AM Carlos O'Donell <carlos@redhat.com> wrote:
>>>>
>>>> On Wed, Sep 21, 2022 at 10:51:08AM -0300, Adhemerval Zanella via Libc-alpha wrote:
>>>>> The compiler might transform __stpcpy calls (which are routed to
>>>>> __builtin_stpcpy as an optimization) to strcpy and x86_64 strcpy
>>>>> multiarch implementation does not build any working symbol due
>>>>> ISA_SHOULD_BUILD not being evaluated for IS_IN(rtld).
>>>>
>>>> Ohhhhhh... that is interesting. This changes the strcpy used in rtld for
>>>> all x86_64 build options, and I'm going to ACK this, but we may need to
>>>> revisit this if it shows up in a profile.
>>>
>>> Will this lead to both strcpy and stpcpy in ld.so? Currently there is only
>>> stpcpy in ld.so.
>>
>> I think that is expected behavior if compiler creates a reference for a
>> supported string function in the loader (rtld build pulls the
>> implementation).
> 
> strcpy is only generated in dl-profile.os:
> 
>    text    data     bss     dec     hex filename
>    2716       0      72    2788     ae4 dl-profile.os (-O2)
>    2265       0      72    2337     921 dl-profile.os (-Os)
>    1840       0       0    1840     730  strcpy.os
> 
> Should we compile dl-profile.c with -O2 with
> 
> CFLAGS-dl-profile.c += $(if $(findstring -Os,$(+cflags)), -O2)
> 
> It will create a smaller ld.so.

It might be an option to a subsequent patch, but I would avoid trying to outsmart
the compiler here since it might generate a smaller in other version and it would
most likely be arch-specific (adding even more building complexity).