Re: [PATCH v5 7/8] MdePkg GCC/X64: avoid 'hidden' visibility for module entry points

["Gao, Liming" <liming.gao@intel.com>]

Ard:
  I will verify it. And, I would ask why only X64 requires it? IA32, ARM and AARCH64 doesn't specially handle it?

Thanks
Liming
> -----Original Message-----
> From: Ard Biesheuvel [mailto:ard.biesheuvel@linaro.org]
> Sent: Tuesday, August 02, 2016 12:12 AM
> To: Gao, Liming <liming.gao@intel.com>
> Cc: Shi, Steven <steven.shi@intel.com>; Zhu, Yonghong
> <yonghong.zhu@intel.com>; Justen, Jordan L <jordan.l.justen@intel.com>;
> edk2-devel@lists.01.org; lersek@redhat.com; leif.lindholm@linaro.org
> Subject: Re: [edk2] [PATCH v5 7/8] MdePkg GCC/X64: avoid 'hidden' visibility
> for module entry points
> 
> On 1 August 2016 at 17:51, Ard Biesheuvel <ard.biesheuvel@linaro.org>
> wrote:
> > On 1 August 2016 at 16:56, Ard Biesheuvel <ard.biesheuvel@linaro.org>
> wrote:
> >> On 1 August 2016 at 16:49, Ard Biesheuvel <ard.biesheuvel@linaro.org>
> wrote:
> >>> On 1 August 2016 at 16:18, Gao, Liming <liming.gao@intel.com> wrote:
> >>>> Ard:
> >>>>   I don't think it is good way to define GCC_VISIBILITY_PROTECTED and
> apply it in EntryPointLib. We only need to expose _ModuleEntryPoint. It has
> been specified in LINK_FLAGS in tools_def.txt. Could we also specify its
> attribute in CC_FLAGS or LINK_FLAGS in tools_def.txt?
> >>>>
> >>>
> >>> It seems this does the trick as well
> >>>
> >>> diff --git a/BaseTools/Scripts/GccBase.lds
> b/BaseTools/Scripts/GccBase.lds
> >>> index 281af8a9bd33..02387d4f8d6f 100644
> >>> --- a/BaseTools/Scripts/GccBase.lds
> >>> +++ b/BaseTools/Scripts/GccBase.lds
> >>> @@ -80,3 +80,7 @@ SECTIONS {
> >>>      *(COMMON)
> >>>    }
> >>>  }
> >>> +
> >>> +VERSION {
> >>> +  { global: _ModuleEntryPoint*; };
> >>> +};
> >>>
> >>>
> >>> Note that * at the end: this is necessary since _ModuleEntryPoint will
> >>> be called _ModuleEntryPoint.lto_priv.xxx in the LTO objects.
> >>>
> >>
> >> Hmm, looks like I spoke too soon. I don't know what I did wrong, but
> >> this does not actually work.
> >>
> >
> > The only alternative I can think of is to add a static non-lto object
> > to the tree that refers to _ModuleEntryPoint, similar to the way I
> > handle the ARM intrinsics in patch #5
> >
> 
> As it turns out, the LTO linker does not need to visibility pragma to
> prevent it from emitting GOT based relocations. This makes sense,
> considering that the LTO linker can see that no symbol references are
> ever satisfied across dynamic object boundaries. That means I could
> work around this in the following way:
> 
> diff --git a/BaseTools/Conf/tools_def.template
> b/BaseTools/Conf/tools_def.template
> index 314adaf6bfa8..983e2fea7390 100644
> --- a/BaseTools/Conf/tools_def.template
> +++ b/BaseTools/Conf/tools_def.template
> @@ -4462,7 +4462,7 @@ DEFINE GCC49_ARM_ASLDLINK_FLAGS      =
> DEF(GCC48_ARM_ASLDLINK_FLAGS)
>  DEFINE GCC49_AARCH64_ASLDLINK_FLAGS  =
> DEF(GCC48_AARCH64_ASLDLINK_FLAGS)
> 
>  DEFINE GCC5_IA32_CC_FLAGS            = DEF(GCC49_IA32_CC_FLAGS) -flto
> -fno-builtin
> -DEFINE GCC5_X64_CC_FLAGS             = DEF(GCC49_X64_CC_FLAGS) -flto
> -fno-builtin
> +DEFINE GCC5_X64_CC_FLAGS             = DEF(GCC49_X64_CC_FLAGS) -flto
> -fno-builtin -DUSING_LTO
>  DEFINE GCC5_IA32_X64_DLINK_COMMON    =
> DEF(GCC49_IA32_X64_DLINK_COMMON)
>  DEFINE GCC5_IA32_X64_ASLDLINK_FLAGS  =
> DEF(GCC49_IA32_X64_ASLDLINK_FLAGS)
>  DEFINE GCC5_IA32_X64_DLINK_FLAGS     =
> DEF(GCC49_IA32_X64_DLINK_FLAGS) -flto
> diff --git a/MdePkg/Include/X64/ProcessorBind.h
> b/MdePkg/Include/X64/ProcessorBind.h
> index 666cc8e8bd16..77fab7055afc 100644
> --- a/MdePkg/Include/X64/ProcessorBind.h
> +++ b/MdePkg/Include/X64/ProcessorBind.h
> @@ -27,12 +27,15 @@
>  #pragma pack()
>  #endif
> 
> -#if defined(__GNUC__) && defined(__pic__)
> +#if defined(__GNUC__) && defined(__pic__) && !defined(USING_LTO)
>  //
>  // Mark all symbol declarations and references as hidden, meaning they will
>  // not be subject to symbol preemption. This allows the compiler to refer to
>  // symbols directly using relative references rather than via the GOT, which
>  // contains absolute symbol addresses that are subject to runtime relocation.
> +// The LTO linker will not emit GOT based relocations anyway, so there is no
> +// need to set the pragma in that case (and doing so will cause issues of its
> +// own)
>  //
>  #pragma GCC visibility push (hidden)
>  #endif
> 
> and I can drop this patch.