diff mbox series

LoongArch: Do not add DF_STATIC_TLS for TLS LE

Message ID 20231228145802.74719-1-ishitatsuyuki@gmail.com
State Accepted
Headers
Series LoongArch: Do not add DF_STATIC_TLS for TLS LE |

Checks

Context Check Description
snail/binutils-gdb-check success Github commit url

Commit Message

Tatsuyuki Ishi Dec. 28, 2023, 2:58 p.m. UTC 
  TLS LE is exclusively for executables, while DF_STATIC_TLS is for DLLs.
DF_STATIC_TLS should only be set for TLS IE (and when it's DLL), not LE.
---
 bfd/elfnn-loongarch.c | 2 --
 1 file changed, 2 deletions(-)

Comments

Tatsuyuki Ishi Jan. 22, 2024, 5:45 a.m. UTC | #1 
> On Dec 28, 2023, at 23:58, Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
> 
> TLS LE is exclusively for executables, while DF_STATIC_TLS is for DLLs.
> DF_STATIC_TLS should only be set for TLS IE (and when it's DLL), not LE.
> ---
> bfd/elfnn-loongarch.c | 2 --
> 1 file changed, 2 deletions(-)
> 
> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> index bd448cda453..64c34e99261 100644
> --- a/bfd/elfnn-loongarch.c
> +++ b/bfd/elfnn-loongarch.c
> @@ -862,8 +862,6 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> 	  if (!bfd_link_executable (info))
> 	    return false;
> 
> -	  info->flags |= DF_STATIC_TLS;
> -
> 	  if (!loongarch_elf_record_tls_and_got_reference (abfd, info, h,
> 							   r_symndx,
> 							   GOT_TLS_LE))
> -- 
> 2.40.1
> 
> 

Any interest in reviewing / merging this and the other two patches sent together?
The DF_STATIC_TLS change is pretty short, the formatting patch is trivial.
As for the last patch introducing a comment change, I’m not sure what Mengqing’s stance is, but my intention for the comment change is to provide a better context for the reader rather than comparing to a solution that is not currently implemented in the codebase.

Tatsuyuki.
Fangrui Song Jan. 22, 2024, 5:48 a.m. UTC | #2 
On Sun, Jan 21, 2024 at 9:45 PM Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
>
> > On Dec 28, 2023, at 23:58, Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
> >
> > TLS LE is exclusively for executables, while DF_STATIC_TLS is for DLLs.
> > DF_STATIC_TLS should only be set for TLS IE (and when it's DLL), not LE.
> > ---
> > bfd/elfnn-loongarch.c | 2 --
> > 1 file changed, 2 deletions(-)
> >
> > diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> > index bd448cda453..64c34e99261 100644
> > --- a/bfd/elfnn-loongarch.c
> > +++ b/bfd/elfnn-loongarch.c
> > @@ -862,8 +862,6 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> >         if (!bfd_link_executable (info))
> >           return false;
> >
> > -       info->flags |= DF_STATIC_TLS;
> > -
> >         if (!loongarch_elf_record_tls_and_got_reference (abfd, info, h,
> >                                                          r_symndx,
> >                                                          GOT_TLS_LE))
> > --
> > 2.40.1
> >
> >
>
> Any interest in reviewing / merging this and the other two patches sent together?
> The DF_STATIC_TLS change is pretty short, the formatting patch is trivial.
> As for the last patch introducing a comment change, I’m not sure what Mengqing’s stance is, but my intention for the comment change is to provide a better context for the reader rather than comparing to a solution that is not currently implemented in the codebase.
>
> Tatsuyuki.

I think this is correct and should be applied.
Xi Ruoyao Jan. 22, 2024, 6:51 a.m. UTC | #3 
On Sun, 2024-01-21 at 21:48 -0800, Fangrui Song wrote:
> On Sun, Jan 21, 2024 at 9:45 PM Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
> > 
> > > On Dec 28, 2023, at 23:58, Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
> > > 
> > > TLS LE is exclusively for executables, while DF_STATIC_TLS is for DLLs.
> > > DF_STATIC_TLS should only be set for TLS IE (and when it's DLL), not LE.
> > > ---
> > > bfd/elfnn-loongarch.c | 2 --
> > > 1 file changed, 2 deletions(-)
> > > 
> > > diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> > > index bd448cda453..64c34e99261 100644
> > > --- a/bfd/elfnn-loongarch.c
> > > +++ b/bfd/elfnn-loongarch.c
> > > @@ -862,8 +862,6 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> > >         if (!bfd_link_executable (info))
> > >           return false;
> > > 
> > > -       info->flags |= DF_STATIC_TLS;
> > > -
> > >         if (!loongarch_elf_record_tls_and_got_reference (abfd, info, h,
> > >                                                          r_symndx,
> > >                                                          GOT_TLS_LE))
> > > --
> > > 2.40.1
> > > 
> > > 
> > 
> > Any interest in reviewing / merging this and the other two patches sent together?
> > The DF_STATIC_TLS change is pretty short, the formatting patch is trivial.
> > As for the last patch introducing a comment change, I’m not sure what Mengqing’s stance is, but my intention for the comment change is to provide a better context for the reader rather than comparing to a solution that is not currently implemented in the codebase.
> > 
> > Tatsuyuki.
> 
> I think this is correct and should be applied.

Will this cause a breakage in practice?  If so we need to apply it for 2.42 branch too.
Fangrui Song Jan. 22, 2024, 7:11 a.m. UTC | #4 
On Sun, Jan 21, 2024 at 10:51 PM Xi Ruoyao <xry111@xry111.site> wrote:
>
> On Sun, 2024-01-21 at 21:48 -0800, Fangrui Song wrote:
> > On Sun, Jan 21, 2024 at 9:45 PM Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
> > >
> > > > On Dec 28, 2023, at 23:58, Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
> > > >
> > > > TLS LE is exclusively for executables, while DF_STATIC_TLS is for DLLs.
> > > > DF_STATIC_TLS should only be set for TLS IE (and when it's DLL), not LE.
> > > > ---
> > > > bfd/elfnn-loongarch.c | 2 --
> > > > 1 file changed, 2 deletions(-)
> > > >
> > > > diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> > > > index bd448cda453..64c34e99261 100644
> > > > --- a/bfd/elfnn-loongarch.c
> > > > +++ b/bfd/elfnn-loongarch.c
> > > > @@ -862,8 +862,6 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> > > >         if (!bfd_link_executable (info))
> > > >           return false;
> > > >
> > > > -       info->flags |= DF_STATIC_TLS;
> > > > -
> > > >         if (!loongarch_elf_record_tls_and_got_reference (abfd, info, h,
> > > >                                                          r_symndx,
> > > >                                                          GOT_TLS_LE))
> > > > --
> > > > 2.40.1
> > > >
> > > >
> > >
> > > Any interest in reviewing / merging this and the other two patches sent together?
> > > The DF_STATIC_TLS change is pretty short, the formatting patch is trivial.
> > > As for the last patch introducing a comment change, I’m not sure what Mengqing’s stance is, but my intention for the comment change is to provide a better context for the reader rather than comparing to a solution that is not currently implemented in the codebase.
> > >
> > > Tatsuyuki.
> >
> > I think this is correct and should be applied.
>
> Will this cause a breakage in practice?  If so we need to apply it for 2.42 branch too.

No. DF_STATIC_TLS is a not-so-useful dynamic tag. It helps a `readelf
-d` reader to know whether the DSO uses initial-exec.

For lld, I made a similar change in Nov 2021
(https://github.com/llvm/llvm-project/commit/6ca8fde226e907db13bc538e721af8724f0e92d0).
lld before Dec 2022 did not set DF_STATIC_TLS for non-x86
architectures (AArch64 and PPC).
I changed it to set DF_STATIC_TLS.
For both changes nobody has noticed anything.
mengqinggang Jan. 22, 2024, 8:27 a.m. UTC | #5 
Thank you very much, I will apply this patch soon.

For the last patch, I think we can add new comments while retaining the 
original ones.
The original comments describe another implementation method and can be 
compared with the existing ones.
If I wants to allocate GOT entry separately for IE/GD  in the future, I 
can refer to these comments.

在 2024/1/22 下午1:45, Tatsuyuki Ishi 写道:
>> On Dec 28, 2023, at 23:58, Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:
>>
>> TLS LE is exclusively for executables, while DF_STATIC_TLS is for DLLs.
>> DF_STATIC_TLS should only be set for TLS IE (and when it's DLL), not LE.
>> ---
>> bfd/elfnn-loongarch.c | 2 --
>> 1 file changed, 2 deletions(-)
>>
>> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
>> index bd448cda453..64c34e99261 100644
>> --- a/bfd/elfnn-loongarch.c
>> +++ b/bfd/elfnn-loongarch.c
>> @@ -862,8 +862,6 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
>> 	  if (!bfd_link_executable (info))
>> 	    return false;
>>
>> -	  info->flags |= DF_STATIC_TLS;
>> -
>> 	  if (!loongarch_elf_record_tls_and_got_reference (abfd, info, h,
>> 							   r_symndx,
>> 							   GOT_TLS_LE))
>> -- 
>> 2.40.1
>>
>>
> Any interest in reviewing / merging this and the other two patches sent together?
> The DF_STATIC_TLS change is pretty short, the formatting patch is trivial.
> As for the last patch introducing a comment change, I’m not sure what Mengqing’s stance is, but my intention for the comment change is to provide a better context for the reader rather than comparing to a solution that is not currently implemented in the codebase.
>
> Tatsuyuki.
mengqinggang Jan. 22, 2024, 9:46 a.m. UTC | #6 
Thank you very much, it has been applied.


在 2023/12/28 下午10:58, Tatsuyuki Ishi 写道:
> TLS LE is exclusively for executables, while DF_STATIC_TLS is for DLLs.
> DF_STATIC_TLS should only be set for TLS IE (and when it's DLL), not LE.
> ---
>   bfd/elfnn-loongarch.c | 2 --
>   1 file changed, 2 deletions(-)
>
> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
> index bd448cda453..64c34e99261 100644
> --- a/bfd/elfnn-loongarch.c
> +++ b/bfd/elfnn-loongarch.c
> @@ -862,8 +862,6 @@ loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
>   	  if (!bfd_link_executable (info))
>   	    return false;
>   
> -	  info->flags |= DF_STATIC_TLS;
> -
>   	  if (!loongarch_elf_record_tls_and_got_reference (abfd, info, h,
>   							   r_symndx,
>   							   GOT_TLS_LE))
diff mbox series

Patch

diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
index bd448cda453..64c34e99261 100644
--- a/bfd/elfnn-loongarch.c
+++ b/bfd/elfnn-loongarch.c
@@ -862,8 +862,6 @@  loongarch_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
 	  if (!bfd_link_executable (info))
 	    return false;
 
-	  info->flags |= DF_STATIC_TLS;
-
 	  if (!loongarch_elf_record_tls_and_got_reference (abfd, info, h,
 							   r_symndx,
 							   GOT_TLS_LE))