Re: [PATCH] elf: Disable some subtests of ifuncmain1, ifuncmain5 for !PIE

Re: [PATCH] elf: Disable some subtests of ifuncmain1, ifuncmain5 for !PIE

[Sunil Pandey]

[-- Attachment #1: Type: text/plain, Size: 2717 bytes --]

I would like to backport this patch to release branches(from 2.33 to 2.28)
Any comments or objections?

--Sunil


On Fri, Nov 4, 2022 at 11:27 AM Florian Weimer via Libc-alpha <
libc-alpha@sourceware.org> wrote:

> Tested on i686-linux-gnu, x86_64-linux-gnu (with glibc defaults on a
> toolchain which is not PIE-by-default).  Built with build-many-glibcs.py
> defaults (so GCC 12 and binutils 2.39).  The x86 failures are gone.
>
> ---
>  elf/ifuncmain1.c | 13 +++++++++++++
>  elf/ifuncmain5.c |  9 +++++++++
>  2 files changed, 22 insertions(+)
>
> diff --git a/elf/ifuncmain1.c b/elf/ifuncmain1.c
> index 747fc02648..6effce3d77 100644
> --- a/elf/ifuncmain1.c
> +++ b/elf/ifuncmain1.c
> @@ -19,7 +19,14 @@ typedef int (*foo_p) (void);
>  #endif
>
>  foo_p foo_ptr = foo;
> +
> +/* Address-significant access to protected symbols is not supported in
> +   position-dependent mode on several architectures because GCC
> +   generates relocations that assume that the address is local to the
> +   main program.  */
> +#ifdef __PIE__
>  foo_p foo_procted_ptr = foo_protected;
> +#endif
>
>  extern foo_p get_foo_p (void);
>  extern foo_p get_foo_hidden_p (void);
> @@ -37,12 +44,16 @@ main (void)
>    if ((*foo_ptr) () != -1)
>      abort ();
>
> +#ifdef __PIE__
>    if (foo_procted_ptr != foo_protected)
>      abort ();
> +#endif
>    if (foo_protected () != 0)
>      abort ();
> +#ifdef __PIE__
>    if ((*foo_procted_ptr) () != 0)
>      abort ();
> +#endif
>
>    p = get_foo_p ();
>    if (p != foo)
> @@ -55,8 +66,10 @@ main (void)
>      abort ();
>
>    p = get_foo_protected_p ();
> +#ifdef __PIE__
>    if (p != foo_protected)
>      abort ();
> +#endif
>    if (ret_foo_protected != 0 || (*p) () != ret_foo_protected)
>      abort ();
>
> diff --git a/elf/ifuncmain5.c b/elf/ifuncmain5.c
> index f398085cb4..6fda768fb6 100644
> --- a/elf/ifuncmain5.c
> +++ b/elf/ifuncmain5.c
> @@ -14,12 +14,19 @@ get_foo (void)
>    return foo;
>  }
>
> +
> +/* Address-significant access to protected symbols is not supported in
> +   position-dependent mode on several architectures because GCC
> +   generates relocations that assume that the address is local to the
> +   main program.  */
> +#ifdef __PIE__
>  foo_p
>  __attribute__ ((noinline))
>  get_foo_protected (void)
>  {
>    return foo_protected;
>  }
> +#endif
>
>  int
>  main (void)
> @@ -30,9 +37,11 @@ main (void)
>    if ((*p) () != -1)
>      abort ();
>
> +#ifdef __PIE__
>    p = get_foo_protected ();
>    if ((*p) () != 0)
>      abort ();
> +#endif
>
>    return 0;
>  }
>
> base-commit: 2ff48a4025515e93d722947a9eabb114f4a65b22
>
>

Re: [PATCH] elf: Disable some subtests of ifuncmain1, ifuncmain5 for !PIE

[H.J. Lu]

On Thu, May 9, 2024 at 2:25 PM Sunil Pandey <skpgkp2@gmail.com> wrote:
>
> I would like to backport this patch to release branches(from 2.33 to 2.28)
> Any comments or objections?

Sounds good to me.

Thanks.

> --Sunil
>
>
> On Fri, Nov 4, 2022 at 11:27 AM Florian Weimer via Libc-alpha <libc-alpha@sourceware.org> wrote:
>>
>> Tested on i686-linux-gnu, x86_64-linux-gnu (with glibc defaults on a
>> toolchain which is not PIE-by-default).  Built with build-many-glibcs.py
>> defaults (so GCC 12 and binutils 2.39).  The x86 failures are gone.
>>
>> ---
>>  elf/ifuncmain1.c | 13 +++++++++++++
>>  elf/ifuncmain5.c |  9 +++++++++
>>  2 files changed, 22 insertions(+)
>>
>> diff --git a/elf/ifuncmain1.c b/elf/ifuncmain1.c
>> index 747fc02648..6effce3d77 100644
>> --- a/elf/ifuncmain1.c
>> +++ b/elf/ifuncmain1.c
>> @@ -19,7 +19,14 @@ typedef int (*foo_p) (void);
>>  #endif
>>
>>  foo_p foo_ptr = foo;
>> +
>> +/* Address-significant access to protected symbols is not supported in
>> +   position-dependent mode on several architectures because GCC
>> +   generates relocations that assume that the address is local to the
>> +   main program.  */
>> +#ifdef __PIE__
>>  foo_p foo_procted_ptr = foo_protected;
>> +#endif
>>
>>  extern foo_p get_foo_p (void);
>>  extern foo_p get_foo_hidden_p (void);
>> @@ -37,12 +44,16 @@ main (void)
>>    if ((*foo_ptr) () != -1)
>>      abort ();
>>
>> +#ifdef __PIE__
>>    if (foo_procted_ptr != foo_protected)
>>      abort ();
>> +#endif
>>    if (foo_protected () != 0)
>>      abort ();
>> +#ifdef __PIE__
>>    if ((*foo_procted_ptr) () != 0)
>>      abort ();
>> +#endif
>>
>>    p = get_foo_p ();
>>    if (p != foo)
>> @@ -55,8 +66,10 @@ main (void)
>>      abort ();
>>
>>    p = get_foo_protected_p ();
>> +#ifdef __PIE__
>>    if (p != foo_protected)
>>      abort ();
>> +#endif
>>    if (ret_foo_protected != 0 || (*p) () != ret_foo_protected)
>>      abort ();
>>
>> diff --git a/elf/ifuncmain5.c b/elf/ifuncmain5.c
>> index f398085cb4..6fda768fb6 100644
>> --- a/elf/ifuncmain5.c
>> +++ b/elf/ifuncmain5.c
>> @@ -14,12 +14,19 @@ get_foo (void)
>>    return foo;
>>  }
>>
>> +
>> +/* Address-significant access to protected symbols is not supported in
>> +   position-dependent mode on several architectures because GCC
>> +   generates relocations that assume that the address is local to the
>> +   main program.  */
>> +#ifdef __PIE__
>>  foo_p
>>  __attribute__ ((noinline))
>>  get_foo_protected (void)
>>  {
>>    return foo_protected;
>>  }
>> +#endif
>>
>>  int
>>  main (void)
>> @@ -30,9 +37,11 @@ main (void)
>>    if ((*p) () != -1)
>>      abort ();
>>
>> +#ifdef __PIE__
>>    p = get_foo_protected ();
>>    if ((*p) () != 0)
>>      abort ();
>> +#endif
>>
>>    return 0;
>>  }
>>
>> base-commit: 2ff48a4025515e93d722947a9eabb114f4a65b22
>>
>
>



-- 
H.J.

Re: [PATCH] elf: Disable some subtests of ifuncmain1, ifuncmain5 for !PIE

[Florian Weimer]

* Sunil Pandey:

> I would like to backport this patch to release branches(from 2.33 to
> 2.28) Any comments or objections?

Seems reasonable.  I stopped at 2.34 recently because that was the
earliest release I needed to backport other things to, there's no
fundamental reason to backport this further back.

Thanks,
Florian

Re: [PATCH] elf: Disable some subtests of ifuncmain1, ifuncmain5 for !PIE

[Robbe Loes]

[-- Attachment #1: Type: text/plain, Size: 756 bytes --]

That's for real, that's for chose!

________________________________
Van: Florian Weimer <fweimer@redhat.com>
Verzonden: donderdag 9 mei 2024 23:30
Aan: Sunil Pandey <skpgkp2@gmail.com>
CC: Hongjiu Lu <hjl.tools@gmail.com>; Libc-stable Mailing List <libc-stable@sourceware.org>; GNU C Library <libc-alpha@sourceware.org>
Onderwerp: Re: [PATCH] elf: Disable some subtests of ifuncmain1, ifuncmain5 for !PIE

* Sunil Pandey:

> I would like to backport this patch to release branches(from 2.33 to
> 2.28) Any comments or objections?

Seems reasonable.  I stopped at 2.34 recently because that was the
earliest release I needed to backport other things to, there's no
fundamental reason to backport this further back.

Thanks,
Florian