riscv: align stack before calling _dl_init [BZ #28703]

  Align the stack pointer to 128 bits during the call to _dl_init() as
specified by the RISC-V ABI [1]. This fixes the elf/tst-align2 test.

Fixes bug 28703.

[1] https://github.com/riscv-non-isa/riscv-elf-psabi-doc
---
 sysdeps/riscv/dl-machine.h | 6 ++++++
 1 file changed, 6 insertions(+)
  

Even though sp should be sufficiently aligned upon program entry, the
_dl_skip_args jig can misalign the stack.  So, LGTM.

On Wed, Dec 15, 2021 at 3:12 PM Aurelien Jarno <aurelien@aurel32.net> wrote:
>
> Align the stack pointer to 128 bits during the call to _dl_init() as
> specified by the RISC-V ABI [1]. This fixes the elf/tst-align2 test.
>
> Fixes bug 28703.
>
> [1] https://github.com/riscv-non-isa/riscv-elf-psabi-doc
> ---
>  sysdeps/riscv/dl-machine.h | 6 ++++++
>  1 file changed, 6 insertions(+)
>
> diff --git a/sysdeps/riscv/dl-machine.h b/sysdeps/riscv/dl-machine.h
> index ce2b3c3875..f457a41c70 100644
> --- a/sysdeps/riscv/dl-machine.h
> +++ b/sysdeps/riscv/dl-machine.h
> @@ -125,8 +125,14 @@ elf_machine_dynamic (void)
>         sll a3, a1, " STRINGXP (PTRLOG) "\n\
>         add a3, a3, a2\n\
>         add a3, a3, " STRINGXP (SZREG) "\n\
> +       # Stash the stack pointer in s1.\n\
> +       mv s1, sp\n\
> +       # Align stack to 128 bits for the _dl_init call.\n\
> +       andi sp, sp,-16\n\
>         # Call the function to run the initializers.\n\
>         jal _dl_init\n\
> +       # Restore the stack pointer for _start.\n\
> +       mv sp, s1\n\
>         # Pass our finalizer function to _start.\n\
>         lla a0, _dl_fini\n\
>         # Jump to the user entry point.\n\
> --
> 2.33.0
>
  

* Aurelien Jarno:

> +	# Align stack to 128 bits for the _dl_init call.\n\
> +	andi sp, sp,-16\n\
>  	# Call the function to run the initializers.\n\
>  	jal _dl_init\n\

How is that extra alignment observable to user code?

Thanks,
Florian
  

On 2021-12-16 07:51, Florian Weimer wrote:
> * Aurelien Jarno:
> 
> > +	# Align stack to 128 bits for the _dl_init call.\n\
> > +	andi sp, sp,-16\n\
> >  	# Call the function to run the initializers.\n\
> >  	jal _dl_init\n\
> 
> How is that extra alignment observable to user code?

This extra alignment is visible to the constructors, called from
_dl_init. This is how this is tested in elf/tst-align2.c, however the
issue wasn't detected due to BZ #27901.
  

On Dez 16 2021, Aurelien Jarno wrote:

> diff --git a/sysdeps/riscv/dl-machine.h b/sysdeps/riscv/dl-machine.h
> index ce2b3c3875..f457a41c70 100644
> --- a/sysdeps/riscv/dl-machine.h
> +++ b/sysdeps/riscv/dl-machine.h
> @@ -125,8 +125,14 @@ elf_machine_dynamic (void)
>  	sll a3, a1, " STRINGXP (PTRLOG) "\n\
>  	add a3, a3, a2\n\
>  	add a3, a3, " STRINGXP (SZREG) "\n\
> +	# Stash the stack pointer in s1.\n\
> +	mv s1, sp\n\
> +	# Align stack to 128 bits for the _dl_init call.\n\
> +	andi sp, sp,-16\n\
>  	# Call the function to run the initializers.\n\
>  	jal _dl_init\n\
> +	# Restore the stack pointer for _start.\n\
> +	mv sp, s1\n\
>  	# Pass our finalizer function to _start.\n\
>  	lla a0, _dl_fini\n\
>  	# Jump to the user entry point.\n\

Shouldn't the adjustment already be made after the _dl_skip_args fixup,
permanently?
  

On 2021-12-16 09:48, Andreas Schwab wrote:
> On Dez 16 2021, Aurelien Jarno wrote:
> 
> > diff --git a/sysdeps/riscv/dl-machine.h b/sysdeps/riscv/dl-machine.h
> > index ce2b3c3875..f457a41c70 100644
> > --- a/sysdeps/riscv/dl-machine.h
> > +++ b/sysdeps/riscv/dl-machine.h
> > @@ -125,8 +125,14 @@ elf_machine_dynamic (void)
> >  	sll a3, a1, " STRINGXP (PTRLOG) "\n\
> >  	add a3, a3, a2\n\
> >  	add a3, a3, " STRINGXP (SZREG) "\n\
> > +	# Stash the stack pointer in s1.\n\
> > +	mv s1, sp\n\
> > +	# Align stack to 128 bits for the _dl_init call.\n\
> > +	andi sp, sp,-16\n\
> >  	# Call the function to run the initializers.\n\
> >  	jal _dl_init\n\
> > +	# Restore the stack pointer for _start.\n\
> > +	mv sp, s1\n\
> >  	# Pass our finalizer function to _start.\n\
> >  	lla a0, _dl_fini\n\
> >  	# Jump to the user entry point.\n\
> 
> Shouldn't the adjustment already be made after the _dl_skip_args fixup,
> permanently?

I have basically followed what is done on i386, mips or x86_64. I can
give a try to keep the alignment permanent and see if things still work.
  

* Aurelien Jarno:

> On 2021-12-16 07:51, Florian Weimer wrote:
>> * Aurelien Jarno:
>> 
>> > +	# Align stack to 128 bits for the _dl_init call.\n\
>> > +	andi sp, sp,-16\n\
>> >  	# Call the function to run the initializers.\n\
>> >  	jal _dl_init\n\
>> 
>> How is that extra alignment observable to user code?
>
> This extra alignment is visible to the constructors, called from
> _dl_init. This is how this is tested in elf/tst-align2.c, however the
> issue wasn't detected due to BZ #27901.

Ah, sorry, I misread “128 bits” as “128 bytes”.  16-byte alignment will
be preserved by the compiler.

Thanks,
Florian
  

On Dez 16 2021, Aurelien Jarno wrote:

> On 2021-12-16 09:48, Andreas Schwab wrote:
>> On Dez 16 2021, Aurelien Jarno wrote:
>> 
>> > diff --git a/sysdeps/riscv/dl-machine.h b/sysdeps/riscv/dl-machine.h
>> > index ce2b3c3875..f457a41c70 100644
>> > --- a/sysdeps/riscv/dl-machine.h
>> > +++ b/sysdeps/riscv/dl-machine.h
>> > @@ -125,8 +125,14 @@ elf_machine_dynamic (void)
>> >  	sll a3, a1, " STRINGXP (PTRLOG) "\n\
>> >  	add a3, a3, a2\n\
>> >  	add a3, a3, " STRINGXP (SZREG) "\n\
>> > +	# Stash the stack pointer in s1.\n\
>> > +	mv s1, sp\n\
>> > +	# Align stack to 128 bits for the _dl_init call.\n\
>> > +	andi sp, sp,-16\n\
>> >  	# Call the function to run the initializers.\n\
>> >  	jal _dl_init\n\
>> > +	# Restore the stack pointer for _start.\n\
>> > +	mv sp, s1\n\
>> >  	# Pass our finalizer function to _start.\n\
>> >  	lla a0, _dl_fini\n\
>> >  	# Jump to the user entry point.\n\
>> 
>> Shouldn't the adjustment already be made after the _dl_skip_args fixup,
>> permanently?
>
> I have basically followed what is done on i386, mips or x86_64. I can
> give a try to keep the alignment permanent and see if things still work.

It looks like your version is ok, since _start is doing its own
alignment (and it needs sp to point to argc).
  

On 2021-12-16 11:42, Florian Weimer via Libc-alpha wrote:
> * Aurelien Jarno:
> 
> > On 2021-12-16 07:51, Florian Weimer wrote:
> >> * Aurelien Jarno:
> >> 
> >> > +	# Align stack to 128 bits for the _dl_init call.\n\
> >> > +	andi sp, sp,-16\n\
> >> >  	# Call the function to run the initializers.\n\
> >> >  	jal _dl_init\n\
> >> 
> >> How is that extra alignment observable to user code?
> >
> > This extra alignment is visible to the constructors, called from
> > _dl_init. This is how this is tested in elf/tst-align2.c, however the
> > issue wasn't detected due to BZ #27901.
> 
> Ah, sorry, I misread “128 bits” as “128 bytes”.  16-byte alignment will
> be preserved by the compiler.

I agree that's a bit unusual to use bits instead of bytes there, but I
used the same terminology as in the RISC-V ABI.