[PATCH] RISC-V: support for vector register accesses via ptrace() in RISC-V Linux native

[Andy Chiu]

On Fri, Aug 11, 2023 at 5:21 AM Maciej W. Rozycki <macro at orcam.me.uk> wrote:
>
> On Fri, 11 Aug 2023, Andy Chiu wrote:
>
> > > >  No, how do you expect it to work with a core dump (that can be examined
> > > > on a different system, or with a cross-debugger)?  You need to change the
> > > > API I'm afraid; it's unusable anyway.  It's a pity the toolchain community
> > > > wasn't consulted if you weren't sure how to design the interface.  Better
> > > > yet it would have been to implement the GDB side before the kernel part
> > > > has been committed.
> >
> > I just took some look into the code and here is what I came up with.
> > Actually, you know VLENB in a core dump file. The size of
> > NT_RISCV_VECTOR in a core dump file just equals sizeof(struct
> > __riscv_v_ext_state), which is 40B, plus VLENB * 32. So, the debugger
> > can actually calculate VLENB and resolve placement of V registers by
> > subtracting 40 from the size of NT_RISCV_VECTOR in a core dump file.
>
>  Fair enough, I didn't dive into Linux code deeply enough to figure out
> that the size of an NT_RISCV_VECTOR core file note is indeed dynamically
> calculated.  Most notes are of a fixed size, but we also have generic
> support for variable-size ones in GDB, so handling this case should be
> reasonably straightforward.
>
>  OTOH VLENB is a program-visible register, so I think it will best be
> provided explicitly regardless rather than having to be reconstructed from
> the size of the note; I would find that awkward.

Agreed.

>
>  NB I have been a bit concerned about the unusually huge allocation size
> of 256KiB+ for the register buffer required for ptrace(2), but I guess
> we'll have to live with it, because any solution that makes it dynamic
> would also complicate the interface.  At least we won't waste filesystem
> space for any extraneous allocation in core dumps.

It is possible to mitigate this consideration with the proposed
solution[1], by calling the ptrace twice. First we make a ptrace call
to obtain VLENB in struct __riscv_v_ext_state by setting the argument
iov.len = sizeof(struct __riscv_v_ext_state). Then, we can allocate a
buffer based on the result of the previous ptrace to get the full
Vector registers dump.

>
>   Maciej

[1]: https://sourceware.org/pipermail/gdb-patches/2023-August/201507.html

Andy