[PATCH v6 03/17] riscv: vector: Use vlenb from DT

[Conor Dooley]

On Thu, May 16, 2024 at 10:00:12PM +0800, Andy Chiu wrote:
> On Sat, May 4, 2024 at 2:21 AM Charlie Jenkins <charlie at rivosinc.com> wrote:

> > +               if (elf_hwcap & COMPAT_HWCAP_ISA_V && has_riscv_homogeneous_vlenb() < 0) {
> > +                       pr_warn("Unsupported heterogeneous vlen detected, vector extension disabled.\
> > +                       elf_hwcap &= ~COMPAT_HWCAP_ISA_V;
> > +               }
> 
> We only touch COMPAT_HWCAP_ISA_V and the failed case only turns off the
> rectified V. So here we have nothing to do with the Xtheadvector.

There's nothing t-head related in the tree at this point, so doing
anything with it would cause build issues.

> However, I am still confused because I think Xtheadvector would also
> need to call into this check, so as to setup vlenb.


> Apart from that, it seems like some vendor stating Xtheadvector is
> actually vector-0.7.

The T-Head implementation is 0.7.x, but I am not really sure what you
mean by this comment.

> Please correct me if I speak anything wrong. One
> thing I noticed is that Xtheadvector wouldn't trap on reading
> th.vlenb but vector-0.7 would. If that is the case, should we require
> Xtheadvector to specify `riscv,vlenb` on the device tree?

In the world of Linux, "vector-0.7" isn't a thing. There's only 1.0, and
after this patchset, "xtheadvector". My understanding, from discussion
on earlier versions of this series the trap is actually accessing
th.vlenb register, despite the documentation stating that it is
unprivileged:
https://github.com/T-head-Semi/thead-extension-spec/blob/master/xtheadvector.adoc
I assume Charlie tried it but was trapping, as v1 had a comment:
+		 * Although xtheadvector states that th.vlenb exists and
+		 * overlaps with the vector 1.0 extension overlaps, an illegal
+		 * instruction is raised if read. These systems all currently
+		 * have a fixed vector length of 128, so hardcode that value.

Cheers,
Conor.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20240516/a540d643/attachment.sig>