[PATCH v4 1/1] riscv: Add RVV memset via multiarch/IFUNC

[Peter Bergner]

On 12/4/25 7:40 AM, Yao Zihong wrote:
> Checked the intended semantics of hwprobe, AT_HWCAP and prctl
> with Andy Chiu on the kernel Bugzilla [1]. The kernel-side clarification is:
> - hwprobe reports per-CPU presence of RVV,  
> - while AT_HWCAP and prctl(PR_RISCV_V_GET_CONTROL) report per-process
>   availability.
> 
> So hwprobe is intentionally not equivalent to AT_HWCAP, and it is expected
> that hwprobe will continue to report RVV even when a process has RVV disabled
> via prctl().
[snip]
> [1] https://bugzilla.kernel.org/show_bug.cgi?id=220795

Yes, we cannot change hwprobe's current behavior, but I hope we can add
a new "modifier" hwprobe key to tell the kernel to return info more
similar to how AT_HWCAP* was supposed to work.  





> Given this clarification, and earlier discussion agreed that the
> prctl-disabled-RVV corner case should not block this minimal pilot for RVV
> memset, would proceeding with v4 as-is(presence-only resolver) be acceptable?

Palmer, Jeff, myself and others have come to consensus that yes, we can
move forward with this.  I have posted a review of the v4 patch.

Peter