Re: [PATCH v4 06/16] riscv: Add Ziccamoa, Ziccif, Ziccrse, and Za64rs to cpufeature and hwprobe

[Conor Dooley <conor.dooley@microchip.com>]

[-- Attachment #1: Type: text/plain, Size: 2577 bytes --]

On Thu, Jun 11, 2026 at 04:12:43PM -0400, Guodong Xu wrote:
> From: Andrew Jones <andrew.jones@oss.qualcomm.com>
> 
> Add Ziccamoa, Ziccif, and Za64rs to riscv_isa_ext[] so they can be
> parsed from devicetree/ACPI ISA strings. Ziccrse is already present
> in cpufeature; this patch only adds its hwprobe exposure.
> 
> Expose all four extensions via hwprobe through new bits in
> RISCV_HWPROBE_KEY_IMA_EXT_1 (RISCV_HWPROBE_EXT_ZICCAMOA, _ZICCIF,
> _ZICCRSE, _ZA64RS), so userspace can probe each of these
> RVA23U64-mandatory extensions individually.
> 
> Rationale for the validation dependencies added for Ziccamoa and Za64rs:
> 
> 1) Ziccamoa depends on Zaamo.  The RVA23 profile prose was updated
> post-ratification to spell out the Zaamo reference: commit
> 2b218613752d in riscv/riscv-profiles ("Improve description of
> Ziccamoa (#224)") reworded the rva23-profile.adoc (and other profiles
> that include Ziccamoa) text from "must support all atomics in A" to
> "must support all atomics in the Zaamo extension" [1].
> 
> 2) Za64rs depends on Zalrsc. The unprivileged ISA manual src/zars.adoc,
> integrated in commit ebe06adc22cd ("Integrate profiles as Volume III
> (#2771)"), defines Za64rs as: "The Za64rs extension requires that the
> reservation sets used by the instructions in the Zalrsc extension be
> contiguous, naturally aligned, and at most 64 bytes in size" [2].

I think I made the point on either an earlier version of this, or a
similar thread, that the point of the validate callback stuff is to make
sure that the kernel is correctly configured to use the extension in
question or an extension it depends on. It's not the kernel's job to
make sure that the firmware has not reported having an extension without
one that it depends on (at least it is not in devicetree land, and I can
only assume that ACPI is by and large the same.

ziccamoa and za64rs don't depend on kernel configuration and neither do
zaamo and zalrsc, so these validate callbacks should be removed.

Cheers,
Conor.

> +static int riscv_ext_zaamo_depends(const struct riscv_isa_ext_data *data,
> +				   const unsigned long *isa_bitmap)
> +{
> +	if (__riscv_isa_extension_available(isa_bitmap, RISCV_ISA_EXT_ZAAMO))
> +		return 0;
> +
> +	return -EPROBE_DEFER;
> +}
> +
> +static int riscv_ext_zalrsc_depends(const struct riscv_isa_ext_data *data,
> +				    const unsigned long *isa_bitmap)
> +{
> +	if (__riscv_isa_extension_available(isa_bitmap, RISCV_ISA_EXT_ZALRSC))
> +		return 0;
> +
> +	return -EPROBE_DEFER;
> +}

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]