[PATCH 04/18] KVM: arm64: pkvm: Remove struct cpu_sve_state

[Vladimir Murzin]

On 5/28/26 16:09, Mark Rutland wrote:
> On Wed, May 27, 2026 at 05:11:26PM +0100, Vladimir Murzin wrote:
>> On 5/27/26 17:02, Mark Rutland wrote:
>>> On Wed, May 27, 2026 at 12:58:47PM +0100, Vladimir Murzin wrote:
>>>> On 5/21/26 14:25, Mark Rutland wrote:
>>>>> -struct cpu_sve_state {
>>>>> -	__u64 zcr_el1;
>>>>> -
>>>>> -	/*
>>>>> -	 * Ordering is important since __sve_save_state/__sve_restore_state
>>>>> -	 * relies on it.
>>>>> -	 */
>>>>> -	__u32 fpsr;
>>>>> -	__u32 fpcr;
>>>>> -
>>>>> -	/* Must be SVE_VQ_BYTES (128 bit) aligned. */
>>>>> -	__u8 sve_regs[];
>>>> It seems that the requirement (driven by SVE ldr/str) is
>>>> satisfied with the new sve_regs pointing to the start of the
>>>> page.
>>>>
>>>> I'm not sure whether we want to keep the comment (or perhaps
>>>> enforce this with explicit checks) so that future refactoring
>>>> doesn't lead to time spent debugging alignment faults...
>>> AFAICT alignment has never been functionally necessary. The LDR (vector)
>>> and STR (vector) instructions only mandate alignment when the relevant
>>> SCTLR_ELx has SCTLR_ELx.A==1. For kernel and hyp code we configure
>>> SCTLR_ELx.A==0, so there's no alignment requirement.
>>>
>>> Per ARM DDI 0487 M.b, section C8.2.437 "LDR (vector)":
>>>
>>>   The load is performed as contiguous byte accesses, with no endian
>>>   conversion and no guarantee of single-copy atomicity larger than a
>>>   byte. However, *if alignment is checked*, then the base register must be
>>>   aligned to 16 bytes.
>>>
>>> Per ARM DDI 0487 M.b, section C8.2.777 "STR (vector)":
>>>
>>>   The store is performed as contiguous byte accesses, with no endian
>>>   conversion and no guarantee of single-copy atomicity larger than a
>>>   byte. However, *if alignment is checked*, then the base register must be
>>>   aligned to 16 bytes.
>>>
>>> ... and in both cases the pseudocode shows that AlignmentEnforced()
>>> depends on the value of SCTLR_ELx.A.
>>>
>>> Given that, I don't think we need the comment.
>>>
>> Thanks for explanation! Maybe worth mentioning in commit message that it never was
>> a requirement? 
> I've added the following to the commi message:
> 
>   I've dropped the comment regarding buffer alignment as AFAICT this was
>   never necessary. The LDR/STR (vector) instructions only require this
>   alignment when SCTLR_ELx.A==1, which is not the case for the kernel or
>   hyp code. Nothing else depends on the alignment.
> 

Brilliant!

Thanks
Vladimir

> Mark.
>