Re: [RFC PATCH 06/12] perf: Support extension of sample_regs

[Liang, Kan]

On 2025-06-18 10:45 a.m., Peter Zijlstra wrote:
> On Wed, Jun 18, 2025 at 09:52:12AM -0400, Liang, Kan wrote:
> 
>> I didn't know we have the alignment requirement for the output.
> 
> Perf buffer is in u64 units.

Sure.

> 
>> If so,
>>
>> PERF_SAMPLE_SIMD_REGS := {
>> 	u64 vectors_mask;
>> 	u64 pred_mask;
>> 	u64 vector_length:16,
>> 	    pred_length:16,
>> 	    reserved:32;
>> 	u64 data[];
>> }
> 
> I really don't think we need this complication; XSAVE is a real pain in
> the arse, but I think we have enough bits in XSTATE_BV to tell what is
> what.

I just realized that XSAVE takes the vector as a whole.
It's impossible that XMM0 can be retrieved but XMM1 fails via XSAVE.
I will use the suggested output format.

PERF_SAMPLE_SIMD_REGS := {
	u16 nr_vectors;
	u16 vector_length;
	u16 nr_pred;
	u16 pred_length;
	u64 data[];
}

Thanks,
Kan