[PATCH v8 13/14] RISC-V: KVM: add support for FWFT SBI extension

[Clément Léger]

On 23/05/2025 20:02, Atish Patra wrote:
> On 5/23/25 9:27 AM, Radim KrÄmáŠwrote:
>> 2025-05-23T17:29:49+02:00, Clément Léger <cleger at rivosinc.com>:
>>> On 23/05/2025 15:05, Radim Krčmář wrote:
>>>> 2025-05-23T12:19:30+02:00, Clément Léger <cleger at rivosinc.com>:
>>>>> +++ b/arch/riscv/kvm/vcpu_sbi_fwft.c
>>>>> +static const enum sbi_fwft_feature_t kvm_fwft_defined_features[] = {
>>>>> +    SBI_FWFT_MISALIGNED_EXC_DELEG,
>>>>> +    SBI_FWFT_LANDING_PAD,
>>>>> +    SBI_FWFT_SHADOW_STACK,
>>>>> +    SBI_FWFT_DOUBLE_TRAP,
>>>>> +    SBI_FWFT_PTE_AD_HW_UPDATING,
>>>>> +    SBI_FWFT_POINTER_MASKING_PMLEN,
>>>>> +};
>>>>
>>>> How will userspace control which subset of these features is allowed in
>>>> the guest?
>>>>
>>>> (We can reuse the KVM SBI extension interface if we don't want to add a
>>>>   FWFT specific ONE_REG.)
>>>
>>> Hi Radim,
>>>
>>> I didn't looked at that part. But most likely using the kvm one reg
>>> interface seems ok like what is done for STA ? We could have per feature
>>> override with one reg per feature.
>>
>> Sounds fine.
>>
> 
> Yeah. We can have a follow up series for SBI FWFT state that allows user
> space to toggle each state individually.
> 
>>> Is this something blocking though ? We'd like to merge FWFT once SBI 3.0
>>> is ratified so that would be nice not delaying it too much. I'll take a
>>> look at it to see if it isn't too long to implement.
>>
>> Not blocking, but I would at least default FWFT to disabled, because
>> current userspace cannot handle [14/14].  (Well... save/restore was
>> probably broken even before, but let's try to not make it worse. :])
>>
> 
> User space can not enable or disable misaligned access delegation as
> there is no interface for now rightly pointed by you. I guess supporting
> that would be quicker than fixing the broader guest save/restore
> anyways. Isn't it ?
> 
> We can have the patches ready for the next MW for FWFT one reg interface.

Yeah sure I'll work on that in the meantime.

> 
>> Thanks.
>>
>> _______________________________________________
>> linux-riscv mailing list
>> linux-riscv at lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-riscv
>