Re: [PATCH v2 14/15] RISC-V: KVM: add support for FWFT SBI extension

[Deepak Gupta <debug@rivosinc.com>]

On Tue, Feb 11, 2025 at 11:31:28AM +0100, Clément Léger wrote:
>
>
>On 11/02/2025 06:43, Deepak Gupta wrote:
>>> +static int kvm_sbi_fwft_get(struct kvm_vcpu *vcpu, unsigned long
>>> feature,
>>> +                unsigned long *value)
>>> +{
>>> +    int ret;
>>> +    struct kvm_sbi_fwft_config *conf;
>>> +
>>> +    ret = kvm_fwft_get_feature(vcpu, feature, &conf);
>>> +    if (ret)
>>> +        return ret;
>>> +
>>> +    return conf->feature->get(vcpu, conf, value);
>>> +}
>>> +
>>> +static int kvm_sbi_ext_fwft_handler(struct kvm_vcpu *vcpu, struct
>>> kvm_run *run,
>>> +                    struct kvm_vcpu_sbi_return *retdata)
>>> +{
>>> +    int ret = 0;
>>> +    struct kvm_cpu_context *cp = &vcpu->arch.guest_context;
>>> +    unsigned long funcid = cp->a6;
>>> +
>>> +    switch (funcid) {
>>> +    case SBI_EXT_FWFT_SET:
>>> +        ret = kvm_sbi_fwft_set(vcpu, cp->a0, cp->a1, cp->a2);
>>> +        break;
>>> +    case SBI_EXT_FWFT_GET:
>>> +        ret = kvm_sbi_fwft_get(vcpu, cp->a0, &retdata->out_val);
>>> +        break;
>>> +    default:
>>> +        ret = SBI_ERR_NOT_SUPPORTED;
>>> +        break;
>>> +    }
>>> +
>>> +    retdata->err_val = ret;
>>> +
>>> +    return 0;
>>> +}
>>> +
>>> +static int kvm_sbi_ext_fwft_init(struct kvm_vcpu *vcpu)
>>> +{
>>> +    struct kvm_sbi_fwft *fwft = vcpu_to_fwft(vcpu);
>>> +    const struct kvm_sbi_fwft_feature *feature;
>>> +    struct kvm_sbi_fwft_config *conf;
>>> +    int i;
>>> +
>>> +    fwft->configs = kcalloc(ARRAY_SIZE(features), sizeof(struct
>>> kvm_sbi_fwft_config),
>>> +                GFP_KERNEL);
>> nit:
>>
>> I understand that in next patch you grow the static array`features`. But
>> in this patch
>> `ARRAY_SIZE(features)` evaluates to 0, thus kcalloc will be returning a
>> pointer
>> to some slab block (IIRC, kcalloc will not return NULL if size
>> eventually evals to 0)
>>
>> This probably won't result in some bad stuff. But still there is a
>> pointer in
>> fwft->configs which is pointing to some random stuff if `features` turns
>> out to be
>> empty.
>>
>> Let me know if I got that right or missing something.
>
>So I actually searched into the kmalloc code to see what hapopens with a
>zero size allocation and it actually return ZERO_SIZE_PTR:
>
>/*
> * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
> *
> * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
> *
> * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL
>can.
> * Both make kfree a no-op.
> */
>
>Which seems like it's not really random and will fault if accessed. I
>think that's enough for that commit (which will be bisectable if needed
>then).
>

Awesome. Thanks for looking into it.

>Clément