Re: [PATCH v2 05/10] x86/HVM/SVM: Add AVIC initialization code

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

> +
> +#define AVIC_DOORBELL           0xc001011b

MSR_AVIC_DOORBELL (and it should probably go to include/asm-x86/msr-index.h)

> +
> +#define AVIC_HPA_SHIFT  12

Is there a reason not to use regular PAGE_SHIFT?

> +#define AVIC_HPA_MASK           (((1ULL << 40) - 1) << AVIC_HPA_SHIFT)
> +#define AVIC_VAPIC_BAR_MASK     AVIC_HPA_MASK
> +
> +/*
> + * Note:
> + * Currently, svm-avic mode is not supported with nested virtualization.
> + * Therefore, it is not yet currently enabled by default. Once the support
> + * is in-place, this should be enabled by default.
> + */
> +bool svm_avic = 0;
> +boolean_param("svm-avic", svm_avic);
> +
> +static struct avic_phy_apic_id_ent *
> +avic_get_phy_apic_id_ent(const struct vcpu *v, unsigned int index)

Something like phys_apicid might be more descriptive.

> +{
> +    struct avic_phy_apic_id_ent *avic_phy_apic_id_table;
> +    struct svm_domain *d = &v->domain->arch.hvm_domain.svm;
> +
> +    if ( !d->avic_phy_apic_id_table_mfn )
> +        return NULL;
> +
> +    /*
> +    * Note: APIC ID = 0xff is used for broadcast.
> +    *       APIC ID > 0xff is reserved.
> +    */
> +    if ( index >= 0xff )
> +        return NULL;
> +
> +    avic_phy_apic_id_table = mfn_to_virt(d->avic_phy_apic_id_table_mfn);
> +
> +    return &avic_phy_apic_id_table[index];
> +}


> +}
> +
> +static inline u32 *
> +avic_get_bk_page_entry(const struct vcpu *v, u32 offset)
> +{
> +    const struct vlapic *vlapic = vcpu_vlapic(v);
> +    char *tmp;
> +
> +    if ( !vlapic || !vlapic->regs_page )

Should this be changed to an ASSERT? I think this is only called from
places that already tested this or from places that you wouldn't get to
unless these both were true (like VMEXIT handler).

-boris


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel