Re: [PART2 RFC v1 5/9] iommu/amd: Introduce amd_iommu_update_ga()

[Radim KrÄmÃÅ]

2016-04-08 07:49-0500, Suravee Suthikulpanit:
> From: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx>
> 
> This patch introduces a new IOMMU interface, amd_iommu_update_ga(),
> which allows KVM (SVM) to update existing posted interrupt IOMMU IRTE when
> load/unload vcpu.
> 
> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx>
> ---
> diff --git a/drivers/iommu/amd_iommu.c b/drivers/iommu/amd_iommu.c
> @@ -4330,4 +4330,74 @@ int amd_iommu_create_irq_domain(struct amd_iommu *iommu)
> +int amd_iommu_update_ga(u32 vcpu_id, u32 cpu, u32 ga_tag,

It'd be nicer to generate the tag on SVM side and pass it whole -- IOMMU
doesn't have to care how hypervisors use the tag.

> +			u64 base, bool is_run)
> +{
> +	unsigned long flags;
> +	struct amd_iommu *iommu;
> +
> +	if (amd_iommu_guest_ir < AMD_IOMMU_GUEST_IR_GA)
> +		return 0;
> +
> +	for_each_iommu(iommu) {
> +		struct amd_ir_data *ir_data;
> +
> +		spin_lock_irqsave(&iommu->ga_hash_lock, flags);
> +
> +		hash_for_each_possible(iommu->ga_hash, ir_data, hnode,
> +				       AMD_IOMMU_GATAG(ga_tag, vcpu_id)) {

All tags can map into the same bucket.  Code below doesn't check that
the ir_data belongs to the tag and will modify unrelated IRTEs.

Have you considered a per-VCPU list of IRTEs on the SVM side?

> +			struct iommu_dev_data *dev_data;
> +			if (!ir_data)

(ir_data can't be NULL.)

> +				break;
> +
> +			dev_data = search_dev_data(ir_data->irq_2_irte.devid);
> +
> +			if (!dev_data || !dev_data->guest_mode)
> +				continue;

(guest_mode can be also read from the irte.)

> +			set_irte_ga(iommu, ir_data->irq_2_irte.devid,
> +				    base, cpu, is_run);

set_irte_ga() is pretty expensive -- do we need to invalidate the irt
when changing cpu and is_run?

2.2.5.2 Interrupt Virtualization Tables with Guest Virtual APIC Enabled,
point 9, bullet 5 says that IRTE is read from memory before considering
IsRun, GATag and Destination, which makes me think that avoiding races
can be faster in the common case.