Re: [PATCH v2 3/6] iommu/amd: Initial support for AMD IOMMU v2 page table

[Vasant Hegde <vasant.hegde@amd.com>]

Hi Robin,


On 7/15/2022 5:25 PM, Robin Murphy wrote:
> On 2022-07-13 06:30, Vasant Hegde wrote:
> [...]
>> +static struct io_pgtable *v2_alloc_pgtable(struct io_pgtable_cfg *cfg, void *cookie)
>> +{
>> +    struct amd_io_pgtable *pgtable = io_pgtable_cfg_to_data(cfg);
>> +    struct protection_domain *pdom = (struct protection_domain *)cookie;
>> +    int ret;
>> +
>> +    pgtable->pgd = alloc_pgtable_page();
>> +    if (!pgtable->pgd)
>> +        return NULL;
>> +
>> +    ret = amd_iommu_domain_set_gcr3(&pdom->domain, 0, iommu_virt_to_phys(pgtable->pgd));
>> +    if (ret)
>> +        goto err_free_pgd;
>> +
>> +    pgtable->iop.ops.map          = iommu_v2_map_page;
>> +    pgtable->iop.ops.unmap        = iommu_v2_unmap_page;
> 
> It would be good to use this opportunity to convert AMD over to map_pages/unmap_pages - not only for the future goal of removing map/unmap (it's not worth maintaining two interfaces where one is a trivial subset of the other), but also since it should be appreciably more efficient for you in practice, especially for v2 with the more limited set of basic page sizes.
> 

Makes sense. I will work on this and post v3 soon.

Thanks
-Vasant