Re: [PATCH v2 3/5] iommu/vt-d: Extend dmar_domain to support nested domain

[Baolu Lu <baolu.lu@linux.intel.com>]

On 3/20/23 9:54 PM, Jason Gunthorpe wrote:
> On Thu, Mar 09, 2023 at 12:22:05AM -0800, Yi Liu wrote:
>> From: Lu Baolu <baolu.lu@linux.intel.com>
>>
>> The nested domain fields are exclusive to those that used for a DMA
>> remapping domain. Use union to avoid memory waste.
>>
>> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
>> Signed-off-by: Yi Liu <yi.l.liu@intel.com>
>> ---
>>   drivers/iommu/intel/iommu.h | 35 +++++++++++++++++++++++++++++------
>>   1 file changed, 29 insertions(+), 6 deletions(-)
> 
> Using unions like this often devolves into a mess.
> 
> You'd be better to have more structures
> 
> struct intel_iommu_domain {
>     struct iommu_domain domain;
>     [general fields about attachment]
> };
> 
> struct intel_iopte_domain {
>      struct intel_iommu_domain domain;
>      [stuff describing the io page table data, pgd, format, etc]
> };
> 
> strut intel_s1_domain {
>       struct intel_iommu_domain domain;
>       struct dmar_domain *s2_domain;
>       /* user page table pointer (in GPA) */
>       unsigned long s1_pgtbl;
>       /* page table attributes */
>       struct iommu_hwpt_intel_vtd s1_cfg;
> };
> static_assert(offset_of(struct intel_s1_domain, domain.domain) ==
>                offset_of(struct intel_iommu_domain, domain));
> 
> The per-domain ops allow to make this work sensibly

Yes. This will make the data structures clearer.

However, this will lead to significant code changes. I think it would be
more appropriate to put it in a separate refactoring series later.

Best regards,
baolu