Re: [PATCH v9 04/11] iommu: Add sva iommu_domain support

[Baolu Lu]

On 2022/6/29 09:54, Tian, Kevin wrote:
> > From: Baolu Lu <baolu.lu@xxxxxxxxxxxxxxx>
> > Sent: Tuesday, June 28, 2022 7:34 PM
> >
> > On 2022/6/28 16:50, Tian, Kevin wrote:
> > > > From: Baolu Lu<baolu.lu@xxxxxxxxxxxxxxx>
> > > > Sent: Tuesday, June 28, 2022 1:41 PM
> > > > > >     struct iommu_domain {
> > > > > >     	unsigned type;
> > > > > >     	const struct iommu_domain_ops *ops;
> > > > > >     	unsigned long pgsize_bitmap;	/* Bitmap of page sizes in use */
> > > > > > -	iommu_fault_handler_t handler;
> > > > > > -	void *handler_token;
> > > > > >     	struct iommu_domain_geometry geometry;
> > > > > >     	struct iommu_dma_cookie *iova_cookie;
> > > > > > +	union {
> > > > > > +		struct {	/* IOMMU_DOMAIN_DMA */
> > > > > > +			iommu_fault_handler_t handler;
> > > > > > +			void *handler_token;
> > > > > > +		};
> > > > > why is it DMA domain specific? What about unmanaged
> > > > > domain? Unrecoverable fault can happen on any type
> > > > > including SVA. Hence I think above should be domain type
> > > > > agnostic.
> > > > The report_iommu_fault() should be replaced by the new
> > > > iommu_report_device_fault(). Jean has already started this work.
> > > >
> > > > https://lore.kernel.org/linux-iommu/Yo4Nw9QyllT1RZbd@myrica/
> > > >
> > > > Currently this is only for DMA domains, hence Robin suggested to make it
> > > > exclude with the SVA domain things.
> > > >
> > > > https://lore.kernel.org/linux-iommu/f3170016-4d7f-e78e-db48-
> > > > 68305f683349@xxxxxxx/
> > > Then it's worthy a comment that those two fields are for
> > > some legacy fault reporting stuff and DMA type only.
> >
> > The iommu_fault and SVA fields are exclusive. The former is used for
> > unrecoverable DMA remapping faults, while the latter is only interested
> > in the recoverable page faults.
> >
> > I will update the commit message with above explanation. Does this work
> > for you?
>
> Not exactly. Your earlier explanation is about old vs. new API thus
> leaving the existing fault handler with current only user is fine.
>
> but this is not related to unrecoverable vs. recoverable. As I said
> unrecoverable could happen on all domain types. Tying it to
> DMA-only doesn't make sense and I think in the end the new
> iommu_report_device_fault() will need support both. Is it not the
> case?

You are right.

The report_iommu_fault() should be replaced by the new
iommu_report_device_fault(). Leave the existing fault handler with the
existing users and the newly added SVA members should exclude it.

Best regards,
baolu