Re: [PATCH v4 01/17] iommu: Add hwpt_type with user_data for domain_alloc_user op

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, Oct 18, 2023 at 01:37:20PM -0300, Jason Gunthorpe wrote:
> On Mon, Oct 16, 2023 at 11:17:56AM -0700, Nicolin Chen wrote:
> > On Mon, Oct 16, 2023 at 08:54:07AM -0300, Jason Gunthorpe wrote:
> > > On Mon, Oct 16, 2023 at 11:28:15AM +0800, Yi Liu wrote:
> > > > On 2023/10/14 01:56, Nicolin Chen wrote:
> > > > > On Fri, Oct 13, 2023 at 11:04:56AM -0300, Jason Gunthorpe wrote:
> > > > > > On Fri, Oct 13, 2023 at 12:33:13PM +0800, Yi Liu wrote:
> > > > > > 
> > > > > > > not really. Below the users of the struct iommu_user_data in my current
> > > > > > > iommufd_nesting branch. Only the domain_alloc_user op has type as there
> > > > > > > can be multiple vendor specific alloc data types. Basically, I'm ok to
> > > > > > > make the change you suggested, just not sure if it is good to add type
> > > > > > > as it is only needed by one path.
> > > > > > 
> > > > > > I don't think we should ever have an opaque data blob without a type
> > > > > > tag..
> > > > > 
> > > > > I can add those "missing" data types, and then a driver will be
> > > > > responsible for sanitizing the type along with the data_len.
> > > > > 
> > > > > I notice that the enum iommu_hwpt_data_type in the posted patch
> > > > > is confined to the alloc_user uAPI. Perhaps we should share it
> > > > > with invalidate too:
> > > > 
> > > > invalidation path does not need a type field today as the data
> > > > type is vendor specific, vendor driver should know the data type
> > > > when calls in.
> > > 
> > > I'm not keen on that, what if a driver needs another type in the
> > > future?  You'd want to make the invalidation data format part of the
> > > domain allocation?
> > 
> > The invalidation data has hwpt_id so it's tied to a hwpt and its 
> > hwpt->domain. Would it be reasonable to have a different type of
> > invalidation data for the same type of hwpt?
> 
> Yeah, maybe? Go down the road 10 years and we might have SMMUv3
> invalidation format v1 and v2 or something?
> 
> Like we don't know what the HW side will do, they might extend the
> command queue to have bigger commands and negotiate with the driver if
> the bigger/smaller format is used. We've done that in our HW a couple
> of times now.

I see. We'll have the type. Thanks!



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux