Re: bind pasid table API

[Jacob Pan <jacob.jun.pan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>]

On Thu, 28 Sep 2017 14:07:05 +0200
Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org> wrote:

> On Wed, Sep 27, 2017 at 10:51:55AM -0700, Jacob Pan wrote:
> > On Wed, 27 Sep 2017 15:40:41 +0200
> > Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org> wrote:  
> > > 	enum pasid_table_model {
> > > 		PASID_TABLE_INTEL,
> > > 		PASID_TABLE_ARM,
> > > 		PASID_TABLE_AMD,
> > > 		/* ... */
> > > 	};
> > >   
> > I guess one vendor could have multiple pasid table format. so
> > perhaps the name could reflect the format as well?  
> 
> Yes, that makes sense.
> 
> > > 	struct pasid_table_config {
> > > 		__u64 base_ptr;
> > > 		__u8 pasid_bits;
> > > 		union {
> > > 			struct {
> > > 				/* Intel specific fields */
> > > 			} intel;
> > > 
> > > 			struct {
> > > 				/* ARM specific fields */
> > > 			} arm;
> > > 
> > > 
> > > 			struct {
> > > 				/* AMD specific fields */
> > > 			} amd;
> > > 
> > > 			/* ... */
> > > 		};
> > > 	};
> > > 
> > > How does that look?
> > >   
> > It should work for us for now but I am not sure how stable the
> > vendor specific fields will be, this is UAPI. BTW, do you also
> > intend to include the pasid_table_model # in pasid_table_config?  
> 
> If its going to be UAPI then we probably also need a version and a
> length/size field. But since individual formats do not change, their
> sub-struct should be stable.
> 
just to confirm, we have two separate APIs for bind_pasid_table_ptr and
bind_page_table. For the former, we need to passdown the following info
from qemu/vfio. For now
struct pasid_table_config {
	__u32 version;
#define PASID_TABLE_CONFIG_VERSION_1 (1)
	__u32 bytes;
	__u64 base_ptr;
	__u8 pasid_bits;
	enum pasid_table_model model;
	union {
		struct {
			/* Intel specific fields */
		} intel;

		struct {
			/* ARM specific fields */
		} arm;

		struct {
			/* AMD specific fields */
		} amd;

		/* ... */
	};
};

> Regards,
> 
> 	Joerg
> 

[Jacob Pan]