On Mon, Oct 23, 2023 at 10:59:35AM -0300, Jason Gunthorpe wrote:
> On Sun, Oct 22, 2023 at 05:18:03PM -0700, Nicolin Chen wrote:
>
> > > > And where should we add this comment? Kdoc of
> > > > the alloc uAPI?
> > >
> > > Maybe right in front of the only enforce_cc op callback?
> >
> > OK. How about this? Might be a bit verbose though:
> > /*
> > * enforce_cache_coherenc must be determined during the HWPT allocation.
> > * Note that a HWPT (non-CC) created for a device (non-CC) can be later
> > * reused by another device (either non-CC or CC). However, A HWPT (CC)
> > * created for a device (CC) cannot be reused by another device (non-CC)
> > * but only devices (CC). Instead user space in this case would need to
> > * allocate a separate HWPT (non-CC).
> > */
>
> Ok, fix the spello in enforce_cache_coherenc
Oops.
I also found the existing piece sorta says a similar thing:
* Set the coherency mode before we do iopt_table_add_domain() as some
* iommus have a per-PTE bit that controls it and need to decide before
* doing any maps.
So, did this and sending v3:
- * enforce_cache_coherenc must be determined during the HWPT allocation.
+ * The cache coherency mode must be configured here and unchanged later.
