Re: [git pull] drm patches for 2.6.27-rc1

[Nick Piggin <nickpiggin@yahoo.com.au>]

On Sunday 19 October 2008 06:31, Linus Torvalds wrote:
> On Sat, 18 Oct 2008, Keith Packard wrote:
> > The basic plan is to have four new functions (yes, I'm making up names
> > here):
> >
> > struct io_mapping *io_reserve_pci_resource(struct pci_dev *dev,
> >                                            int bar,
> >                                            int prot);
> > void io_mapping_free(struct io_mapping *mapping);
> >
> > void *io_map_atomic(struct io_mapping *mapping, unsigned long pfn);
> > void io_unmap_atomic(struct io_mapping *mapping, unsigned long pfn);
>
> The important thing is that mappings need to be per-CPU, so the above may
> work, but only if it's designed so that "io_reserve_pci_resource()" will
> actually reserve space for 'nr_possible_cpu' page mappings, and then the
> "io_[un]map_atomic()" functions do per-CPU mappings.
>
> Anything else is a disaster, because anything else implies TLB shootdown.

TLB shootdown need only be implied if the behaviour required is to
unmap the virtual address *and* cause any other CPU that subsequently
touches it to fault.

For kva, that would be a bug anyway (use after free). The only thing
it implies is that a TLB shootdown happens at some point before the
address get reused.

Still, it's always going to be faster than a global mapping, if done
properly. I was thinking about doing a vmap_atomic thing generically
in the vmap layer... why exactly do we need the FIXMAP stuff for it?