Re: GRU driver feedback

[Nick Piggin]

On Wednesday 30 July 2008 15:50, Nick Piggin wrote:
> On Wednesday 30 July 2008 04:53, Robin Holt wrote:

> > In the case where unmap_region is clearing page tables, the caller to
> > unmap_region is expected to be holding the mmap_sem writably.  Jacks
> > fault handler will immediately return when it fails on the
> > down_read_trylock().
>
> No, you are right of course. I had in my mind the problems faced by
> lockless get_user_pages, in which case I was worried about the page table
> existence, but missed the fact that you're holding mmap_sem to provide
> existence (which it would, as you note, although one day we may want to
> reclaim page tables or something that doesn't take mmap_sem, so a big
> comment would be nice here).

The other thing is... then GRU should get rid of the local_irq_disable
in the atomic pte lookup. By definition it is worthless if we can be
operating on an mm that is not running on current (and if I understand
correctly, sn2 can avoid sending tlb flush IPIs completely sometimes?)
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/