Re: Filtering bits in set_pte_at()
From: Hugh Dickins
Date: Mon Nov 02 2009 - 18:45:57 EST
On Tue, 3 Nov 2009, Benjamin Herrenschmidt wrote:
> On Mon, 2009-11-02 at 13:27 +0000, Hugh Dickins wrote:
> > You're being a very good citizen to want to bring this so forcefully
> > to the attention of any user of set_pte_at(); but given how few care,
> > and the other such functions you'd want to change too, am I being
> > disgracefully lazy to suggest that you simply change the occasional
> > update_mmu_cache(vma, address, pte);
> > to
> > /* powerpc's set_pte_at might have adjusted the pte */
> > update_mmu_cache(vma, address, *ptep);
> > ? Which would make no difference to those architectures whose
> > update_mmu_cache() is an empty macro. And fix the mm/hugetlb.c
> > instance in a similar way?
> That would do fine. In fact, I've always been slightly annoyed by
> set_pte_at() not taking the PTE pointer for other reasons such as on
> 64-K pages, we have a "hidden" part of the PTE that is at PTE address +
> 32K, or we may want to get to the PTE page for some reason (some arch
> store things there) etc...
> IE. update_mmu_cache() would be more generally useful if it took the
> ptep instead of the pte. Of course, I'm sure some embedded archs are
> going to cry for the added load here ...
> I like your idea. I'll look into doing a patch converting it and will
> post it here.
Well, I wasn't proposing
update_mmu_cache(vma, address, ptep);
update_mmu_cache(vma, address, *ptep);
which may not meet your future idea, but is much less churn for now
i.e. no change to any of the arch's update_mmu_cache(),
just a change to some of its callsites.
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/