Re: [RFC][PATCH 1/5] mm: Introduce VM_PINNED and interfaces

[Peter Zijlstra <peterz@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 1049 bytes --]

On Wed, May 28, 2014 at 09:48:43PM -0400, Rik van Riel wrote:
> On 05/26/2014 10:56 AM, Peter Zijlstra wrote:
> 
> >  include/linux/mm.h       |    3 +
> >  include/linux/mm_types.h |    5 +
> >  kernel/fork.c            |    2 
> >  mm/mlock.c               |  133 ++++++++++++++++++++++++++++++++++++++++++-----
> >  mm/mmap.c                |   18 ++++--
> >  5 files changed, 141 insertions(+), 20 deletions(-)
> 
> I'm guessing you will also want a patch that adds some code to
> rmap.c, madvise.c, and a few other places to actually enforce
> the VM_PINNED semantics?

Eventually, yes. As it stands they're not needed, because perf pages
aren't in the pagecache and IB goes a big get_user_pages() and the
elevated refcount stops everything dead.

But yes, once we go do fancy things like migrate the pages into
UNMOVABLE blocks, then we need to have VM_PINNED itself mean something,
instead of it describing something.

But before we go down there, I'd like to get the IB bits sorted and
general agreement on the path.

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]