On Mon, Mar 06, 2017 at 04:13:28PM -0800, Darrick J. Wong wrote: > On Tue, Mar 07, 2017 at 01:07:54AM +0100, Christoph Hellwig wrote: > > I like killing it, but shouldn't we just try a normal kmem_zalloc? > > At least for the fallback it's the right thing, and even for an > > order 2 allocation it seems like a useful first try. > > I'm confused -- kmem_zalloc_large tries kmem_zalloc with KM_MAYFAIL and > only falls back to __vmalloc if it doesn't get anything. Yup, that's right. > Or maybe I've misunderstood, and you're asking if we should try > kmem_zalloc(4 pages), then kmem_zalloc(1 page), and only then switch to > the __vmalloc calls? Just call kmem_zalloc_large() for 4 pages without a fallback on failure - that's exactly how we handle allocations for things like the 64k xattr buffers.... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
