On Fri, Dec 20, 2013 at 05:31:29AM +0900, Linus Torvalds wrote: > On Fri, Dec 20, 2013 at 5:11 AM, Linus Torvalds > <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > > > So here's the same patch, but with stuff moved around a bit, and the > > "oops, couldn't create page" part fixed. > > > > Bit it's still totally and entirely untested. > > Btw, I think this actually fixes a bug, in that it doesn't leak the > page reference count if the do_mmap_pgoff() call fails. > > That said, that looks like just a memory leak, not explaining the > problem Dave sees. And maybe I'm missing something. > > And no, I still haven't actually tested this at all. Is there an aio > tester that is worth trying? There are a few tests in the libaio source, some in xfstests and fio, and a few I've got sitting around. To specifically exercise the page migration code path, there's a test at http://www.kvack.org/~bcrl/aio-numa-test.c . -ben > Linus -- "Thought is the essence of where you are now." -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>