Messages in this thread | | | Date | Wed, 11 Apr 2012 08:48:24 +0200 | From | Marek Szyprowski <> | Subject | RE: [PATCHv24 00/16] Contiguous Memory Allocator |
| |
Hi,
On Tuesday, April 10, 2012 7:20 PM Sandeep Patil wrote:
> >> This is (yet another) update of CMA patches. > > > > > > How well CMA is supposed to work if you have mlocked processes? I've > > been testing these patches, and noticed that by creating a small mlocked > > process you start to get plenty of test_pages_isolated() failure warnings, > > and bigger allocations will always fail. > > CMIIW, I think mlocked pages are never migrated. The reason is because > __isolate_lru_pages() does not isolate Unevictable pages right now. > > Minchan added support to allow this but the patch was dropped. > > See the discussion at : https://lkml.org/lkml/2011/8/29/295
Right, we are aware of this limitation. We are working on solving it but we didn't consider it a blocker for the core CMA patches. Such issues can be easily fixed with the incremental patches.
Best regards -- Marek Szyprowski Samsung Poland R&D Center
| |