On Tue, Jan 06, 2015 at 09:33:15AM -0600, Christoph Lameter wrote: > On Tue, 6 Jan 2015, Andreas Mohr wrote: > > by merely queuing them into a simple submission queue > > which then will be delay-applied by main-context > > either once main-context enters a certain "quiet" state (e.g. context > > switch?), > > or once main-context needs to actively take into account > > This is basically the same approach as you mentioned before and would > multiply the resources needed. I think we are close to be able to avoid > allocations from interrupt contexts. Someone would need to perform an > audit to see what is left to be done. If so then lots of allocator paths > both in the page allocator and slab allocator can be dramatically > simplified.
OK, so we seem to be already well near the finishing line of single-context operation. In case of multi-context access, in general I'd guess that challenges are similar to "traditional" multi-thread-capable heap allocator implementations in userspace, where I'm sure large amounts of research papers (some dead-tree?) have been written about how to achieve scalable low-contention multi-thread access arbitration to the same shared underlying memory resource (but since in Linux circles some implementations exceed usual accumulated research knowledge / Best Practice, such papers may or may not be of much help ;) Andreas Mohr -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
