On 12/11/18 11:51 AM, Matthew Wilcox wrote: > On Tue, Dec 11, 2018 at 11:32:54AM -0700, Jens Axboe wrote: >> Don't see any regressions. But if we're fiddling with it anyway, can't >> we do something smarter? Make the fast path just index a table, and put >> all the big hammers in setup/destroy. We're spending a non-substantial >> amount of time doing lookups, that's really no different before and >> after the patch. > > Thanks for checking it out. > > I think the fast path does just index a table. Until you have more than > 64 pointers in the XArray, it's just xa->head->slots[i]. And then up > to 4096 pointers, it's xa->head->slots[i >> 6]->slots[i]. It has the > advantage that if you only have one kioctx (which is surely many programs > using AIO), it's just xa->head, so even better than a table lookup. > > It'll start to deteriorate after 4096 kioctxs, with one extra indirection > for every 6 bits, but by that point, we'd've been straining the memory > allocator to allocate a large table anyway. I agree, and nobody cares about 4k kioctxs, you're way into the weeds at that point anyway. So as the followup said, I think we're fine as-is for this particular case. -- Jens Axboe
