We had an off-by-one error which meant that we never marked the first page in a read as accessed. This was visible as a slowdown when re-reading a file as pages were being evicted from cache too soon. In reviewing this code, we noticed a second bug where a multi-page folio would be marked as accessed multiple times when doing reads that were less than the size of the folio. Abstract the comparison of whether two file positions are in the same folio into a new function, fixing both of these bugs.
Unified · Split
| @@ -2629,6 +2629,13 @@ err: · 2628 unmodified lines | ||
| 2629 | 2629 | return err; |
| 2630 | 2630 | } |
| 2631 | 2631 | |
| 2632 | +static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio) | |
| 2633 | +{ | |
| 2634 | + unsigned int shift = folio_shift(folio); | |
| 2635 | + | |
| 2636 | + return (pos1 >> shift == pos2 >> shift); | |
| 2637 | +} | |
| 2638 | + | |
| 2632 | 2639 | /** |
| 2633 | 2640 | * filemap_read - Read data from the page cache. |
| 2634 | 2641 | * @iocb: The iocb to read. |
| @@ -2700,11 +2707,11 @@ ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, · 65 unmodified lines | ||
| 2700 | 2707 | writably_mapped = mapping_writably_mapped(mapping); |
| 2701 | 2708 | |
| 2702 | 2709 | /* |
| 2703 | − * When a sequential read accesses a page several times, only | |
| 2710 | + * When a read accesses the same folio several times, only | |
| 2704 | 2711 | * mark it as accessed the first time. |
| 2705 | 2712 | */ |
| 2706 | − if (iocb->ki_pos >> PAGE_SHIFT != | |
| 2707 | − ra->prev_pos >> PAGE_SHIFT) | |
| 2713 | + if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1, | |
| 2714 | + fbatch.folios[0])) | |
| 2708 | 2715 | folio_mark_accessed(fbatch.folios[0]); |
| 2709 | 2716 | |
| 2710 | 2717 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |