| rjw | 1f88458 | 2022-01-06 17:20:42 +0800 | [diff] [blame] | 1 | Page fragments |
| 2 | -------------- |
| 3 | |
| 4 | A page fragment is an arbitrary-length arbitrary-offset area of memory |
| 5 | which resides within a 0 or higher order compound page. Multiple |
| 6 | fragments within that page are individually refcounted, in the page's |
| 7 | reference counter. |
| 8 | |
| 9 | The page_frag functions, page_frag_alloc and page_frag_free, provide a |
| 10 | simple allocation framework for page fragments. This is used by the |
| 11 | network stack and network device drivers to provide a backing region of |
| 12 | memory for use as either an sk_buff->head, or to be used in the "frags" |
| 13 | portion of skb_shared_info. |
| 14 | |
| 15 | In order to make use of the page fragment APIs a backing page fragment |
| 16 | cache is needed. This provides a central point for the fragment allocation |
| 17 | and tracks allows multiple calls to make use of a cached page. The |
| 18 | advantage to doing this is that multiple calls to get_page can be avoided |
| 19 | which can be expensive at allocation time. However due to the nature of |
| 20 | this caching it is required that any calls to the cache be protected by |
| 21 | either a per-cpu limitation, or a per-cpu limitation and forcing interrupts |
| 22 | to be disabled when executing the fragment allocation. |
| 23 | |
| 24 | The network stack uses two separate caches per CPU to handle fragment |
| 25 | allocation. The netdev_alloc_cache is used by callers making use of the |
| 26 | __netdev_alloc_frag and __netdev_alloc_skb calls. The napi_alloc_cache is |
| 27 | used by callers of the __napi_alloc_frag and __napi_alloc_skb calls. The |
| 28 | main difference between these two calls is the context in which they may be |
| 29 | called. The "netdev" prefixed functions are usable in any context as these |
| 30 | functions will disable interrupts, while the "napi" prefixed functions are |
| 31 | only usable within the softirq context. |
| 32 | |
| 33 | Many network device drivers use a similar methodology for allocating page |
| 34 | fragments, but the page fragments are cached at the ring or descriptor |
| 35 | level. In order to enable these cases it is necessary to provide a generic |
| 36 | way of tearing down a page cache. For this reason __page_frag_cache_drain |
| 37 | was implemented. It allows for freeing multiple references from a single |
| 38 | page via a single call. The advantage to doing this is that it allows for |
| 39 | cleaning up the multiple references that were added to a page in order to |
| 40 | avoid calling get_page per allocation. |
| 41 | |
| 42 | Alexander Duyck, Nov 29, 2016. |