net: page_pool: Add documentation on page_pool API
authorIlias Apalodimas <ilias.apalodimas@linaro.org>
Fri, 21 Feb 2020 09:15:19 +0000 (11:15 +0200)
committerDavid S. Miller <davem@davemloft.net>
Fri, 21 Feb 2020 16:13:30 +0000 (08:13 -0800)
Add documentation explaining the basic functionality and design
principles of the API

Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Documentation/networking/page_pool.rst [new file with mode: 0644]

diff --git a/Documentation/networking/page_pool.rst b/Documentation/networking/page_pool.rst
new file mode 100644 (file)
index 0000000..43088dd
--- /dev/null
@@ -0,0 +1,159 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+=============
+Page Pool API
+=============
+
+The page_pool allocator is optimized for the XDP mode that uses one frame
+per-page, but it can fallback on the regular page allocator APIs.
+
+Basic use involves replacing alloc_pages() calls with the
+page_pool_alloc_pages() call.  Drivers should use page_pool_dev_alloc_pages()
+replacing dev_alloc_pages().
+
+API keeps track of inflight pages, in order to let API user know
+when it is safe to free a page_pool object.  Thus, API users
+must run page_pool_release_page() when a page is leaving the page_pool or
+call page_pool_put_page() where appropriate in order to maintain correct
+accounting.
+
+API user must call page_pool_put_page() once on a page, as it
+will either recycle the page, or in case of refcnt > 1, it will
+release the DMA mapping and inflight state accounting.
+
+Architecture overview
+=====================
+
+.. code-block:: none
+
+    +------------------+
+    |       Driver     |
+    +------------------+
+            ^
+            |
+            |
+            |
+            v
+    +--------------------------------------------+
+    |                request memory              |
+    +--------------------------------------------+
+        ^                                  ^
+        |                                  |
+        | Pool empty                       | Pool has entries
+        |                                  |
+        v                                  v
+    +-----------------------+     +------------------------+
+    | alloc (and map) pages |     |  get page from cache   |
+    +-----------------------+     +------------------------+
+                                    ^                    ^
+                                    |                    |
+                                    | cache available    | No entries, refill
+                                    |                    | from ptr-ring
+                                    |                    |
+                                    v                    v
+                          +-----------------+     +------------------+
+                          |   Fast cache    |     |  ptr-ring cache  |
+                          +-----------------+     +------------------+
+
+API interface
+=============
+The number of pools created **must** match the number of hardware queues
+unless hardware restrictions make that impossible. This would otherwise beat the
+purpose of page pool, which is allocate pages fast from cache without locking.
+This lockless guarantee naturally comes from running under a NAPI softirq.
+The protection doesn't strictly have to be NAPI, any guarantee that allocating
+a page will cause no race conditions is enough.
+
+* page_pool_create(): Create a pool.
+    * flags:      PP_FLAG_DMA_MAP, PP_FLAG_DMA_SYNC_DEV
+    * order:      2^order pages on allocation
+    * pool_size:  size of the ptr_ring
+    * nid:        preferred NUMA node for allocation
+    * dev:        struct device. Used on DMA operations
+    * dma_dir:    DMA direction
+    * max_len:    max DMA sync memory size
+    * offset:     DMA address offset
+
+* page_pool_put_page(): The outcome of this depends on the page refcnt. If the
+  driver bumps the refcnt > 1 this will unmap the page. If the page refcnt is 1
+  the allocator owns the page and will try to recycle it in one of the pool
+  caches. If PP_FLAG_DMA_SYNC_DEV is set, the page will be synced for_device
+  using dma_sync_single_range_for_device().
+
+* page_pool_put_full_page(): Similar to page_pool_put_page(), but will DMA sync
+  for the entire memory area configured in area pool->max_len.
+
+* page_pool_recycle_direct(): Similar to page_pool_put_full_page() but caller
+  must guarantee safe context (e.g NAPI), since it will recycle the page
+  directly into the pool fast cache.
+
+* page_pool_release_page(): Unmap the page (if mapped) and account for it on
+  inflight counters.
+
+* page_pool_dev_alloc_pages(): Get a page from the page allocator or page_pool
+  caches.
+
+* page_pool_get_dma_addr(): Retrieve the stored DMA address.
+
+* page_pool_get_dma_dir(): Retrieve the stored DMA direction.
+
+Coding examples
+===============
+
+Registration
+------------
+
+.. code-block:: c
+
+    /* Page pool registration */
+    struct page_pool_params pp_params = { 0 };
+    struct xdp_rxq_info xdp_rxq;
+    int err;
+
+    pp_params.order = 0;
+    /* internal DMA mapping in page_pool */
+    pp_params.flags = PP_FLAG_DMA_MAP;
+    pp_params.pool_size = DESC_NUM;
+    pp_params.nid = NUMA_NO_NODE;
+    pp_params.dev = priv->dev;
+    pp_params.dma_dir = xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
+    page_pool = page_pool_create(&pp_params);
+
+    err = xdp_rxq_info_reg(&xdp_rxq, ndev, 0);
+    if (err)
+        goto err_out;
+
+    err = xdp_rxq_info_reg_mem_model(&xdp_rxq, MEM_TYPE_PAGE_POOL, page_pool);
+    if (err)
+        goto err_out;
+
+NAPI poller
+-----------
+
+
+.. code-block:: c
+
+    /* NAPI Rx poller */
+    enum dma_data_direction dma_dir;
+
+    dma_dir = page_pool_get_dma_dir(dring->page_pool);
+    while (done < budget) {
+        if (some error)
+            page_pool_recycle_direct(page_pool, page);
+        if (packet_is_xdp) {
+            if XDP_DROP:
+                page_pool_recycle_direct(page_pool, page);
+        } else (packet_is_skb) {
+            page_pool_release_page(page_pool, page);
+            new_page = page_pool_dev_alloc_pages(page_pool);
+        }
+    }
+
+Driver unload
+-------------
+
+.. code-block:: c
+
+    /* Driver unload */
+    page_pool_put_full_page(page_pool, page, false);
+    xdp_rxq_info_unreg(&xdp_rxq);