1/* SPDX-License-Identifier: GPL-2.0
2 *
3 * page_pool/helpers.h
4 *	Author:	Jesper Dangaard Brouer <netoptimizer@brouer.com>
5 *	Copyright (C) 2016 Red Hat, Inc.
6 */
7
8/**
9 * DOC: page_pool allocator
10 *
11 * The page_pool allocator is optimized for the XDP mode that
12 * uses one frame per-page, but it can fallback on the
13 * regular page allocator APIs.
14 *
15 * Basic use involves replacing alloc_pages() calls with the
16 * page_pool_alloc_pages() call.  Drivers should use
17 * page_pool_dev_alloc_pages() replacing dev_alloc_pages().
18 *
19 * The API keeps track of in-flight pages, in order to let API users know
20 * when it is safe to free a page_pool object.  Thus, API users
21 * must call page_pool_put_page() to free the page, or attach
22 * the page to a page_pool-aware object like skbs marked with
23 * skb_mark_for_recycle().
24 *
25 * API users must call page_pool_put_page() once on a page, as it
26 * will either recycle the page, or in case of refcnt > 1, it will
27 * release the DMA mapping and in-flight state accounting.
28 */
29#ifndef _NET_PAGE_POOL_HELPERS_H
30#define _NET_PAGE_POOL_HELPERS_H
31
32#include <net/page_pool/types.h>
33
34#ifdef CONFIG_PAGE_POOL_STATS
35int page_pool_ethtool_stats_get_count(void);
36u8 *page_pool_ethtool_stats_get_strings(u8 *data);
37u64 *page_pool_ethtool_stats_get(u64 *data, void *stats);
38
39/*
40 * Drivers that wish to harvest page pool stats and report them to users
41 * (perhaps via ethtool, debugfs, or another mechanism) can allocate a
42 * struct page_pool_stats call page_pool_get_stats to get stats for the specified pool.
43 */
44bool page_pool_get_stats(struct page_pool *pool,
45			 struct page_pool_stats *stats);
46#else
47static inline int page_pool_ethtool_stats_get_count(void)
48{
49	return 0;
50}
51
52static inline u8 *page_pool_ethtool_stats_get_strings(u8 *data)
53{
54	return data;
55}
56
57static inline u64 *page_pool_ethtool_stats_get(u64 *data, void *stats)
58{
59	return data;
60}
61#endif
62
63/**
64 * page_pool_dev_alloc_pages() - allocate a page.
65 * @pool:	pool from which to allocate
66 *
67 * Get a page from the page allocator or page_pool caches.
68 */
69static inline struct page *page_pool_dev_alloc_pages(struct page_pool *pool)
70{
71	gfp_t gfp = (GFP_ATOMIC | __GFP_NOWARN);
72
73	return page_pool_alloc_pages(pool, gfp);
74}
75
76static inline struct page *page_pool_dev_alloc_frag(struct page_pool *pool,
77						    unsigned int *offset,
78						    unsigned int size)
79{
80	gfp_t gfp = (GFP_ATOMIC | __GFP_NOWARN);
81
82	return page_pool_alloc_frag(pool, offset, size, gfp);
83}
84
85/**
86 * page_pool_get_dma_dir() - Retrieve the stored DMA direction.
87 * @pool:	pool from which page was allocated
88 *
89 * Get the stored dma direction. A driver might decide to store this locally
90 * and avoid the extra cache line from page_pool to determine the direction.
91 */
92static
93inline enum dma_data_direction page_pool_get_dma_dir(struct page_pool *pool)
94{
95	return pool->p.dma_dir;
96}
97
98/* pp_frag_count represents the number of writers who can update the page
99 * either by updating skb->data or via DMA mappings for the device.
100 * We can't rely on the page refcnt for that as we don't know who might be
101 * holding page references and we can't reliably destroy or sync DMA mappings
102 * of the fragments.
103 *
104 * When pp_frag_count reaches 0 we can either recycle the page if the page
105 * refcnt is 1 or return it back to the memory allocator and destroy any
106 * mappings we have.
107 */
108static inline void page_pool_fragment_page(struct page *page, long nr)
109{
110	atomic_long_set(&page->pp_frag_count, nr);
111}
112
113static inline long page_pool_defrag_page(struct page *page, long nr)
114{
115	long ret;
116
117	/* If nr == pp_frag_count then we have cleared all remaining
118	 * references to the page. No need to actually overwrite it, instead
119	 * we can leave this to be overwritten by the calling function.
120	 *
121	 * The main advantage to doing this is that an atomic_read is
122	 * generally a much cheaper operation than an atomic update,
123	 * especially when dealing with a page that may be partitioned
124	 * into only 2 or 3 pieces.
125	 */
126	if (atomic_long_read(&page->pp_frag_count) == nr)
127		return 0;
128
129	ret = atomic_long_sub_return(nr, &page->pp_frag_count);
130	WARN_ON(ret < 0);
131	return ret;
132}
133
134static inline bool page_pool_is_last_frag(struct page_pool *pool,
135					  struct page *page)
136{
137	/* If fragments aren't enabled or count is 0 we were the last user */
138	return !(pool->p.flags & PP_FLAG_PAGE_FRAG) ||
139	       (page_pool_defrag_page(page, 1) == 0);
140}
141
142/**
143 * page_pool_put_page() - release a reference to a page pool page
144 * @pool:	pool from which page was allocated
145 * @page:	page to release a reference on
146 * @dma_sync_size: how much of the page may have been touched by the device
147 * @allow_direct: released by the consumer, allow lockless caching
148 *
149 * The outcome of this depends on the page refcnt. If the driver bumps
150 * the refcnt > 1 this will unmap the page. If the page refcnt is 1
151 * the allocator owns the page and will try to recycle it in one of the pool
152 * caches. If PP_FLAG_DMA_SYNC_DEV is set, the page will be synced for_device
153 * using dma_sync_single_range_for_device().
154 */
155static inline void page_pool_put_page(struct page_pool *pool,
156				      struct page *page,
157				      unsigned int dma_sync_size,
158				      bool allow_direct)
159{
160	/* When page_pool isn't compiled-in, net/core/xdp.c doesn't
161	 * allow registering MEM_TYPE_PAGE_POOL, but shield linker.
162	 */
163#ifdef CONFIG_PAGE_POOL
164	if (!page_pool_is_last_frag(pool, page))
165		return;
166
167	page_pool_put_defragged_page(pool, page, dma_sync_size, allow_direct);
168#endif
169}
170
171/**
172 * page_pool_put_full_page() - release a reference on a page pool page
173 * @pool:	pool from which page was allocated
174 * @page:	page to release a reference on
175 * @allow_direct: released by the consumer, allow lockless caching
176 *
177 * Similar to page_pool_put_page(), but will DMA sync the entire memory area
178 * as configured in &page_pool_params.max_len.
179 */
180static inline void page_pool_put_full_page(struct page_pool *pool,
181					   struct page *page, bool allow_direct)
182{
183	page_pool_put_page(pool, page, -1, allow_direct);
184}
185
186/**
187 * page_pool_recycle_direct() - release a reference on a page pool page
188 * @pool:	pool from which page was allocated
189 * @page:	page to release a reference on
190 *
191 * Similar to page_pool_put_full_page() but caller must guarantee safe context
192 * (e.g NAPI), since it will recycle the page directly into the pool fast cache.
193 */
194static inline void page_pool_recycle_direct(struct page_pool *pool,
195					    struct page *page)
196{
197	page_pool_put_full_page(pool, page, true);
198}
199
200#define PAGE_POOL_DMA_USE_PP_FRAG_COUNT	\
201		(sizeof(dma_addr_t) > sizeof(unsigned long))
202
203/**
204 * page_pool_get_dma_addr() - Retrieve the stored DMA address.
205 * @page:	page allocated from a page pool
206 *
207 * Fetch the DMA address of the page. The page pool to which the page belongs
208 * must had been created with PP_FLAG_DMA_MAP.
209 */
210static inline dma_addr_t page_pool_get_dma_addr(struct page *page)
211{
212	dma_addr_t ret = page->dma_addr;
213
214	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
215		ret |= (dma_addr_t)page->dma_addr_upper << 16 << 16;
216
217	return ret;
218}
219
220static inline void page_pool_set_dma_addr(struct page *page, dma_addr_t addr)
221{
222	page->dma_addr = addr;
223	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT)
224		page->dma_addr_upper = upper_32_bits(addr);
225}
226
227static inline bool page_pool_put(struct page_pool *pool)
228{
229	return refcount_dec_and_test(&pool->user_cnt);
230}
231
232static inline void page_pool_nid_changed(struct page_pool *pool, int new_nid)
233{
234	if (unlikely(pool->p.nid != new_nid))
235		page_pool_update_nid(pool, new_nid);
236}
237
238#endif /* _NET_PAGE_POOL_HELPERS_H */
239