xref: /linux/include/net/xsk_buff_pool.h (revision 5d7f3375cd574fc83f0cd30839b21455c0e79ae2)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright(c) 2020 Intel Corporation. */
3 
4 #ifndef XSK_BUFF_POOL_H_
5 #define XSK_BUFF_POOL_H_
6 
7 #include <linux/if_xdp.h>
8 #include <linux/types.h>
9 #include <linux/dma-mapping.h>
10 #include <linux/bpf.h>
11 #include <net/xdp.h>
12 
13 struct xsk_buff_pool;
14 struct xdp_rxq_info;
15 struct xsk_cb_desc;
16 struct xsk_queue;
17 struct xdp_desc;
18 struct xdp_umem;
19 struct xdp_sock;
20 struct device;
21 struct page;
22 
23 #define XSK_PRIV_MAX 24
24 
25 struct xdp_buff_xsk {
26 	struct xdp_buff xdp;
27 	u8 cb[XSK_PRIV_MAX];
28 	dma_addr_t dma;
29 	dma_addr_t frame_dma;
30 	struct xsk_buff_pool *pool;
31 	struct list_head list_node;
32 } __aligned_largest;
33 
34 #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb))
35 #define XSK_TX_COMPL_FITS(t) BUILD_BUG_ON(sizeof(struct xsk_tx_metadata_compl) > sizeof(t))
36 
37 struct xsk_dma_map {
38 	dma_addr_t *dma_pages;
39 	struct device *dev;
40 	struct net_device *netdev;
41 	refcount_t users;
42 	struct list_head list; /* Protected by the RTNL_LOCK */
43 	u32 dma_pages_cnt;
44 };
45 
46 struct xsk_buff_pool {
47 	/* Members only used in the control path first. */
48 	struct device *dev;
49 	struct net_device *netdev;
50 	struct list_head xsk_tx_list;
51 	/* Assigned pools, including deferred releases; ops protected. */
52 	struct hlist_node dev_node;
53 	/* Protects modifications to the xsk_tx_list */
54 	spinlock_t xsk_tx_list_lock;
55 	refcount_t users;
56 	struct xdp_umem *umem;
57 	struct work_struct work;
58 	/* Protects generic receive in shared and non-shared umem mode. */
59 	spinlock_t rx_lock;
60 	struct list_head free_list;
61 	struct list_head xskb_list;
62 	u32 heads_cnt;
63 	u16 queue_id;
64 
65 	/* Data path members as close to free_heads at the end as possible. */
66 	struct xsk_queue *fq ____cacheline_aligned_in_smp;
67 	struct xsk_queue *cq;
68 	/* For performance reasons, each buff pool has its own array of dma_pages
69 	 * even when they are identical.
70 	 */
71 	dma_addr_t *dma_pages;
72 	struct xdp_buff_xsk *heads;
73 	struct xdp_desc *tx_descs;
74 	u64 chunk_mask;
75 	u64 addrs_cnt;
76 	u32 free_list_cnt;
77 	u32 dma_pages_cnt;
78 	u32 free_heads_cnt;
79 	u32 headroom;
80 	u32 chunk_size;
81 	u32 chunk_shift;
82 	u32 frame_len;
83 	u32 tx_descs_nentries;
84 	u32 reclaim_descs;
85 	u32 tx_zc_pending_descs;
86 	u32 xdp_zc_max_segs;
87 	u8 tx_metadata_len; /* inherited from umem */
88 	u8 cached_need_wakeup;
89 	bool uses_need_wakeup;
90 	bool unaligned;
91 	bool tx_sw_csum;
92 	void *addrs;
93 	/* Mutual exclusion of the completion ring in the SKB mode.
94 	 * Protect: NAPI TX thread and sendmsg error paths in the SKB
95 	 * destructor callback.
96 	 */
97 	spinlock_t cq_prod_lock;
98 	struct xdp_buff_xsk *free_heads[];
99 };
100 
101 /* Masks for xdp_umem_page flags.
102  * The low 12-bits of the addr will be 0 since this is the page address, so we
103  * can use them for flags.
104  */
105 #define XSK_NEXT_PG_CONTIG_SHIFT 0
106 #define XSK_NEXT_PG_CONTIG_MASK BIT_ULL(XSK_NEXT_PG_CONTIG_SHIFT)
107 
108 /* AF_XDP core. */
109 struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
110 						struct xdp_umem *umem,
111 						u32 max_segs);
112 int xp_assign_dev(struct xsk_buff_pool *pool, struct net_device *dev,
113 		  u16 queue_id, u16 flags);
114 int xp_assign_dev_shared(struct xsk_buff_pool *pool, struct xdp_sock *umem_xs,
115 			 struct net_device *dev, u16 queue_id);
116 int xp_alloc_tx_descs(struct xsk_buff_pool *pool, struct xdp_sock *xs,
117 		      u32 max_segs);
118 void xp_destroy(struct xsk_buff_pool *pool);
119 void xp_get_pool(struct xsk_buff_pool *pool);
120 bool xp_put_pool(struct xsk_buff_pool *pool);
121 void xp_clear_dev(struct xsk_buff_pool *pool);
122 void xp_add_xsk(struct xsk_buff_pool *pool, struct xdp_sock *xs);
123 void xp_del_xsk(struct xsk_buff_pool *pool, struct xdp_sock *xs);
124 void xp_clear_dev_all(struct net_device *dev);
125 
126 /* AF_XDP, and XDP core. */
127 void xp_free(struct xdp_buff_xsk *xskb);
128 
129 static inline void xp_init_xskb_addr(struct xdp_buff_xsk *xskb, struct xsk_buff_pool *pool,
130 				     u64 addr)
131 {
132 	xskb->xdp.data_hard_start = pool->addrs + addr + pool->headroom;
133 }
134 
135 static inline void xp_init_xskb_dma(struct xdp_buff_xsk *xskb, struct xsk_buff_pool *pool,
136 				    dma_addr_t *dma_pages, u64 addr)
137 {
138 	xskb->frame_dma = (dma_pages[addr >> PAGE_SHIFT] & ~XSK_NEXT_PG_CONTIG_MASK) +
139 		(addr & ~PAGE_MASK);
140 	xskb->dma = xskb->frame_dma + pool->headroom + XDP_PACKET_HEADROOM;
141 }
142 
143 /* AF_XDP ZC drivers, via xdp_sock_buff.h */
144 void xp_set_rxq_info(struct xsk_buff_pool *pool, struct xdp_rxq_info *rxq);
145 void xp_fill_cb(struct xsk_buff_pool *pool, struct xsk_cb_desc *desc);
146 int xp_dma_map(struct xsk_buff_pool *pool, struct device *dev,
147 	       unsigned long attrs, struct page **pages, u32 nr_pages);
148 void xp_dma_unmap(struct xsk_buff_pool *pool, unsigned long attrs);
149 struct xdp_buff *xp_alloc(struct xsk_buff_pool *pool);
150 u32 xp_alloc_batch(struct xsk_buff_pool *pool, struct xdp_buff **xdp, u32 max);
151 bool xp_can_alloc(struct xsk_buff_pool *pool, u32 count);
152 void *xp_raw_get_data(struct xsk_buff_pool *pool, u64 addr);
153 dma_addr_t xp_raw_get_dma(struct xsk_buff_pool *pool, u64 addr);
154 
155 struct xdp_desc_ctx {
156 	dma_addr_t dma;
157 	struct xsk_tx_metadata *meta;
158 };
159 
160 struct xdp_desc_ctx xp_raw_get_ctx(const struct xsk_buff_pool *pool, u64 addr,
161 				   u32 options);
162 
163 static inline dma_addr_t xp_get_dma(struct xdp_buff_xsk *xskb)
164 {
165 	return xskb->dma;
166 }
167 
168 static inline dma_addr_t xp_get_frame_dma(struct xdp_buff_xsk *xskb)
169 {
170 	return xskb->frame_dma;
171 }
172 
173 static inline void xp_dma_sync_for_cpu(struct xdp_buff_xsk *xskb)
174 {
175 	dma_sync_single_for_cpu(xskb->pool->dev, xskb->dma,
176 				xskb->pool->frame_len,
177 				DMA_BIDIRECTIONAL);
178 }
179 
180 static inline void xp_dma_sync_for_device(struct xsk_buff_pool *pool,
181 					  dma_addr_t dma, size_t size)
182 {
183 	dma_sync_single_for_device(pool->dev, dma, size, DMA_BIDIRECTIONAL);
184 }
185 
186 static inline bool xp_desc_crosses_non_contig_pg(struct xsk_buff_pool *pool,
187 						 u64 addr, u32 len)
188 {
189 	bool cross_pg = (addr & (PAGE_SIZE - 1)) + len > PAGE_SIZE;
190 
191 	if (likely(!cross_pg))
192 		return false;
193 
194 	return pool->dma_pages &&
195 	       !(pool->dma_pages[addr >> PAGE_SHIFT] & XSK_NEXT_PG_CONTIG_MASK);
196 }
197 
198 static inline bool xp_mb_desc(const struct xdp_desc *desc)
199 {
200 	return desc->options & XDP_PKT_CONTD;
201 }
202 
203 static inline u64 xp_aligned_extract_addr(struct xsk_buff_pool *pool, u64 addr)
204 {
205 	return addr & pool->chunk_mask;
206 }
207 
208 static inline u64 xp_unaligned_extract_addr(u64 addr)
209 {
210 	return addr & XSK_UNALIGNED_BUF_ADDR_MASK;
211 }
212 
213 static inline u64 xp_unaligned_extract_offset(u64 addr)
214 {
215 	return addr >> XSK_UNALIGNED_BUF_OFFSET_SHIFT;
216 }
217 
218 static inline u64 xp_unaligned_add_offset_to_addr(u64 addr)
219 {
220 	return xp_unaligned_extract_addr(addr) +
221 		xp_unaligned_extract_offset(addr);
222 }
223 
224 static inline u32 xp_aligned_extract_idx(struct xsk_buff_pool *pool, u64 addr)
225 {
226 	return xp_aligned_extract_addr(pool, addr) >> pool->chunk_shift;
227 }
228 
229 static inline void xp_release(struct xdp_buff_xsk *xskb)
230 {
231 	if (xskb->pool->unaligned)
232 		xskb->pool->free_heads[xskb->pool->free_heads_cnt++] = xskb;
233 }
234 
235 static inline u64 xp_get_handle(struct xdp_buff_xsk *xskb,
236 				struct xsk_buff_pool *pool)
237 {
238 	u64 orig_addr = xskb->xdp.data - pool->addrs;
239 	u64 offset;
240 
241 	if (!pool->unaligned)
242 		return orig_addr;
243 
244 	offset = xskb->xdp.data - xskb->xdp.data_hard_start;
245 	offset += pool->headroom;
246 	orig_addr -= offset;
247 	return orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT);
248 }
249 
250 static inline bool xp_tx_metadata_enabled(const struct xsk_buff_pool *pool)
251 {
252 	return pool->tx_metadata_len > 0;
253 }
254 
255 #endif /* XSK_BUFF_POOL_H_ */
256