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 <net/xdp.h>
11
12struct xsk_buff_pool;
13struct xdp_rxq_info;
14struct xsk_queue;
15struct xdp_desc;
16struct xdp_umem;
17struct xdp_sock;
18struct device;
19struct page;
20
21struct xdp_buff_xsk {
22	struct xdp_buff xdp;
23	dma_addr_t dma;
24	dma_addr_t frame_dma;
25	struct xsk_buff_pool *pool;
26	bool unaligned;
27	u64 orig_addr;
28	struct list_head free_list_node;
29};
30
31struct xsk_dma_map {
32	dma_addr_t *dma_pages;
33	struct device *dev;
34	struct net_device *netdev;
35	refcount_t users;
36	struct list_head list; /* Protected by the RTNL_LOCK */
37	u32 dma_pages_cnt;
38	bool dma_need_sync;
39};
40
41struct xsk_buff_pool {
42	/* Members only used in the control path first. */
43	struct device *dev;
44	struct net_device *netdev;
45	struct list_head xsk_tx_list;
46	/* Protects modifications to the xsk_tx_list */
47	spinlock_t xsk_tx_list_lock;
48	refcount_t users;
49	struct xdp_umem *umem;
50	struct work_struct work;
51	struct list_head free_list;
52	u32 heads_cnt;
53	u16 queue_id;
54
55	/* Data path members as close to free_heads at the end as possible. */
56	struct xsk_queue *fq ____cacheline_aligned_in_smp;
57	struct xsk_queue *cq;
58	/* For performance reasons, each buff pool has its own array of dma_pages
59	 * even when they are identical.
60	 */
61	dma_addr_t *dma_pages;
62	struct xdp_buff_xsk *heads;
63	u64 chunk_mask;
64	u64 addrs_cnt;
65	u32 free_list_cnt;
66	u32 dma_pages_cnt;
67	u32 free_heads_cnt;
68	u32 headroom;
69	u32 chunk_size;
70	u32 frame_len;
71	u8 cached_need_wakeup;
72	bool uses_need_wakeup;
73	bool dma_need_sync;
74	bool unaligned;
75	void *addrs;
76	/* Mutual exclusion of the completion ring in the SKB mode. Two cases to protect:
77	 * NAPI TX thread and sendmsg error paths in the SKB destructor callback and when
78	 * sockets share a single cq when the same netdev and queue id is shared.
79	 */
80	spinlock_t cq_lock;
81	struct xdp_buff_xsk *free_heads[];
82};
83
84/* AF_XDP core. */
85struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
86						struct xdp_umem *umem);
87int xp_assign_dev(struct xsk_buff_pool *pool, struct net_device *dev,
88		  u16 queue_id, u16 flags);
89int xp_assign_dev_shared(struct xsk_buff_pool *pool, struct xdp_sock *umem_xs,
90			 struct net_device *dev, u16 queue_id);
91void xp_destroy(struct xsk_buff_pool *pool);
92void xp_release(struct xdp_buff_xsk *xskb);
93void xp_get_pool(struct xsk_buff_pool *pool);
94bool xp_put_pool(struct xsk_buff_pool *pool);
95void xp_clear_dev(struct xsk_buff_pool *pool);
96void xp_add_xsk(struct xsk_buff_pool *pool, struct xdp_sock *xs);
97void xp_del_xsk(struct xsk_buff_pool *pool, struct xdp_sock *xs);
98
99/* AF_XDP, and XDP core. */
100void xp_free(struct xdp_buff_xsk *xskb);
101
102/* AF_XDP ZC drivers, via xdp_sock_buff.h */
103void xp_set_rxq_info(struct xsk_buff_pool *pool, struct xdp_rxq_info *rxq);
104int xp_dma_map(struct xsk_buff_pool *pool, struct device *dev,
105	       unsigned long attrs, struct page **pages, u32 nr_pages);
106void xp_dma_unmap(struct xsk_buff_pool *pool, unsigned long attrs);
107struct xdp_buff *xp_alloc(struct xsk_buff_pool *pool);
108bool xp_can_alloc(struct xsk_buff_pool *pool, u32 count);
109void *xp_raw_get_data(struct xsk_buff_pool *pool, u64 addr);
110dma_addr_t xp_raw_get_dma(struct xsk_buff_pool *pool, u64 addr);
111static inline dma_addr_t xp_get_dma(struct xdp_buff_xsk *xskb)
112{
113	return xskb->dma;
114}
115
116static inline dma_addr_t xp_get_frame_dma(struct xdp_buff_xsk *xskb)
117{
118	return xskb->frame_dma;
119}
120
121void xp_dma_sync_for_cpu_slow(struct xdp_buff_xsk *xskb);
122static inline void xp_dma_sync_for_cpu(struct xdp_buff_xsk *xskb)
123{
124	xp_dma_sync_for_cpu_slow(xskb);
125}
126
127void xp_dma_sync_for_device_slow(struct xsk_buff_pool *pool, dma_addr_t dma,
128				 size_t size);
129static inline void xp_dma_sync_for_device(struct xsk_buff_pool *pool,
130					  dma_addr_t dma, size_t size)
131{
132	if (!pool->dma_need_sync)
133		return;
134
135	xp_dma_sync_for_device_slow(pool, dma, size);
136}
137
138/* Masks for xdp_umem_page flags.
139 * The low 12-bits of the addr will be 0 since this is the page address, so we
140 * can use them for flags.
141 */
142#define XSK_NEXT_PG_CONTIG_SHIFT 0
143#define XSK_NEXT_PG_CONTIG_MASK BIT_ULL(XSK_NEXT_PG_CONTIG_SHIFT)
144
145static inline bool xp_desc_crosses_non_contig_pg(struct xsk_buff_pool *pool,
146						 u64 addr, u32 len)
147{
148	bool cross_pg = (addr & (PAGE_SIZE - 1)) + len > PAGE_SIZE;
149
150	if (likely(!cross_pg))
151		return false;
152
153	return pool->dma_pages_cnt &&
154	       !(pool->dma_pages[addr >> PAGE_SHIFT] & XSK_NEXT_PG_CONTIG_MASK);
155}
156
157static inline u64 xp_aligned_extract_addr(struct xsk_buff_pool *pool, u64 addr)
158{
159	return addr & pool->chunk_mask;
160}
161
162static inline u64 xp_unaligned_extract_addr(u64 addr)
163{
164	return addr & XSK_UNALIGNED_BUF_ADDR_MASK;
165}
166
167static inline u64 xp_unaligned_extract_offset(u64 addr)
168{
169	return addr >> XSK_UNALIGNED_BUF_OFFSET_SHIFT;
170}
171
172static inline u64 xp_unaligned_add_offset_to_addr(u64 addr)
173{
174	return xp_unaligned_extract_addr(addr) +
175		xp_unaligned_extract_offset(addr);
176}
177
178#endif /* XSK_BUFF_POOL_H_ */
179