xref: /linux/net/xdp/xsk_queue.h (revision 3286f88f31da060ac2789cee247153961ba57e49)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* XDP user-space ring structure
3  * Copyright(c) 2018 Intel Corporation.
4  */
5 
6 #ifndef _LINUX_XSK_QUEUE_H
7 #define _LINUX_XSK_QUEUE_H
8 
9 #include <linux/types.h>
10 #include <linux/if_xdp.h>
11 #include <net/xdp_sock.h>
12 #include <net/xsk_buff_pool.h>
13 
14 #include "xsk.h"
15 
16 struct xdp_ring {
17 	u32 producer ____cacheline_aligned_in_smp;
18 	/* Hinder the adjacent cache prefetcher to prefetch the consumer
19 	 * pointer if the producer pointer is touched and vice versa.
20 	 */
21 	u32 pad1 ____cacheline_aligned_in_smp;
22 	u32 consumer ____cacheline_aligned_in_smp;
23 	u32 pad2 ____cacheline_aligned_in_smp;
24 	u32 flags;
25 	u32 pad3 ____cacheline_aligned_in_smp;
26 };
27 
28 /* Used for the RX and TX queues for packets */
29 struct xdp_rxtx_ring {
30 	struct xdp_ring ptrs;
31 	struct xdp_desc desc[] ____cacheline_aligned_in_smp;
32 };
33 
34 /* Used for the fill and completion queues for buffers */
35 struct xdp_umem_ring {
36 	struct xdp_ring ptrs;
37 	u64 desc[] ____cacheline_aligned_in_smp;
38 };
39 
40 struct xsk_queue {
41 	u32 ring_mask;
42 	u32 nentries;
43 	u32 cached_prod;
44 	u32 cached_cons;
45 	struct xdp_ring *ring;
46 	u64 invalid_descs;
47 	u64 queue_empty_descs;
48 	size_t ring_vmalloc_size;
49 };
50 
51 /* The structure of the shared state of the rings are a simple
52  * circular buffer, as outlined in
53  * Documentation/core-api/circular-buffers.rst. For the Rx and
54  * completion ring, the kernel is the producer and user space is the
55  * consumer. For the Tx and fill rings, the kernel is the consumer and
56  * user space is the producer.
57  *
58  * producer                         consumer
59  *
60  * if (LOAD ->consumer) {  (A)      LOAD.acq ->producer  (C)
61  *    STORE $data                   LOAD $data
62  *    STORE.rel ->producer (B)      STORE.rel ->consumer (D)
63  * }
64  *
65  * (A) pairs with (D), and (B) pairs with (C).
66  *
67  * Starting with (B), it protects the data from being written after
68  * the producer pointer. If this barrier was missing, the consumer
69  * could observe the producer pointer being set and thus load the data
70  * before the producer has written the new data. The consumer would in
71  * this case load the old data.
72  *
73  * (C) protects the consumer from speculatively loading the data before
74  * the producer pointer actually has been read. If we do not have this
75  * barrier, some architectures could load old data as speculative loads
76  * are not discarded as the CPU does not know there is a dependency
77  * between ->producer and data.
78  *
79  * (A) is a control dependency that separates the load of ->consumer
80  * from the stores of $data. In case ->consumer indicates there is no
81  * room in the buffer to store $data we do not. The dependency will
82  * order both of the stores after the loads. So no barrier is needed.
83  *
84  * (D) protects the load of the data to be observed to happen after the
85  * store of the consumer pointer. If we did not have this memory
86  * barrier, the producer could observe the consumer pointer being set
87  * and overwrite the data with a new value before the consumer got the
88  * chance to read the old value. The consumer would thus miss reading
89  * the old entry and very likely read the new entry twice, once right
90  * now and again after circling through the ring.
91  */
92 
93 /* The operations on the rings are the following:
94  *
95  * producer                           consumer
96  *
97  * RESERVE entries                    PEEK in the ring for entries
98  * WRITE data into the ring           READ data from the ring
99  * SUBMIT entries                     RELEASE entries
100  *
101  * The producer reserves one or more entries in the ring. It can then
102  * fill in these entries and finally submit them so that they can be
103  * seen and read by the consumer.
104  *
105  * The consumer peeks into the ring to see if the producer has written
106  * any new entries. If so, the consumer can then read these entries
107  * and when it is done reading them release them back to the producer
108  * so that the producer can use these slots to fill in new entries.
109  *
110  * The function names below reflect these operations.
111  */
112 
113 /* Functions that read and validate content from consumer rings. */
114 
115 static inline void __xskq_cons_read_addr_unchecked(struct xsk_queue *q, u32 cached_cons, u64 *addr)
116 {
117 	struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring;
118 	u32 idx = cached_cons & q->ring_mask;
119 
120 	*addr = ring->desc[idx];
121 }
122 
123 static inline bool xskq_cons_read_addr_unchecked(struct xsk_queue *q, u64 *addr)
124 {
125 	if (q->cached_cons != q->cached_prod) {
126 		__xskq_cons_read_addr_unchecked(q, q->cached_cons, addr);
127 		return true;
128 	}
129 
130 	return false;
131 }
132 
133 static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool,
134 					    struct xdp_desc *desc)
135 {
136 	u64 offset = desc->addr & (pool->chunk_size - 1);
137 
138 	if (offset + desc->len > pool->chunk_size)
139 		return false;
140 
141 	if (desc->addr >= pool->addrs_cnt)
142 		return false;
143 
144 	if (desc->options)
145 		return false;
146 	return true;
147 }
148 
149 static inline bool xp_unaligned_validate_desc(struct xsk_buff_pool *pool,
150 					      struct xdp_desc *desc)
151 {
152 	u64 addr = xp_unaligned_add_offset_to_addr(desc->addr);
153 
154 	if (desc->len > pool->chunk_size)
155 		return false;
156 
157 	if (addr >= pool->addrs_cnt || addr + desc->len > pool->addrs_cnt ||
158 	    xp_desc_crosses_non_contig_pg(pool, addr, desc->len))
159 		return false;
160 
161 	if (desc->options)
162 		return false;
163 	return true;
164 }
165 
166 static inline bool xp_validate_desc(struct xsk_buff_pool *pool,
167 				    struct xdp_desc *desc)
168 {
169 	return pool->unaligned ? xp_unaligned_validate_desc(pool, desc) :
170 		xp_aligned_validate_desc(pool, desc);
171 }
172 
173 static inline bool xskq_cons_is_valid_desc(struct xsk_queue *q,
174 					   struct xdp_desc *d,
175 					   struct xsk_buff_pool *pool)
176 {
177 	if (!xp_validate_desc(pool, d)) {
178 		q->invalid_descs++;
179 		return false;
180 	}
181 	return true;
182 }
183 
184 static inline bool xskq_cons_read_desc(struct xsk_queue *q,
185 				       struct xdp_desc *desc,
186 				       struct xsk_buff_pool *pool)
187 {
188 	while (q->cached_cons != q->cached_prod) {
189 		struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring;
190 		u32 idx = q->cached_cons & q->ring_mask;
191 
192 		*desc = ring->desc[idx];
193 		if (xskq_cons_is_valid_desc(q, desc, pool))
194 			return true;
195 
196 		q->cached_cons++;
197 	}
198 
199 	return false;
200 }
201 
202 static inline void xskq_cons_release_n(struct xsk_queue *q, u32 cnt)
203 {
204 	q->cached_cons += cnt;
205 }
206 
207 static inline u32 xskq_cons_read_desc_batch(struct xsk_queue *q, struct xsk_buff_pool *pool,
208 					    u32 max)
209 {
210 	u32 cached_cons = q->cached_cons, nb_entries = 0;
211 	struct xdp_desc *descs = pool->tx_descs;
212 
213 	while (cached_cons != q->cached_prod && nb_entries < max) {
214 		struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring;
215 		u32 idx = cached_cons & q->ring_mask;
216 
217 		descs[nb_entries] = ring->desc[idx];
218 		if (unlikely(!xskq_cons_is_valid_desc(q, &descs[nb_entries], pool))) {
219 			/* Skip the entry */
220 			cached_cons++;
221 			continue;
222 		}
223 
224 		nb_entries++;
225 		cached_cons++;
226 	}
227 
228 	/* Release valid plus any invalid entries */
229 	xskq_cons_release_n(q, cached_cons - q->cached_cons);
230 	return nb_entries;
231 }
232 
233 /* Functions for consumers */
234 
235 static inline void __xskq_cons_release(struct xsk_queue *q)
236 {
237 	smp_store_release(&q->ring->consumer, q->cached_cons); /* D, matchees A */
238 }
239 
240 static inline void __xskq_cons_peek(struct xsk_queue *q)
241 {
242 	/* Refresh the local pointer */
243 	q->cached_prod = smp_load_acquire(&q->ring->producer);  /* C, matches B */
244 }
245 
246 static inline void xskq_cons_get_entries(struct xsk_queue *q)
247 {
248 	__xskq_cons_release(q);
249 	__xskq_cons_peek(q);
250 }
251 
252 static inline u32 xskq_cons_nb_entries(struct xsk_queue *q, u32 max)
253 {
254 	u32 entries = q->cached_prod - q->cached_cons;
255 
256 	if (entries >= max)
257 		return max;
258 
259 	__xskq_cons_peek(q);
260 	entries = q->cached_prod - q->cached_cons;
261 
262 	return entries >= max ? max : entries;
263 }
264 
265 static inline bool xskq_cons_has_entries(struct xsk_queue *q, u32 cnt)
266 {
267 	return xskq_cons_nb_entries(q, cnt) >= cnt;
268 }
269 
270 static inline bool xskq_cons_peek_addr_unchecked(struct xsk_queue *q, u64 *addr)
271 {
272 	if (q->cached_prod == q->cached_cons)
273 		xskq_cons_get_entries(q);
274 	return xskq_cons_read_addr_unchecked(q, addr);
275 }
276 
277 static inline bool xskq_cons_peek_desc(struct xsk_queue *q,
278 				       struct xdp_desc *desc,
279 				       struct xsk_buff_pool *pool)
280 {
281 	if (q->cached_prod == q->cached_cons)
282 		xskq_cons_get_entries(q);
283 	return xskq_cons_read_desc(q, desc, pool);
284 }
285 
286 /* To improve performance in the xskq_cons_release functions, only update local state here.
287  * Reflect this to global state when we get new entries from the ring in
288  * xskq_cons_get_entries() and whenever Rx or Tx processing are completed in the NAPI loop.
289  */
290 static inline void xskq_cons_release(struct xsk_queue *q)
291 {
292 	q->cached_cons++;
293 }
294 
295 static inline u32 xskq_cons_present_entries(struct xsk_queue *q)
296 {
297 	/* No barriers needed since data is not accessed */
298 	return READ_ONCE(q->ring->producer) - READ_ONCE(q->ring->consumer);
299 }
300 
301 /* Functions for producers */
302 
303 static inline u32 xskq_prod_nb_free(struct xsk_queue *q, u32 max)
304 {
305 	u32 free_entries = q->nentries - (q->cached_prod - q->cached_cons);
306 
307 	if (free_entries >= max)
308 		return max;
309 
310 	/* Refresh the local tail pointer */
311 	q->cached_cons = READ_ONCE(q->ring->consumer);
312 	free_entries = q->nentries - (q->cached_prod - q->cached_cons);
313 
314 	return free_entries >= max ? max : free_entries;
315 }
316 
317 static inline bool xskq_prod_is_full(struct xsk_queue *q)
318 {
319 	return xskq_prod_nb_free(q, 1) ? false : true;
320 }
321 
322 static inline void xskq_prod_cancel(struct xsk_queue *q)
323 {
324 	q->cached_prod--;
325 }
326 
327 static inline int xskq_prod_reserve(struct xsk_queue *q)
328 {
329 	if (xskq_prod_is_full(q))
330 		return -ENOSPC;
331 
332 	/* A, matches D */
333 	q->cached_prod++;
334 	return 0;
335 }
336 
337 static inline int xskq_prod_reserve_addr(struct xsk_queue *q, u64 addr)
338 {
339 	struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring;
340 
341 	if (xskq_prod_is_full(q))
342 		return -ENOSPC;
343 
344 	/* A, matches D */
345 	ring->desc[q->cached_prod++ & q->ring_mask] = addr;
346 	return 0;
347 }
348 
349 static inline void xskq_prod_write_addr_batch(struct xsk_queue *q, struct xdp_desc *descs,
350 					      u32 nb_entries)
351 {
352 	struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring;
353 	u32 i, cached_prod;
354 
355 	/* A, matches D */
356 	cached_prod = q->cached_prod;
357 	for (i = 0; i < nb_entries; i++)
358 		ring->desc[cached_prod++ & q->ring_mask] = descs[i].addr;
359 	q->cached_prod = cached_prod;
360 }
361 
362 static inline int xskq_prod_reserve_desc(struct xsk_queue *q,
363 					 u64 addr, u32 len)
364 {
365 	struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring;
366 	u32 idx;
367 
368 	if (xskq_prod_is_full(q))
369 		return -ENOBUFS;
370 
371 	/* A, matches D */
372 	idx = q->cached_prod++ & q->ring_mask;
373 	ring->desc[idx].addr = addr;
374 	ring->desc[idx].len = len;
375 
376 	return 0;
377 }
378 
379 static inline void __xskq_prod_submit(struct xsk_queue *q, u32 idx)
380 {
381 	smp_store_release(&q->ring->producer, idx); /* B, matches C */
382 }
383 
384 static inline void xskq_prod_submit(struct xsk_queue *q)
385 {
386 	__xskq_prod_submit(q, q->cached_prod);
387 }
388 
389 static inline void xskq_prod_submit_addr(struct xsk_queue *q, u64 addr)
390 {
391 	struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring;
392 	u32 idx = q->ring->producer;
393 
394 	ring->desc[idx++ & q->ring_mask] = addr;
395 
396 	__xskq_prod_submit(q, idx);
397 }
398 
399 static inline void xskq_prod_submit_n(struct xsk_queue *q, u32 nb_entries)
400 {
401 	__xskq_prod_submit(q, q->ring->producer + nb_entries);
402 }
403 
404 static inline bool xskq_prod_is_empty(struct xsk_queue *q)
405 {
406 	/* No barriers needed since data is not accessed */
407 	return READ_ONCE(q->ring->consumer) == READ_ONCE(q->ring->producer);
408 }
409 
410 /* For both producers and consumers */
411 
412 static inline u64 xskq_nb_invalid_descs(struct xsk_queue *q)
413 {
414 	return q ? q->invalid_descs : 0;
415 }
416 
417 static inline u64 xskq_nb_queue_empty_descs(struct xsk_queue *q)
418 {
419 	return q ? q->queue_empty_descs : 0;
420 }
421 
422 struct xsk_queue *xskq_create(u32 nentries, bool umem_queue);
423 void xskq_destroy(struct xsk_queue *q_ops);
424 
425 #endif /* _LINUX_XSK_QUEUE_H */
426