xref: /dpdk/lib/vhost/vhost.h (revision c9902a15bd005b6d4fe072cf7b60fe4ee679155f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2018 Intel Corporation
3  */
4 
5 #ifndef _VHOST_NET_CDEV_H_
6 #define _VHOST_NET_CDEV_H_
7 #include <stdint.h>
8 #include <stdio.h>
9 #include <stdbool.h>
10 #include <sys/types.h>
11 #include <sys/queue.h>
12 #include <unistd.h>
13 #include <linux/vhost.h>
14 #include <linux/virtio_net.h>
15 #include <sys/socket.h>
16 #include <linux/if.h>
17 
18 #include <rte_log.h>
19 #include <rte_ether.h>
20 #include <rte_rwlock.h>
21 #include <rte_malloc.h>
22 
23 #include "rte_vhost.h"
24 #include "rte_vdpa.h"
25 #include "rte_vdpa_dev.h"
26 
27 #include "rte_vhost_async.h"
28 
29 /* Used to indicate that the device is running on a data core */
30 #define VIRTIO_DEV_RUNNING ((uint32_t)1 << 0)
31 /* Used to indicate that the device is ready to operate */
32 #define VIRTIO_DEV_READY ((uint32_t)1 << 1)
33 /* Used to indicate that the built-in vhost net device backend is enabled */
34 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET ((uint32_t)1 << 2)
35 /* Used to indicate that the device has its own data path and configured */
36 #define VIRTIO_DEV_VDPA_CONFIGURED ((uint32_t)1 << 3)
37 /* Used to indicate that the feature negotiation failed */
38 #define VIRTIO_DEV_FEATURES_FAILED ((uint32_t)1 << 4)
39 /* Used to indicate that the virtio_net tx code should fill TX ol_flags */
40 #define VIRTIO_DEV_LEGACY_OL_FLAGS ((uint32_t)1 << 5)
41 
42 /* Backend value set by guest. */
43 #define VIRTIO_DEV_STOPPED -1
44 
45 #define BUF_VECTOR_MAX 256
46 
47 #define VHOST_LOG_CACHE_NR 32
48 
49 #define MAX_PKT_BURST 32
50 
51 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
52 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 4)
53 
54 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)	\
55 	((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
56 		VRING_DESC_F_WRITE)
57 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)	\
58 	((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
59 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
60 					 VRING_DESC_F_INDIRECT)
61 
62 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
63 			    sizeof(struct vring_packed_desc))
64 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
65 
66 #ifdef VHOST_GCC_UNROLL_PRAGMA
67 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
68 	for (iter = val; iter < size; iter++)
69 #endif
70 
71 #ifdef VHOST_CLANG_UNROLL_PRAGMA
72 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
73 	for (iter = val; iter < size; iter++)
74 #endif
75 
76 #ifdef VHOST_ICC_UNROLL_PRAGMA
77 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
78 	for (iter = val; iter < size; iter++)
79 #endif
80 
81 #ifndef vhost_for_each_try_unroll
82 #define vhost_for_each_try_unroll(iter, val, num) \
83 	for (iter = val; iter < num; iter++)
84 #endif
85 
86 /**
87  * Structure contains buffer address, length and descriptor index
88  * from vring to do scatter RX.
89  */
90 struct buf_vector {
91 	uint64_t buf_iova;
92 	uint64_t buf_addr;
93 	uint32_t buf_len;
94 	uint32_t desc_idx;
95 };
96 
97 /*
98  * Structure contains the info for each batched memory copy.
99  */
100 struct batch_copy_elem {
101 	void *dst;
102 	void *src;
103 	uint32_t len;
104 	uint64_t log_addr;
105 };
106 
107 /*
108  * Structure that contains the info for batched dirty logging.
109  */
110 struct log_cache_entry {
111 	uint32_t offset;
112 	unsigned long val;
113 };
114 
115 struct vring_used_elem_packed {
116 	uint16_t id;
117 	uint16_t flags;
118 	uint32_t len;
119 	uint32_t count;
120 };
121 
122 /**
123  * Structure contains variables relevant to RX/TX virtqueues.
124  */
125 struct vhost_virtqueue {
126 	union {
127 		struct vring_desc	*desc;
128 		struct vring_packed_desc   *desc_packed;
129 	};
130 	union {
131 		struct vring_avail	*avail;
132 		struct vring_packed_desc_event *driver_event;
133 	};
134 	union {
135 		struct vring_used	*used;
136 		struct vring_packed_desc_event *device_event;
137 	};
138 	uint16_t		size;
139 
140 	uint16_t		last_avail_idx;
141 	uint16_t		last_used_idx;
142 	/* Last used index we notify to front end. */
143 	uint16_t		signalled_used;
144 	bool			signalled_used_valid;
145 #define VIRTIO_INVALID_EVENTFD		(-1)
146 #define VIRTIO_UNINITIALIZED_EVENTFD	(-2)
147 
148 	bool			enabled;
149 	bool			access_ok;
150 	bool			ready;
151 
152 	rte_spinlock_t		access_lock;
153 
154 
155 	union {
156 		struct vring_used_elem  *shadow_used_split;
157 		struct vring_used_elem_packed *shadow_used_packed;
158 	};
159 	uint16_t                shadow_used_idx;
160 	/* Record packed ring enqueue latest desc cache aligned index */
161 	uint16_t		shadow_aligned_idx;
162 	/* Record packed ring first dequeue desc index */
163 	uint16_t		shadow_last_used_idx;
164 
165 	uint16_t		batch_copy_nb_elems;
166 	struct batch_copy_elem	*batch_copy_elems;
167 	int			numa_node;
168 	bool			used_wrap_counter;
169 	bool			avail_wrap_counter;
170 
171 	/* Physical address of used ring, for logging */
172 	uint16_t		log_cache_nb_elem;
173 	uint64_t		log_guest_addr;
174 	struct log_cache_entry	*log_cache;
175 
176 	rte_rwlock_t	iotlb_lock;
177 	rte_rwlock_t	iotlb_pending_lock;
178 	struct rte_mempool *iotlb_pool;
179 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
180 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
181 	int				iotlb_cache_nr;
182 
183 	/* Used to notify the guest (trigger interrupt) */
184 	int			callfd;
185 	/* Currently unused as polling mode is enabled */
186 	int			kickfd;
187 
188 	/* inflight share memory info */
189 	union {
190 		struct rte_vhost_inflight_info_split *inflight_split;
191 		struct rte_vhost_inflight_info_packed *inflight_packed;
192 	};
193 	struct rte_vhost_resubmit_info *resubmit_inflight;
194 	uint64_t		global_counter;
195 
196 	/* operation callbacks for async dma */
197 	struct rte_vhost_async_channel_ops	async_ops;
198 
199 	struct rte_vhost_iov_iter *it_pool;
200 	struct iovec *vec_pool;
201 
202 	/* async data transfer status */
203 	struct async_inflight_info *async_pkts_info;
204 	uint16_t	async_pkts_idx;
205 	uint16_t	async_pkts_inflight_n;
206 	uint16_t	async_last_pkts_n;
207 	union {
208 		struct vring_used_elem  *async_descs_split;
209 		struct vring_used_elem_packed *async_buffers_packed;
210 	};
211 	union {
212 		uint16_t async_desc_idx_split;
213 		uint16_t async_buffer_idx_packed;
214 	};
215 	union {
216 		uint16_t last_async_desc_idx_split;
217 		uint16_t last_async_buffer_idx_packed;
218 	};
219 
220 	/* vq async features */
221 	bool		async_registered;
222 
223 	int			notif_enable;
224 #define VIRTIO_UNINITIALIZED_NOTIF	(-1)
225 
226 	struct vhost_vring_addr ring_addrs;
227 } __rte_cache_aligned;
228 
229 /* Virtio device status as per Virtio specification */
230 #define VIRTIO_DEVICE_STATUS_RESET		0x00
231 #define VIRTIO_DEVICE_STATUS_ACK		0x01
232 #define VIRTIO_DEVICE_STATUS_DRIVER		0x02
233 #define VIRTIO_DEVICE_STATUS_DRIVER_OK		0x04
234 #define VIRTIO_DEVICE_STATUS_FEATURES_OK	0x08
235 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET	0x40
236 #define VIRTIO_DEVICE_STATUS_FAILED		0x80
237 
238 #define VHOST_MAX_VRING			0x100
239 #define VHOST_MAX_QUEUE_PAIRS		0x80
240 
241 /* Declare IOMMU related bits for older kernels */
242 #ifndef VIRTIO_F_IOMMU_PLATFORM
243 
244 #define VIRTIO_F_IOMMU_PLATFORM 33
245 
246 struct vhost_iotlb_msg {
247 	__u64 iova;
248 	__u64 size;
249 	__u64 uaddr;
250 #define VHOST_ACCESS_RO      0x1
251 #define VHOST_ACCESS_WO      0x2
252 #define VHOST_ACCESS_RW      0x3
253 	__u8 perm;
254 #define VHOST_IOTLB_MISS           1
255 #define VHOST_IOTLB_UPDATE         2
256 #define VHOST_IOTLB_INVALIDATE     3
257 #define VHOST_IOTLB_ACCESS_FAIL    4
258 	__u8 type;
259 };
260 
261 #define VHOST_IOTLB_MSG 0x1
262 
263 struct vhost_msg {
264 	int type;
265 	union {
266 		struct vhost_iotlb_msg iotlb;
267 		__u8 padding[64];
268 	};
269 };
270 #endif
271 
272 /*
273  * Define virtio 1.0 for older kernels
274  */
275 #ifndef VIRTIO_F_VERSION_1
276  #define VIRTIO_F_VERSION_1 32
277 #endif
278 
279 /* Declare packed ring related bits for older kernels */
280 #ifndef VIRTIO_F_RING_PACKED
281 
282 #define VIRTIO_F_RING_PACKED 34
283 
284 struct vring_packed_desc {
285 	uint64_t addr;
286 	uint32_t len;
287 	uint16_t id;
288 	uint16_t flags;
289 };
290 
291 struct vring_packed_desc_event {
292 	uint16_t off_wrap;
293 	uint16_t flags;
294 };
295 #endif
296 
297 /*
298  * Declare below packed ring defines unconditionally
299  * as Kernel header might use different names.
300  */
301 #define VRING_DESC_F_AVAIL	(1ULL << 7)
302 #define VRING_DESC_F_USED	(1ULL << 15)
303 
304 #define VRING_EVENT_F_ENABLE 0x0
305 #define VRING_EVENT_F_DISABLE 0x1
306 #define VRING_EVENT_F_DESC 0x2
307 
308 /*
309  * Available and used descs are in same order
310  */
311 #ifndef VIRTIO_F_IN_ORDER
312 #define VIRTIO_F_IN_ORDER      35
313 #endif
314 
315 /* Features supported by this builtin vhost-user net driver. */
316 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
317 				(1ULL << VIRTIO_F_ANY_LAYOUT) | \
318 				(1ULL << VIRTIO_NET_F_CTRL_VQ) | \
319 				(1ULL << VIRTIO_NET_F_CTRL_RX) | \
320 				(1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
321 				(1ULL << VIRTIO_NET_F_MQ)      | \
322 				(1ULL << VIRTIO_F_VERSION_1)   | \
323 				(1ULL << VHOST_F_LOG_ALL)      | \
324 				(1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
325 				(1ULL << VIRTIO_NET_F_GSO) | \
326 				(1ULL << VIRTIO_NET_F_HOST_TSO4) | \
327 				(1ULL << VIRTIO_NET_F_HOST_TSO6) | \
328 				(1ULL << VIRTIO_NET_F_HOST_UFO) | \
329 				(1ULL << VIRTIO_NET_F_HOST_ECN) | \
330 				(1ULL << VIRTIO_NET_F_CSUM)    | \
331 				(1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
332 				(1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
333 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
334 				(1ULL << VIRTIO_NET_F_GUEST_UFO) | \
335 				(1ULL << VIRTIO_NET_F_GUEST_ECN) | \
336 				(1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
337 				(1ULL << VIRTIO_RING_F_EVENT_IDX) | \
338 				(1ULL << VIRTIO_NET_F_MTU)  | \
339 				(1ULL << VIRTIO_F_IN_ORDER) | \
340 				(1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
341 				(1ULL << VIRTIO_F_RING_PACKED))
342 
343 
344 struct guest_page {
345 	uint64_t guest_phys_addr;
346 	uint64_t host_phys_addr;
347 	uint64_t size;
348 };
349 
350 struct inflight_mem_info {
351 	int		fd;
352 	void		*addr;
353 	uint64_t	size;
354 };
355 
356 /**
357  * Device structure contains all configuration information relating
358  * to the device.
359  */
360 struct virtio_net {
361 	/* Frontend (QEMU) memory and memory region information */
362 	struct rte_vhost_memory	*mem;
363 	uint64_t		features;
364 	uint64_t		protocol_features;
365 	int			vid;
366 	uint32_t		flags;
367 	uint16_t		vhost_hlen;
368 	/* to tell if we need broadcast rarp packet */
369 	int16_t			broadcast_rarp;
370 	uint32_t		nr_vring;
371 	int			async_copy;
372 	int			extbuf;
373 	int			linearbuf;
374 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
375 	struct inflight_mem_info *inflight_info;
376 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
377 	char			ifname[IF_NAME_SZ];
378 	uint64_t		log_size;
379 	uint64_t		log_base;
380 	uint64_t		log_addr;
381 	struct rte_ether_addr	mac;
382 	uint16_t		mtu;
383 	uint8_t			status;
384 
385 	struct vhost_device_ops const *notify_ops;
386 
387 	uint32_t		nr_guest_pages;
388 	uint32_t		max_guest_pages;
389 	struct guest_page       *guest_pages;
390 
391 	int			slave_req_fd;
392 	rte_spinlock_t		slave_req_lock;
393 
394 	int			postcopy_ufd;
395 	int			postcopy_listening;
396 
397 	struct rte_vdpa_device *vdpa_dev;
398 
399 	/* context data for the external message handlers */
400 	void			*extern_data;
401 	/* pre and post vhost user message handlers for the device */
402 	struct rte_vhost_user_extern_ops extern_ops;
403 } __rte_cache_aligned;
404 
405 static __rte_always_inline bool
406 vq_is_packed(struct virtio_net *dev)
407 {
408 	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
409 }
410 
411 static inline bool
412 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
413 {
414 	uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
415 
416 	return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
417 		wrap_counter != !!(flags & VRING_DESC_F_USED);
418 }
419 
420 static inline void
421 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
422 {
423 	vq->last_used_idx += num;
424 	if (vq->last_used_idx >= vq->size) {
425 		vq->used_wrap_counter ^= 1;
426 		vq->last_used_idx -= vq->size;
427 	}
428 }
429 
430 static inline void
431 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
432 {
433 	vq->last_avail_idx += num;
434 	if (vq->last_avail_idx >= vq->size) {
435 		vq->avail_wrap_counter ^= 1;
436 		vq->last_avail_idx -= vq->size;
437 	}
438 }
439 
440 void __vhost_log_cache_write(struct virtio_net *dev,
441 		struct vhost_virtqueue *vq,
442 		uint64_t addr, uint64_t len);
443 void __vhost_log_cache_write_iova(struct virtio_net *dev,
444 		struct vhost_virtqueue *vq,
445 		uint64_t iova, uint64_t len);
446 void __vhost_log_cache_sync(struct virtio_net *dev,
447 		struct vhost_virtqueue *vq);
448 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
449 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
450 			    uint64_t iova, uint64_t len);
451 
452 static __rte_always_inline void
453 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
454 {
455 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
456 		__vhost_log_write(dev, addr, len);
457 }
458 
459 static __rte_always_inline void
460 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
461 {
462 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
463 		__vhost_log_cache_sync(dev, vq);
464 }
465 
466 static __rte_always_inline void
467 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
468 			uint64_t addr, uint64_t len)
469 {
470 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
471 		__vhost_log_cache_write(dev, vq, addr, len);
472 }
473 
474 static __rte_always_inline void
475 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
476 			uint64_t offset, uint64_t len)
477 {
478 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
479 		if (unlikely(vq->log_guest_addr == 0))
480 			return;
481 		__vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
482 					len);
483 	}
484 }
485 
486 static __rte_always_inline void
487 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
488 		     uint64_t offset, uint64_t len)
489 {
490 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
491 		if (unlikely(vq->log_guest_addr == 0))
492 			return;
493 		__vhost_log_write(dev, vq->log_guest_addr + offset, len);
494 	}
495 }
496 
497 static __rte_always_inline void
498 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
499 			   uint64_t iova, uint64_t len)
500 {
501 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
502 		return;
503 
504 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
505 		__vhost_log_cache_write_iova(dev, vq, iova, len);
506 	else
507 		__vhost_log_cache_write(dev, vq, iova, len);
508 }
509 
510 static __rte_always_inline void
511 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
512 			   uint64_t iova, uint64_t len)
513 {
514 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
515 		return;
516 
517 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
518 		__vhost_log_write_iova(dev, vq, iova, len);
519 	else
520 		__vhost_log_write(dev, iova, len);
521 }
522 
523 extern int vhost_config_log_level;
524 extern int vhost_data_log_level;
525 
526 #define VHOST_LOG_CONFIG(level, fmt, args...)			\
527 	rte_log(RTE_LOG_ ## level, vhost_config_log_level,	\
528 		"VHOST_CONFIG: " fmt, ##args)
529 
530 #define VHOST_LOG_DATA(level, fmt, args...) \
531 	(void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?	\
532 	 rte_log(RTE_LOG_ ## level,  vhost_data_log_level,	\
533 		"VHOST_DATA : " fmt, ##args) :			\
534 	 0)
535 
536 #ifdef RTE_LIBRTE_VHOST_DEBUG
537 #define VHOST_MAX_PRINT_BUFF 6072
538 #define PRINT_PACKET(device, addr, size, header) do { \
539 	char *pkt_addr = (char *)(addr); \
540 	unsigned int index; \
541 	char packet[VHOST_MAX_PRINT_BUFF]; \
542 	\
543 	if ((header)) \
544 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
545 	else \
546 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
547 	for (index = 0; index < (size); index++) { \
548 		snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
549 			"%02hhx ", pkt_addr[index]); \
550 	} \
551 	snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
552 	\
553 	VHOST_LOG_DATA(DEBUG, "%s", packet); \
554 } while (0)
555 #else
556 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
557 #endif
558 
559 #define MAX_VHOST_DEVICE	1024
560 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
561 
562 #define VHOST_BINARY_SEARCH_THRESH 256
563 
564 static __rte_always_inline int guest_page_addrcmp(const void *p1,
565 						const void *p2)
566 {
567 	const struct guest_page *page1 = (const struct guest_page *)p1;
568 	const struct guest_page *page2 = (const struct guest_page *)p2;
569 
570 	if (page1->guest_phys_addr > page2->guest_phys_addr)
571 		return 1;
572 	if (page1->guest_phys_addr < page2->guest_phys_addr)
573 		return -1;
574 
575 	return 0;
576 }
577 
578 static __rte_always_inline rte_iova_t
579 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
580 	uint64_t gpa_size, uint64_t *hpa_size)
581 {
582 	uint32_t i;
583 	struct guest_page *page;
584 	struct guest_page key;
585 
586 	*hpa_size = gpa_size;
587 	if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
588 		key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
589 		page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
590 			       sizeof(struct guest_page), guest_page_addrcmp);
591 		if (page) {
592 			if (gpa + gpa_size <=
593 					page->guest_phys_addr + page->size) {
594 				return gpa - page->guest_phys_addr +
595 					page->host_phys_addr;
596 			} else if (gpa < page->guest_phys_addr +
597 						page->size) {
598 				*hpa_size = page->guest_phys_addr +
599 					page->size - gpa;
600 				return gpa - page->guest_phys_addr +
601 					page->host_phys_addr;
602 			}
603 		}
604 	} else {
605 		for (i = 0; i < dev->nr_guest_pages; i++) {
606 			page = &dev->guest_pages[i];
607 
608 			if (gpa >= page->guest_phys_addr) {
609 				if (gpa + gpa_size <=
610 					page->guest_phys_addr + page->size) {
611 					return gpa - page->guest_phys_addr +
612 						page->host_phys_addr;
613 				} else if (gpa < page->guest_phys_addr +
614 							page->size) {
615 					*hpa_size = page->guest_phys_addr +
616 						page->size - gpa;
617 					return gpa - page->guest_phys_addr +
618 						page->host_phys_addr;
619 				}
620 			}
621 		}
622 	}
623 
624 	*hpa_size = 0;
625 	return 0;
626 }
627 
628 /* Convert guest physical address to host physical address */
629 static __rte_always_inline rte_iova_t
630 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
631 {
632 	rte_iova_t hpa;
633 	uint64_t hpa_size;
634 
635 	hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
636 	return hpa_size == size ? hpa : 0;
637 }
638 
639 static __rte_always_inline uint64_t
640 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
641 {
642 	struct rte_vhost_mem_region *r;
643 	uint32_t i;
644 
645 	if (unlikely(!dev || !dev->mem))
646 		return 0;
647 
648 	for (i = 0; i < dev->mem->nregions; i++) {
649 		r = &dev->mem->regions[i];
650 
651 		if (vva >= r->host_user_addr &&
652 		    vva + len <  r->host_user_addr + r->size) {
653 			return r->guest_phys_addr + vva - r->host_user_addr;
654 		}
655 	}
656 	return 0;
657 }
658 
659 static __rte_always_inline struct virtio_net *
660 get_device(int vid)
661 {
662 	struct virtio_net *dev = vhost_devices[vid];
663 
664 	if (unlikely(!dev)) {
665 		VHOST_LOG_CONFIG(ERR,
666 			"(%d) device not found.\n", vid);
667 	}
668 
669 	return dev;
670 }
671 
672 int vhost_new_device(void);
673 void cleanup_device(struct virtio_net *dev, int destroy);
674 void reset_device(struct virtio_net *dev);
675 void vhost_destroy_device(int);
676 void vhost_destroy_device_notify(struct virtio_net *dev);
677 
678 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
679 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
680 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
681 
682 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
683 
684 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
685 
686 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
687 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags);
688 void vhost_enable_extbuf(int vid);
689 void vhost_enable_linearbuf(int vid);
690 int vhost_enable_guest_notification(struct virtio_net *dev,
691 		struct vhost_virtqueue *vq, int enable);
692 
693 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
694 
695 /*
696  * Backend-specific cleanup.
697  *
698  * TODO: fix it; we have one backend now
699  */
700 void vhost_backend_cleanup(struct virtio_net *dev);
701 
702 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
703 			uint64_t iova, uint64_t *len, uint8_t perm);
704 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
705 			struct vhost_virtqueue *vq,
706 			uint64_t desc_addr, uint64_t desc_len);
707 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
708 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
709 		uint64_t log_addr);
710 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
711 
712 static __rte_always_inline uint64_t
713 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
714 			uint64_t iova, uint64_t *len, uint8_t perm)
715 {
716 	if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
717 		return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
718 
719 	return __vhost_iova_to_vva(dev, vq, iova, len, perm);
720 }
721 
722 #define vhost_avail_event(vr) \
723 	(*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
724 #define vhost_used_event(vr) \
725 	(*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
726 
727 /*
728  * The following is used with VIRTIO_RING_F_EVENT_IDX.
729  * Assuming a given event_idx value from the other size, if we have
730  * just incremented index from old to new_idx, should we trigger an
731  * event?
732  */
733 static __rte_always_inline int
734 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
735 {
736 	return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
737 }
738 
739 static __rte_always_inline void
740 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
741 {
742 	/* Flush used->idx update before we read avail->flags. */
743 	rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
744 
745 	/* Don't kick guest if we don't reach index specified by guest. */
746 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
747 		uint16_t old = vq->signalled_used;
748 		uint16_t new = vq->last_used_idx;
749 		bool signalled_used_valid = vq->signalled_used_valid;
750 
751 		vq->signalled_used = new;
752 		vq->signalled_used_valid = true;
753 
754 		VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
755 			__func__,
756 			vhost_used_event(vq),
757 			old, new);
758 
759 		if ((vhost_need_event(vhost_used_event(vq), new, old) &&
760 					(vq->callfd >= 0)) ||
761 				unlikely(!signalled_used_valid)) {
762 			eventfd_write(vq->callfd, (eventfd_t) 1);
763 			if (dev->notify_ops->guest_notified)
764 				dev->notify_ops->guest_notified(dev->vid);
765 		}
766 	} else {
767 		/* Kick the guest if necessary. */
768 		if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
769 				&& (vq->callfd >= 0)) {
770 			eventfd_write(vq->callfd, (eventfd_t)1);
771 			if (dev->notify_ops->guest_notified)
772 				dev->notify_ops->guest_notified(dev->vid);
773 		}
774 	}
775 }
776 
777 static __rte_always_inline void
778 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
779 {
780 	uint16_t old, new, off, off_wrap;
781 	bool signalled_used_valid, kick = false;
782 
783 	/* Flush used desc update. */
784 	rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
785 
786 	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
787 		if (vq->driver_event->flags !=
788 				VRING_EVENT_F_DISABLE)
789 			kick = true;
790 		goto kick;
791 	}
792 
793 	old = vq->signalled_used;
794 	new = vq->last_used_idx;
795 	vq->signalled_used = new;
796 	signalled_used_valid = vq->signalled_used_valid;
797 	vq->signalled_used_valid = true;
798 
799 	if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
800 		if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
801 			kick = true;
802 		goto kick;
803 	}
804 
805 	if (unlikely(!signalled_used_valid)) {
806 		kick = true;
807 		goto kick;
808 	}
809 
810 	rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
811 
812 	off_wrap = vq->driver_event->off_wrap;
813 	off = off_wrap & ~(1 << 15);
814 
815 	if (new <= old)
816 		old -= vq->size;
817 
818 	if (vq->used_wrap_counter != off_wrap >> 15)
819 		off -= vq->size;
820 
821 	if (vhost_need_event(off, new, old))
822 		kick = true;
823 kick:
824 	if (kick) {
825 		eventfd_write(vq->callfd, (eventfd_t)1);
826 		if (dev->notify_ops->guest_notified)
827 			dev->notify_ops->guest_notified(dev->vid);
828 	}
829 }
830 
831 static __rte_always_inline void
832 free_ind_table(void *idesc)
833 {
834 	rte_free(idesc);
835 }
836 
837 static __rte_always_inline void
838 restore_mbuf(struct rte_mbuf *m)
839 {
840 	uint32_t mbuf_size, priv_size;
841 
842 	while (m) {
843 		priv_size = rte_pktmbuf_priv_size(m->pool);
844 		mbuf_size = sizeof(struct rte_mbuf) + priv_size;
845 		/* start of buffer is after mbuf structure and priv data */
846 
847 		m->buf_addr = (char *)m + mbuf_size;
848 		m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
849 		m = m->next;
850 	}
851 }
852 
853 static __rte_always_inline bool
854 mbuf_is_consumed(struct rte_mbuf *m)
855 {
856 	while (m) {
857 		if (rte_mbuf_refcnt_read(m) > 1)
858 			return false;
859 		m = m->next;
860 	}
861 
862 	return true;
863 }
864 
865 #endif /* _VHOST_NET_CDEV_H_ */
866