xref: /dpdk/lib/vhost/vhost.h (revision d38febb08d57fec29fed27a2d12a507fc6fcdfa1)
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 	uint32_t	async_threshold;
223 
224 	int			notif_enable;
225 #define VIRTIO_UNINITIALIZED_NOTIF	(-1)
226 
227 	struct vhost_vring_addr ring_addrs;
228 } __rte_cache_aligned;
229 
230 /* Virtio device status as per Virtio specification */
231 #define VIRTIO_DEVICE_STATUS_RESET		0x00
232 #define VIRTIO_DEVICE_STATUS_ACK		0x01
233 #define VIRTIO_DEVICE_STATUS_DRIVER		0x02
234 #define VIRTIO_DEVICE_STATUS_DRIVER_OK		0x04
235 #define VIRTIO_DEVICE_STATUS_FEATURES_OK	0x08
236 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET	0x40
237 #define VIRTIO_DEVICE_STATUS_FAILED		0x80
238 
239 #define VHOST_MAX_VRING			0x100
240 #define VHOST_MAX_QUEUE_PAIRS		0x80
241 
242 /* Declare IOMMU related bits for older kernels */
243 #ifndef VIRTIO_F_IOMMU_PLATFORM
244 
245 #define VIRTIO_F_IOMMU_PLATFORM 33
246 
247 struct vhost_iotlb_msg {
248 	__u64 iova;
249 	__u64 size;
250 	__u64 uaddr;
251 #define VHOST_ACCESS_RO      0x1
252 #define VHOST_ACCESS_WO      0x2
253 #define VHOST_ACCESS_RW      0x3
254 	__u8 perm;
255 #define VHOST_IOTLB_MISS           1
256 #define VHOST_IOTLB_UPDATE         2
257 #define VHOST_IOTLB_INVALIDATE     3
258 #define VHOST_IOTLB_ACCESS_FAIL    4
259 	__u8 type;
260 };
261 
262 #define VHOST_IOTLB_MSG 0x1
263 
264 struct vhost_msg {
265 	int type;
266 	union {
267 		struct vhost_iotlb_msg iotlb;
268 		__u8 padding[64];
269 	};
270 };
271 #endif
272 
273 /*
274  * Define virtio 1.0 for older kernels
275  */
276 #ifndef VIRTIO_F_VERSION_1
277  #define VIRTIO_F_VERSION_1 32
278 #endif
279 
280 /* Declare packed ring related bits for older kernels */
281 #ifndef VIRTIO_F_RING_PACKED
282 
283 #define VIRTIO_F_RING_PACKED 34
284 
285 struct vring_packed_desc {
286 	uint64_t addr;
287 	uint32_t len;
288 	uint16_t id;
289 	uint16_t flags;
290 };
291 
292 struct vring_packed_desc_event {
293 	uint16_t off_wrap;
294 	uint16_t flags;
295 };
296 #endif
297 
298 /*
299  * Declare below packed ring defines unconditionally
300  * as Kernel header might use different names.
301  */
302 #define VRING_DESC_F_AVAIL	(1ULL << 7)
303 #define VRING_DESC_F_USED	(1ULL << 15)
304 
305 #define VRING_EVENT_F_ENABLE 0x0
306 #define VRING_EVENT_F_DISABLE 0x1
307 #define VRING_EVENT_F_DESC 0x2
308 
309 /*
310  * Available and used descs are in same order
311  */
312 #ifndef VIRTIO_F_IN_ORDER
313 #define VIRTIO_F_IN_ORDER      35
314 #endif
315 
316 /* Features supported by this builtin vhost-user net driver. */
317 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
318 				(1ULL << VIRTIO_F_ANY_LAYOUT) | \
319 				(1ULL << VIRTIO_NET_F_CTRL_VQ) | \
320 				(1ULL << VIRTIO_NET_F_CTRL_RX) | \
321 				(1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
322 				(1ULL << VIRTIO_NET_F_MQ)      | \
323 				(1ULL << VIRTIO_F_VERSION_1)   | \
324 				(1ULL << VHOST_F_LOG_ALL)      | \
325 				(1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
326 				(1ULL << VIRTIO_NET_F_GSO) | \
327 				(1ULL << VIRTIO_NET_F_HOST_TSO4) | \
328 				(1ULL << VIRTIO_NET_F_HOST_TSO6) | \
329 				(1ULL << VIRTIO_NET_F_HOST_UFO) | \
330 				(1ULL << VIRTIO_NET_F_HOST_ECN) | \
331 				(1ULL << VIRTIO_NET_F_CSUM)    | \
332 				(1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
333 				(1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
334 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
335 				(1ULL << VIRTIO_NET_F_GUEST_UFO) | \
336 				(1ULL << VIRTIO_NET_F_GUEST_ECN) | \
337 				(1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
338 				(1ULL << VIRTIO_RING_F_EVENT_IDX) | \
339 				(1ULL << VIRTIO_NET_F_MTU)  | \
340 				(1ULL << VIRTIO_F_IN_ORDER) | \
341 				(1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
342 				(1ULL << VIRTIO_F_RING_PACKED))
343 
344 
345 struct guest_page {
346 	uint64_t guest_phys_addr;
347 	uint64_t host_phys_addr;
348 	uint64_t size;
349 };
350 
351 struct inflight_mem_info {
352 	int		fd;
353 	void		*addr;
354 	uint64_t	size;
355 };
356 
357 /**
358  * Device structure contains all configuration information relating
359  * to the device.
360  */
361 struct virtio_net {
362 	/* Frontend (QEMU) memory and memory region information */
363 	struct rte_vhost_memory	*mem;
364 	uint64_t		features;
365 	uint64_t		protocol_features;
366 	int			vid;
367 	uint32_t		flags;
368 	uint16_t		vhost_hlen;
369 	/* to tell if we need broadcast rarp packet */
370 	int16_t			broadcast_rarp;
371 	uint32_t		nr_vring;
372 	int			async_copy;
373 	int			extbuf;
374 	int			linearbuf;
375 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
376 	struct inflight_mem_info *inflight_info;
377 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
378 	char			ifname[IF_NAME_SZ];
379 	uint64_t		log_size;
380 	uint64_t		log_base;
381 	uint64_t		log_addr;
382 	struct rte_ether_addr	mac;
383 	uint16_t		mtu;
384 	uint8_t			status;
385 
386 	struct vhost_device_ops const *notify_ops;
387 
388 	uint32_t		nr_guest_pages;
389 	uint32_t		max_guest_pages;
390 	struct guest_page       *guest_pages;
391 
392 	int			slave_req_fd;
393 	rte_spinlock_t		slave_req_lock;
394 
395 	int			postcopy_ufd;
396 	int			postcopy_listening;
397 
398 	struct rte_vdpa_device *vdpa_dev;
399 
400 	/* context data for the external message handlers */
401 	void			*extern_data;
402 	/* pre and post vhost user message handlers for the device */
403 	struct rte_vhost_user_extern_ops extern_ops;
404 } __rte_cache_aligned;
405 
406 static __rte_always_inline bool
407 vq_is_packed(struct virtio_net *dev)
408 {
409 	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
410 }
411 
412 static inline bool
413 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
414 {
415 	uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
416 
417 	return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
418 		wrap_counter != !!(flags & VRING_DESC_F_USED);
419 }
420 
421 static inline void
422 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
423 {
424 	vq->last_used_idx += num;
425 	if (vq->last_used_idx >= vq->size) {
426 		vq->used_wrap_counter ^= 1;
427 		vq->last_used_idx -= vq->size;
428 	}
429 }
430 
431 static inline void
432 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
433 {
434 	vq->last_avail_idx += num;
435 	if (vq->last_avail_idx >= vq->size) {
436 		vq->avail_wrap_counter ^= 1;
437 		vq->last_avail_idx -= vq->size;
438 	}
439 }
440 
441 void __vhost_log_cache_write(struct virtio_net *dev,
442 		struct vhost_virtqueue *vq,
443 		uint64_t addr, uint64_t len);
444 void __vhost_log_cache_write_iova(struct virtio_net *dev,
445 		struct vhost_virtqueue *vq,
446 		uint64_t iova, uint64_t len);
447 void __vhost_log_cache_sync(struct virtio_net *dev,
448 		struct vhost_virtqueue *vq);
449 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
450 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
451 			    uint64_t iova, uint64_t len);
452 
453 static __rte_always_inline void
454 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
455 {
456 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
457 		__vhost_log_write(dev, addr, len);
458 }
459 
460 static __rte_always_inline void
461 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
462 {
463 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
464 		__vhost_log_cache_sync(dev, vq);
465 }
466 
467 static __rte_always_inline void
468 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
469 			uint64_t addr, uint64_t len)
470 {
471 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
472 		__vhost_log_cache_write(dev, vq, addr, len);
473 }
474 
475 static __rte_always_inline void
476 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
477 			uint64_t offset, uint64_t len)
478 {
479 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
480 		if (unlikely(vq->log_guest_addr == 0))
481 			return;
482 		__vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
483 					len);
484 	}
485 }
486 
487 static __rte_always_inline void
488 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
489 		     uint64_t offset, uint64_t len)
490 {
491 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
492 		if (unlikely(vq->log_guest_addr == 0))
493 			return;
494 		__vhost_log_write(dev, vq->log_guest_addr + offset, len);
495 	}
496 }
497 
498 static __rte_always_inline void
499 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
500 			   uint64_t iova, uint64_t len)
501 {
502 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
503 		return;
504 
505 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
506 		__vhost_log_cache_write_iova(dev, vq, iova, len);
507 	else
508 		__vhost_log_cache_write(dev, vq, iova, len);
509 }
510 
511 static __rte_always_inline void
512 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
513 			   uint64_t iova, uint64_t len)
514 {
515 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
516 		return;
517 
518 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
519 		__vhost_log_write_iova(dev, vq, iova, len);
520 	else
521 		__vhost_log_write(dev, iova, len);
522 }
523 
524 extern int vhost_config_log_level;
525 extern int vhost_data_log_level;
526 
527 #define VHOST_LOG_CONFIG(level, fmt, args...)			\
528 	rte_log(RTE_LOG_ ## level, vhost_config_log_level,	\
529 		"VHOST_CONFIG: " fmt, ##args)
530 
531 #define VHOST_LOG_DATA(level, fmt, args...) \
532 	(void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?	\
533 	 rte_log(RTE_LOG_ ## level,  vhost_data_log_level,	\
534 		"VHOST_DATA : " fmt, ##args) :			\
535 	 0)
536 
537 #ifdef RTE_LIBRTE_VHOST_DEBUG
538 #define VHOST_MAX_PRINT_BUFF 6072
539 #define PRINT_PACKET(device, addr, size, header) do { \
540 	char *pkt_addr = (char *)(addr); \
541 	unsigned int index; \
542 	char packet[VHOST_MAX_PRINT_BUFF]; \
543 	\
544 	if ((header)) \
545 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
546 	else \
547 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
548 	for (index = 0; index < (size); index++) { \
549 		snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
550 			"%02hhx ", pkt_addr[index]); \
551 	} \
552 	snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
553 	\
554 	VHOST_LOG_DATA(DEBUG, "%s", packet); \
555 } while (0)
556 #else
557 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
558 #endif
559 
560 #define MAX_VHOST_DEVICE	1024
561 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
562 
563 #define VHOST_BINARY_SEARCH_THRESH 256
564 
565 static __rte_always_inline int guest_page_addrcmp(const void *p1,
566 						const void *p2)
567 {
568 	const struct guest_page *page1 = (const struct guest_page *)p1;
569 	const struct guest_page *page2 = (const struct guest_page *)p2;
570 
571 	if (page1->guest_phys_addr > page2->guest_phys_addr)
572 		return 1;
573 	if (page1->guest_phys_addr < page2->guest_phys_addr)
574 		return -1;
575 
576 	return 0;
577 }
578 
579 static __rte_always_inline rte_iova_t
580 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
581 	uint64_t gpa_size, uint64_t *hpa_size)
582 {
583 	uint32_t i;
584 	struct guest_page *page;
585 	struct guest_page key;
586 
587 	*hpa_size = gpa_size;
588 	if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
589 		key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
590 		page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
591 			       sizeof(struct guest_page), guest_page_addrcmp);
592 		if (page) {
593 			if (gpa + gpa_size <=
594 					page->guest_phys_addr + page->size) {
595 				return gpa - page->guest_phys_addr +
596 					page->host_phys_addr;
597 			} else if (gpa < page->guest_phys_addr +
598 						page->size) {
599 				*hpa_size = page->guest_phys_addr +
600 					page->size - gpa;
601 				return gpa - page->guest_phys_addr +
602 					page->host_phys_addr;
603 			}
604 		}
605 	} else {
606 		for (i = 0; i < dev->nr_guest_pages; i++) {
607 			page = &dev->guest_pages[i];
608 
609 			if (gpa >= page->guest_phys_addr) {
610 				if (gpa + gpa_size <=
611 					page->guest_phys_addr + page->size) {
612 					return gpa - page->guest_phys_addr +
613 						page->host_phys_addr;
614 				} else if (gpa < page->guest_phys_addr +
615 							page->size) {
616 					*hpa_size = page->guest_phys_addr +
617 						page->size - gpa;
618 					return gpa - page->guest_phys_addr +
619 						page->host_phys_addr;
620 				}
621 			}
622 		}
623 	}
624 
625 	*hpa_size = 0;
626 	return 0;
627 }
628 
629 /* Convert guest physical address to host physical address */
630 static __rte_always_inline rte_iova_t
631 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
632 {
633 	rte_iova_t hpa;
634 	uint64_t hpa_size;
635 
636 	hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
637 	return hpa_size == size ? hpa : 0;
638 }
639 
640 static __rte_always_inline uint64_t
641 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
642 {
643 	struct rte_vhost_mem_region *r;
644 	uint32_t i;
645 
646 	if (unlikely(!dev || !dev->mem))
647 		return 0;
648 
649 	for (i = 0; i < dev->mem->nregions; i++) {
650 		r = &dev->mem->regions[i];
651 
652 		if (vva >= r->host_user_addr &&
653 		    vva + len <  r->host_user_addr + r->size) {
654 			return r->guest_phys_addr + vva - r->host_user_addr;
655 		}
656 	}
657 	return 0;
658 }
659 
660 static __rte_always_inline struct virtio_net *
661 get_device(int vid)
662 {
663 	struct virtio_net *dev = vhost_devices[vid];
664 
665 	if (unlikely(!dev)) {
666 		VHOST_LOG_CONFIG(ERR,
667 			"(%d) device not found.\n", vid);
668 	}
669 
670 	return dev;
671 }
672 
673 int vhost_new_device(void);
674 void cleanup_device(struct virtio_net *dev, int destroy);
675 void reset_device(struct virtio_net *dev);
676 void vhost_destroy_device(int);
677 void vhost_destroy_device_notify(struct virtio_net *dev);
678 
679 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
680 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
681 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
682 
683 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
684 
685 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
686 
687 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
688 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags);
689 void vhost_enable_extbuf(int vid);
690 void vhost_enable_linearbuf(int vid);
691 int vhost_enable_guest_notification(struct virtio_net *dev,
692 		struct vhost_virtqueue *vq, int enable);
693 
694 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
695 
696 /*
697  * Backend-specific cleanup.
698  *
699  * TODO: fix it; we have one backend now
700  */
701 void vhost_backend_cleanup(struct virtio_net *dev);
702 
703 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
704 			uint64_t iova, uint64_t *len, uint8_t perm);
705 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
706 			struct vhost_virtqueue *vq,
707 			uint64_t desc_addr, uint64_t desc_len);
708 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
709 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
710 		uint64_t log_addr);
711 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
712 
713 static __rte_always_inline uint64_t
714 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
715 			uint64_t iova, uint64_t *len, uint8_t perm)
716 {
717 	if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
718 		return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
719 
720 	return __vhost_iova_to_vva(dev, vq, iova, len, perm);
721 }
722 
723 #define vhost_avail_event(vr) \
724 	(*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
725 #define vhost_used_event(vr) \
726 	(*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
727 
728 /*
729  * The following is used with VIRTIO_RING_F_EVENT_IDX.
730  * Assuming a given event_idx value from the other size, if we have
731  * just incremented index from old to new_idx, should we trigger an
732  * event?
733  */
734 static __rte_always_inline int
735 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
736 {
737 	return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
738 }
739 
740 static __rte_always_inline void
741 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
742 {
743 	/* Flush used->idx update before we read avail->flags. */
744 	rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
745 
746 	/* Don't kick guest if we don't reach index specified by guest. */
747 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
748 		uint16_t old = vq->signalled_used;
749 		uint16_t new = vq->last_used_idx;
750 		bool signalled_used_valid = vq->signalled_used_valid;
751 
752 		vq->signalled_used = new;
753 		vq->signalled_used_valid = true;
754 
755 		VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
756 			__func__,
757 			vhost_used_event(vq),
758 			old, new);
759 
760 		if ((vhost_need_event(vhost_used_event(vq), new, old) &&
761 					(vq->callfd >= 0)) ||
762 				unlikely(!signalled_used_valid)) {
763 			eventfd_write(vq->callfd, (eventfd_t) 1);
764 			if (dev->notify_ops->guest_notified)
765 				dev->notify_ops->guest_notified(dev->vid);
766 		}
767 	} else {
768 		/* Kick the guest if necessary. */
769 		if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
770 				&& (vq->callfd >= 0)) {
771 			eventfd_write(vq->callfd, (eventfd_t)1);
772 			if (dev->notify_ops->guest_notified)
773 				dev->notify_ops->guest_notified(dev->vid);
774 		}
775 	}
776 }
777 
778 static __rte_always_inline void
779 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
780 {
781 	uint16_t old, new, off, off_wrap;
782 	bool signalled_used_valid, kick = false;
783 
784 	/* Flush used desc update. */
785 	rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
786 
787 	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
788 		if (vq->driver_event->flags !=
789 				VRING_EVENT_F_DISABLE)
790 			kick = true;
791 		goto kick;
792 	}
793 
794 	old = vq->signalled_used;
795 	new = vq->last_used_idx;
796 	vq->signalled_used = new;
797 	signalled_used_valid = vq->signalled_used_valid;
798 	vq->signalled_used_valid = true;
799 
800 	if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
801 		if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
802 			kick = true;
803 		goto kick;
804 	}
805 
806 	if (unlikely(!signalled_used_valid)) {
807 		kick = true;
808 		goto kick;
809 	}
810 
811 	rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
812 
813 	off_wrap = vq->driver_event->off_wrap;
814 	off = off_wrap & ~(1 << 15);
815 
816 	if (new <= old)
817 		old -= vq->size;
818 
819 	if (vq->used_wrap_counter != off_wrap >> 15)
820 		off -= vq->size;
821 
822 	if (vhost_need_event(off, new, old))
823 		kick = true;
824 kick:
825 	if (kick) {
826 		eventfd_write(vq->callfd, (eventfd_t)1);
827 		if (dev->notify_ops->guest_notified)
828 			dev->notify_ops->guest_notified(dev->vid);
829 	}
830 }
831 
832 static __rte_always_inline void
833 free_ind_table(void *idesc)
834 {
835 	rte_free(idesc);
836 }
837 
838 static __rte_always_inline void
839 restore_mbuf(struct rte_mbuf *m)
840 {
841 	uint32_t mbuf_size, priv_size;
842 
843 	while (m) {
844 		priv_size = rte_pktmbuf_priv_size(m->pool);
845 		mbuf_size = sizeof(struct rte_mbuf) + priv_size;
846 		/* start of buffer is after mbuf structure and priv data */
847 
848 		m->buf_addr = (char *)m + mbuf_size;
849 		m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
850 		m = m->next;
851 	}
852 }
853 
854 static __rte_always_inline bool
855 mbuf_is_consumed(struct rte_mbuf *m)
856 {
857 	while (m) {
858 		if (rte_mbuf_refcnt_read(m) > 1)
859 			return false;
860 		m = m->next;
861 	}
862 
863 	return true;
864 }
865 
866 #endif /* _VHOST_NET_CDEV_H_ */
867