xref: /dpdk/lib/vhost/vhost.h (revision 7af3e7aaf9141e78858a38c0d6e32b46a196c08f)
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/queue.h>
11 #include <unistd.h>
12 #include <linux/virtio_net.h>
13 #include <sys/socket.h>
14 #include <linux/if.h>
15 
16 #include <rte_log.h>
17 #include <rte_ether.h>
18 #include <rte_malloc.h>
19 #include <rte_dmadev.h>
20 
21 #include "rte_vhost.h"
22 #include "vdpa_driver.h"
23 
24 #include "rte_vhost_async.h"
25 
26 /* Used to indicate that the device is running on a data core */
27 #define VIRTIO_DEV_RUNNING ((uint32_t)1 << 0)
28 /* Used to indicate that the device is ready to operate */
29 #define VIRTIO_DEV_READY ((uint32_t)1 << 1)
30 /* Used to indicate that the built-in vhost net device backend is enabled */
31 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET ((uint32_t)1 << 2)
32 /* Used to indicate that the device has its own data path and configured */
33 #define VIRTIO_DEV_VDPA_CONFIGURED ((uint32_t)1 << 3)
34 /* Used to indicate that the feature negotiation failed */
35 #define VIRTIO_DEV_FEATURES_FAILED ((uint32_t)1 << 4)
36 /* Used to indicate that the virtio_net tx code should fill TX ol_flags */
37 #define VIRTIO_DEV_LEGACY_OL_FLAGS ((uint32_t)1 << 5)
38 /*  Used to indicate the application has requested statistics collection */
39 #define VIRTIO_DEV_STATS_ENABLED ((uint32_t)1 << 6)
40 
41 /* Backend value set by guest. */
42 #define VIRTIO_DEV_STOPPED -1
43 
44 #define BUF_VECTOR_MAX 256
45 
46 #define VHOST_LOG_CACHE_NR 32
47 
48 #define MAX_PKT_BURST 32
49 
50 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST)
51 #define VHOST_MAX_ASYNC_VEC 2048
52 #define VIRTIO_MAX_RX_PKTLEN 9728U
53 #define VHOST_DMA_MAX_COPY_COMPLETE ((VIRTIO_MAX_RX_PKTLEN / RTE_MBUF_DEFAULT_DATAROOM) \
54 		* MAX_PKT_BURST)
55 
56 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)	\
57 	((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
58 		VRING_DESC_F_WRITE)
59 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)	\
60 	((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
61 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
62 					 VRING_DESC_F_INDIRECT)
63 
64 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
65 			    sizeof(struct vring_packed_desc))
66 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
67 
68 #ifdef VHOST_GCC_UNROLL_PRAGMA
69 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
70 	for (iter = val; iter < size; iter++)
71 #endif
72 
73 #ifdef VHOST_CLANG_UNROLL_PRAGMA
74 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
75 	for (iter = val; iter < size; iter++)
76 #endif
77 
78 #ifdef VHOST_ICC_UNROLL_PRAGMA
79 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
80 	for (iter = val; iter < size; iter++)
81 #endif
82 
83 #ifndef vhost_for_each_try_unroll
84 #define vhost_for_each_try_unroll(iter, val, num) \
85 	for (iter = val; iter < num; iter++)
86 #endif
87 
88 /**
89  * Structure contains buffer address, length and descriptor index
90  * from vring to do scatter RX.
91  */
92 struct buf_vector {
93 	uint64_t buf_iova;
94 	uint64_t buf_addr;
95 	uint32_t buf_len;
96 	uint32_t desc_idx;
97 };
98 
99 /*
100  * Structure contains the info for each batched memory copy.
101  */
102 struct batch_copy_elem {
103 	void *dst;
104 	void *src;
105 	uint32_t len;
106 	uint64_t log_addr;
107 };
108 
109 /*
110  * Structure that contains the info for batched dirty logging.
111  */
112 struct log_cache_entry {
113 	uint32_t offset;
114 	unsigned long val;
115 };
116 
117 struct vring_used_elem_packed {
118 	uint16_t id;
119 	uint16_t flags;
120 	uint32_t len;
121 	uint32_t count;
122 };
123 
124 /**
125  * Virtqueue statistics
126  */
127 struct virtqueue_stats {
128 	uint64_t packets;
129 	uint64_t bytes;
130 	uint64_t multicast;
131 	uint64_t broadcast;
132 	/* Size bins in array as RFC 2819, undersized [0], 64 [1], etc */
133 	uint64_t size_bins[8];
134 	uint64_t guest_notifications;
135 	uint64_t iotlb_hits;
136 	uint64_t iotlb_misses;
137 	uint64_t inflight_submitted;
138 	uint64_t inflight_completed;
139 };
140 
141 /**
142  * iovec
143  */
144 struct vhost_iovec {
145 	void *src_addr;
146 	void *dst_addr;
147 	size_t len;
148 };
149 
150 /**
151  * iovec iterator
152  */
153 struct vhost_iov_iter {
154 	/** pointer to the iovec array */
155 	struct vhost_iovec *iov;
156 	/** number of iovec in this iterator */
157 	unsigned long nr_segs;
158 };
159 
160 struct async_dma_vchan_info {
161 	/* circular array to track if packet copy completes */
162 	bool **pkts_cmpl_flag_addr;
163 
164 	/* max elements in 'pkts_cmpl_flag_addr' */
165 	uint16_t ring_size;
166 	/* ring index mask for 'pkts_cmpl_flag_addr' */
167 	uint16_t ring_mask;
168 
169 	/**
170 	 * DMA virtual channel lock. Although it is able to bind DMA
171 	 * virtual channels to data plane threads, vhost control plane
172 	 * thread could call data plane functions too, thus causing
173 	 * DMA device contention.
174 	 *
175 	 * For example, in VM exit case, vhost control plane thread needs
176 	 * to clear in-flight packets before disable vring, but there could
177 	 * be anotther data plane thread is enqueuing packets to the same
178 	 * vring with the same DMA virtual channel. As dmadev PMD functions
179 	 * are lock-free, the control plane and data plane threads could
180 	 * operate the same DMA virtual channel at the same time.
181 	 */
182 	rte_spinlock_t dma_lock;
183 };
184 
185 struct async_dma_info {
186 	struct async_dma_vchan_info *vchans;
187 	/* number of registered virtual channels */
188 	uint16_t nr_vchans;
189 };
190 
191 extern struct async_dma_info dma_copy_track[RTE_DMADEV_DEFAULT_MAX];
192 
193 /**
194  * inflight async packet information
195  */
196 struct async_inflight_info {
197 	struct rte_mbuf *mbuf;
198 	uint16_t descs; /* num of descs inflight */
199 	uint16_t nr_buffers; /* num of buffers inflight for packed ring */
200 	struct virtio_net_hdr nethdr;
201 };
202 
203 struct vhost_async {
204 	struct vhost_iov_iter iov_iter[VHOST_MAX_ASYNC_IT];
205 	struct vhost_iovec iovec[VHOST_MAX_ASYNC_VEC];
206 	uint16_t iter_idx;
207 	uint16_t iovec_idx;
208 
209 	/* data transfer status */
210 	struct async_inflight_info *pkts_info;
211 	/**
212 	 * Packet reorder array. "true" indicates that DMA device
213 	 * completes all copies for the packet.
214 	 *
215 	 * Note that this array could be written by multiple threads
216 	 * simultaneously. For example, in the case of thread0 and
217 	 * thread1 RX packets from NIC and then enqueue packets to
218 	 * vring0 and vring1 with own DMA device DMA0 and DMA1, it's
219 	 * possible for thread0 to get completed copies belonging to
220 	 * vring1 from DMA0, while thread0 is calling rte_vhost_poll
221 	 * _enqueue_completed() for vring0 and thread1 is calling
222 	 * rte_vhost_submit_enqueue_burst() for vring1. In this case,
223 	 * vq->access_lock cannot protect pkts_cmpl_flag of vring1.
224 	 *
225 	 * However, since offloading is per-packet basis, each packet
226 	 * flag will only be written by one thread. And single byte
227 	 * write is atomic, so no lock for pkts_cmpl_flag is needed.
228 	 */
229 	bool *pkts_cmpl_flag;
230 	uint16_t pkts_idx;
231 	uint16_t pkts_inflight_n;
232 	union {
233 		struct vring_used_elem  *descs_split;
234 		struct vring_used_elem_packed *buffers_packed;
235 	};
236 	union {
237 		uint16_t desc_idx_split;
238 		uint16_t buffer_idx_packed;
239 	};
240 	union {
241 		uint16_t last_desc_idx_split;
242 		uint16_t last_buffer_idx_packed;
243 	};
244 };
245 
246 /**
247  * Structure contains variables relevant to RX/TX virtqueues.
248  */
249 struct vhost_virtqueue {
250 	union {
251 		struct vring_desc	*desc;
252 		struct vring_packed_desc   *desc_packed;
253 	};
254 	union {
255 		struct vring_avail	*avail;
256 		struct vring_packed_desc_event *driver_event;
257 	};
258 	union {
259 		struct vring_used	*used;
260 		struct vring_packed_desc_event *device_event;
261 	};
262 	uint16_t		size;
263 
264 	uint16_t		last_avail_idx;
265 	uint16_t		last_used_idx;
266 	/* Last used index we notify to front end. */
267 	uint16_t		signalled_used;
268 	bool			signalled_used_valid;
269 #define VIRTIO_INVALID_EVENTFD		(-1)
270 #define VIRTIO_UNINITIALIZED_EVENTFD	(-2)
271 
272 	bool			enabled;
273 	bool			access_ok;
274 	bool			ready;
275 
276 	rte_spinlock_t		access_lock;
277 
278 
279 	union {
280 		struct vring_used_elem  *shadow_used_split;
281 		struct vring_used_elem_packed *shadow_used_packed;
282 	};
283 	uint16_t                shadow_used_idx;
284 	/* Record packed ring enqueue latest desc cache aligned index */
285 	uint16_t		shadow_aligned_idx;
286 	/* Record packed ring first dequeue desc index */
287 	uint16_t		shadow_last_used_idx;
288 
289 	uint16_t		batch_copy_nb_elems;
290 	struct batch_copy_elem	*batch_copy_elems;
291 	int			numa_node;
292 	bool			used_wrap_counter;
293 	bool			avail_wrap_counter;
294 
295 	/* Physical address of used ring, for logging */
296 	uint16_t		log_cache_nb_elem;
297 	uint64_t		log_guest_addr;
298 	struct log_cache_entry	*log_cache;
299 
300 	rte_rwlock_t	iotlb_lock;
301 	rte_rwlock_t	iotlb_pending_lock;
302 	struct rte_mempool *iotlb_pool;
303 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
304 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
305 	int				iotlb_cache_nr;
306 
307 	/* Used to notify the guest (trigger interrupt) */
308 	int			callfd;
309 	/* Currently unused as polling mode is enabled */
310 	int			kickfd;
311 
312 	/* inflight share memory info */
313 	union {
314 		struct rte_vhost_inflight_info_split *inflight_split;
315 		struct rte_vhost_inflight_info_packed *inflight_packed;
316 	};
317 	struct rte_vhost_resubmit_info *resubmit_inflight;
318 	uint64_t		global_counter;
319 
320 	struct vhost_async	*async;
321 
322 	int			notif_enable;
323 #define VIRTIO_UNINITIALIZED_NOTIF	(-1)
324 
325 	struct vhost_vring_addr ring_addrs;
326 	struct virtqueue_stats	stats;
327 } __rte_cache_aligned;
328 
329 /* Virtio device status as per Virtio specification */
330 #define VIRTIO_DEVICE_STATUS_RESET		0x00
331 #define VIRTIO_DEVICE_STATUS_ACK		0x01
332 #define VIRTIO_DEVICE_STATUS_DRIVER		0x02
333 #define VIRTIO_DEVICE_STATUS_DRIVER_OK		0x04
334 #define VIRTIO_DEVICE_STATUS_FEATURES_OK	0x08
335 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET	0x40
336 #define VIRTIO_DEVICE_STATUS_FAILED		0x80
337 
338 #define VHOST_MAX_VRING			0x100
339 #define VHOST_MAX_QUEUE_PAIRS		0x80
340 
341 /* Declare IOMMU related bits for older kernels */
342 #ifndef VIRTIO_F_IOMMU_PLATFORM
343 
344 #define VIRTIO_F_IOMMU_PLATFORM 33
345 
346 struct vhost_iotlb_msg {
347 	__u64 iova;
348 	__u64 size;
349 	__u64 uaddr;
350 #define VHOST_ACCESS_RO      0x1
351 #define VHOST_ACCESS_WO      0x2
352 #define VHOST_ACCESS_RW      0x3
353 	__u8 perm;
354 #define VHOST_IOTLB_MISS           1
355 #define VHOST_IOTLB_UPDATE         2
356 #define VHOST_IOTLB_INVALIDATE     3
357 #define VHOST_IOTLB_ACCESS_FAIL    4
358 	__u8 type;
359 };
360 
361 #define VHOST_IOTLB_MSG 0x1
362 
363 struct vhost_msg {
364 	int type;
365 	union {
366 		struct vhost_iotlb_msg iotlb;
367 		__u8 padding[64];
368 	};
369 };
370 #endif
371 
372 /*
373  * Define virtio 1.0 for older kernels
374  */
375 #ifndef VIRTIO_F_VERSION_1
376  #define VIRTIO_F_VERSION_1 32
377 #endif
378 
379 /* Declare packed ring related bits for older kernels */
380 #ifndef VIRTIO_F_RING_PACKED
381 
382 #define VIRTIO_F_RING_PACKED 34
383 
384 struct vring_packed_desc {
385 	uint64_t addr;
386 	uint32_t len;
387 	uint16_t id;
388 	uint16_t flags;
389 };
390 
391 struct vring_packed_desc_event {
392 	uint16_t off_wrap;
393 	uint16_t flags;
394 };
395 #endif
396 
397 /*
398  * Declare below packed ring defines unconditionally
399  * as Kernel header might use different names.
400  */
401 #define VRING_DESC_F_AVAIL	(1ULL << 7)
402 #define VRING_DESC_F_USED	(1ULL << 15)
403 
404 #define VRING_EVENT_F_ENABLE 0x0
405 #define VRING_EVENT_F_DISABLE 0x1
406 #define VRING_EVENT_F_DESC 0x2
407 
408 /*
409  * Available and used descs are in same order
410  */
411 #ifndef VIRTIO_F_IN_ORDER
412 #define VIRTIO_F_IN_ORDER      35
413 #endif
414 
415 /* Features supported by this builtin vhost-user net driver. */
416 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
417 				(1ULL << VIRTIO_F_ANY_LAYOUT) | \
418 				(1ULL << VIRTIO_NET_F_CTRL_VQ) | \
419 				(1ULL << VIRTIO_NET_F_CTRL_RX) | \
420 				(1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
421 				(1ULL << VIRTIO_NET_F_MQ)      | \
422 				(1ULL << VIRTIO_F_VERSION_1)   | \
423 				(1ULL << VHOST_F_LOG_ALL)      | \
424 				(1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
425 				(1ULL << VIRTIO_NET_F_GSO) | \
426 				(1ULL << VIRTIO_NET_F_HOST_TSO4) | \
427 				(1ULL << VIRTIO_NET_F_HOST_TSO6) | \
428 				(1ULL << VIRTIO_NET_F_HOST_UFO) | \
429 				(1ULL << VIRTIO_NET_F_HOST_ECN) | \
430 				(1ULL << VIRTIO_NET_F_CSUM)    | \
431 				(1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
432 				(1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
433 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
434 				(1ULL << VIRTIO_NET_F_GUEST_UFO) | \
435 				(1ULL << VIRTIO_NET_F_GUEST_ECN) | \
436 				(1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
437 				(1ULL << VIRTIO_RING_F_EVENT_IDX) | \
438 				(1ULL << VIRTIO_NET_F_MTU)  | \
439 				(1ULL << VIRTIO_F_IN_ORDER) | \
440 				(1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
441 				(1ULL << VIRTIO_F_RING_PACKED))
442 
443 
444 struct guest_page {
445 	uint64_t guest_phys_addr;
446 	uint64_t host_iova;
447 	uint64_t host_user_addr;
448 	uint64_t size;
449 };
450 
451 struct inflight_mem_info {
452 	int		fd;
453 	void		*addr;
454 	uint64_t	size;
455 };
456 
457 /**
458  * Device structure contains all configuration information relating
459  * to the device.
460  */
461 struct virtio_net {
462 	/* Frontend (QEMU) memory and memory region information */
463 	struct rte_vhost_memory	*mem;
464 	uint64_t		features;
465 	uint64_t		protocol_features;
466 	int			vid;
467 	uint32_t		flags;
468 	uint16_t		vhost_hlen;
469 	/* to tell if we need broadcast rarp packet */
470 	int16_t			broadcast_rarp;
471 	uint32_t		nr_vring;
472 	int			async_copy;
473 
474 	int			extbuf;
475 	int			linearbuf;
476 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
477 	struct inflight_mem_info *inflight_info;
478 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
479 	char			ifname[IF_NAME_SZ];
480 	uint64_t		log_size;
481 	uint64_t		log_base;
482 	uint64_t		log_addr;
483 	struct rte_ether_addr	mac;
484 	uint16_t		mtu;
485 	uint8_t			status;
486 
487 	struct rte_vhost_device_ops const *notify_ops;
488 
489 	uint32_t		nr_guest_pages;
490 	uint32_t		max_guest_pages;
491 	struct guest_page       *guest_pages;
492 
493 	int			slave_req_fd;
494 	rte_spinlock_t		slave_req_lock;
495 
496 	int			postcopy_ufd;
497 	int			postcopy_listening;
498 
499 	struct rte_vdpa_device *vdpa_dev;
500 
501 	/* context data for the external message handlers */
502 	void			*extern_data;
503 	/* pre and post vhost user message handlers for the device */
504 	struct rte_vhost_user_extern_ops extern_ops;
505 } __rte_cache_aligned;
506 
507 static __rte_always_inline bool
508 vq_is_packed(struct virtio_net *dev)
509 {
510 	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
511 }
512 
513 static inline bool
514 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
515 {
516 	uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
517 
518 	return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
519 		wrap_counter != !!(flags & VRING_DESC_F_USED);
520 }
521 
522 static inline void
523 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
524 {
525 	vq->last_used_idx += num;
526 	if (vq->last_used_idx >= vq->size) {
527 		vq->used_wrap_counter ^= 1;
528 		vq->last_used_idx -= vq->size;
529 	}
530 }
531 
532 static inline void
533 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
534 {
535 	vq->last_avail_idx += num;
536 	if (vq->last_avail_idx >= vq->size) {
537 		vq->avail_wrap_counter ^= 1;
538 		vq->last_avail_idx -= vq->size;
539 	}
540 }
541 
542 void __vhost_log_cache_write(struct virtio_net *dev,
543 		struct vhost_virtqueue *vq,
544 		uint64_t addr, uint64_t len);
545 void __vhost_log_cache_write_iova(struct virtio_net *dev,
546 		struct vhost_virtqueue *vq,
547 		uint64_t iova, uint64_t len);
548 void __vhost_log_cache_sync(struct virtio_net *dev,
549 		struct vhost_virtqueue *vq);
550 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
551 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
552 			    uint64_t iova, uint64_t len);
553 
554 static __rte_always_inline void
555 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
556 {
557 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
558 		__vhost_log_write(dev, addr, len);
559 }
560 
561 static __rte_always_inline void
562 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
563 {
564 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
565 		__vhost_log_cache_sync(dev, vq);
566 }
567 
568 static __rte_always_inline void
569 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
570 			uint64_t addr, uint64_t len)
571 {
572 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
573 		__vhost_log_cache_write(dev, vq, addr, len);
574 }
575 
576 static __rte_always_inline void
577 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
578 			uint64_t offset, uint64_t len)
579 {
580 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
581 		if (unlikely(vq->log_guest_addr == 0))
582 			return;
583 		__vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
584 					len);
585 	}
586 }
587 
588 static __rte_always_inline void
589 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
590 		     uint64_t offset, uint64_t len)
591 {
592 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
593 		if (unlikely(vq->log_guest_addr == 0))
594 			return;
595 		__vhost_log_write(dev, vq->log_guest_addr + offset, len);
596 	}
597 }
598 
599 static __rte_always_inline void
600 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
601 			   uint64_t iova, uint64_t len)
602 {
603 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
604 		return;
605 
606 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
607 		__vhost_log_cache_write_iova(dev, vq, iova, len);
608 	else
609 		__vhost_log_cache_write(dev, vq, iova, len);
610 }
611 
612 static __rte_always_inline void
613 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
614 			   uint64_t iova, uint64_t len)
615 {
616 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
617 		return;
618 
619 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
620 		__vhost_log_write_iova(dev, vq, iova, len);
621 	else
622 		__vhost_log_write(dev, iova, len);
623 }
624 
625 extern int vhost_config_log_level;
626 extern int vhost_data_log_level;
627 
628 #define VHOST_LOG_CONFIG(level, fmt, args...)			\
629 	rte_log(RTE_LOG_ ## level, vhost_config_log_level,	\
630 		"VHOST_CONFIG: " fmt, ##args)
631 
632 #define VHOST_LOG_DATA(level, fmt, args...) \
633 	(void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ?	\
634 	 rte_log(RTE_LOG_ ## level,  vhost_data_log_level,	\
635 		"VHOST_DATA : " fmt, ##args) :			\
636 	 0)
637 
638 #ifdef RTE_LIBRTE_VHOST_DEBUG
639 #define VHOST_MAX_PRINT_BUFF 6072
640 #define PRINT_PACKET(device, addr, size, header) do { \
641 	char *pkt_addr = (char *)(addr); \
642 	unsigned int index; \
643 	char packet[VHOST_MAX_PRINT_BUFF]; \
644 	\
645 	if ((header)) \
646 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
647 	else \
648 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
649 	for (index = 0; index < (size); index++) { \
650 		snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
651 			"%02hhx ", pkt_addr[index]); \
652 	} \
653 	snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
654 	\
655 	VHOST_LOG_DATA(DEBUG, "%s", packet); \
656 } while (0)
657 #else
658 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
659 #endif
660 
661 extern struct virtio_net *vhost_devices[RTE_MAX_VHOST_DEVICE];
662 
663 #define VHOST_BINARY_SEARCH_THRESH 256
664 
665 static __rte_always_inline int guest_page_addrcmp(const void *p1,
666 						const void *p2)
667 {
668 	const struct guest_page *page1 = (const struct guest_page *)p1;
669 	const struct guest_page *page2 = (const struct guest_page *)p2;
670 
671 	if (page1->guest_phys_addr > page2->guest_phys_addr)
672 		return 1;
673 	if (page1->guest_phys_addr < page2->guest_phys_addr)
674 		return -1;
675 
676 	return 0;
677 }
678 
679 static __rte_always_inline int guest_page_rangecmp(const void *p1, const void *p2)
680 {
681 	const struct guest_page *page1 = (const struct guest_page *)p1;
682 	const struct guest_page *page2 = (const struct guest_page *)p2;
683 
684 	if (page1->guest_phys_addr >= page2->guest_phys_addr) {
685 		if (page1->guest_phys_addr < page2->guest_phys_addr + page2->size)
686 			return 0;
687 		else
688 			return 1;
689 	} else
690 		return -1;
691 }
692 
693 static __rte_always_inline rte_iova_t
694 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
695 	uint64_t gpa_size, uint64_t *hpa_size)
696 {
697 	uint32_t i;
698 	struct guest_page *page;
699 	struct guest_page key;
700 
701 	*hpa_size = gpa_size;
702 	if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
703 		key.guest_phys_addr = gpa;
704 		page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
705 			       sizeof(struct guest_page), guest_page_rangecmp);
706 		if (page) {
707 			if (gpa + gpa_size <=
708 					page->guest_phys_addr + page->size) {
709 				return gpa - page->guest_phys_addr +
710 					page->host_iova;
711 			} else if (gpa < page->guest_phys_addr +
712 						page->size) {
713 				*hpa_size = page->guest_phys_addr +
714 					page->size - gpa;
715 				return gpa - page->guest_phys_addr +
716 					page->host_iova;
717 			}
718 		}
719 	} else {
720 		for (i = 0; i < dev->nr_guest_pages; i++) {
721 			page = &dev->guest_pages[i];
722 
723 			if (gpa >= page->guest_phys_addr) {
724 				if (gpa + gpa_size <=
725 					page->guest_phys_addr + page->size) {
726 					return gpa - page->guest_phys_addr +
727 						page->host_iova;
728 				} else if (gpa < page->guest_phys_addr +
729 							page->size) {
730 					*hpa_size = page->guest_phys_addr +
731 						page->size - gpa;
732 					return gpa - page->guest_phys_addr +
733 						page->host_iova;
734 				}
735 			}
736 		}
737 	}
738 
739 	*hpa_size = 0;
740 	return 0;
741 }
742 
743 /* Convert guest physical address to host physical address */
744 static __rte_always_inline rte_iova_t
745 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
746 {
747 	rte_iova_t hpa;
748 	uint64_t hpa_size;
749 
750 	hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
751 	return hpa_size == size ? hpa : 0;
752 }
753 
754 static __rte_always_inline uint64_t
755 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
756 {
757 	struct rte_vhost_mem_region *r;
758 	uint32_t i;
759 
760 	if (unlikely(!dev || !dev->mem))
761 		return 0;
762 
763 	for (i = 0; i < dev->mem->nregions; i++) {
764 		r = &dev->mem->regions[i];
765 
766 		if (vva >= r->host_user_addr &&
767 		    vva + len <  r->host_user_addr + r->size) {
768 			return r->guest_phys_addr + vva - r->host_user_addr;
769 		}
770 	}
771 	return 0;
772 }
773 
774 static __rte_always_inline struct virtio_net *
775 get_device(int vid)
776 {
777 	struct virtio_net *dev = vhost_devices[vid];
778 
779 	if (unlikely(!dev)) {
780 		VHOST_LOG_CONFIG(ERR,
781 			"(%d) device not found.\n", vid);
782 	}
783 
784 	return dev;
785 }
786 
787 int vhost_new_device(void);
788 void cleanup_device(struct virtio_net *dev, int destroy);
789 void reset_device(struct virtio_net *dev);
790 void vhost_destroy_device(int);
791 void vhost_destroy_device_notify(struct virtio_net *dev);
792 
793 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
794 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
795 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
796 
797 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
798 
799 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
800 
801 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
802 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags, bool stats_enabled);
803 void vhost_enable_extbuf(int vid);
804 void vhost_enable_linearbuf(int vid);
805 int vhost_enable_guest_notification(struct virtio_net *dev,
806 		struct vhost_virtqueue *vq, int enable);
807 
808 struct rte_vhost_device_ops const *vhost_driver_callback_get(const char *path);
809 
810 /*
811  * Backend-specific cleanup.
812  *
813  * TODO: fix it; we have one backend now
814  */
815 void vhost_backend_cleanup(struct virtio_net *dev);
816 
817 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
818 			uint64_t iova, uint64_t *len, uint8_t perm);
819 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
820 			struct vhost_virtqueue *vq,
821 			uint64_t desc_addr, uint64_t desc_len);
822 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
823 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
824 		uint64_t log_addr);
825 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
826 
827 static __rte_always_inline uint64_t
828 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
829 			uint64_t iova, uint64_t *len, uint8_t perm)
830 {
831 	if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
832 		return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
833 
834 	return __vhost_iova_to_vva(dev, vq, iova, len, perm);
835 }
836 
837 #define vhost_avail_event(vr) \
838 	(*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
839 #define vhost_used_event(vr) \
840 	(*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
841 
842 /*
843  * The following is used with VIRTIO_RING_F_EVENT_IDX.
844  * Assuming a given event_idx value from the other size, if we have
845  * just incremented index from old to new_idx, should we trigger an
846  * event?
847  */
848 static __rte_always_inline int
849 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
850 {
851 	return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
852 }
853 
854 static __rte_always_inline void
855 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
856 {
857 	/* Flush used->idx update before we read avail->flags. */
858 	rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
859 
860 	/* Don't kick guest if we don't reach index specified by guest. */
861 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
862 		uint16_t old = vq->signalled_used;
863 		uint16_t new = vq->last_used_idx;
864 		bool signalled_used_valid = vq->signalled_used_valid;
865 
866 		vq->signalled_used = new;
867 		vq->signalled_used_valid = true;
868 
869 		VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
870 			__func__,
871 			vhost_used_event(vq),
872 			old, new);
873 
874 		if ((vhost_need_event(vhost_used_event(vq), new, old) &&
875 					(vq->callfd >= 0)) ||
876 				unlikely(!signalled_used_valid)) {
877 			eventfd_write(vq->callfd, (eventfd_t) 1);
878 			if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
879 				vq->stats.guest_notifications++;
880 			if (dev->notify_ops->guest_notified)
881 				dev->notify_ops->guest_notified(dev->vid);
882 		}
883 	} else {
884 		/* Kick the guest if necessary. */
885 		if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
886 				&& (vq->callfd >= 0)) {
887 			eventfd_write(vq->callfd, (eventfd_t)1);
888 			if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
889 				vq->stats.guest_notifications++;
890 			if (dev->notify_ops->guest_notified)
891 				dev->notify_ops->guest_notified(dev->vid);
892 		}
893 	}
894 }
895 
896 static __rte_always_inline void
897 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
898 {
899 	uint16_t old, new, off, off_wrap;
900 	bool signalled_used_valid, kick = false;
901 
902 	/* Flush used desc update. */
903 	rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
904 
905 	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
906 		if (vq->driver_event->flags !=
907 				VRING_EVENT_F_DISABLE)
908 			kick = true;
909 		goto kick;
910 	}
911 
912 	old = vq->signalled_used;
913 	new = vq->last_used_idx;
914 	vq->signalled_used = new;
915 	signalled_used_valid = vq->signalled_used_valid;
916 	vq->signalled_used_valid = true;
917 
918 	if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
919 		if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
920 			kick = true;
921 		goto kick;
922 	}
923 
924 	if (unlikely(!signalled_used_valid)) {
925 		kick = true;
926 		goto kick;
927 	}
928 
929 	rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
930 
931 	off_wrap = vq->driver_event->off_wrap;
932 	off = off_wrap & ~(1 << 15);
933 
934 	if (new <= old)
935 		old -= vq->size;
936 
937 	if (vq->used_wrap_counter != off_wrap >> 15)
938 		off -= vq->size;
939 
940 	if (vhost_need_event(off, new, old))
941 		kick = true;
942 kick:
943 	if (kick) {
944 		eventfd_write(vq->callfd, (eventfd_t)1);
945 		if (dev->notify_ops->guest_notified)
946 			dev->notify_ops->guest_notified(dev->vid);
947 	}
948 }
949 
950 static __rte_always_inline void
951 free_ind_table(void *idesc)
952 {
953 	rte_free(idesc);
954 }
955 
956 static __rte_always_inline void
957 restore_mbuf(struct rte_mbuf *m)
958 {
959 	uint32_t mbuf_size, priv_size;
960 
961 	while (m) {
962 		priv_size = rte_pktmbuf_priv_size(m->pool);
963 		mbuf_size = sizeof(struct rte_mbuf) + priv_size;
964 		/* start of buffer is after mbuf structure and priv data */
965 
966 		m->buf_addr = (char *)m + mbuf_size;
967 		m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
968 		m = m->next;
969 	}
970 }
971 
972 static __rte_always_inline bool
973 mbuf_is_consumed(struct rte_mbuf *m)
974 {
975 	while (m) {
976 		if (rte_mbuf_refcnt_read(m) > 1)
977 			return false;
978 		m = m->next;
979 	}
980 
981 	return true;
982 }
983 #endif /* _VHOST_NET_CDEV_H_ */
984