xref: /dpdk/lib/vhost/vhost.h (revision e9fd1ebf981f361844aea9ec94e17f4bda5e1479)
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 <stdlib.h>
11 #include <sys/queue.h>
12 #include <unistd.h>
13 #include <linux/virtio_net.h>
14 #include <sys/socket.h>
15 #include <linux/if.h>
16 #include <sys/mman.h>
17 
18 #include <rte_log.h>
19 #include <rte_ether.h>
20 #include <rte_malloc.h>
21 #include <rte_dmadev.h>
22 
23 #include "rte_vhost.h"
24 #include "vdpa_driver.h"
25 
26 #include "rte_vhost_async.h"
27 
28 /* Used to indicate that the device is running on a data core */
29 #define VIRTIO_DEV_RUNNING ((uint32_t)1 << 0)
30 /* Used to indicate that the device is ready to operate */
31 #define VIRTIO_DEV_READY ((uint32_t)1 << 1)
32 /* Used to indicate that the built-in vhost net device backend is enabled */
33 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET ((uint32_t)1 << 2)
34 /* Used to indicate that the device has its own data path and configured */
35 #define VIRTIO_DEV_VDPA_CONFIGURED ((uint32_t)1 << 3)
36 /* Used to indicate that the feature negotiation failed */
37 #define VIRTIO_DEV_FEATURES_FAILED ((uint32_t)1 << 4)
38 /* Used to indicate that the virtio_net tx code should fill TX ol_flags */
39 #define VIRTIO_DEV_LEGACY_OL_FLAGS ((uint32_t)1 << 5)
40 /*  Used to indicate the application has requested statistics collection */
41 #define VIRTIO_DEV_STATS_ENABLED ((uint32_t)1 << 6)
42 /*  Used to indicate the application has requested iommu support */
43 #define VIRTIO_DEV_SUPPORT_IOMMU ((uint32_t)1 << 7)
44 
45 /* Backend value set by guest. */
46 #define VIRTIO_DEV_STOPPED -1
47 
48 #define BUF_VECTOR_MAX 256
49 
50 #define VHOST_LOG_CACHE_NR 32
51 
52 #define MAX_PKT_BURST 32
53 
54 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST)
55 #define VHOST_MAX_ASYNC_VEC 2048
56 #define VIRTIO_MAX_RX_PKTLEN 9728U
57 #define VHOST_DMA_MAX_COPY_COMPLETE ((VIRTIO_MAX_RX_PKTLEN / RTE_MBUF_DEFAULT_DATAROOM) \
58 		* MAX_PKT_BURST)
59 
60 #define PACKED_DESC_ENQUEUE_USED_FLAG(w)	\
61 	((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
62 		VRING_DESC_F_WRITE)
63 #define PACKED_DESC_DEQUEUE_USED_FLAG(w)	\
64 	((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
65 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
66 					 VRING_DESC_F_INDIRECT)
67 
68 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
69 			    sizeof(struct vring_packed_desc))
70 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
71 
72 #ifdef VHOST_GCC_UNROLL_PRAGMA
73 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
74 	for (iter = val; iter < size; iter++)
75 #endif
76 
77 #ifdef VHOST_CLANG_UNROLL_PRAGMA
78 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
79 	for (iter = val; iter < size; iter++)
80 #endif
81 
82 #ifdef VHOST_ICC_UNROLL_PRAGMA
83 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
84 	for (iter = val; iter < size; iter++)
85 #endif
86 
87 #ifndef vhost_for_each_try_unroll
88 #define vhost_for_each_try_unroll(iter, val, num) \
89 	for (iter = val; iter < num; iter++)
90 #endif
91 
92 struct virtio_net;
93 struct vhost_virtqueue;
94 
95 typedef void (*vhost_iotlb_remove_notify)(uint64_t addr, uint64_t off, uint64_t size);
96 
97 typedef int (*vhost_iotlb_miss_cb)(struct virtio_net *dev, uint64_t iova, uint8_t perm);
98 
99 typedef int (*vhost_vring_inject_irq_cb)(struct virtio_net *dev, struct vhost_virtqueue *vq);
100 /**
101  * Structure that contains backend-specific ops.
102  */
103 struct vhost_backend_ops {
104 	vhost_iotlb_remove_notify iotlb_remove_notify;
105 	vhost_iotlb_miss_cb iotlb_miss;
106 	vhost_vring_inject_irq_cb inject_irq;
107 };
108 
109 /**
110  * Structure contains buffer address, length and descriptor index
111  * from vring to do scatter RX.
112  */
113 struct buf_vector {
114 	uint64_t buf_iova;
115 	uint64_t buf_addr;
116 	uint32_t buf_len;
117 	uint32_t desc_idx;
118 };
119 
120 /*
121  * Structure contains the info for each batched memory copy.
122  */
123 struct batch_copy_elem {
124 	void *dst;
125 	void *src;
126 	uint32_t len;
127 	uint64_t log_addr;
128 };
129 
130 /*
131  * Structure that contains the info for batched dirty logging.
132  */
133 struct log_cache_entry {
134 	uint32_t offset;
135 	unsigned long val;
136 };
137 
138 struct vring_used_elem_packed {
139 	uint16_t id;
140 	uint16_t flags;
141 	uint32_t len;
142 	uint32_t count;
143 };
144 
145 /**
146  * Virtqueue statistics
147  */
148 struct virtqueue_stats {
149 	uint64_t packets;
150 	uint64_t bytes;
151 	uint64_t multicast;
152 	uint64_t broadcast;
153 	/* Size bins in array as RFC 2819, undersized [0], 64 [1], etc */
154 	uint64_t size_bins[8];
155 	uint64_t iotlb_hits;
156 	uint64_t iotlb_misses;
157 	uint64_t inflight_submitted;
158 	uint64_t inflight_completed;
159 	uint64_t mbuf_alloc_failed;
160 	uint64_t guest_notifications_suppressed;
161 	/* Counters below are atomic, and should be incremented as such. */
162 	RTE_ATOMIC(uint64_t) guest_notifications;
163 	RTE_ATOMIC(uint64_t) guest_notifications_offloaded;
164 	RTE_ATOMIC(uint64_t) guest_notifications_error;
165 };
166 
167 /**
168  * iovec
169  */
170 struct vhost_iovec {
171 	void *src_addr;
172 	void *dst_addr;
173 	size_t len;
174 };
175 
176 /**
177  * iovec iterator
178  */
179 struct vhost_iov_iter {
180 	/** pointer to the iovec array */
181 	struct vhost_iovec *iov;
182 	/** number of iovec in this iterator */
183 	unsigned long nr_segs;
184 };
185 
186 struct async_dma_vchan_info {
187 	/* circular array to track if packet copy completes */
188 	bool **pkts_cmpl_flag_addr;
189 
190 	/* max elements in 'pkts_cmpl_flag_addr' */
191 	uint16_t ring_size;
192 	/* ring index mask for 'pkts_cmpl_flag_addr' */
193 	uint16_t ring_mask;
194 
195 	/**
196 	 * DMA virtual channel lock. Although it is able to bind DMA
197 	 * virtual channels to data plane threads, vhost control plane
198 	 * thread could call data plane functions too, thus causing
199 	 * DMA device contention.
200 	 *
201 	 * For example, in VM exit case, vhost control plane thread needs
202 	 * to clear in-flight packets before disable vring, but there could
203 	 * be anotther data plane thread is enqueuing packets to the same
204 	 * vring with the same DMA virtual channel. As dmadev PMD functions
205 	 * are lock-free, the control plane and data plane threads could
206 	 * operate the same DMA virtual channel at the same time.
207 	 */
208 	rte_spinlock_t dma_lock;
209 };
210 
211 struct async_dma_info {
212 	struct async_dma_vchan_info *vchans;
213 	/* number of registered virtual channels */
214 	uint16_t nr_vchans;
215 };
216 
217 extern struct async_dma_info dma_copy_track[RTE_DMADEV_DEFAULT_MAX];
218 
219 /**
220  * inflight async packet information
221  */
222 struct async_inflight_info {
223 	struct rte_mbuf *mbuf;
224 	uint16_t descs; /* num of descs inflight */
225 	uint16_t nr_buffers; /* num of buffers inflight for packed ring */
226 	struct virtio_net_hdr nethdr;
227 };
228 
229 struct vhost_async {
230 	struct vhost_iov_iter iov_iter[VHOST_MAX_ASYNC_IT];
231 	struct vhost_iovec iovec[VHOST_MAX_ASYNC_VEC];
232 	uint16_t iter_idx;
233 	uint16_t iovec_idx;
234 
235 	/* data transfer status */
236 	struct async_inflight_info *pkts_info;
237 	/**
238 	 * Packet reorder array. "true" indicates that DMA device
239 	 * completes all copies for the packet.
240 	 *
241 	 * Note that this array could be written by multiple threads
242 	 * simultaneously. For example, in the case of thread0 and
243 	 * thread1 RX packets from NIC and then enqueue packets to
244 	 * vring0 and vring1 with own DMA device DMA0 and DMA1, it's
245 	 * possible for thread0 to get completed copies belonging to
246 	 * vring1 from DMA0, while thread0 is calling rte_vhost_poll
247 	 * _enqueue_completed() for vring0 and thread1 is calling
248 	 * rte_vhost_submit_enqueue_burst() for vring1. In this case,
249 	 * vq->access_lock cannot protect pkts_cmpl_flag of vring1.
250 	 *
251 	 * However, since offloading is per-packet basis, each packet
252 	 * flag will only be written by one thread. And single byte
253 	 * write is atomic, so no lock for pkts_cmpl_flag is needed.
254 	 */
255 	bool *pkts_cmpl_flag;
256 	uint16_t pkts_idx;
257 	uint16_t pkts_inflight_n;
258 	union {
259 		struct vring_used_elem  *descs_split;
260 		struct vring_used_elem_packed *buffers_packed;
261 	};
262 	union {
263 		uint16_t desc_idx_split;
264 		uint16_t buffer_idx_packed;
265 	};
266 	union {
267 		uint16_t last_desc_idx_split;
268 		uint16_t last_buffer_idx_packed;
269 	};
270 };
271 
272 /**
273  * Structure contains variables relevant to RX/TX virtqueues.
274  */
275 struct vhost_virtqueue {
276 	union {
277 		struct vring_desc	*desc;
278 		struct vring_packed_desc   *desc_packed;
279 	};
280 	union {
281 		struct vring_avail	*avail;
282 		struct vring_packed_desc_event *driver_event;
283 	};
284 	union {
285 		struct vring_used	*used;
286 		struct vring_packed_desc_event *device_event;
287 	};
288 	uint16_t		size;
289 
290 	uint16_t		last_avail_idx;
291 	uint16_t		last_used_idx;
292 	/* Last used index we notify to front end. */
293 	uint16_t		signalled_used;
294 	bool			signalled_used_valid;
295 #define VIRTIO_INVALID_EVENTFD		(-1)
296 #define VIRTIO_UNINITIALIZED_EVENTFD	(-2)
297 
298 	bool			enabled;
299 	/* Protected by vq->access_lock */
300 	bool			access_ok __rte_guarded_var;
301 	bool			ready;
302 
303 	rte_rwlock_t		access_lock;
304 
305 
306 	union {
307 		struct vring_used_elem  *shadow_used_split;
308 		struct vring_used_elem_packed *shadow_used_packed;
309 	};
310 	uint16_t                shadow_used_idx;
311 	/* Record packed ring enqueue latest desc cache aligned index */
312 	uint16_t		shadow_aligned_idx;
313 	/* Record packed ring first dequeue desc index */
314 	uint16_t		shadow_last_used_idx;
315 
316 	uint16_t		batch_copy_nb_elems;
317 	struct batch_copy_elem	*batch_copy_elems;
318 	int			numa_node;
319 	bool			used_wrap_counter;
320 	bool			avail_wrap_counter;
321 
322 	/* Physical address of used ring, for logging */
323 	uint16_t		log_cache_nb_elem;
324 	uint64_t		log_guest_addr;
325 	struct log_cache_entry	*log_cache;
326 
327 	rte_rwlock_t	iotlb_lock;
328 
329 	/* Used to notify the guest (trigger interrupt) */
330 	int			callfd;
331 	/* Currently unused as polling mode is enabled */
332 	int			kickfd;
333 
334 	/* Index of this vq in dev->virtqueue[] */
335 	uint32_t		index;
336 
337 	/* inflight share memory info */
338 	union {
339 		struct rte_vhost_inflight_info_split *inflight_split;
340 		struct rte_vhost_inflight_info_packed *inflight_packed;
341 	};
342 	struct rte_vhost_resubmit_info *resubmit_inflight;
343 	uint64_t		global_counter;
344 
345 	struct vhost_async	*async __rte_guarded_var;
346 
347 	int			notif_enable;
348 #define VIRTIO_UNINITIALIZED_NOTIF	(-1)
349 
350 	struct vhost_vring_addr ring_addrs;
351 	struct virtqueue_stats	stats;
352 
353 	RTE_ATOMIC(bool) irq_pending;
354 } __rte_cache_aligned;
355 
356 /* Virtio device status as per Virtio specification */
357 #define VIRTIO_DEVICE_STATUS_RESET		0x00
358 #define VIRTIO_DEVICE_STATUS_ACK		0x01
359 #define VIRTIO_DEVICE_STATUS_DRIVER		0x02
360 #define VIRTIO_DEVICE_STATUS_DRIVER_OK		0x04
361 #define VIRTIO_DEVICE_STATUS_FEATURES_OK	0x08
362 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET	0x40
363 #define VIRTIO_DEVICE_STATUS_FAILED		0x80
364 
365 #define VHOST_MAX_VRING			0x100
366 #define VHOST_MAX_QUEUE_PAIRS		0x80
367 
368 /* Declare IOMMU related bits for older kernels */
369 #ifndef VIRTIO_F_IOMMU_PLATFORM
370 
371 #define VIRTIO_F_IOMMU_PLATFORM 33
372 
373 struct vhost_iotlb_msg {
374 	__u64 iova;
375 	__u64 size;
376 	__u64 uaddr;
377 #define VHOST_ACCESS_RO      0x1
378 #define VHOST_ACCESS_WO      0x2
379 #define VHOST_ACCESS_RW      0x3
380 	__u8 perm;
381 #define VHOST_IOTLB_MISS           1
382 #define VHOST_IOTLB_UPDATE         2
383 #define VHOST_IOTLB_INVALIDATE     3
384 #define VHOST_IOTLB_ACCESS_FAIL    4
385 	__u8 type;
386 };
387 
388 #define VHOST_IOTLB_MSG 0x1
389 
390 struct vhost_msg {
391 	int type;
392 	union {
393 		struct vhost_iotlb_msg iotlb;
394 		__u8 padding[64];
395 	};
396 };
397 #endif
398 
399 /*
400  * Define virtio 1.0 for older kernels
401  */
402 #ifndef VIRTIO_F_VERSION_1
403  #define VIRTIO_F_VERSION_1 32
404 #endif
405 
406 /* Declare packed ring related bits for older kernels */
407 #ifndef VIRTIO_F_RING_PACKED
408 
409 #define VIRTIO_F_RING_PACKED 34
410 
411 struct vring_packed_desc {
412 	uint64_t addr;
413 	uint32_t len;
414 	uint16_t id;
415 	uint16_t flags;
416 };
417 
418 struct vring_packed_desc_event {
419 	uint16_t off_wrap;
420 	uint16_t flags;
421 };
422 #endif
423 
424 /*
425  * Declare below packed ring defines unconditionally
426  * as Kernel header might use different names.
427  */
428 #define VRING_DESC_F_AVAIL	(1ULL << 7)
429 #define VRING_DESC_F_USED	(1ULL << 15)
430 
431 #define VRING_EVENT_F_ENABLE 0x0
432 #define VRING_EVENT_F_DISABLE 0x1
433 #define VRING_EVENT_F_DESC 0x2
434 
435 /*
436  * Available and used descs are in same order
437  */
438 #ifndef VIRTIO_F_IN_ORDER
439 #define VIRTIO_F_IN_ORDER      35
440 #endif
441 
442 /* Features supported by this builtin vhost-user net driver. */
443 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
444 				(1ULL << VIRTIO_F_ANY_LAYOUT) | \
445 				(1ULL << VIRTIO_NET_F_CTRL_VQ) | \
446 				(1ULL << VIRTIO_NET_F_MQ)      | \
447 				(1ULL << VIRTIO_F_VERSION_1)   | \
448 				(1ULL << VIRTIO_NET_F_GSO) | \
449 				(1ULL << VIRTIO_NET_F_HOST_TSO4) | \
450 				(1ULL << VIRTIO_NET_F_HOST_TSO6) | \
451 				(1ULL << VIRTIO_NET_F_HOST_UFO) | \
452 				(1ULL << VIRTIO_NET_F_HOST_ECN) | \
453 				(1ULL << VIRTIO_NET_F_CSUM)    | \
454 				(1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
455 				(1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
456 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
457 				(1ULL << VIRTIO_NET_F_GUEST_UFO) | \
458 				(1ULL << VIRTIO_NET_F_GUEST_ECN) | \
459 				(1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
460 				(1ULL << VIRTIO_RING_F_EVENT_IDX) | \
461 				(1ULL << VIRTIO_F_IN_ORDER) | \
462 				(1ULL << VIRTIO_F_IOMMU_PLATFORM))
463 
464 
465 struct guest_page {
466 	uint64_t guest_phys_addr;
467 	uint64_t host_iova;
468 	uint64_t host_user_addr;
469 	uint64_t size;
470 };
471 
472 struct inflight_mem_info {
473 	int		fd;
474 	void		*addr;
475 	uint64_t	size;
476 };
477 
478 /**
479  * Device structure contains all configuration information relating
480  * to the device.
481  */
482 struct virtio_net {
483 	/* Frontend (QEMU) memory and memory region information */
484 	struct rte_vhost_memory	*mem;
485 	uint64_t		features;
486 	uint64_t		protocol_features;
487 	int			vid;
488 	uint32_t		flags;
489 	uint16_t		vhost_hlen;
490 	/* to tell if we need broadcast rarp packet */
491 	RTE_ATOMIC(int16_t)	broadcast_rarp;
492 	uint32_t		nr_vring;
493 	int			async_copy;
494 
495 	int			extbuf;
496 	int			linearbuf;
497 	struct vhost_virtqueue	*virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
498 
499 	rte_rwlock_t	iotlb_pending_lock;
500 	struct vhost_iotlb_entry *iotlb_pool;
501 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
502 	TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
503 	int				iotlb_cache_nr;
504 	rte_spinlock_t	iotlb_free_lock;
505 	SLIST_HEAD(, vhost_iotlb_entry) iotlb_free_list;
506 
507 	struct inflight_mem_info *inflight_info;
508 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
509 	char			ifname[IF_NAME_SZ];
510 	uint64_t		log_size;
511 	uint64_t		log_base;
512 	uint64_t		log_addr;
513 	struct rte_ether_addr	mac;
514 	uint16_t		mtu;
515 	uint8_t			status;
516 
517 	struct rte_vhost_device_ops const *notify_ops;
518 
519 	uint32_t		nr_guest_pages;
520 	uint32_t		max_guest_pages;
521 	struct guest_page       *guest_pages;
522 
523 	int			backend_req_fd;
524 	rte_spinlock_t		backend_req_lock;
525 
526 	int			postcopy_ufd;
527 	int			postcopy_listening;
528 	int			vduse_ctrl_fd;
529 	int			vduse_dev_fd;
530 
531 	struct vhost_virtqueue	*cvq;
532 
533 	struct rte_vdpa_device *vdpa_dev;
534 
535 	/* context data for the external message handlers */
536 	void			*extern_data;
537 	/* pre and post vhost user message handlers for the device */
538 	struct rte_vhost_user_extern_ops extern_ops;
539 
540 	struct vhost_backend_ops *backend_ops;
541 } __rte_cache_aligned;
542 
543 static inline void
544 vq_assert_lock__(struct virtio_net *dev, struct vhost_virtqueue *vq, const char *func)
545 	__rte_assert_exclusive_lock(&vq->access_lock)
546 {
547 	if (unlikely(!rte_rwlock_write_is_locked(&vq->access_lock)))
548 		rte_panic("VHOST_CONFIG: (%s) %s() called without access lock taken.\n",
549 			dev->ifname, func);
550 }
551 #define vq_assert_lock(dev, vq) vq_assert_lock__(dev, vq, __func__)
552 
553 static __rte_always_inline bool
554 vq_is_packed(struct virtio_net *dev)
555 {
556 	return dev->features & (1ull << VIRTIO_F_RING_PACKED);
557 }
558 
559 static inline bool
560 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
561 {
562 	uint16_t flags = rte_atomic_load_explicit((unsigned short __rte_atomic *)&desc->flags,
563 		rte_memory_order_acquire);
564 
565 	return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
566 		wrap_counter != !!(flags & VRING_DESC_F_USED);
567 }
568 
569 static inline void
570 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
571 {
572 	vq->last_used_idx += num;
573 	if (vq->last_used_idx >= vq->size) {
574 		vq->used_wrap_counter ^= 1;
575 		vq->last_used_idx -= vq->size;
576 	}
577 }
578 
579 static inline void
580 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
581 {
582 	vq->last_avail_idx += num;
583 	if (vq->last_avail_idx >= vq->size) {
584 		vq->avail_wrap_counter ^= 1;
585 		vq->last_avail_idx -= vq->size;
586 	}
587 }
588 
589 void __vhost_log_cache_write(struct virtio_net *dev,
590 		struct vhost_virtqueue *vq,
591 		uint64_t addr, uint64_t len);
592 void __vhost_log_cache_write_iova(struct virtio_net *dev,
593 		struct vhost_virtqueue *vq,
594 		uint64_t iova, uint64_t len)
595 	__rte_shared_locks_required(&vq->iotlb_lock);
596 void __vhost_log_cache_sync(struct virtio_net *dev,
597 		struct vhost_virtqueue *vq);
598 
599 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
600 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
601 			    uint64_t iova, uint64_t len)
602 	__rte_shared_locks_required(&vq->iotlb_lock);
603 
604 static __rte_always_inline void
605 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
606 {
607 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
608 		__vhost_log_write(dev, addr, len);
609 }
610 
611 static __rte_always_inline void
612 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
613 {
614 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
615 		__vhost_log_cache_sync(dev, vq);
616 }
617 
618 static __rte_always_inline void
619 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
620 			uint64_t addr, uint64_t len)
621 {
622 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
623 		__vhost_log_cache_write(dev, vq, addr, len);
624 }
625 
626 static __rte_always_inline void
627 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
628 			uint64_t offset, uint64_t len)
629 {
630 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
631 		if (unlikely(vq->log_guest_addr == 0))
632 			return;
633 		__vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
634 					len);
635 	}
636 }
637 
638 static __rte_always_inline void
639 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
640 		     uint64_t offset, uint64_t len)
641 {
642 	if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
643 		if (unlikely(vq->log_guest_addr == 0))
644 			return;
645 		__vhost_log_write(dev, vq->log_guest_addr + offset, len);
646 	}
647 }
648 
649 static __rte_always_inline void
650 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
651 			   uint64_t iova, uint64_t len)
652 	__rte_shared_locks_required(&vq->iotlb_lock)
653 {
654 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
655 		return;
656 
657 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
658 		__vhost_log_cache_write_iova(dev, vq, iova, len);
659 	else
660 		__vhost_log_cache_write(dev, vq, iova, len);
661 }
662 
663 static __rte_always_inline void
664 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
665 			   uint64_t iova, uint64_t len)
666 	__rte_shared_locks_required(&vq->iotlb_lock)
667 {
668 	if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
669 		return;
670 
671 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
672 		__vhost_log_write_iova(dev, vq, iova, len);
673 	else
674 		__vhost_log_write(dev, iova, len);
675 }
676 
677 extern int vhost_config_log_level;
678 #define RTE_LOGTYPE_VHOST_CONFIG vhost_config_log_level
679 extern int vhost_data_log_level;
680 #define RTE_LOGTYPE_VHOST_DATA vhost_data_log_level
681 
682 #define VHOST_CONFIG_LOG(prefix, level, ...) \
683 	RTE_LOG_LINE_PREFIX(level, VHOST_CONFIG, "(%s) ", prefix, __VA_ARGS__)
684 
685 #define VHOST_DATA_LOG(prefix, level, ...) \
686 	RTE_LOG_DP_LINE_PREFIX(level, VHOST_DATA, "(%s) ", prefix, __VA_ARGS__)
687 
688 #ifdef RTE_LIBRTE_VHOST_DEBUG
689 #define VHOST_MAX_PRINT_BUFF 6072
690 #define PRINT_PACKET(device, addr, size, header) do { \
691 	char *pkt_addr = (char *)(addr); \
692 	unsigned int index; \
693 	char packet[VHOST_MAX_PRINT_BUFF]; \
694 	\
695 	if ((header)) \
696 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
697 	else \
698 		snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
699 	for (index = 0; index < (size); index++) { \
700 		snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
701 			"%02hhx ", pkt_addr[index]); \
702 	} \
703 	snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
704 	\
705 	RTE_LOG_DP(DEBUG, VHOST_DATA, "(%s) %s", dev->ifname, packet); \
706 } while (0)
707 #else
708 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
709 #endif
710 
711 extern struct virtio_net *vhost_devices[RTE_MAX_VHOST_DEVICE];
712 
713 #define VHOST_BINARY_SEARCH_THRESH 256
714 
715 static __rte_always_inline int guest_page_addrcmp(const void *p1,
716 						const void *p2)
717 {
718 	const struct guest_page *page1 = (const struct guest_page *)p1;
719 	const struct guest_page *page2 = (const struct guest_page *)p2;
720 
721 	if (page1->guest_phys_addr > page2->guest_phys_addr)
722 		return 1;
723 	if (page1->guest_phys_addr < page2->guest_phys_addr)
724 		return -1;
725 
726 	return 0;
727 }
728 
729 static __rte_always_inline int guest_page_rangecmp(const void *p1, const void *p2)
730 {
731 	const struct guest_page *page1 = (const struct guest_page *)p1;
732 	const struct guest_page *page2 = (const struct guest_page *)p2;
733 
734 	if (page1->guest_phys_addr >= page2->guest_phys_addr) {
735 		if (page1->guest_phys_addr < page2->guest_phys_addr + page2->size)
736 			return 0;
737 		else
738 			return 1;
739 	} else
740 		return -1;
741 }
742 
743 static __rte_always_inline rte_iova_t
744 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
745 	uint64_t gpa_size, uint64_t *hpa_size)
746 {
747 	uint32_t i;
748 	struct guest_page *page;
749 	struct guest_page key;
750 
751 	*hpa_size = gpa_size;
752 	if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
753 		key.guest_phys_addr = gpa;
754 		page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
755 			       sizeof(struct guest_page), guest_page_rangecmp);
756 		if (page) {
757 			if (gpa + gpa_size <=
758 					page->guest_phys_addr + page->size) {
759 				return gpa - page->guest_phys_addr +
760 					page->host_iova;
761 			} else if (gpa < page->guest_phys_addr +
762 						page->size) {
763 				*hpa_size = page->guest_phys_addr +
764 					page->size - gpa;
765 				return gpa - page->guest_phys_addr +
766 					page->host_iova;
767 			}
768 		}
769 	} else {
770 		for (i = 0; i < dev->nr_guest_pages; i++) {
771 			page = &dev->guest_pages[i];
772 
773 			if (gpa >= page->guest_phys_addr) {
774 				if (gpa + gpa_size <=
775 					page->guest_phys_addr + page->size) {
776 					return gpa - page->guest_phys_addr +
777 						page->host_iova;
778 				} else if (gpa < page->guest_phys_addr +
779 							page->size) {
780 					*hpa_size = page->guest_phys_addr +
781 						page->size - gpa;
782 					return gpa - page->guest_phys_addr +
783 						page->host_iova;
784 				}
785 			}
786 		}
787 	}
788 
789 	*hpa_size = 0;
790 	return 0;
791 }
792 
793 /* Convert guest physical address to host physical address */
794 static __rte_always_inline rte_iova_t
795 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
796 {
797 	rte_iova_t hpa;
798 	uint64_t hpa_size;
799 
800 	hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
801 	return hpa_size == size ? hpa : 0;
802 }
803 
804 static __rte_always_inline uint64_t
805 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
806 {
807 	struct rte_vhost_mem_region *r;
808 	uint32_t i;
809 
810 	if (unlikely(!dev || !dev->mem))
811 		return 0;
812 
813 	for (i = 0; i < dev->mem->nregions; i++) {
814 		r = &dev->mem->regions[i];
815 
816 		if (vva >= r->host_user_addr &&
817 		    vva + len <  r->host_user_addr + r->size) {
818 			return r->guest_phys_addr + vva - r->host_user_addr;
819 		}
820 	}
821 	return 0;
822 }
823 
824 static __rte_always_inline struct virtio_net *
825 get_device(int vid)
826 {
827 	struct virtio_net *dev = NULL;
828 
829 	if (likely(vid >= 0 && vid < RTE_MAX_VHOST_DEVICE))
830 		dev = vhost_devices[vid];
831 
832 	if (unlikely(!dev)) {
833 		VHOST_CONFIG_LOG("device", ERR, "(%d) device not found.", vid);
834 	}
835 
836 	return dev;
837 }
838 
839 int vhost_new_device(struct vhost_backend_ops *ops);
840 void cleanup_device(struct virtio_net *dev, int destroy);
841 void reset_device(struct virtio_net *dev);
842 void vhost_destroy_device(int);
843 void vhost_destroy_device_notify(struct virtio_net *dev);
844 
845 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
846 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
847 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
848 
849 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
850 
851 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
852 
853 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
854 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags, bool stats_enabled,
855 	bool support_iommu);
856 void vhost_enable_extbuf(int vid);
857 void vhost_enable_linearbuf(int vid);
858 int vhost_enable_guest_notification(struct virtio_net *dev,
859 		struct vhost_virtqueue *vq, int enable);
860 
861 struct rte_vhost_device_ops const *vhost_driver_callback_get(const char *path);
862 
863 /*
864  * Backend-specific cleanup.
865  *
866  * TODO: fix it; we have one backend now
867  */
868 void vhost_backend_cleanup(struct virtio_net *dev);
869 
870 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
871 			uint64_t iova, uint64_t *len, uint8_t perm)
872 	__rte_shared_locks_required(&vq->iotlb_lock);
873 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
874 			struct vhost_virtqueue *vq,
875 			uint64_t desc_addr, uint64_t desc_len)
876 	__rte_shared_locks_required(&vq->iotlb_lock);
877 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
878 	__rte_exclusive_locks_required(&vq->access_lock)
879 	__rte_shared_locks_required(&vq->iotlb_lock);
880 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
881 		uint64_t log_addr)
882 	__rte_shared_locks_required(&vq->iotlb_lock);
883 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
884 	__rte_exclusive_locks_required(&vq->access_lock);
885 
886 static __rte_always_inline uint64_t
887 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
888 			uint64_t iova, uint64_t *len, uint8_t perm)
889 	__rte_shared_locks_required(&vq->iotlb_lock)
890 {
891 	if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
892 		return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
893 
894 	return __vhost_iova_to_vva(dev, vq, iova, len, perm);
895 }
896 
897 #define vhost_avail_event(vr) \
898 	(*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
899 #define vhost_used_event(vr) \
900 	(*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
901 
902 /*
903  * The following is used with VIRTIO_RING_F_EVENT_IDX.
904  * Assuming a given event_idx value from the other size, if we have
905  * just incremented index from old to new_idx, should we trigger an
906  * event?
907  */
908 static __rte_always_inline int
909 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
910 {
911 	return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
912 }
913 
914 static __rte_always_inline void
915 vhost_vring_inject_irq(struct virtio_net *dev, struct vhost_virtqueue *vq)
916 {
917 	bool expected = false;
918 
919 	if (dev->notify_ops->guest_notify) {
920 		if (rte_atomic_compare_exchange_strong_explicit(&vq->irq_pending, &expected, true,
921 				  rte_memory_order_release, rte_memory_order_relaxed)) {
922 			if (dev->notify_ops->guest_notify(dev->vid, vq->index)) {
923 				if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
924 					rte_atomic_fetch_add_explicit(
925 						&vq->stats.guest_notifications_offloaded,
926 						1, rte_memory_order_relaxed);
927 				return;
928 			}
929 
930 			/* Offloading failed, fallback to direct IRQ injection */
931 			rte_atomic_store_explicit(&vq->irq_pending, false,
932 				rte_memory_order_release);
933 		} else {
934 			vq->stats.guest_notifications_suppressed++;
935 			return;
936 		}
937 	}
938 
939 	if (dev->backend_ops->inject_irq(dev, vq)) {
940 		if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
941 			rte_atomic_fetch_add_explicit(&vq->stats.guest_notifications_error,
942 				1, rte_memory_order_relaxed);
943 		return;
944 	}
945 
946 	if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
947 		rte_atomic_fetch_add_explicit(&vq->stats.guest_notifications,
948 			1, rte_memory_order_relaxed);
949 	if (dev->notify_ops->guest_notified)
950 		dev->notify_ops->guest_notified(dev->vid);
951 }
952 
953 static __rte_always_inline void
954 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
955 {
956 	/* Flush used->idx update before we read avail->flags. */
957 	rte_atomic_thread_fence(rte_memory_order_seq_cst);
958 
959 	/* Don't kick guest if we don't reach index specified by guest. */
960 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
961 		uint16_t old = vq->signalled_used;
962 		uint16_t new = vq->last_used_idx;
963 		bool signalled_used_valid = vq->signalled_used_valid;
964 
965 		vq->signalled_used = new;
966 		vq->signalled_used_valid = true;
967 
968 		VHOST_DATA_LOG(dev->ifname, DEBUG,
969 			"%s: used_event_idx=%d, old=%d, new=%d",
970 			__func__, vhost_used_event(vq), old, new);
971 
972 		if (vhost_need_event(vhost_used_event(vq), new, old) ||
973 				unlikely(!signalled_used_valid))
974 			vhost_vring_inject_irq(dev, vq);
975 	} else {
976 		/* Kick the guest if necessary. */
977 		if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT))
978 			vhost_vring_inject_irq(dev, vq);
979 	}
980 }
981 
982 static __rte_always_inline void
983 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
984 {
985 	uint16_t old, new, off, off_wrap;
986 	bool signalled_used_valid, kick = false;
987 
988 	/* Flush used desc update. */
989 	rte_atomic_thread_fence(rte_memory_order_seq_cst);
990 
991 	if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
992 		if (vq->driver_event->flags !=
993 				VRING_EVENT_F_DISABLE)
994 			kick = true;
995 		goto kick;
996 	}
997 
998 	old = vq->signalled_used;
999 	new = vq->last_used_idx;
1000 	vq->signalled_used = new;
1001 	signalled_used_valid = vq->signalled_used_valid;
1002 	vq->signalled_used_valid = true;
1003 
1004 	if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
1005 		if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
1006 			kick = true;
1007 		goto kick;
1008 	}
1009 
1010 	if (unlikely(!signalled_used_valid)) {
1011 		kick = true;
1012 		goto kick;
1013 	}
1014 
1015 	rte_atomic_thread_fence(rte_memory_order_acquire);
1016 
1017 	off_wrap = vq->driver_event->off_wrap;
1018 	off = off_wrap & ~(1 << 15);
1019 
1020 	if (new <= old)
1021 		old -= vq->size;
1022 
1023 	if (vq->used_wrap_counter != off_wrap >> 15)
1024 		off -= vq->size;
1025 
1026 	if (vhost_need_event(off, new, old))
1027 		kick = true;
1028 kick:
1029 	if (kick)
1030 		vhost_vring_inject_irq(dev, vq);
1031 }
1032 
1033 static __rte_always_inline void
1034 free_ind_table(void *idesc)
1035 {
1036 	rte_free(idesc);
1037 }
1038 
1039 static __rte_always_inline void
1040 restore_mbuf(struct rte_mbuf *m)
1041 {
1042 	uint32_t mbuf_size, priv_size;
1043 
1044 	while (m) {
1045 		priv_size = rte_pktmbuf_priv_size(m->pool);
1046 		mbuf_size = sizeof(struct rte_mbuf) + priv_size;
1047 		/* start of buffer is after mbuf structure and priv data */
1048 
1049 		m->buf_addr = (char *)m + mbuf_size;
1050 		rte_mbuf_iova_set(m, rte_mempool_virt2iova(m) + mbuf_size);
1051 		m = m->next;
1052 	}
1053 }
1054 
1055 static __rte_always_inline bool
1056 mbuf_is_consumed(struct rte_mbuf *m)
1057 {
1058 	while (m) {
1059 		if (rte_mbuf_refcnt_read(m) > 1)
1060 			return false;
1061 		m = m->next;
1062 	}
1063 
1064 	return true;
1065 }
1066 
1067 void mem_set_dump(struct virtio_net *dev, void *ptr, size_t size, bool enable, uint64_t alignment);
1068 
1069 #endif /* _VHOST_NET_CDEV_H_ */
1070