1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation
5 #ifndef _VHOST_NET_CDEV_H_
6 #define _VHOST_NET_CDEV_H_
10 #include <sys/types.h>
11 #include <sys/queue.h>
13 #include <linux/vhost.h>
14 #include <linux/virtio_net.h>
15 #include <sys/socket.h>
19 #include <rte_ether.h>
20 #include <rte_rwlock.h>
21 #include <rte_malloc.h>
23 #include "rte_vhost.h"
26 /* Used to indicate that the device is running on a data core */
27 #define VIRTIO_DEV_RUNNING 1
28 /* Used to indicate that the device is ready to operate */
29 #define VIRTIO_DEV_READY 2
30 /* Used to indicate that the built-in vhost net device backend is enabled */
31 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET 4
32 /* Used to indicate that the device has its own data path and configured */
33 #define VIRTIO_DEV_VDPA_CONFIGURED 8
35 /* Backend value set by guest. */
36 #define VIRTIO_DEV_STOPPED -1
38 #define BUF_VECTOR_MAX 256
40 #define VHOST_LOG_CACHE_NR 32
43 * Structure contains buffer address, length and descriptor index
44 * from vring to do scatter RX.
54 * A structure to hold some fields needed in zero copy code path,
55 * mainly for associating an mbuf with the right desc_idx.
58 struct rte_mbuf *mbuf;
63 TAILQ_ENTRY(zcopy_mbuf) next;
65 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
68 * Structure contains the info for each batched memory copy.
70 struct batch_copy_elem {
78 * Structure that contains the info for batched dirty logging.
80 struct log_cache_entry {
85 struct vring_used_elem_packed {
92 * Structure contains variables relevant to RX/TX virtqueues.
94 struct vhost_virtqueue {
96 struct vring_desc *desc;
97 struct vring_packed_desc *desc_packed;
100 struct vring_avail *avail;
101 struct vring_packed_desc_event *driver_event;
104 struct vring_used *used;
105 struct vring_packed_desc_event *device_event;
109 uint16_t last_avail_idx;
110 uint16_t last_used_idx;
111 /* Last used index we notify to front end. */
112 uint16_t signalled_used;
113 bool signalled_used_valid;
114 #define VIRTIO_INVALID_EVENTFD (-1)
115 #define VIRTIO_UNINITIALIZED_EVENTFD (-2)
117 /* Backend value to determine if device should started/stopped */
121 rte_spinlock_t access_lock;
123 /* Used to notify the guest (trigger interrupt) */
125 /* Currently unused as polling mode is enabled */
128 /* Physical address of used ring, for logging */
129 uint64_t log_guest_addr;
133 uint16_t last_zmbuf_idx;
134 struct zcopy_mbuf *zmbufs;
135 struct zcopy_mbuf_list zmbuf_list;
138 struct vring_used_elem *shadow_used_split;
139 struct vring_used_elem_packed *shadow_used_packed;
141 uint16_t shadow_used_idx;
142 struct vhost_vring_addr ring_addrs;
144 struct batch_copy_elem *batch_copy_elems;
145 uint16_t batch_copy_nb_elems;
146 bool used_wrap_counter;
147 bool avail_wrap_counter;
149 struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
150 uint16_t log_cache_nb_elem;
152 rte_rwlock_t iotlb_lock;
153 rte_rwlock_t iotlb_pending_lock;
154 struct rte_mempool *iotlb_pool;
155 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
157 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
158 } __rte_cache_aligned;
160 /* Old kernels have no such macros defined */
161 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
162 #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
165 #ifndef VIRTIO_NET_F_MQ
166 #define VIRTIO_NET_F_MQ 22
169 #define VHOST_MAX_VRING 0x100
170 #define VHOST_MAX_QUEUE_PAIRS 0x80
172 #ifndef VIRTIO_NET_F_MTU
173 #define VIRTIO_NET_F_MTU 3
176 #ifndef VIRTIO_F_ANY_LAYOUT
177 #define VIRTIO_F_ANY_LAYOUT 27
180 /* Declare IOMMU related bits for older kernels */
181 #ifndef VIRTIO_F_IOMMU_PLATFORM
183 #define VIRTIO_F_IOMMU_PLATFORM 33
185 struct vhost_iotlb_msg {
189 #define VHOST_ACCESS_RO 0x1
190 #define VHOST_ACCESS_WO 0x2
191 #define VHOST_ACCESS_RW 0x3
193 #define VHOST_IOTLB_MISS 1
194 #define VHOST_IOTLB_UPDATE 2
195 #define VHOST_IOTLB_INVALIDATE 3
196 #define VHOST_IOTLB_ACCESS_FAIL 4
200 #define VHOST_IOTLB_MSG 0x1
205 struct vhost_iotlb_msg iotlb;
212 * Define virtio 1.0 for older kernels
214 #ifndef VIRTIO_F_VERSION_1
215 #define VIRTIO_F_VERSION_1 32
218 /* Declare packed ring related bits for older kernels */
219 #ifndef VIRTIO_F_RING_PACKED
221 #define VIRTIO_F_RING_PACKED 34
223 struct vring_packed_desc {
230 struct vring_packed_desc_event {
237 * Declare below packed ring defines unconditionally
238 * as Kernel header might use different names.
240 #define VRING_DESC_F_AVAIL (1ULL << 7)
241 #define VRING_DESC_F_USED (1ULL << 15)
243 #define VRING_EVENT_F_ENABLE 0x0
244 #define VRING_EVENT_F_DISABLE 0x1
245 #define VRING_EVENT_F_DESC 0x2
248 * Available and used descs are in same order
250 #ifndef VIRTIO_F_IN_ORDER
251 #define VIRTIO_F_IN_ORDER 35
254 /* Features supported by this builtin vhost-user net driver. */
255 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
256 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
257 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
258 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
259 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
260 (1ULL << VIRTIO_NET_F_MQ) | \
261 (1ULL << VIRTIO_F_VERSION_1) | \
262 (1ULL << VHOST_F_LOG_ALL) | \
263 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
264 (1ULL << VIRTIO_NET_F_GSO) | \
265 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
266 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
267 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
268 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
269 (1ULL << VIRTIO_NET_F_CSUM) | \
270 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
271 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
272 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
273 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
274 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
275 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
276 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
277 (1ULL << VIRTIO_NET_F_MTU) | \
278 (1ULL << VIRTIO_F_IN_ORDER) | \
279 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
280 (1ULL << VIRTIO_F_RING_PACKED))
284 uint64_t guest_phys_addr;
285 uint64_t host_phys_addr;
290 * Device structure contains all configuration information relating
294 /* Frontend (QEMU) memory and memory region information */
295 struct rte_vhost_memory *mem;
297 uint64_t protocol_features;
301 /* to tell if we need broadcast rarp packet */
302 rte_atomic16_t broadcast_rarp;
304 int dequeue_zero_copy;
305 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
306 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
307 char ifname[IF_NAME_SZ];
311 struct rte_ether_addr mac;
314 struct vhost_device_ops const *notify_ops;
316 uint32_t nr_guest_pages;
317 uint32_t max_guest_pages;
318 struct guest_page *guest_pages;
321 rte_spinlock_t slave_req_lock;
324 int postcopy_listening;
327 * Device id to identify a specific backend device.
328 * It's set to -1 for the default software implementation.
332 /* context data for the external message handlers */
334 /* pre and post vhost user message handlers for the device */
335 struct rte_vhost_user_extern_ops extern_ops;
336 } __rte_cache_aligned;
338 static __rte_always_inline bool
339 vq_is_packed(struct virtio_net *dev)
341 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
345 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
347 uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
349 return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
350 wrap_counter != !!(flags & VRING_DESC_F_USED);
353 void __vhost_log_cache_write(struct virtio_net *dev,
354 struct vhost_virtqueue *vq,
355 uint64_t addr, uint64_t len);
356 void __vhost_log_cache_write_iova(struct virtio_net *dev,
357 struct vhost_virtqueue *vq,
358 uint64_t iova, uint64_t len);
359 void __vhost_log_cache_sync(struct virtio_net *dev,
360 struct vhost_virtqueue *vq);
361 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
362 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
363 uint64_t iova, uint64_t len);
365 static __rte_always_inline void
366 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
368 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
369 __vhost_log_write(dev, addr, len);
372 static __rte_always_inline void
373 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
375 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
376 __vhost_log_cache_sync(dev, vq);
379 static __rte_always_inline void
380 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
381 uint64_t addr, uint64_t len)
383 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
384 __vhost_log_cache_write(dev, vq, addr, len);
387 static __rte_always_inline void
388 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
389 uint64_t offset, uint64_t len)
391 vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
394 static __rte_always_inline void
395 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
396 uint64_t offset, uint64_t len)
398 vhost_log_write(dev, vq->log_guest_addr + offset, len);
401 static __rte_always_inline void
402 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
403 uint64_t iova, uint64_t len)
405 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
408 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
409 __vhost_log_cache_write_iova(dev, vq, iova, len);
411 __vhost_log_cache_write(dev, vq, iova, len);
414 static __rte_always_inline void
415 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
416 uint64_t iova, uint64_t len)
418 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
421 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
422 __vhost_log_write_iova(dev, vq, iova, len);
424 __vhost_log_write(dev, iova, len);
427 /* Macros for printing using RTE_LOG */
428 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
429 #define RTE_LOGTYPE_VHOST_DATA RTE_LOGTYPE_USER1
431 #ifdef RTE_LIBRTE_VHOST_DEBUG
432 #define VHOST_MAX_PRINT_BUFF 6072
433 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
434 RTE_LOG(DEBUG, log_type, fmt, ##args)
435 #define PRINT_PACKET(device, addr, size, header) do { \
436 char *pkt_addr = (char *)(addr); \
437 unsigned int index; \
438 char packet[VHOST_MAX_PRINT_BUFF]; \
441 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
443 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
444 for (index = 0; index < (size); index++) { \
445 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
446 "%02hhx ", pkt_addr[index]); \
448 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
450 VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
453 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
454 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
457 extern uint64_t VHOST_FEATURES;
458 #define MAX_VHOST_DEVICE 1024
459 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
461 /* Convert guest physical address to host physical address */
462 static __rte_always_inline rte_iova_t
463 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
466 struct guest_page *page;
468 for (i = 0; i < dev->nr_guest_pages; i++) {
469 page = &dev->guest_pages[i];
471 if (gpa >= page->guest_phys_addr &&
472 gpa + size < page->guest_phys_addr + page->size) {
473 return gpa - page->guest_phys_addr +
474 page->host_phys_addr;
481 static __rte_always_inline uint64_t
482 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
484 struct rte_vhost_mem_region *r;
487 if (unlikely(!dev || !dev->mem))
490 for (i = 0; i < dev->mem->nregions; i++) {
491 r = &dev->mem->regions[i];
493 if (vva >= r->host_user_addr &&
494 vva + len < r->host_user_addr + r->size) {
495 return r->guest_phys_addr + vva - r->host_user_addr;
501 static __rte_always_inline struct virtio_net *
504 struct virtio_net *dev = vhost_devices[vid];
506 if (unlikely(!dev)) {
507 RTE_LOG(ERR, VHOST_CONFIG,
508 "(%d) device not found.\n", vid);
514 int vhost_new_device(void);
515 void cleanup_device(struct virtio_net *dev, int destroy);
516 void reset_device(struct virtio_net *dev);
517 void vhost_destroy_device(int);
518 void vhost_destroy_device_notify(struct virtio_net *dev);
520 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
521 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
523 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
525 void vhost_attach_vdpa_device(int vid, int did);
527 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
528 void vhost_enable_dequeue_zero_copy(int vid);
529 void vhost_set_builtin_virtio_net(int vid, bool enable);
531 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
534 * Backend-specific cleanup.
536 * TODO: fix it; we have one backend now
538 void vhost_backend_cleanup(struct virtio_net *dev);
540 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
541 uint64_t iova, uint64_t *len, uint8_t perm);
542 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
543 struct vhost_virtqueue *vq,
544 uint64_t desc_addr, uint64_t desc_len);
545 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
546 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
548 static __rte_always_inline uint64_t
549 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
550 uint64_t iova, uint64_t *len, uint8_t perm)
552 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
553 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
555 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
558 #define vhost_avail_event(vr) \
559 (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
560 #define vhost_used_event(vr) \
561 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
564 * The following is used with VIRTIO_RING_F_EVENT_IDX.
565 * Assuming a given event_idx value from the other size, if we have
566 * just incremented index from old to new_idx, should we trigger an
569 static __rte_always_inline int
570 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
572 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
575 static __rte_always_inline void
576 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
578 /* Flush used->idx update before we read avail->flags. */
581 /* Don't kick guest if we don't reach index specified by guest. */
582 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
583 uint16_t old = vq->signalled_used;
584 uint16_t new = vq->last_used_idx;
585 bool signalled_used_valid = vq->signalled_used_valid;
587 vq->signalled_used = new;
588 vq->signalled_used_valid = true;
590 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
592 vhost_used_event(vq),
595 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
596 (vq->callfd >= 0)) ||
597 unlikely(!signalled_used_valid)) {
598 eventfd_write(vq->callfd, (eventfd_t) 1);
599 if (dev->notify_ops->guest_notified)
600 dev->notify_ops->guest_notified(dev->vid);
603 /* Kick the guest if necessary. */
604 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
605 && (vq->callfd >= 0)) {
606 eventfd_write(vq->callfd, (eventfd_t)1);
607 if (dev->notify_ops->guest_notified)
608 dev->notify_ops->guest_notified(dev->vid);
613 static __rte_always_inline void
614 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
616 uint16_t old, new, off, off_wrap;
617 bool signalled_used_valid, kick = false;
619 /* Flush used desc update. */
622 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
623 if (vq->driver_event->flags !=
624 VRING_EVENT_F_DISABLE)
629 old = vq->signalled_used;
630 new = vq->last_used_idx;
631 vq->signalled_used = new;
632 signalled_used_valid = vq->signalled_used_valid;
633 vq->signalled_used_valid = true;
635 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
636 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
641 if (unlikely(!signalled_used_valid)) {
648 off_wrap = vq->driver_event->off_wrap;
649 off = off_wrap & ~(1 << 15);
654 if (vq->used_wrap_counter != off_wrap >> 15)
657 if (vhost_need_event(off, new, old))
661 eventfd_write(vq->callfd, (eventfd_t)1);
662 if (dev->notify_ops->guest_notified)
663 dev->notify_ops->guest_notified(dev->vid);
667 static __rte_always_inline void
668 free_ind_table(void *idesc)
673 static __rte_always_inline void
674 restore_mbuf(struct rte_mbuf *m)
676 uint32_t mbuf_size, priv_size;
679 priv_size = rte_pktmbuf_priv_size(m->pool);
680 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
681 /* start of buffer is after mbuf structure and priv data */
683 m->buf_addr = (char *)m + mbuf_size;
684 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
689 static __rte_always_inline bool
690 mbuf_is_consumed(struct rte_mbuf *m)
693 if (rte_mbuf_refcnt_read(m) > 1)
701 static __rte_always_inline void
702 put_zmbuf(struct zcopy_mbuf *zmbuf)
707 #endif /* _VHOST_NET_CDEV_H_ */