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;
289 struct inflight_mem_info {
296 * Device structure contains all configuration information relating
300 /* Frontend (QEMU) memory and memory region information */
301 struct rte_vhost_memory *mem;
303 uint64_t protocol_features;
307 /* to tell if we need broadcast rarp packet */
308 rte_atomic16_t broadcast_rarp;
310 int dequeue_zero_copy;
311 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
312 struct inflight_mem_info *inflight_info;
313 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
314 char ifname[IF_NAME_SZ];
318 struct rte_ether_addr mac;
321 struct vhost_device_ops const *notify_ops;
323 uint32_t nr_guest_pages;
324 uint32_t max_guest_pages;
325 struct guest_page *guest_pages;
328 rte_spinlock_t slave_req_lock;
331 int postcopy_listening;
334 * Device id to identify a specific backend device.
335 * It's set to -1 for the default software implementation.
339 /* context data for the external message handlers */
341 /* pre and post vhost user message handlers for the device */
342 struct rte_vhost_user_extern_ops extern_ops;
343 } __rte_cache_aligned;
345 static __rte_always_inline bool
346 vq_is_packed(struct virtio_net *dev)
348 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
352 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
354 uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
356 return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
357 wrap_counter != !!(flags & VRING_DESC_F_USED);
360 void __vhost_log_cache_write(struct virtio_net *dev,
361 struct vhost_virtqueue *vq,
362 uint64_t addr, uint64_t len);
363 void __vhost_log_cache_write_iova(struct virtio_net *dev,
364 struct vhost_virtqueue *vq,
365 uint64_t iova, uint64_t len);
366 void __vhost_log_cache_sync(struct virtio_net *dev,
367 struct vhost_virtqueue *vq);
368 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
369 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
370 uint64_t iova, uint64_t len);
372 static __rte_always_inline void
373 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
375 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
376 __vhost_log_write(dev, addr, len);
379 static __rte_always_inline void
380 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
382 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
383 __vhost_log_cache_sync(dev, vq);
386 static __rte_always_inline void
387 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
388 uint64_t addr, uint64_t len)
390 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
391 __vhost_log_cache_write(dev, vq, addr, len);
394 static __rte_always_inline void
395 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
396 uint64_t offset, uint64_t len)
398 vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
401 static __rte_always_inline void
402 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
403 uint64_t offset, uint64_t len)
405 vhost_log_write(dev, vq->log_guest_addr + offset, len);
408 static __rte_always_inline void
409 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
410 uint64_t iova, uint64_t len)
412 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
415 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
416 __vhost_log_cache_write_iova(dev, vq, iova, len);
418 __vhost_log_cache_write(dev, vq, iova, len);
421 static __rte_always_inline void
422 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
423 uint64_t iova, uint64_t len)
425 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
428 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
429 __vhost_log_write_iova(dev, vq, iova, len);
431 __vhost_log_write(dev, iova, len);
434 /* Macros for printing using RTE_LOG */
435 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
436 #define RTE_LOGTYPE_VHOST_DATA RTE_LOGTYPE_USER1
438 #ifdef RTE_LIBRTE_VHOST_DEBUG
439 #define VHOST_MAX_PRINT_BUFF 6072
440 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
441 RTE_LOG(DEBUG, log_type, fmt, ##args)
442 #define PRINT_PACKET(device, addr, size, header) do { \
443 char *pkt_addr = (char *)(addr); \
444 unsigned int index; \
445 char packet[VHOST_MAX_PRINT_BUFF]; \
448 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
450 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
451 for (index = 0; index < (size); index++) { \
452 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
453 "%02hhx ", pkt_addr[index]); \
455 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
457 VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
460 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
461 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
464 extern uint64_t VHOST_FEATURES;
465 #define MAX_VHOST_DEVICE 1024
466 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
468 /* Convert guest physical address to host physical address */
469 static __rte_always_inline rte_iova_t
470 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
473 struct guest_page *page;
475 for (i = 0; i < dev->nr_guest_pages; i++) {
476 page = &dev->guest_pages[i];
478 if (gpa >= page->guest_phys_addr &&
479 gpa + size < page->guest_phys_addr + page->size) {
480 return gpa - page->guest_phys_addr +
481 page->host_phys_addr;
488 static __rte_always_inline uint64_t
489 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
491 struct rte_vhost_mem_region *r;
494 if (unlikely(!dev || !dev->mem))
497 for (i = 0; i < dev->mem->nregions; i++) {
498 r = &dev->mem->regions[i];
500 if (vva >= r->host_user_addr &&
501 vva + len < r->host_user_addr + r->size) {
502 return r->guest_phys_addr + vva - r->host_user_addr;
508 static __rte_always_inline struct virtio_net *
511 struct virtio_net *dev = vhost_devices[vid];
513 if (unlikely(!dev)) {
514 RTE_LOG(ERR, VHOST_CONFIG,
515 "(%d) device not found.\n", vid);
521 int vhost_new_device(void);
522 void cleanup_device(struct virtio_net *dev, int destroy);
523 void reset_device(struct virtio_net *dev);
524 void vhost_destroy_device(int);
525 void vhost_destroy_device_notify(struct virtio_net *dev);
527 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
528 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
530 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
532 void vhost_attach_vdpa_device(int vid, int did);
534 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
535 void vhost_enable_dequeue_zero_copy(int vid);
536 void vhost_set_builtin_virtio_net(int vid, bool enable);
538 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
541 * Backend-specific cleanup.
543 * TODO: fix it; we have one backend now
545 void vhost_backend_cleanup(struct virtio_net *dev);
547 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
548 uint64_t iova, uint64_t *len, uint8_t perm);
549 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
550 struct vhost_virtqueue *vq,
551 uint64_t desc_addr, uint64_t desc_len);
552 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
553 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
555 static __rte_always_inline uint64_t
556 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
557 uint64_t iova, uint64_t *len, uint8_t perm)
559 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
560 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
562 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
565 #define vhost_avail_event(vr) \
566 (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
567 #define vhost_used_event(vr) \
568 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
571 * The following is used with VIRTIO_RING_F_EVENT_IDX.
572 * Assuming a given event_idx value from the other size, if we have
573 * just incremented index from old to new_idx, should we trigger an
576 static __rte_always_inline int
577 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
579 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
582 static __rte_always_inline void
583 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
585 /* Flush used->idx update before we read avail->flags. */
588 /* Don't kick guest if we don't reach index specified by guest. */
589 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
590 uint16_t old = vq->signalled_used;
591 uint16_t new = vq->last_used_idx;
592 bool signalled_used_valid = vq->signalled_used_valid;
594 vq->signalled_used = new;
595 vq->signalled_used_valid = true;
597 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
599 vhost_used_event(vq),
602 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
603 (vq->callfd >= 0)) ||
604 unlikely(!signalled_used_valid)) {
605 eventfd_write(vq->callfd, (eventfd_t) 1);
606 if (dev->notify_ops->guest_notified)
607 dev->notify_ops->guest_notified(dev->vid);
610 /* Kick the guest if necessary. */
611 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
612 && (vq->callfd >= 0)) {
613 eventfd_write(vq->callfd, (eventfd_t)1);
614 if (dev->notify_ops->guest_notified)
615 dev->notify_ops->guest_notified(dev->vid);
620 static __rte_always_inline void
621 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
623 uint16_t old, new, off, off_wrap;
624 bool signalled_used_valid, kick = false;
626 /* Flush used desc update. */
629 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
630 if (vq->driver_event->flags !=
631 VRING_EVENT_F_DISABLE)
636 old = vq->signalled_used;
637 new = vq->last_used_idx;
638 vq->signalled_used = new;
639 signalled_used_valid = vq->signalled_used_valid;
640 vq->signalled_used_valid = true;
642 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
643 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
648 if (unlikely(!signalled_used_valid)) {
655 off_wrap = vq->driver_event->off_wrap;
656 off = off_wrap & ~(1 << 15);
661 if (vq->used_wrap_counter != off_wrap >> 15)
664 if (vhost_need_event(off, new, old))
668 eventfd_write(vq->callfd, (eventfd_t)1);
669 if (dev->notify_ops->guest_notified)
670 dev->notify_ops->guest_notified(dev->vid);
674 static __rte_always_inline void
675 free_ind_table(void *idesc)
680 static __rte_always_inline void
681 restore_mbuf(struct rte_mbuf *m)
683 uint32_t mbuf_size, priv_size;
686 priv_size = rte_pktmbuf_priv_size(m->pool);
687 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
688 /* start of buffer is after mbuf structure and priv data */
690 m->buf_addr = (char *)m + mbuf_size;
691 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
696 static __rte_always_inline bool
697 mbuf_is_consumed(struct rte_mbuf *m)
700 if (rte_mbuf_refcnt_read(m) > 1)
708 static __rte_always_inline void
709 put_zmbuf(struct zcopy_mbuf *zmbuf)
714 #endif /* _VHOST_NET_CDEV_H_ */