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>
22 #include "rte_vhost.h"
25 /* Used to indicate that the device is running on a data core */
26 #define VIRTIO_DEV_RUNNING 1
27 /* Used to indicate that the device is ready to operate */
28 #define VIRTIO_DEV_READY 2
29 /* Used to indicate that the built-in vhost net device backend is enabled */
30 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET 4
31 /* Used to indicate that the device has its own data path and configured */
32 #define VIRTIO_DEV_VDPA_CONFIGURED 8
34 /* Backend value set by guest. */
35 #define VIRTIO_DEV_STOPPED -1
37 #define BUF_VECTOR_MAX 256
39 #define VHOST_LOG_CACHE_NR 32
42 * Structure contains buffer address, length and descriptor index
43 * from vring to do scatter RX.
53 * A structure to hold some fields needed in zero copy code path,
54 * mainly for associating an mbuf with the right desc_idx.
57 struct rte_mbuf *mbuf;
62 TAILQ_ENTRY(zcopy_mbuf) next;
64 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
67 * Structure contains the info for each batched memory copy.
69 struct batch_copy_elem {
77 * Structure that contains the info for batched dirty logging.
79 struct log_cache_entry {
84 struct vring_used_elem_packed {
91 * Structure contains variables relevant to RX/TX virtqueues.
93 struct vhost_virtqueue {
95 struct vring_desc *desc;
96 struct vring_packed_desc *desc_packed;
99 struct vring_avail *avail;
100 struct vring_packed_desc_event *driver_event;
103 struct vring_used *used;
104 struct vring_packed_desc_event *device_event;
108 uint16_t last_avail_idx;
109 uint16_t last_used_idx;
110 /* Last used index we notify to front end. */
111 uint16_t signalled_used;
112 bool signalled_used_valid;
113 #define VIRTIO_INVALID_EVENTFD (-1)
114 #define VIRTIO_UNINITIALIZED_EVENTFD (-2)
116 /* Backend value to determine if device should started/stopped */
120 rte_spinlock_t access_lock;
122 /* Used to notify the guest (trigger interrupt) */
124 /* Currently unused as polling mode is enabled */
127 /* Physical address of used ring, for logging */
128 uint64_t log_guest_addr;
132 uint16_t last_zmbuf_idx;
133 struct zcopy_mbuf *zmbufs;
134 struct zcopy_mbuf_list zmbuf_list;
137 struct vring_used_elem *shadow_used_split;
138 struct vring_used_elem_packed *shadow_used_packed;
140 uint16_t shadow_used_idx;
141 struct vhost_vring_addr ring_addrs;
143 struct batch_copy_elem *batch_copy_elems;
144 uint16_t batch_copy_nb_elems;
145 bool used_wrap_counter;
146 bool avail_wrap_counter;
148 struct log_cache_entry log_cache[VHOST_LOG_CACHE_NR];
149 uint16_t log_cache_nb_elem;
151 rte_rwlock_t iotlb_lock;
152 rte_rwlock_t iotlb_pending_lock;
153 struct rte_mempool *iotlb_pool;
154 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
156 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
157 } __rte_cache_aligned;
159 /* Old kernels have no such macros defined */
160 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
161 #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
164 #ifndef VIRTIO_NET_F_MQ
165 #define VIRTIO_NET_F_MQ 22
168 #define VHOST_MAX_VRING 0x100
169 #define VHOST_MAX_QUEUE_PAIRS 0x80
171 #ifndef VIRTIO_NET_F_MTU
172 #define VIRTIO_NET_F_MTU 3
175 #ifndef VIRTIO_F_ANY_LAYOUT
176 #define VIRTIO_F_ANY_LAYOUT 27
179 /* Declare IOMMU related bits for older kernels */
180 #ifndef VIRTIO_F_IOMMU_PLATFORM
182 #define VIRTIO_F_IOMMU_PLATFORM 33
184 struct vhost_iotlb_msg {
188 #define VHOST_ACCESS_RO 0x1
189 #define VHOST_ACCESS_WO 0x2
190 #define VHOST_ACCESS_RW 0x3
192 #define VHOST_IOTLB_MISS 1
193 #define VHOST_IOTLB_UPDATE 2
194 #define VHOST_IOTLB_INVALIDATE 3
195 #define VHOST_IOTLB_ACCESS_FAIL 4
199 #define VHOST_IOTLB_MSG 0x1
204 struct vhost_iotlb_msg iotlb;
211 * Define virtio 1.0 for older kernels
213 #ifndef VIRTIO_F_VERSION_1
214 #define VIRTIO_F_VERSION_1 32
217 /* Declare packed ring related bits for older kernels */
218 #ifndef VIRTIO_F_RING_PACKED
220 #define VIRTIO_F_RING_PACKED 34
222 #define VRING_DESC_F_NEXT 1
223 #define VRING_DESC_F_WRITE 2
224 #define VRING_DESC_F_INDIRECT 4
226 #define VRING_DESC_F_AVAIL (1ULL << 7)
227 #define VRING_DESC_F_USED (1ULL << 15)
229 struct vring_packed_desc {
236 #define VRING_EVENT_F_ENABLE 0x0
237 #define VRING_EVENT_F_DISABLE 0x1
238 #define VRING_EVENT_F_DESC 0x2
240 struct vring_packed_desc_event {
247 * Available and used descs are in same order
249 #ifndef VIRTIO_F_IN_ORDER
250 #define VIRTIO_F_IN_ORDER 35
253 /* Features supported by this builtin vhost-user net driver. */
254 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
255 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
256 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
257 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
258 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
259 (1ULL << VIRTIO_NET_F_MQ) | \
260 (1ULL << VIRTIO_F_VERSION_1) | \
261 (1ULL << VHOST_F_LOG_ALL) | \
262 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
263 (1ULL << VIRTIO_NET_F_GSO) | \
264 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
265 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
266 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
267 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
268 (1ULL << VIRTIO_NET_F_CSUM) | \
269 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
270 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
271 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
272 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
273 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
274 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
275 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
276 (1ULL << VIRTIO_NET_F_MTU) | \
277 (1ULL << VIRTIO_F_IN_ORDER) | \
278 (1ULL << VIRTIO_F_IOMMU_PLATFORM))
282 uint64_t guest_phys_addr;
283 uint64_t host_phys_addr;
287 /* The possible results of a message handling function */
289 /* Message handling failed */
291 /* Message handling successful */
293 /* Message handling successful and reply prepared */
298 * function prototype for the vhost backend to handler specific vhost user
299 * messages prior to the master message handling
306 * If the handler requires skipping the master message handling, this variable
307 * shall be written 1, otherwise 0.
309 * VH_RESULT_OK on success, VH_RESULT_REPLY on success with reply,
310 * VH_RESULT_ERR on failure
312 typedef enum vh_result (*vhost_msg_pre_handle)(int vid, void *msg,
313 uint32_t *skip_master);
316 * function prototype for the vhost backend to handler specific vhost user
317 * messages after the master message handling is done
324 * VH_RESULT_OK on success, VH_RESULT_REPLY on success with reply,
325 * VH_RESULT_ERR on failure
327 typedef enum vh_result (*vhost_msg_post_handle)(int vid, void *msg);
330 * pre and post vhost user message handlers
332 struct vhost_user_extern_ops {
333 vhost_msg_pre_handle pre_msg_handle;
334 vhost_msg_post_handle post_msg_handle;
338 * Device structure contains all configuration information relating
342 /* Frontend (QEMU) memory and memory region information */
343 struct rte_vhost_memory *mem;
345 uint64_t protocol_features;
349 /* to tell if we need broadcast rarp packet */
350 rte_atomic16_t broadcast_rarp;
352 int dequeue_zero_copy;
353 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
354 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
355 char ifname[IF_NAME_SZ];
359 struct ether_addr mac;
362 struct vhost_device_ops const *notify_ops;
364 uint32_t nr_guest_pages;
365 uint32_t max_guest_pages;
366 struct guest_page *guest_pages;
369 rte_spinlock_t slave_req_lock;
372 * Device id to identify a specific backend device.
373 * It's set to -1 for the default software implementation.
377 /* private data for virtio device */
379 /* pre and post vhost user message handlers for the device */
380 struct vhost_user_extern_ops extern_ops;
381 } __rte_cache_aligned;
383 static __rte_always_inline bool
384 vq_is_packed(struct virtio_net *dev)
386 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
390 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
392 return wrap_counter == !!(desc->flags & VRING_DESC_F_AVAIL) &&
393 wrap_counter != !!(desc->flags & VRING_DESC_F_USED);
396 #define VHOST_LOG_PAGE 4096
399 * Atomically set a bit in memory.
401 static __rte_always_inline void
402 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
404 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
406 * __sync_ built-ins are deprecated, but __atomic_ ones
407 * are sub-optimized in older GCC versions.
409 __sync_fetch_and_or_1(addr, (1U << nr));
411 __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
415 static __rte_always_inline void
416 vhost_log_page(uint8_t *log_base, uint64_t page)
418 vhost_set_bit(page % 8, &log_base[page / 8]);
421 static __rte_always_inline void
422 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
426 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
427 !dev->log_base || !len))
430 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
433 /* To make sure guest memory updates are committed before logging */
436 page = addr / VHOST_LOG_PAGE;
437 while (page * VHOST_LOG_PAGE < addr + len) {
438 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
443 static __rte_always_inline void
444 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
446 unsigned long *log_base;
449 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
453 log_base = (unsigned long *)(uintptr_t)dev->log_base;
456 * It is expected a write memory barrier has been issued
457 * before this function is called.
460 for (i = 0; i < vq->log_cache_nb_elem; i++) {
461 struct log_cache_entry *elem = vq->log_cache + i;
463 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
465 * '__sync' builtins are deprecated, but '__atomic' ones
466 * are sub-optimized in older GCC versions.
468 __sync_fetch_and_or(log_base + elem->offset, elem->val);
470 __atomic_fetch_or(log_base + elem->offset, elem->val,
477 vq->log_cache_nb_elem = 0;
480 static __rte_always_inline void
481 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
484 uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
485 uint32_t offset = page / (sizeof(unsigned long) << 3);
488 for (i = 0; i < vq->log_cache_nb_elem; i++) {
489 struct log_cache_entry *elem = vq->log_cache + i;
491 if (elem->offset == offset) {
492 elem->val |= (1UL << bit_nr);
497 if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
499 * No more room for a new log cache entry,
500 * so write the dirty log map directly.
503 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
508 vq->log_cache[i].offset = offset;
509 vq->log_cache[i].val = (1UL << bit_nr);
510 vq->log_cache_nb_elem++;
513 static __rte_always_inline void
514 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
515 uint64_t addr, uint64_t len)
519 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
520 !dev->log_base || !len))
523 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
526 page = addr / VHOST_LOG_PAGE;
527 while (page * VHOST_LOG_PAGE < addr + len) {
528 vhost_log_cache_page(dev, vq, page);
533 static __rte_always_inline void
534 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
535 uint64_t offset, uint64_t len)
537 vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
540 static __rte_always_inline void
541 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
542 uint64_t offset, uint64_t len)
544 vhost_log_write(dev, vq->log_guest_addr + offset, len);
547 /* Macros for printing using RTE_LOG */
548 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
549 #define RTE_LOGTYPE_VHOST_DATA RTE_LOGTYPE_USER1
551 #ifdef RTE_LIBRTE_VHOST_DEBUG
552 #define VHOST_MAX_PRINT_BUFF 6072
553 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
554 RTE_LOG(DEBUG, log_type, fmt, ##args)
555 #define PRINT_PACKET(device, addr, size, header) do { \
556 char *pkt_addr = (char *)(addr); \
557 unsigned int index; \
558 char packet[VHOST_MAX_PRINT_BUFF]; \
561 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
563 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
564 for (index = 0; index < (size); index++) { \
565 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
566 "%02hhx ", pkt_addr[index]); \
568 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
570 VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
573 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
574 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
577 extern uint64_t VHOST_FEATURES;
578 #define MAX_VHOST_DEVICE 1024
579 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
581 /* Convert guest physical address to host physical address */
582 static __rte_always_inline rte_iova_t
583 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
586 struct guest_page *page;
588 for (i = 0; i < dev->nr_guest_pages; i++) {
589 page = &dev->guest_pages[i];
591 if (gpa >= page->guest_phys_addr &&
592 gpa + size < page->guest_phys_addr + page->size) {
593 return gpa - page->guest_phys_addr +
594 page->host_phys_addr;
601 static __rte_always_inline struct virtio_net *
604 struct virtio_net *dev = vhost_devices[vid];
606 if (unlikely(!dev)) {
607 RTE_LOG(ERR, VHOST_CONFIG,
608 "(%d) device not found.\n", vid);
614 int vhost_new_device(void);
615 void cleanup_device(struct virtio_net *dev, int destroy);
616 void reset_device(struct virtio_net *dev);
617 void vhost_destroy_device(int);
618 void vhost_destroy_device_notify(struct virtio_net *dev);
620 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
621 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
623 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
625 void vhost_attach_vdpa_device(int vid, int did);
626 void vhost_detach_vdpa_device(int vid);
628 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
629 void vhost_enable_dequeue_zero_copy(int vid);
630 void vhost_set_builtin_virtio_net(int vid, bool enable);
632 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
635 * Backend-specific cleanup.
637 * TODO: fix it; we have one backend now
639 void vhost_backend_cleanup(struct virtio_net *dev);
641 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
642 uint64_t iova, uint64_t *len, uint8_t perm);
643 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
644 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
646 static __rte_always_inline uint64_t
647 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
648 uint64_t iova, uint64_t *len, uint8_t perm)
650 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
651 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
653 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
656 #define vhost_avail_event(vr) \
657 (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
658 #define vhost_used_event(vr) \
659 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
662 * The following is used with VIRTIO_RING_F_EVENT_IDX.
663 * Assuming a given event_idx value from the other size, if we have
664 * just incremented index from old to new_idx, should we trigger an
667 static __rte_always_inline int
668 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
670 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
673 static __rte_always_inline void
674 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
676 /* Flush used->idx update before we read avail->flags. */
679 /* Don't kick guest if we don't reach index specified by guest. */
680 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
681 uint16_t old = vq->signalled_used;
682 uint16_t new = vq->last_used_idx;
684 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
686 vhost_used_event(vq),
688 if (vhost_need_event(vhost_used_event(vq), new, old)
689 && (vq->callfd >= 0)) {
690 vq->signalled_used = vq->last_used_idx;
691 eventfd_write(vq->callfd, (eventfd_t) 1);
694 /* Kick the guest if necessary. */
695 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
696 && (vq->callfd >= 0))
697 eventfd_write(vq->callfd, (eventfd_t)1);
701 static __rte_always_inline void
702 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
704 uint16_t old, new, off, off_wrap;
705 bool signalled_used_valid, kick = false;
707 /* Flush used desc update. */
710 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
711 if (vq->driver_event->flags !=
712 VRING_EVENT_F_DISABLE)
717 old = vq->signalled_used;
718 new = vq->last_used_idx;
719 vq->signalled_used = new;
720 signalled_used_valid = vq->signalled_used_valid;
721 vq->signalled_used_valid = true;
723 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
724 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
729 if (unlikely(!signalled_used_valid)) {
736 off_wrap = vq->driver_event->off_wrap;
737 off = off_wrap & ~(1 << 15);
742 if (vq->used_wrap_counter != off_wrap >> 15)
745 if (vhost_need_event(off, new, old))
749 eventfd_write(vq->callfd, (eventfd_t)1);
752 #endif /* _VHOST_NET_CDEV_H_ */