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;
288 * function prototype for the vhost backend to handler specific vhost user
289 * messages prior to the master message handling
295 * @param require_reply
296 * If the handler requires sending a reply, this varaible shall be written 1,
299 * If the handler requires skipping the master message handling, this variable
300 * shall be written 1, otherwise 0.
302 * 0 on success, -1 on failure
304 typedef int (*vhost_msg_pre_handle)(int vid, void *msg,
305 uint32_t *require_reply, uint32_t *skip_master);
308 * function prototype for the vhost backend to handler specific vhost user
309 * messages after the master message handling is done
315 * @param require_reply
316 * If the handler requires sending a reply, this varaible shall be written 1,
319 * 0 on success, -1 on failure
321 typedef int (*vhost_msg_post_handle)(int vid, void *msg,
322 uint32_t *require_reply);
325 * pre and post vhost user message handlers
327 struct vhost_user_extern_ops {
328 vhost_msg_pre_handle pre_msg_handle;
329 vhost_msg_post_handle post_msg_handle;
333 * Device structure contains all configuration information relating
337 /* Frontend (QEMU) memory and memory region information */
338 struct rte_vhost_memory *mem;
340 uint64_t protocol_features;
344 /* to tell if we need broadcast rarp packet */
345 rte_atomic16_t broadcast_rarp;
347 int dequeue_zero_copy;
348 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
349 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
350 char ifname[IF_NAME_SZ];
354 struct ether_addr mac;
357 struct vhost_device_ops const *notify_ops;
359 uint32_t nr_guest_pages;
360 uint32_t max_guest_pages;
361 struct guest_page *guest_pages;
364 rte_spinlock_t slave_req_lock;
367 * Device id to identify a specific backend device.
368 * It's set to -1 for the default software implementation.
372 /* private data for virtio device */
374 /* pre and post vhost user message handlers for the device */
375 struct vhost_user_extern_ops extern_ops;
376 } __rte_cache_aligned;
378 static __rte_always_inline bool
379 vq_is_packed(struct virtio_net *dev)
381 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
385 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
387 return wrap_counter == !!(desc->flags & VRING_DESC_F_AVAIL) &&
388 wrap_counter != !!(desc->flags & VRING_DESC_F_USED);
391 #define VHOST_LOG_PAGE 4096
394 * Atomically set a bit in memory.
396 static __rte_always_inline void
397 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
399 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
401 * __sync_ built-ins are deprecated, but __atomic_ ones
402 * are sub-optimized in older GCC versions.
404 __sync_fetch_and_or_1(addr, (1U << nr));
406 __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
410 static __rte_always_inline void
411 vhost_log_page(uint8_t *log_base, uint64_t page)
413 vhost_set_bit(page % 8, &log_base[page / 8]);
416 static __rte_always_inline void
417 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
421 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
422 !dev->log_base || !len))
425 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
428 /* To make sure guest memory updates are committed before logging */
431 page = addr / VHOST_LOG_PAGE;
432 while (page * VHOST_LOG_PAGE < addr + len) {
433 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
438 static __rte_always_inline void
439 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
441 unsigned long *log_base;
444 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
448 log_base = (unsigned long *)(uintptr_t)dev->log_base;
451 * It is expected a write memory barrier has been issued
452 * before this function is called.
455 for (i = 0; i < vq->log_cache_nb_elem; i++) {
456 struct log_cache_entry *elem = vq->log_cache + i;
458 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
460 * '__sync' builtins are deprecated, but '__atomic' ones
461 * are sub-optimized in older GCC versions.
463 __sync_fetch_and_or(log_base + elem->offset, elem->val);
465 __atomic_fetch_or(log_base + elem->offset, elem->val,
472 vq->log_cache_nb_elem = 0;
475 static __rte_always_inline void
476 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
479 uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
480 uint32_t offset = page / (sizeof(unsigned long) << 3);
483 for (i = 0; i < vq->log_cache_nb_elem; i++) {
484 struct log_cache_entry *elem = vq->log_cache + i;
486 if (elem->offset == offset) {
487 elem->val |= (1UL << bit_nr);
492 if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
494 * No more room for a new log cache entry,
495 * so write the dirty log map directly.
498 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
503 vq->log_cache[i].offset = offset;
504 vq->log_cache[i].val = (1UL << bit_nr);
505 vq->log_cache_nb_elem++;
508 static __rte_always_inline void
509 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
510 uint64_t addr, uint64_t len)
514 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
515 !dev->log_base || !len))
518 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
521 page = addr / VHOST_LOG_PAGE;
522 while (page * VHOST_LOG_PAGE < addr + len) {
523 vhost_log_cache_page(dev, vq, page);
528 static __rte_always_inline void
529 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
530 uint64_t offset, uint64_t len)
532 vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset, len);
535 static __rte_always_inline void
536 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
537 uint64_t offset, uint64_t len)
539 vhost_log_write(dev, vq->log_guest_addr + offset, len);
542 /* Macros for printing using RTE_LOG */
543 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
544 #define RTE_LOGTYPE_VHOST_DATA RTE_LOGTYPE_USER1
546 #ifdef RTE_LIBRTE_VHOST_DEBUG
547 #define VHOST_MAX_PRINT_BUFF 6072
548 #define VHOST_LOG_DEBUG(log_type, fmt, args...) \
549 RTE_LOG(DEBUG, log_type, fmt, ##args)
550 #define PRINT_PACKET(device, addr, size, header) do { \
551 char *pkt_addr = (char *)(addr); \
552 unsigned int index; \
553 char packet[VHOST_MAX_PRINT_BUFF]; \
556 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
558 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
559 for (index = 0; index < (size); index++) { \
560 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
561 "%02hhx ", pkt_addr[index]); \
563 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
565 VHOST_LOG_DEBUG(VHOST_DATA, "%s", packet); \
568 #define VHOST_LOG_DEBUG(log_type, fmt, args...) do {} while (0)
569 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
572 extern uint64_t VHOST_FEATURES;
573 #define MAX_VHOST_DEVICE 1024
574 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
576 /* Convert guest physical address to host physical address */
577 static __rte_always_inline rte_iova_t
578 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
581 struct guest_page *page;
583 for (i = 0; i < dev->nr_guest_pages; i++) {
584 page = &dev->guest_pages[i];
586 if (gpa >= page->guest_phys_addr &&
587 gpa + size < page->guest_phys_addr + page->size) {
588 return gpa - page->guest_phys_addr +
589 page->host_phys_addr;
596 static __rte_always_inline struct virtio_net *
599 struct virtio_net *dev = vhost_devices[vid];
601 if (unlikely(!dev)) {
602 RTE_LOG(ERR, VHOST_CONFIG,
603 "(%d) device not found.\n", vid);
609 int vhost_new_device(void);
610 void cleanup_device(struct virtio_net *dev, int destroy);
611 void reset_device(struct virtio_net *dev);
612 void vhost_destroy_device(int);
613 void vhost_destroy_device_notify(struct virtio_net *dev);
615 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
616 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
618 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
620 void vhost_attach_vdpa_device(int vid, int did);
621 void vhost_detach_vdpa_device(int vid);
623 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
624 void vhost_enable_dequeue_zero_copy(int vid);
625 void vhost_set_builtin_virtio_net(int vid, bool enable);
627 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
630 * Backend-specific cleanup.
632 * TODO: fix it; we have one backend now
634 void vhost_backend_cleanup(struct virtio_net *dev);
636 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
637 uint64_t iova, uint64_t *len, uint8_t perm);
638 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
639 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
641 static __rte_always_inline uint64_t
642 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
643 uint64_t iova, uint64_t *len, uint8_t perm)
645 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
646 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
648 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
651 #define vhost_used_event(vr) \
652 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
655 * The following is used with VIRTIO_RING_F_EVENT_IDX.
656 * Assuming a given event_idx value from the other size, if we have
657 * just incremented index from old to new_idx, should we trigger an
660 static __rte_always_inline int
661 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
663 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
666 static __rte_always_inline void
667 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
669 /* Flush used->idx update before we read avail->flags. */
672 /* Don't kick guest if we don't reach index specified by guest. */
673 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
674 uint16_t old = vq->signalled_used;
675 uint16_t new = vq->last_used_idx;
677 VHOST_LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
679 vhost_used_event(vq),
681 if (vhost_need_event(vhost_used_event(vq), new, old)
682 && (vq->callfd >= 0)) {
683 vq->signalled_used = vq->last_used_idx;
684 eventfd_write(vq->callfd, (eventfd_t) 1);
687 /* Kick the guest if necessary. */
688 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
689 && (vq->callfd >= 0))
690 eventfd_write(vq->callfd, (eventfd_t)1);
694 static __rte_always_inline void
695 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
697 uint16_t old, new, off, off_wrap;
698 bool signalled_used_valid, kick = false;
700 /* Flush used desc update. */
703 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
704 if (vq->driver_event->flags !=
705 VRING_EVENT_F_DISABLE)
710 old = vq->signalled_used;
711 new = vq->last_used_idx;
712 vq->signalled_used = new;
713 signalled_used_valid = vq->signalled_used_valid;
714 vq->signalled_used_valid = true;
716 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
717 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
722 if (unlikely(!signalled_used_valid)) {
729 off_wrap = vq->driver_event->off_wrap;
730 off = off_wrap & ~(1 << 15);
735 if (vq->used_wrap_counter != off_wrap >> 15)
738 if (vhost_need_event(off, new, old))
742 eventfd_write(vq->callfd, (eventfd_t)1);
745 #endif /* _VHOST_NET_CDEV_H_ */