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_malloc.h>
21 #include <rte_dmadev.h>
23 #include "rte_vhost.h"
24 #include "vdpa_driver.h"
26 #include "rte_vhost_async.h"
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)
43 /* Backend value set by guest. */
44 #define VIRTIO_DEV_STOPPED -1
46 #define BUF_VECTOR_MAX 256
48 #define VHOST_LOG_CACHE_NR 32
50 #define MAX_PKT_BURST 32
52 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST)
53 #define VHOST_MAX_ASYNC_VEC 2048
54 #define VIRTIO_MAX_RX_PKTLEN 9728U
55 #define VHOST_DMA_MAX_COPY_COMPLETE ((VIRTIO_MAX_RX_PKTLEN / RTE_MBUF_DEFAULT_DATAROOM) \
58 #define PACKED_DESC_ENQUEUE_USED_FLAG(w) \
59 ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
61 #define PACKED_DESC_DEQUEUE_USED_FLAG(w) \
62 ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
63 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
64 VRING_DESC_F_INDIRECT)
66 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
67 sizeof(struct vring_packed_desc))
68 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
70 #ifdef VHOST_GCC_UNROLL_PRAGMA
71 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
72 for (iter = val; iter < size; iter++)
75 #ifdef VHOST_CLANG_UNROLL_PRAGMA
76 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
77 for (iter = val; iter < size; iter++)
80 #ifdef VHOST_ICC_UNROLL_PRAGMA
81 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
82 for (iter = val; iter < size; iter++)
85 #ifndef vhost_for_each_try_unroll
86 #define vhost_for_each_try_unroll(iter, val, num) \
87 for (iter = val; iter < num; iter++)
91 * Structure contains buffer address, length and descriptor index
92 * from vring to do scatter RX.
102 * Structure contains the info for each batched memory copy.
104 struct batch_copy_elem {
112 * Structure that contains the info for batched dirty logging.
114 struct log_cache_entry {
119 struct vring_used_elem_packed {
127 * Virtqueue statistics
129 struct virtqueue_stats {
134 /* Size bins in array as RFC 2819, undersized [0], 64 [1], etc */
135 uint64_t size_bins[8];
136 uint64_t guest_notifications;
138 uint64_t iotlb_misses;
153 struct vhost_iov_iter {
154 /** pointer to the iovec array */
155 struct vhost_iovec *iov;
156 /** number of iovec in this iterator */
157 unsigned long nr_segs;
160 struct async_dma_vchan_info {
161 /* circular array to track if packet copy completes */
162 bool **pkts_cmpl_flag_addr;
164 /* max elements in 'pkts_cmpl_flag_addr' */
166 /* ring index mask for 'pkts_cmpl_flag_addr' */
170 * DMA virtual channel lock. Although it is able to bind DMA
171 * virtual channels to data plane threads, vhost control plane
172 * thread could call data plane functions too, thus causing
173 * DMA device contention.
175 * For example, in VM exit case, vhost control plane thread needs
176 * to clear in-flight packets before disable vring, but there could
177 * be anotther data plane thread is enqueuing packets to the same
178 * vring with the same DMA virtual channel. As dmadev PMD functions
179 * are lock-free, the control plane and data plane threads could
180 * operate the same DMA virtual channel at the same time.
182 rte_spinlock_t dma_lock;
185 struct async_dma_info {
186 struct async_dma_vchan_info *vchans;
187 /* number of registered virtual channels */
191 extern struct async_dma_info dma_copy_track[RTE_DMADEV_DEFAULT_MAX];
194 * inflight async packet information
196 struct async_inflight_info {
197 struct rte_mbuf *mbuf;
198 uint16_t descs; /* num of descs inflight */
199 uint16_t nr_buffers; /* num of buffers inflight for packed ring */
203 struct vhost_iov_iter iov_iter[VHOST_MAX_ASYNC_IT];
204 struct vhost_iovec iovec[VHOST_MAX_ASYNC_VEC];
208 /* data transfer status */
209 struct async_inflight_info *pkts_info;
211 * Packet reorder array. "true" indicates that DMA device
212 * completes all copies for the packet.
214 * Note that this array could be written by multiple threads
215 * simultaneously. For example, in the case of thread0 and
216 * thread1 RX packets from NIC and then enqueue packets to
217 * vring0 and vring1 with own DMA device DMA0 and DMA1, it's
218 * possible for thread0 to get completed copies belonging to
219 * vring1 from DMA0, while thread0 is calling rte_vhost_poll
220 * _enqueue_completed() for vring0 and thread1 is calling
221 * rte_vhost_submit_enqueue_burst() for vring1. In this case,
222 * vq->access_lock cannot protect pkts_cmpl_flag of vring1.
224 * However, since offloading is per-packet basis, each packet
225 * flag will only be written by one thread. And single byte
226 * write is atomic, so no lock for pkts_cmpl_flag is needed.
228 bool *pkts_cmpl_flag;
230 uint16_t pkts_inflight_n;
232 struct vring_used_elem *descs_split;
233 struct vring_used_elem_packed *buffers_packed;
236 uint16_t desc_idx_split;
237 uint16_t buffer_idx_packed;
240 uint16_t last_desc_idx_split;
241 uint16_t last_buffer_idx_packed;
246 * Structure contains variables relevant to RX/TX virtqueues.
248 struct vhost_virtqueue {
250 struct vring_desc *desc;
251 struct vring_packed_desc *desc_packed;
254 struct vring_avail *avail;
255 struct vring_packed_desc_event *driver_event;
258 struct vring_used *used;
259 struct vring_packed_desc_event *device_event;
263 uint16_t last_avail_idx;
264 uint16_t last_used_idx;
265 /* Last used index we notify to front end. */
266 uint16_t signalled_used;
267 bool signalled_used_valid;
268 #define VIRTIO_INVALID_EVENTFD (-1)
269 #define VIRTIO_UNINITIALIZED_EVENTFD (-2)
275 rte_spinlock_t access_lock;
279 struct vring_used_elem *shadow_used_split;
280 struct vring_used_elem_packed *shadow_used_packed;
282 uint16_t shadow_used_idx;
283 /* Record packed ring enqueue latest desc cache aligned index */
284 uint16_t shadow_aligned_idx;
285 /* Record packed ring first dequeue desc index */
286 uint16_t shadow_last_used_idx;
288 uint16_t batch_copy_nb_elems;
289 struct batch_copy_elem *batch_copy_elems;
291 bool used_wrap_counter;
292 bool avail_wrap_counter;
294 /* Physical address of used ring, for logging */
295 uint16_t log_cache_nb_elem;
296 uint64_t log_guest_addr;
297 struct log_cache_entry *log_cache;
299 rte_rwlock_t iotlb_lock;
300 rte_rwlock_t iotlb_pending_lock;
301 struct rte_mempool *iotlb_pool;
302 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
303 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
306 /* Used to notify the guest (trigger interrupt) */
308 /* Currently unused as polling mode is enabled */
311 /* inflight share memory info */
313 struct rte_vhost_inflight_info_split *inflight_split;
314 struct rte_vhost_inflight_info_packed *inflight_packed;
316 struct rte_vhost_resubmit_info *resubmit_inflight;
317 uint64_t global_counter;
319 struct vhost_async *async;
322 #define VIRTIO_UNINITIALIZED_NOTIF (-1)
324 struct vhost_vring_addr ring_addrs;
325 struct virtqueue_stats stats;
326 } __rte_cache_aligned;
328 /* Virtio device status as per Virtio specification */
329 #define VIRTIO_DEVICE_STATUS_RESET 0x00
330 #define VIRTIO_DEVICE_STATUS_ACK 0x01
331 #define VIRTIO_DEVICE_STATUS_DRIVER 0x02
332 #define VIRTIO_DEVICE_STATUS_DRIVER_OK 0x04
333 #define VIRTIO_DEVICE_STATUS_FEATURES_OK 0x08
334 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET 0x40
335 #define VIRTIO_DEVICE_STATUS_FAILED 0x80
337 #define VHOST_MAX_VRING 0x100
338 #define VHOST_MAX_QUEUE_PAIRS 0x80
340 /* Declare IOMMU related bits for older kernels */
341 #ifndef VIRTIO_F_IOMMU_PLATFORM
343 #define VIRTIO_F_IOMMU_PLATFORM 33
345 struct vhost_iotlb_msg {
349 #define VHOST_ACCESS_RO 0x1
350 #define VHOST_ACCESS_WO 0x2
351 #define VHOST_ACCESS_RW 0x3
353 #define VHOST_IOTLB_MISS 1
354 #define VHOST_IOTLB_UPDATE 2
355 #define VHOST_IOTLB_INVALIDATE 3
356 #define VHOST_IOTLB_ACCESS_FAIL 4
360 #define VHOST_IOTLB_MSG 0x1
365 struct vhost_iotlb_msg iotlb;
372 * Define virtio 1.0 for older kernels
374 #ifndef VIRTIO_F_VERSION_1
375 #define VIRTIO_F_VERSION_1 32
378 /* Declare packed ring related bits for older kernels */
379 #ifndef VIRTIO_F_RING_PACKED
381 #define VIRTIO_F_RING_PACKED 34
383 struct vring_packed_desc {
390 struct vring_packed_desc_event {
397 * Declare below packed ring defines unconditionally
398 * as Kernel header might use different names.
400 #define VRING_DESC_F_AVAIL (1ULL << 7)
401 #define VRING_DESC_F_USED (1ULL << 15)
403 #define VRING_EVENT_F_ENABLE 0x0
404 #define VRING_EVENT_F_DISABLE 0x1
405 #define VRING_EVENT_F_DESC 0x2
408 * Available and used descs are in same order
410 #ifndef VIRTIO_F_IN_ORDER
411 #define VIRTIO_F_IN_ORDER 35
414 /* Features supported by this builtin vhost-user net driver. */
415 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
416 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
417 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
418 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
419 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
420 (1ULL << VIRTIO_NET_F_MQ) | \
421 (1ULL << VIRTIO_F_VERSION_1) | \
422 (1ULL << VHOST_F_LOG_ALL) | \
423 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
424 (1ULL << VIRTIO_NET_F_GSO) | \
425 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
426 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
427 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
428 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
429 (1ULL << VIRTIO_NET_F_CSUM) | \
430 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
431 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
432 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
433 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
434 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
435 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
436 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
437 (1ULL << VIRTIO_NET_F_MTU) | \
438 (1ULL << VIRTIO_F_IN_ORDER) | \
439 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
440 (1ULL << VIRTIO_F_RING_PACKED))
444 uint64_t guest_phys_addr;
446 uint64_t host_user_addr;
450 struct inflight_mem_info {
457 * Device structure contains all configuration information relating
461 /* Frontend (QEMU) memory and memory region information */
462 struct rte_vhost_memory *mem;
464 uint64_t protocol_features;
468 /* to tell if we need broadcast rarp packet */
469 int16_t broadcast_rarp;
475 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
476 struct inflight_mem_info *inflight_info;
477 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
478 char ifname[IF_NAME_SZ];
482 struct rte_ether_addr mac;
486 struct rte_vhost_device_ops const *notify_ops;
488 uint32_t nr_guest_pages;
489 uint32_t max_guest_pages;
490 struct guest_page *guest_pages;
493 rte_spinlock_t slave_req_lock;
496 int postcopy_listening;
498 struct rte_vdpa_device *vdpa_dev;
500 /* context data for the external message handlers */
502 /* pre and post vhost user message handlers for the device */
503 struct rte_vhost_user_extern_ops extern_ops;
504 } __rte_cache_aligned;
506 static __rte_always_inline bool
507 vq_is_packed(struct virtio_net *dev)
509 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
513 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
515 uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
517 return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
518 wrap_counter != !!(flags & VRING_DESC_F_USED);
522 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
524 vq->last_used_idx += num;
525 if (vq->last_used_idx >= vq->size) {
526 vq->used_wrap_counter ^= 1;
527 vq->last_used_idx -= vq->size;
532 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
534 vq->last_avail_idx += num;
535 if (vq->last_avail_idx >= vq->size) {
536 vq->avail_wrap_counter ^= 1;
537 vq->last_avail_idx -= vq->size;
541 void __vhost_log_cache_write(struct virtio_net *dev,
542 struct vhost_virtqueue *vq,
543 uint64_t addr, uint64_t len);
544 void __vhost_log_cache_write_iova(struct virtio_net *dev,
545 struct vhost_virtqueue *vq,
546 uint64_t iova, uint64_t len);
547 void __vhost_log_cache_sync(struct virtio_net *dev,
548 struct vhost_virtqueue *vq);
549 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
550 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
551 uint64_t iova, uint64_t len);
553 static __rte_always_inline void
554 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
556 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
557 __vhost_log_write(dev, addr, len);
560 static __rte_always_inline void
561 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
563 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
564 __vhost_log_cache_sync(dev, vq);
567 static __rte_always_inline void
568 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
569 uint64_t addr, uint64_t len)
571 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
572 __vhost_log_cache_write(dev, vq, addr, len);
575 static __rte_always_inline void
576 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
577 uint64_t offset, uint64_t len)
579 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
580 if (unlikely(vq->log_guest_addr == 0))
582 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
587 static __rte_always_inline void
588 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
589 uint64_t offset, uint64_t len)
591 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
592 if (unlikely(vq->log_guest_addr == 0))
594 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
598 static __rte_always_inline void
599 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
600 uint64_t iova, uint64_t len)
602 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
605 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
606 __vhost_log_cache_write_iova(dev, vq, iova, len);
608 __vhost_log_cache_write(dev, vq, iova, len);
611 static __rte_always_inline void
612 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
613 uint64_t iova, uint64_t len)
615 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
618 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
619 __vhost_log_write_iova(dev, vq, iova, len);
621 __vhost_log_write(dev, iova, len);
624 extern int vhost_config_log_level;
625 extern int vhost_data_log_level;
627 #define VHOST_LOG_CONFIG(level, fmt, args...) \
628 rte_log(RTE_LOG_ ## level, vhost_config_log_level, \
629 "VHOST_CONFIG: " fmt, ##args)
631 #define VHOST_LOG_DATA(level, fmt, args...) \
632 (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ? \
633 rte_log(RTE_LOG_ ## level, vhost_data_log_level, \
634 "VHOST_DATA : " fmt, ##args) : \
637 #ifdef RTE_LIBRTE_VHOST_DEBUG
638 #define VHOST_MAX_PRINT_BUFF 6072
639 #define PRINT_PACKET(device, addr, size, header) do { \
640 char *pkt_addr = (char *)(addr); \
641 unsigned int index; \
642 char packet[VHOST_MAX_PRINT_BUFF]; \
645 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
647 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
648 for (index = 0; index < (size); index++) { \
649 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
650 "%02hhx ", pkt_addr[index]); \
652 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
654 VHOST_LOG_DATA(DEBUG, "%s", packet); \
657 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
660 extern struct virtio_net *vhost_devices[RTE_MAX_VHOST_DEVICE];
662 #define VHOST_BINARY_SEARCH_THRESH 256
664 static __rte_always_inline int guest_page_addrcmp(const void *p1,
667 const struct guest_page *page1 = (const struct guest_page *)p1;
668 const struct guest_page *page2 = (const struct guest_page *)p2;
670 if (page1->guest_phys_addr > page2->guest_phys_addr)
672 if (page1->guest_phys_addr < page2->guest_phys_addr)
678 static __rte_always_inline int guest_page_rangecmp(const void *p1, const void *p2)
680 const struct guest_page *page1 = (const struct guest_page *)p1;
681 const struct guest_page *page2 = (const struct guest_page *)p2;
683 if (page1->guest_phys_addr >= page2->guest_phys_addr) {
684 if (page1->guest_phys_addr < page2->guest_phys_addr + page2->size)
692 static __rte_always_inline rte_iova_t
693 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
694 uint64_t gpa_size, uint64_t *hpa_size)
697 struct guest_page *page;
698 struct guest_page key;
700 *hpa_size = gpa_size;
701 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
702 key.guest_phys_addr = gpa;
703 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
704 sizeof(struct guest_page), guest_page_rangecmp);
706 if (gpa + gpa_size <=
707 page->guest_phys_addr + page->size) {
708 return gpa - page->guest_phys_addr +
710 } else if (gpa < page->guest_phys_addr +
712 *hpa_size = page->guest_phys_addr +
714 return gpa - page->guest_phys_addr +
719 for (i = 0; i < dev->nr_guest_pages; i++) {
720 page = &dev->guest_pages[i];
722 if (gpa >= page->guest_phys_addr) {
723 if (gpa + gpa_size <=
724 page->guest_phys_addr + page->size) {
725 return gpa - page->guest_phys_addr +
727 } else if (gpa < page->guest_phys_addr +
729 *hpa_size = page->guest_phys_addr +
731 return gpa - page->guest_phys_addr +
742 /* Convert guest physical address to host physical address */
743 static __rte_always_inline rte_iova_t
744 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
749 hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
750 return hpa_size == size ? hpa : 0;
753 static __rte_always_inline uint64_t
754 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
756 struct rte_vhost_mem_region *r;
759 if (unlikely(!dev || !dev->mem))
762 for (i = 0; i < dev->mem->nregions; i++) {
763 r = &dev->mem->regions[i];
765 if (vva >= r->host_user_addr &&
766 vva + len < r->host_user_addr + r->size) {
767 return r->guest_phys_addr + vva - r->host_user_addr;
773 static __rte_always_inline struct virtio_net *
776 struct virtio_net *dev = vhost_devices[vid];
778 if (unlikely(!dev)) {
779 VHOST_LOG_CONFIG(ERR,
780 "(%d) device not found.\n", vid);
786 int vhost_new_device(void);
787 void cleanup_device(struct virtio_net *dev, int destroy);
788 void reset_device(struct virtio_net *dev);
789 void vhost_destroy_device(int);
790 void vhost_destroy_device_notify(struct virtio_net *dev);
792 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
793 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
794 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
796 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
798 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
800 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
801 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags, bool stats_enabled);
802 void vhost_enable_extbuf(int vid);
803 void vhost_enable_linearbuf(int vid);
804 int vhost_enable_guest_notification(struct virtio_net *dev,
805 struct vhost_virtqueue *vq, int enable);
807 struct rte_vhost_device_ops const *vhost_driver_callback_get(const char *path);
810 * Backend-specific cleanup.
812 * TODO: fix it; we have one backend now
814 void vhost_backend_cleanup(struct virtio_net *dev);
816 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
817 uint64_t iova, uint64_t *len, uint8_t perm);
818 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
819 struct vhost_virtqueue *vq,
820 uint64_t desc_addr, uint64_t desc_len);
821 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
822 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
824 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
826 static __rte_always_inline uint64_t
827 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
828 uint64_t iova, uint64_t *len, uint8_t perm)
830 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
831 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
833 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
836 #define vhost_avail_event(vr) \
837 (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
838 #define vhost_used_event(vr) \
839 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
842 * The following is used with VIRTIO_RING_F_EVENT_IDX.
843 * Assuming a given event_idx value from the other size, if we have
844 * just incremented index from old to new_idx, should we trigger an
847 static __rte_always_inline int
848 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
850 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
853 static __rte_always_inline void
854 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
856 /* Flush used->idx update before we read avail->flags. */
857 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
859 /* Don't kick guest if we don't reach index specified by guest. */
860 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
861 uint16_t old = vq->signalled_used;
862 uint16_t new = vq->last_used_idx;
863 bool signalled_used_valid = vq->signalled_used_valid;
865 vq->signalled_used = new;
866 vq->signalled_used_valid = true;
868 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
870 vhost_used_event(vq),
873 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
874 (vq->callfd >= 0)) ||
875 unlikely(!signalled_used_valid)) {
876 eventfd_write(vq->callfd, (eventfd_t) 1);
877 if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
878 vq->stats.guest_notifications++;
879 if (dev->notify_ops->guest_notified)
880 dev->notify_ops->guest_notified(dev->vid);
883 /* Kick the guest if necessary. */
884 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
885 && (vq->callfd >= 0)) {
886 eventfd_write(vq->callfd, (eventfd_t)1);
887 if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
888 vq->stats.guest_notifications++;
889 if (dev->notify_ops->guest_notified)
890 dev->notify_ops->guest_notified(dev->vid);
895 static __rte_always_inline void
896 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
898 uint16_t old, new, off, off_wrap;
899 bool signalled_used_valid, kick = false;
901 /* Flush used desc update. */
902 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
904 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
905 if (vq->driver_event->flags !=
906 VRING_EVENT_F_DISABLE)
911 old = vq->signalled_used;
912 new = vq->last_used_idx;
913 vq->signalled_used = new;
914 signalled_used_valid = vq->signalled_used_valid;
915 vq->signalled_used_valid = true;
917 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
918 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
923 if (unlikely(!signalled_used_valid)) {
928 rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
930 off_wrap = vq->driver_event->off_wrap;
931 off = off_wrap & ~(1 << 15);
936 if (vq->used_wrap_counter != off_wrap >> 15)
939 if (vhost_need_event(off, new, old))
943 eventfd_write(vq->callfd, (eventfd_t)1);
944 if (dev->notify_ops->guest_notified)
945 dev->notify_ops->guest_notified(dev->vid);
949 static __rte_always_inline void
950 free_ind_table(void *idesc)
955 static __rte_always_inline void
956 restore_mbuf(struct rte_mbuf *m)
958 uint32_t mbuf_size, priv_size;
961 priv_size = rte_pktmbuf_priv_size(m->pool);
962 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
963 /* start of buffer is after mbuf structure and priv data */
965 m->buf_addr = (char *)m + mbuf_size;
966 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
971 static __rte_always_inline bool
972 mbuf_is_consumed(struct rte_mbuf *m)
975 if (rte_mbuf_refcnt_read(m) > 1)
982 #endif /* _VHOST_NET_CDEV_H_ */