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/queue.h>
12 #include <linux/virtio_net.h>
13 #include <sys/socket.h>
17 #include <rte_ether.h>
18 #include <rte_malloc.h>
19 #include <rte_dmadev.h>
21 #include "rte_vhost.h"
22 #include "vdpa_driver.h"
24 #include "rte_vhost_async.h"
26 /* Used to indicate that the device is running on a data core */
27 #define VIRTIO_DEV_RUNNING ((uint32_t)1 << 0)
28 /* Used to indicate that the device is ready to operate */
29 #define VIRTIO_DEV_READY ((uint32_t)1 << 1)
30 /* Used to indicate that the built-in vhost net device backend is enabled */
31 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET ((uint32_t)1 << 2)
32 /* Used to indicate that the device has its own data path and configured */
33 #define VIRTIO_DEV_VDPA_CONFIGURED ((uint32_t)1 << 3)
34 /* Used to indicate that the feature negotiation failed */
35 #define VIRTIO_DEV_FEATURES_FAILED ((uint32_t)1 << 4)
36 /* Used to indicate that the virtio_net tx code should fill TX ol_flags */
37 #define VIRTIO_DEV_LEGACY_OL_FLAGS ((uint32_t)1 << 5)
38 /* Used to indicate the application has requested statistics collection */
39 #define VIRTIO_DEV_STATS_ENABLED ((uint32_t)1 << 6)
41 /* Backend value set by guest. */
42 #define VIRTIO_DEV_STOPPED -1
44 #define BUF_VECTOR_MAX 256
46 #define VHOST_LOG_CACHE_NR 32
48 #define MAX_PKT_BURST 32
50 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST)
51 #define VHOST_MAX_ASYNC_VEC 2048
52 #define VIRTIO_MAX_RX_PKTLEN 9728U
53 #define VHOST_DMA_MAX_COPY_COMPLETE ((VIRTIO_MAX_RX_PKTLEN / RTE_MBUF_DEFAULT_DATAROOM) \
56 #define PACKED_DESC_ENQUEUE_USED_FLAG(w) \
57 ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
59 #define PACKED_DESC_DEQUEUE_USED_FLAG(w) \
60 ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
61 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
62 VRING_DESC_F_INDIRECT)
64 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
65 sizeof(struct vring_packed_desc))
66 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
68 #ifdef VHOST_GCC_UNROLL_PRAGMA
69 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
70 for (iter = val; iter < size; iter++)
73 #ifdef VHOST_CLANG_UNROLL_PRAGMA
74 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
75 for (iter = val; iter < size; iter++)
78 #ifdef VHOST_ICC_UNROLL_PRAGMA
79 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
80 for (iter = val; iter < size; iter++)
83 #ifndef vhost_for_each_try_unroll
84 #define vhost_for_each_try_unroll(iter, val, num) \
85 for (iter = val; iter < num; iter++)
89 * Structure contains buffer address, length and descriptor index
90 * from vring to do scatter RX.
100 * Structure contains the info for each batched memory copy.
102 struct batch_copy_elem {
110 * Structure that contains the info for batched dirty logging.
112 struct log_cache_entry {
117 struct vring_used_elem_packed {
125 * Virtqueue statistics
127 struct virtqueue_stats {
132 /* Size bins in array as RFC 2819, undersized [0], 64 [1], etc */
133 uint64_t size_bins[8];
134 uint64_t guest_notifications;
136 uint64_t iotlb_misses;
137 uint64_t inflight_submitted;
138 uint64_t inflight_completed;
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 */
200 struct virtio_net_hdr nethdr;
204 struct vhost_iov_iter iov_iter[VHOST_MAX_ASYNC_IT];
205 struct vhost_iovec iovec[VHOST_MAX_ASYNC_VEC];
209 /* data transfer status */
210 struct async_inflight_info *pkts_info;
212 * Packet reorder array. "true" indicates that DMA device
213 * completes all copies for the packet.
215 * Note that this array could be written by multiple threads
216 * simultaneously. For example, in the case of thread0 and
217 * thread1 RX packets from NIC and then enqueue packets to
218 * vring0 and vring1 with own DMA device DMA0 and DMA1, it's
219 * possible for thread0 to get completed copies belonging to
220 * vring1 from DMA0, while thread0 is calling rte_vhost_poll
221 * _enqueue_completed() for vring0 and thread1 is calling
222 * rte_vhost_submit_enqueue_burst() for vring1. In this case,
223 * vq->access_lock cannot protect pkts_cmpl_flag of vring1.
225 * However, since offloading is per-packet basis, each packet
226 * flag will only be written by one thread. And single byte
227 * write is atomic, so no lock for pkts_cmpl_flag is needed.
229 bool *pkts_cmpl_flag;
231 uint16_t pkts_inflight_n;
233 struct vring_used_elem *descs_split;
234 struct vring_used_elem_packed *buffers_packed;
237 uint16_t desc_idx_split;
238 uint16_t buffer_idx_packed;
241 uint16_t last_desc_idx_split;
242 uint16_t last_buffer_idx_packed;
247 * Structure contains variables relevant to RX/TX virtqueues.
249 struct vhost_virtqueue {
251 struct vring_desc *desc;
252 struct vring_packed_desc *desc_packed;
255 struct vring_avail *avail;
256 struct vring_packed_desc_event *driver_event;
259 struct vring_used *used;
260 struct vring_packed_desc_event *device_event;
264 uint16_t last_avail_idx;
265 uint16_t last_used_idx;
266 /* Last used index we notify to front end. */
267 uint16_t signalled_used;
268 bool signalled_used_valid;
269 #define VIRTIO_INVALID_EVENTFD (-1)
270 #define VIRTIO_UNINITIALIZED_EVENTFD (-2)
276 rte_spinlock_t access_lock;
280 struct vring_used_elem *shadow_used_split;
281 struct vring_used_elem_packed *shadow_used_packed;
283 uint16_t shadow_used_idx;
284 /* Record packed ring enqueue latest desc cache aligned index */
285 uint16_t shadow_aligned_idx;
286 /* Record packed ring first dequeue desc index */
287 uint16_t shadow_last_used_idx;
289 uint16_t batch_copy_nb_elems;
290 struct batch_copy_elem *batch_copy_elems;
292 bool used_wrap_counter;
293 bool avail_wrap_counter;
295 /* Physical address of used ring, for logging */
296 uint16_t log_cache_nb_elem;
297 uint64_t log_guest_addr;
298 struct log_cache_entry *log_cache;
300 rte_rwlock_t iotlb_lock;
301 rte_rwlock_t iotlb_pending_lock;
302 struct rte_mempool *iotlb_pool;
303 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
304 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
307 /* Used to notify the guest (trigger interrupt) */
309 /* Currently unused as polling mode is enabled */
312 /* inflight share memory info */
314 struct rte_vhost_inflight_info_split *inflight_split;
315 struct rte_vhost_inflight_info_packed *inflight_packed;
317 struct rte_vhost_resubmit_info *resubmit_inflight;
318 uint64_t global_counter;
320 struct vhost_async *async;
323 #define VIRTIO_UNINITIALIZED_NOTIF (-1)
325 struct vhost_vring_addr ring_addrs;
326 struct virtqueue_stats stats;
327 } __rte_cache_aligned;
329 /* Virtio device status as per Virtio specification */
330 #define VIRTIO_DEVICE_STATUS_RESET 0x00
331 #define VIRTIO_DEVICE_STATUS_ACK 0x01
332 #define VIRTIO_DEVICE_STATUS_DRIVER 0x02
333 #define VIRTIO_DEVICE_STATUS_DRIVER_OK 0x04
334 #define VIRTIO_DEVICE_STATUS_FEATURES_OK 0x08
335 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET 0x40
336 #define VIRTIO_DEVICE_STATUS_FAILED 0x80
338 #define VHOST_MAX_VRING 0x100
339 #define VHOST_MAX_QUEUE_PAIRS 0x80
341 /* Declare IOMMU related bits for older kernels */
342 #ifndef VIRTIO_F_IOMMU_PLATFORM
344 #define VIRTIO_F_IOMMU_PLATFORM 33
346 struct vhost_iotlb_msg {
350 #define VHOST_ACCESS_RO 0x1
351 #define VHOST_ACCESS_WO 0x2
352 #define VHOST_ACCESS_RW 0x3
354 #define VHOST_IOTLB_MISS 1
355 #define VHOST_IOTLB_UPDATE 2
356 #define VHOST_IOTLB_INVALIDATE 3
357 #define VHOST_IOTLB_ACCESS_FAIL 4
361 #define VHOST_IOTLB_MSG 0x1
366 struct vhost_iotlb_msg iotlb;
373 * Define virtio 1.0 for older kernels
375 #ifndef VIRTIO_F_VERSION_1
376 #define VIRTIO_F_VERSION_1 32
379 /* Declare packed ring related bits for older kernels */
380 #ifndef VIRTIO_F_RING_PACKED
382 #define VIRTIO_F_RING_PACKED 34
384 struct vring_packed_desc {
391 struct vring_packed_desc_event {
398 * Declare below packed ring defines unconditionally
399 * as Kernel header might use different names.
401 #define VRING_DESC_F_AVAIL (1ULL << 7)
402 #define VRING_DESC_F_USED (1ULL << 15)
404 #define VRING_EVENT_F_ENABLE 0x0
405 #define VRING_EVENT_F_DISABLE 0x1
406 #define VRING_EVENT_F_DESC 0x2
409 * Available and used descs are in same order
411 #ifndef VIRTIO_F_IN_ORDER
412 #define VIRTIO_F_IN_ORDER 35
415 /* Features supported by this builtin vhost-user net driver. */
416 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
417 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
418 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
419 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
420 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
421 (1ULL << VIRTIO_NET_F_MQ) | \
422 (1ULL << VIRTIO_F_VERSION_1) | \
423 (1ULL << VHOST_F_LOG_ALL) | \
424 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
425 (1ULL << VIRTIO_NET_F_GSO) | \
426 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
427 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
428 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
429 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
430 (1ULL << VIRTIO_NET_F_CSUM) | \
431 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
432 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
433 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
434 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
435 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
436 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
437 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
438 (1ULL << VIRTIO_NET_F_MTU) | \
439 (1ULL << VIRTIO_F_IN_ORDER) | \
440 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
441 (1ULL << VIRTIO_F_RING_PACKED))
445 uint64_t guest_phys_addr;
447 uint64_t host_user_addr;
451 struct inflight_mem_info {
458 * Device structure contains all configuration information relating
462 /* Frontend (QEMU) memory and memory region information */
463 struct rte_vhost_memory *mem;
465 uint64_t protocol_features;
469 /* to tell if we need broadcast rarp packet */
470 int16_t broadcast_rarp;
476 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
477 struct inflight_mem_info *inflight_info;
478 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
479 char ifname[IF_NAME_SZ];
483 struct rte_ether_addr mac;
487 struct rte_vhost_device_ops const *notify_ops;
489 uint32_t nr_guest_pages;
490 uint32_t max_guest_pages;
491 struct guest_page *guest_pages;
494 rte_spinlock_t slave_req_lock;
497 int postcopy_listening;
499 struct rte_vdpa_device *vdpa_dev;
501 /* context data for the external message handlers */
503 /* pre and post vhost user message handlers for the device */
504 struct rte_vhost_user_extern_ops extern_ops;
505 } __rte_cache_aligned;
507 static __rte_always_inline bool
508 vq_is_packed(struct virtio_net *dev)
510 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
514 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
516 uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
518 return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
519 wrap_counter != !!(flags & VRING_DESC_F_USED);
523 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
525 vq->last_used_idx += num;
526 if (vq->last_used_idx >= vq->size) {
527 vq->used_wrap_counter ^= 1;
528 vq->last_used_idx -= vq->size;
533 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
535 vq->last_avail_idx += num;
536 if (vq->last_avail_idx >= vq->size) {
537 vq->avail_wrap_counter ^= 1;
538 vq->last_avail_idx -= vq->size;
542 void __vhost_log_cache_write(struct virtio_net *dev,
543 struct vhost_virtqueue *vq,
544 uint64_t addr, uint64_t len);
545 void __vhost_log_cache_write_iova(struct virtio_net *dev,
546 struct vhost_virtqueue *vq,
547 uint64_t iova, uint64_t len);
548 void __vhost_log_cache_sync(struct virtio_net *dev,
549 struct vhost_virtqueue *vq);
550 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
551 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
552 uint64_t iova, uint64_t len);
554 static __rte_always_inline void
555 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
557 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
558 __vhost_log_write(dev, addr, len);
561 static __rte_always_inline void
562 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
564 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
565 __vhost_log_cache_sync(dev, vq);
568 static __rte_always_inline void
569 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
570 uint64_t addr, uint64_t len)
572 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
573 __vhost_log_cache_write(dev, vq, addr, len);
576 static __rte_always_inline void
577 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
578 uint64_t offset, uint64_t len)
580 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
581 if (unlikely(vq->log_guest_addr == 0))
583 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
588 static __rte_always_inline void
589 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
590 uint64_t offset, uint64_t len)
592 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
593 if (unlikely(vq->log_guest_addr == 0))
595 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
599 static __rte_always_inline void
600 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
601 uint64_t iova, uint64_t len)
603 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
606 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
607 __vhost_log_cache_write_iova(dev, vq, iova, len);
609 __vhost_log_cache_write(dev, vq, iova, len);
612 static __rte_always_inline void
613 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
614 uint64_t iova, uint64_t len)
616 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
619 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
620 __vhost_log_write_iova(dev, vq, iova, len);
622 __vhost_log_write(dev, iova, len);
625 extern int vhost_config_log_level;
626 extern int vhost_data_log_level;
628 #define VHOST_LOG_CONFIG(level, fmt, args...) \
629 rte_log(RTE_LOG_ ## level, vhost_config_log_level, \
630 "VHOST_CONFIG: " fmt, ##args)
632 #define VHOST_LOG_DATA(level, fmt, args...) \
633 (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ? \
634 rte_log(RTE_LOG_ ## level, vhost_data_log_level, \
635 "VHOST_DATA : " fmt, ##args) : \
638 #ifdef RTE_LIBRTE_VHOST_DEBUG
639 #define VHOST_MAX_PRINT_BUFF 6072
640 #define PRINT_PACKET(device, addr, size, header) do { \
641 char *pkt_addr = (char *)(addr); \
642 unsigned int index; \
643 char packet[VHOST_MAX_PRINT_BUFF]; \
646 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
648 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
649 for (index = 0; index < (size); index++) { \
650 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
651 "%02hhx ", pkt_addr[index]); \
653 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
655 VHOST_LOG_DATA(DEBUG, "%s", packet); \
658 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
661 extern struct virtio_net *vhost_devices[RTE_MAX_VHOST_DEVICE];
663 #define VHOST_BINARY_SEARCH_THRESH 256
665 static __rte_always_inline int guest_page_addrcmp(const void *p1,
668 const struct guest_page *page1 = (const struct guest_page *)p1;
669 const struct guest_page *page2 = (const struct guest_page *)p2;
671 if (page1->guest_phys_addr > page2->guest_phys_addr)
673 if (page1->guest_phys_addr < page2->guest_phys_addr)
679 static __rte_always_inline int guest_page_rangecmp(const void *p1, const void *p2)
681 const struct guest_page *page1 = (const struct guest_page *)p1;
682 const struct guest_page *page2 = (const struct guest_page *)p2;
684 if (page1->guest_phys_addr >= page2->guest_phys_addr) {
685 if (page1->guest_phys_addr < page2->guest_phys_addr + page2->size)
693 static __rte_always_inline rte_iova_t
694 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
695 uint64_t gpa_size, uint64_t *hpa_size)
698 struct guest_page *page;
699 struct guest_page key;
701 *hpa_size = gpa_size;
702 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
703 key.guest_phys_addr = gpa;
704 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
705 sizeof(struct guest_page), guest_page_rangecmp);
707 if (gpa + gpa_size <=
708 page->guest_phys_addr + page->size) {
709 return gpa - page->guest_phys_addr +
711 } else if (gpa < page->guest_phys_addr +
713 *hpa_size = page->guest_phys_addr +
715 return gpa - page->guest_phys_addr +
720 for (i = 0; i < dev->nr_guest_pages; i++) {
721 page = &dev->guest_pages[i];
723 if (gpa >= page->guest_phys_addr) {
724 if (gpa + gpa_size <=
725 page->guest_phys_addr + page->size) {
726 return gpa - page->guest_phys_addr +
728 } else if (gpa < page->guest_phys_addr +
730 *hpa_size = page->guest_phys_addr +
732 return gpa - page->guest_phys_addr +
743 /* Convert guest physical address to host physical address */
744 static __rte_always_inline rte_iova_t
745 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
750 hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
751 return hpa_size == size ? hpa : 0;
754 static __rte_always_inline uint64_t
755 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
757 struct rte_vhost_mem_region *r;
760 if (unlikely(!dev || !dev->mem))
763 for (i = 0; i < dev->mem->nregions; i++) {
764 r = &dev->mem->regions[i];
766 if (vva >= r->host_user_addr &&
767 vva + len < r->host_user_addr + r->size) {
768 return r->guest_phys_addr + vva - r->host_user_addr;
774 static __rte_always_inline struct virtio_net *
777 struct virtio_net *dev = vhost_devices[vid];
779 if (unlikely(!dev)) {
780 VHOST_LOG_CONFIG(ERR,
781 "(%d) device not found.\n", vid);
787 int vhost_new_device(void);
788 void cleanup_device(struct virtio_net *dev, int destroy);
789 void reset_device(struct virtio_net *dev);
790 void vhost_destroy_device(int);
791 void vhost_destroy_device_notify(struct virtio_net *dev);
793 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
794 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
795 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
797 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
799 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
801 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
802 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags, bool stats_enabled);
803 void vhost_enable_extbuf(int vid);
804 void vhost_enable_linearbuf(int vid);
805 int vhost_enable_guest_notification(struct virtio_net *dev,
806 struct vhost_virtqueue *vq, int enable);
808 struct rte_vhost_device_ops const *vhost_driver_callback_get(const char *path);
811 * Backend-specific cleanup.
813 * TODO: fix it; we have one backend now
815 void vhost_backend_cleanup(struct virtio_net *dev);
817 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
818 uint64_t iova, uint64_t *len, uint8_t perm);
819 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
820 struct vhost_virtqueue *vq,
821 uint64_t desc_addr, uint64_t desc_len);
822 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
823 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
825 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
827 static __rte_always_inline uint64_t
828 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
829 uint64_t iova, uint64_t *len, uint8_t perm)
831 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
832 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
834 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
837 #define vhost_avail_event(vr) \
838 (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
839 #define vhost_used_event(vr) \
840 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
843 * The following is used with VIRTIO_RING_F_EVENT_IDX.
844 * Assuming a given event_idx value from the other size, if we have
845 * just incremented index from old to new_idx, should we trigger an
848 static __rte_always_inline int
849 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
851 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
854 static __rte_always_inline void
855 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
857 /* Flush used->idx update before we read avail->flags. */
858 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
860 /* Don't kick guest if we don't reach index specified by guest. */
861 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
862 uint16_t old = vq->signalled_used;
863 uint16_t new = vq->last_used_idx;
864 bool signalled_used_valid = vq->signalled_used_valid;
866 vq->signalled_used = new;
867 vq->signalled_used_valid = true;
869 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
871 vhost_used_event(vq),
874 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
875 (vq->callfd >= 0)) ||
876 unlikely(!signalled_used_valid)) {
877 eventfd_write(vq->callfd, (eventfd_t) 1);
878 if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
879 vq->stats.guest_notifications++;
880 if (dev->notify_ops->guest_notified)
881 dev->notify_ops->guest_notified(dev->vid);
884 /* Kick the guest if necessary. */
885 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
886 && (vq->callfd >= 0)) {
887 eventfd_write(vq->callfd, (eventfd_t)1);
888 if (dev->flags & VIRTIO_DEV_STATS_ENABLED)
889 vq->stats.guest_notifications++;
890 if (dev->notify_ops->guest_notified)
891 dev->notify_ops->guest_notified(dev->vid);
896 static __rte_always_inline void
897 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
899 uint16_t old, new, off, off_wrap;
900 bool signalled_used_valid, kick = false;
902 /* Flush used desc update. */
903 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
905 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
906 if (vq->driver_event->flags !=
907 VRING_EVENT_F_DISABLE)
912 old = vq->signalled_used;
913 new = vq->last_used_idx;
914 vq->signalled_used = new;
915 signalled_used_valid = vq->signalled_used_valid;
916 vq->signalled_used_valid = true;
918 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
919 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
924 if (unlikely(!signalled_used_valid)) {
929 rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
931 off_wrap = vq->driver_event->off_wrap;
932 off = off_wrap & ~(1 << 15);
937 if (vq->used_wrap_counter != off_wrap >> 15)
940 if (vhost_need_event(off, new, old))
944 eventfd_write(vq->callfd, (eventfd_t)1);
945 if (dev->notify_ops->guest_notified)
946 dev->notify_ops->guest_notified(dev->vid);
950 static __rte_always_inline void
951 free_ind_table(void *idesc)
956 static __rte_always_inline void
957 restore_mbuf(struct rte_mbuf *m)
959 uint32_t mbuf_size, priv_size;
962 priv_size = rte_pktmbuf_priv_size(m->pool);
963 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
964 /* start of buffer is after mbuf structure and priv data */
966 m->buf_addr = (char *)m + mbuf_size;
967 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
972 static __rte_always_inline bool
973 mbuf_is_consumed(struct rte_mbuf *m)
976 if (rte_mbuf_refcnt_read(m) > 1)
983 #endif /* _VHOST_NET_CDEV_H_ */