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];
150 struct vhost_iov_iter {
151 /** pointer to the iovec array */
152 struct vhost_iovec *iov;
153 /** number of iovec in this iterator */
154 unsigned long nr_segs;
157 struct async_dma_vchan_info {
158 /* circular array to track if packet copy completes */
159 bool **pkts_cmpl_flag_addr;
161 /* max elements in 'pkts_cmpl_flag_addr' */
163 /* ring index mask for 'pkts_cmpl_flag_addr' */
167 * DMA virtual channel lock. Although it is able to bind DMA
168 * virtual channels to data plane threads, vhost control plane
169 * thread could call data plane functions too, thus causing
170 * DMA device contention.
172 * For example, in VM exit case, vhost control plane thread needs
173 * to clear in-flight packets before disable vring, but there could
174 * be anotther data plane thread is enqueuing packets to the same
175 * vring with the same DMA virtual channel. As dmadev PMD functions
176 * are lock-free, the control plane and data plane threads could
177 * operate the same DMA virtual channel at the same time.
179 rte_spinlock_t dma_lock;
182 struct async_dma_info {
183 struct async_dma_vchan_info *vchans;
184 /* number of registered virtual channels */
188 extern struct async_dma_info dma_copy_track[RTE_DMADEV_DEFAULT_MAX];
191 * inflight async packet information
193 struct async_inflight_info {
194 struct rte_mbuf *mbuf;
195 uint16_t descs; /* num of descs inflight */
196 uint16_t nr_buffers; /* num of buffers inflight for packed ring */
200 struct vhost_iov_iter iov_iter[VHOST_MAX_ASYNC_IT];
201 struct vhost_iovec iovec[VHOST_MAX_ASYNC_VEC];
205 /* data transfer status */
206 struct async_inflight_info *pkts_info;
208 * Packet reorder array. "true" indicates that DMA device
209 * completes all copies for the packet.
211 * Note that this array could be written by multiple threads
212 * simultaneously. For example, in the case of thread0 and
213 * thread1 RX packets from NIC and then enqueue packets to
214 * vring0 and vring1 with own DMA device DMA0 and DMA1, it's
215 * possible for thread0 to get completed copies belonging to
216 * vring1 from DMA0, while thread0 is calling rte_vhost_poll
217 * _enqueue_completed() for vring0 and thread1 is calling
218 * rte_vhost_submit_enqueue_burst() for vring1. In this case,
219 * vq->access_lock cannot protect pkts_cmpl_flag of vring1.
221 * However, since offloading is per-packet basis, each packet
222 * flag will only be written by one thread. And single byte
223 * write is atomic, so no lock for pkts_cmpl_flag is needed.
225 bool *pkts_cmpl_flag;
227 uint16_t pkts_inflight_n;
229 struct vring_used_elem *descs_split;
230 struct vring_used_elem_packed *buffers_packed;
233 uint16_t desc_idx_split;
234 uint16_t buffer_idx_packed;
237 uint16_t last_desc_idx_split;
238 uint16_t last_buffer_idx_packed;
243 * Structure contains variables relevant to RX/TX virtqueues.
245 struct vhost_virtqueue {
247 struct vring_desc *desc;
248 struct vring_packed_desc *desc_packed;
251 struct vring_avail *avail;
252 struct vring_packed_desc_event *driver_event;
255 struct vring_used *used;
256 struct vring_packed_desc_event *device_event;
260 uint16_t last_avail_idx;
261 uint16_t last_used_idx;
262 /* Last used index we notify to front end. */
263 uint16_t signalled_used;
264 bool signalled_used_valid;
265 #define VIRTIO_INVALID_EVENTFD (-1)
266 #define VIRTIO_UNINITIALIZED_EVENTFD (-2)
272 rte_spinlock_t access_lock;
276 struct vring_used_elem *shadow_used_split;
277 struct vring_used_elem_packed *shadow_used_packed;
279 uint16_t shadow_used_idx;
280 /* Record packed ring enqueue latest desc cache aligned index */
281 uint16_t shadow_aligned_idx;
282 /* Record packed ring first dequeue desc index */
283 uint16_t shadow_last_used_idx;
285 uint16_t batch_copy_nb_elems;
286 struct batch_copy_elem *batch_copy_elems;
288 bool used_wrap_counter;
289 bool avail_wrap_counter;
291 /* Physical address of used ring, for logging */
292 uint16_t log_cache_nb_elem;
293 uint64_t log_guest_addr;
294 struct log_cache_entry *log_cache;
296 rte_rwlock_t iotlb_lock;
297 rte_rwlock_t iotlb_pending_lock;
298 struct rte_mempool *iotlb_pool;
299 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
300 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
303 /* Used to notify the guest (trigger interrupt) */
305 /* Currently unused as polling mode is enabled */
308 /* inflight share memory info */
310 struct rte_vhost_inflight_info_split *inflight_split;
311 struct rte_vhost_inflight_info_packed *inflight_packed;
313 struct rte_vhost_resubmit_info *resubmit_inflight;
314 uint64_t global_counter;
316 struct vhost_async *async;
319 #define VIRTIO_UNINITIALIZED_NOTIF (-1)
321 struct vhost_vring_addr ring_addrs;
322 struct virtqueue_stats stats;
323 } __rte_cache_aligned;
325 /* Virtio device status as per Virtio specification */
326 #define VIRTIO_DEVICE_STATUS_RESET 0x00
327 #define VIRTIO_DEVICE_STATUS_ACK 0x01
328 #define VIRTIO_DEVICE_STATUS_DRIVER 0x02
329 #define VIRTIO_DEVICE_STATUS_DRIVER_OK 0x04
330 #define VIRTIO_DEVICE_STATUS_FEATURES_OK 0x08
331 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET 0x40
332 #define VIRTIO_DEVICE_STATUS_FAILED 0x80
334 #define VHOST_MAX_VRING 0x100
335 #define VHOST_MAX_QUEUE_PAIRS 0x80
337 /* Declare IOMMU related bits for older kernels */
338 #ifndef VIRTIO_F_IOMMU_PLATFORM
340 #define VIRTIO_F_IOMMU_PLATFORM 33
342 struct vhost_iotlb_msg {
346 #define VHOST_ACCESS_RO 0x1
347 #define VHOST_ACCESS_WO 0x2
348 #define VHOST_ACCESS_RW 0x3
350 #define VHOST_IOTLB_MISS 1
351 #define VHOST_IOTLB_UPDATE 2
352 #define VHOST_IOTLB_INVALIDATE 3
353 #define VHOST_IOTLB_ACCESS_FAIL 4
357 #define VHOST_IOTLB_MSG 0x1
362 struct vhost_iotlb_msg iotlb;
369 * Define virtio 1.0 for older kernels
371 #ifndef VIRTIO_F_VERSION_1
372 #define VIRTIO_F_VERSION_1 32
375 /* Declare packed ring related bits for older kernels */
376 #ifndef VIRTIO_F_RING_PACKED
378 #define VIRTIO_F_RING_PACKED 34
380 struct vring_packed_desc {
387 struct vring_packed_desc_event {
394 * Declare below packed ring defines unconditionally
395 * as Kernel header might use different names.
397 #define VRING_DESC_F_AVAIL (1ULL << 7)
398 #define VRING_DESC_F_USED (1ULL << 15)
400 #define VRING_EVENT_F_ENABLE 0x0
401 #define VRING_EVENT_F_DISABLE 0x1
402 #define VRING_EVENT_F_DESC 0x2
405 * Available and used descs are in same order
407 #ifndef VIRTIO_F_IN_ORDER
408 #define VIRTIO_F_IN_ORDER 35
411 /* Features supported by this builtin vhost-user net driver. */
412 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
413 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
414 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
415 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
416 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
417 (1ULL << VIRTIO_NET_F_MQ) | \
418 (1ULL << VIRTIO_F_VERSION_1) | \
419 (1ULL << VHOST_F_LOG_ALL) | \
420 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
421 (1ULL << VIRTIO_NET_F_GSO) | \
422 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
423 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
424 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
425 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
426 (1ULL << VIRTIO_NET_F_CSUM) | \
427 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
428 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
429 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
430 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
431 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
432 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
433 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
434 (1ULL << VIRTIO_NET_F_MTU) | \
435 (1ULL << VIRTIO_F_IN_ORDER) | \
436 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
437 (1ULL << VIRTIO_F_RING_PACKED))
441 uint64_t guest_phys_addr;
443 uint64_t host_user_addr;
447 struct inflight_mem_info {
454 * Device structure contains all configuration information relating
458 /* Frontend (QEMU) memory and memory region information */
459 struct rte_vhost_memory *mem;
461 uint64_t protocol_features;
465 /* to tell if we need broadcast rarp packet */
466 int16_t broadcast_rarp;
472 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
473 struct inflight_mem_info *inflight_info;
474 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
475 char ifname[IF_NAME_SZ];
479 struct rte_ether_addr mac;
483 struct rte_vhost_device_ops const *notify_ops;
485 uint32_t nr_guest_pages;
486 uint32_t max_guest_pages;
487 struct guest_page *guest_pages;
490 rte_spinlock_t slave_req_lock;
493 int postcopy_listening;
495 struct rte_vdpa_device *vdpa_dev;
497 /* context data for the external message handlers */
499 /* pre and post vhost user message handlers for the device */
500 struct rte_vhost_user_extern_ops extern_ops;
501 } __rte_cache_aligned;
503 static __rte_always_inline bool
504 vq_is_packed(struct virtio_net *dev)
506 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
510 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
512 uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
514 return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
515 wrap_counter != !!(flags & VRING_DESC_F_USED);
519 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
521 vq->last_used_idx += num;
522 if (vq->last_used_idx >= vq->size) {
523 vq->used_wrap_counter ^= 1;
524 vq->last_used_idx -= vq->size;
529 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
531 vq->last_avail_idx += num;
532 if (vq->last_avail_idx >= vq->size) {
533 vq->avail_wrap_counter ^= 1;
534 vq->last_avail_idx -= vq->size;
538 void __vhost_log_cache_write(struct virtio_net *dev,
539 struct vhost_virtqueue *vq,
540 uint64_t addr, uint64_t len);
541 void __vhost_log_cache_write_iova(struct virtio_net *dev,
542 struct vhost_virtqueue *vq,
543 uint64_t iova, uint64_t len);
544 void __vhost_log_cache_sync(struct virtio_net *dev,
545 struct vhost_virtqueue *vq);
546 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
547 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
548 uint64_t iova, uint64_t len);
550 static __rte_always_inline void
551 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
553 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
554 __vhost_log_write(dev, addr, len);
557 static __rte_always_inline void
558 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
560 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
561 __vhost_log_cache_sync(dev, vq);
564 static __rte_always_inline void
565 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
566 uint64_t addr, uint64_t len)
568 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
569 __vhost_log_cache_write(dev, vq, addr, len);
572 static __rte_always_inline void
573 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
574 uint64_t offset, uint64_t len)
576 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
577 if (unlikely(vq->log_guest_addr == 0))
579 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
584 static __rte_always_inline void
585 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
586 uint64_t offset, uint64_t len)
588 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
589 if (unlikely(vq->log_guest_addr == 0))
591 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
595 static __rte_always_inline void
596 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
597 uint64_t iova, uint64_t len)
599 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
602 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
603 __vhost_log_cache_write_iova(dev, vq, iova, len);
605 __vhost_log_cache_write(dev, vq, iova, len);
608 static __rte_always_inline void
609 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
610 uint64_t iova, uint64_t len)
612 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
615 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
616 __vhost_log_write_iova(dev, vq, iova, len);
618 __vhost_log_write(dev, iova, len);
621 extern int vhost_config_log_level;
622 extern int vhost_data_log_level;
624 #define VHOST_LOG_CONFIG(level, fmt, args...) \
625 rte_log(RTE_LOG_ ## level, vhost_config_log_level, \
626 "VHOST_CONFIG: " fmt, ##args)
628 #define VHOST_LOG_DATA(level, fmt, args...) \
629 (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ? \
630 rte_log(RTE_LOG_ ## level, vhost_data_log_level, \
631 "VHOST_DATA : " fmt, ##args) : \
634 #ifdef RTE_LIBRTE_VHOST_DEBUG
635 #define VHOST_MAX_PRINT_BUFF 6072
636 #define PRINT_PACKET(device, addr, size, header) do { \
637 char *pkt_addr = (char *)(addr); \
638 unsigned int index; \
639 char packet[VHOST_MAX_PRINT_BUFF]; \
642 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
644 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
645 for (index = 0; index < (size); index++) { \
646 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
647 "%02hhx ", pkt_addr[index]); \
649 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
651 VHOST_LOG_DATA(DEBUG, "%s", packet); \
654 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
657 extern struct virtio_net *vhost_devices[RTE_MAX_VHOST_DEVICE];
659 #define VHOST_BINARY_SEARCH_THRESH 256
661 static __rte_always_inline int guest_page_addrcmp(const void *p1,
664 const struct guest_page *page1 = (const struct guest_page *)p1;
665 const struct guest_page *page2 = (const struct guest_page *)p2;
667 if (page1->guest_phys_addr > page2->guest_phys_addr)
669 if (page1->guest_phys_addr < page2->guest_phys_addr)
675 static __rte_always_inline int guest_page_rangecmp(const void *p1, const void *p2)
677 const struct guest_page *page1 = (const struct guest_page *)p1;
678 const struct guest_page *page2 = (const struct guest_page *)p2;
680 if (page1->guest_phys_addr >= page2->guest_phys_addr) {
681 if (page1->guest_phys_addr < page2->guest_phys_addr + page2->size)
689 static __rte_always_inline rte_iova_t
690 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
691 uint64_t gpa_size, uint64_t *hpa_size)
694 struct guest_page *page;
695 struct guest_page key;
697 *hpa_size = gpa_size;
698 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
699 key.guest_phys_addr = gpa;
700 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
701 sizeof(struct guest_page), guest_page_rangecmp);
703 if (gpa + gpa_size <=
704 page->guest_phys_addr + page->size) {
705 return gpa - page->guest_phys_addr +
707 } else if (gpa < page->guest_phys_addr +
709 *hpa_size = page->guest_phys_addr +
711 return gpa - page->guest_phys_addr +
716 for (i = 0; i < dev->nr_guest_pages; i++) {
717 page = &dev->guest_pages[i];
719 if (gpa >= page->guest_phys_addr) {
720 if (gpa + gpa_size <=
721 page->guest_phys_addr + page->size) {
722 return gpa - page->guest_phys_addr +
724 } else if (gpa < page->guest_phys_addr +
726 *hpa_size = page->guest_phys_addr +
728 return gpa - page->guest_phys_addr +
739 /* Convert guest physical address to host physical address */
740 static __rte_always_inline rte_iova_t
741 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
746 hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
747 return hpa_size == size ? hpa : 0;
750 static __rte_always_inline uint64_t
751 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
753 struct rte_vhost_mem_region *r;
756 if (unlikely(!dev || !dev->mem))
759 for (i = 0; i < dev->mem->nregions; i++) {
760 r = &dev->mem->regions[i];
762 if (vva >= r->host_user_addr &&
763 vva + len < r->host_user_addr + r->size) {
764 return r->guest_phys_addr + vva - r->host_user_addr;
770 static __rte_always_inline struct virtio_net *
773 struct virtio_net *dev = vhost_devices[vid];
775 if (unlikely(!dev)) {
776 VHOST_LOG_CONFIG(ERR,
777 "(%d) device not found.\n", vid);
783 int vhost_new_device(void);
784 void cleanup_device(struct virtio_net *dev, int destroy);
785 void reset_device(struct virtio_net *dev);
786 void vhost_destroy_device(int);
787 void vhost_destroy_device_notify(struct virtio_net *dev);
789 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
790 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
791 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
793 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
795 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
797 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
798 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags, bool stats_enabled);
799 void vhost_enable_extbuf(int vid);
800 void vhost_enable_linearbuf(int vid);
801 int vhost_enable_guest_notification(struct virtio_net *dev,
802 struct vhost_virtqueue *vq, int enable);
804 struct rte_vhost_device_ops const *vhost_driver_callback_get(const char *path);
807 * Backend-specific cleanup.
809 * TODO: fix it; we have one backend now
811 void vhost_backend_cleanup(struct virtio_net *dev);
813 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
814 uint64_t iova, uint64_t *len, uint8_t perm);
815 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
816 struct vhost_virtqueue *vq,
817 uint64_t desc_addr, uint64_t desc_len);
818 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
819 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
821 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
823 static __rte_always_inline uint64_t
824 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
825 uint64_t iova, uint64_t *len, uint8_t perm)
827 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
828 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
830 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
833 #define vhost_avail_event(vr) \
834 (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
835 #define vhost_used_event(vr) \
836 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
839 * The following is used with VIRTIO_RING_F_EVENT_IDX.
840 * Assuming a given event_idx value from the other size, if we have
841 * just incremented index from old to new_idx, should we trigger an
844 static __rte_always_inline int
845 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
847 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
850 static __rte_always_inline void
851 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
853 /* Flush used->idx update before we read avail->flags. */
854 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
856 /* Don't kick guest if we don't reach index specified by guest. */
857 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
858 uint16_t old = vq->signalled_used;
859 uint16_t new = vq->last_used_idx;
860 bool signalled_used_valid = vq->signalled_used_valid;
862 vq->signalled_used = new;
863 vq->signalled_used_valid = true;
865 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
867 vhost_used_event(vq),
870 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
871 (vq->callfd >= 0)) ||
872 unlikely(!signalled_used_valid)) {
873 eventfd_write(vq->callfd, (eventfd_t) 1);
874 if (dev->notify_ops->guest_notified)
875 dev->notify_ops->guest_notified(dev->vid);
878 /* Kick the guest if necessary. */
879 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
880 && (vq->callfd >= 0)) {
881 eventfd_write(vq->callfd, (eventfd_t)1);
882 if (dev->notify_ops->guest_notified)
883 dev->notify_ops->guest_notified(dev->vid);
888 static __rte_always_inline void
889 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
891 uint16_t old, new, off, off_wrap;
892 bool signalled_used_valid, kick = false;
894 /* Flush used desc update. */
895 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
897 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
898 if (vq->driver_event->flags !=
899 VRING_EVENT_F_DISABLE)
904 old = vq->signalled_used;
905 new = vq->last_used_idx;
906 vq->signalled_used = new;
907 signalled_used_valid = vq->signalled_used_valid;
908 vq->signalled_used_valid = true;
910 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
911 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
916 if (unlikely(!signalled_used_valid)) {
921 rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
923 off_wrap = vq->driver_event->off_wrap;
924 off = off_wrap & ~(1 << 15);
929 if (vq->used_wrap_counter != off_wrap >> 15)
932 if (vhost_need_event(off, new, old))
936 eventfd_write(vq->callfd, (eventfd_t)1);
937 if (dev->notify_ops->guest_notified)
938 dev->notify_ops->guest_notified(dev->vid);
942 static __rte_always_inline void
943 free_ind_table(void *idesc)
948 static __rte_always_inline void
949 restore_mbuf(struct rte_mbuf *m)
951 uint32_t mbuf_size, priv_size;
954 priv_size = rte_pktmbuf_priv_size(m->pool);
955 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
956 /* start of buffer is after mbuf structure and priv data */
958 m->buf_addr = (char *)m + mbuf_size;
959 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
964 static __rte_always_inline bool
965 mbuf_is_consumed(struct rte_mbuf *m)
968 if (rte_mbuf_refcnt_read(m) > 1)
975 #endif /* _VHOST_NET_CDEV_H_ */