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>
21 #include <rte_malloc.h>
23 #include "rte_vhost.h"
25 #include "rte_vdpa_dev.h"
27 #include "rte_vhost_async.h"
29 /* Used to indicate that the device is running on a data core */
30 #define VIRTIO_DEV_RUNNING ((uint32_t)1 << 0)
31 /* Used to indicate that the device is ready to operate */
32 #define VIRTIO_DEV_READY ((uint32_t)1 << 1)
33 /* Used to indicate that the built-in vhost net device backend is enabled */
34 #define VIRTIO_DEV_BUILTIN_VIRTIO_NET ((uint32_t)1 << 2)
35 /* Used to indicate that the device has its own data path and configured */
36 #define VIRTIO_DEV_VDPA_CONFIGURED ((uint32_t)1 << 3)
37 /* Used to indicate that the feature negotiation failed */
38 #define VIRTIO_DEV_FEATURES_FAILED ((uint32_t)1 << 4)
39 /* Used to indicate that the virtio_net tx code should fill TX ol_flags */
40 #define VIRTIO_DEV_LEGACY_OL_FLAGS ((uint32_t)1 << 5)
42 /* Backend value set by guest. */
43 #define VIRTIO_DEV_STOPPED -1
45 #define BUF_VECTOR_MAX 256
47 #define VHOST_LOG_CACHE_NR 32
49 #define MAX_PKT_BURST 32
51 #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2)
52 #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 4)
54 #define PACKED_DESC_ENQUEUE_USED_FLAG(w) \
55 ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED | VRING_DESC_F_WRITE) : \
57 #define PACKED_DESC_DEQUEUE_USED_FLAG(w) \
58 ((w) ? (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) : 0x0)
59 #define PACKED_DESC_SINGLE_DEQUEUE_FLAG (VRING_DESC_F_NEXT | \
60 VRING_DESC_F_INDIRECT)
62 #define PACKED_BATCH_SIZE (RTE_CACHE_LINE_SIZE / \
63 sizeof(struct vring_packed_desc))
64 #define PACKED_BATCH_MASK (PACKED_BATCH_SIZE - 1)
66 #ifdef VHOST_GCC_UNROLL_PRAGMA
67 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("GCC unroll 4") \
68 for (iter = val; iter < size; iter++)
71 #ifdef VHOST_CLANG_UNROLL_PRAGMA
72 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll 4") \
73 for (iter = val; iter < size; iter++)
76 #ifdef VHOST_ICC_UNROLL_PRAGMA
77 #define vhost_for_each_try_unroll(iter, val, size) _Pragma("unroll (4)") \
78 for (iter = val; iter < size; iter++)
81 #ifndef vhost_for_each_try_unroll
82 #define vhost_for_each_try_unroll(iter, val, num) \
83 for (iter = val; iter < num; iter++)
87 * Structure contains buffer address, length and descriptor index
88 * from vring to do scatter RX.
98 * Structure contains the info for each batched memory copy.
100 struct batch_copy_elem {
108 * Structure that contains the info for batched dirty logging.
110 struct log_cache_entry {
115 struct vring_used_elem_packed {
123 * Structure contains variables relevant to RX/TX virtqueues.
125 struct vhost_virtqueue {
127 struct vring_desc *desc;
128 struct vring_packed_desc *desc_packed;
131 struct vring_avail *avail;
132 struct vring_packed_desc_event *driver_event;
135 struct vring_used *used;
136 struct vring_packed_desc_event *device_event;
140 uint16_t last_avail_idx;
141 uint16_t last_used_idx;
142 /* Last used index we notify to front end. */
143 uint16_t signalled_used;
144 bool signalled_used_valid;
145 #define VIRTIO_INVALID_EVENTFD (-1)
146 #define VIRTIO_UNINITIALIZED_EVENTFD (-2)
152 rte_spinlock_t access_lock;
156 struct vring_used_elem *shadow_used_split;
157 struct vring_used_elem_packed *shadow_used_packed;
159 uint16_t shadow_used_idx;
160 /* Record packed ring enqueue latest desc cache aligned index */
161 uint16_t shadow_aligned_idx;
162 /* Record packed ring first dequeue desc index */
163 uint16_t shadow_last_used_idx;
165 uint16_t batch_copy_nb_elems;
166 struct batch_copy_elem *batch_copy_elems;
168 bool used_wrap_counter;
169 bool avail_wrap_counter;
171 /* Physical address of used ring, for logging */
172 uint16_t log_cache_nb_elem;
173 uint64_t log_guest_addr;
174 struct log_cache_entry *log_cache;
176 rte_rwlock_t iotlb_lock;
177 rte_rwlock_t iotlb_pending_lock;
178 struct rte_mempool *iotlb_pool;
179 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
180 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
183 /* Used to notify the guest (trigger interrupt) */
185 /* Currently unused as polling mode is enabled */
188 /* inflight share memory info */
190 struct rte_vhost_inflight_info_split *inflight_split;
191 struct rte_vhost_inflight_info_packed *inflight_packed;
193 struct rte_vhost_resubmit_info *resubmit_inflight;
194 uint64_t global_counter;
196 /* operation callbacks for async dma */
197 struct rte_vhost_async_channel_ops async_ops;
199 struct rte_vhost_iov_iter *it_pool;
200 struct iovec *vec_pool;
202 /* async data transfer status */
203 struct async_inflight_info *async_pkts_info;
204 uint16_t async_pkts_idx;
205 uint16_t async_pkts_inflight_n;
206 uint16_t async_last_pkts_n;
208 struct vring_used_elem *async_descs_split;
209 struct vring_used_elem_packed *async_buffers_packed;
212 uint16_t async_desc_idx_split;
213 uint16_t async_buffer_idx_packed;
216 uint16_t last_async_desc_idx_split;
217 uint16_t last_async_buffer_idx_packed;
220 /* vq async features */
221 bool async_registered;
224 #define VIRTIO_UNINITIALIZED_NOTIF (-1)
226 struct vhost_vring_addr ring_addrs;
227 } __rte_cache_aligned;
229 /* Virtio device status as per Virtio specification */
230 #define VIRTIO_DEVICE_STATUS_RESET 0x00
231 #define VIRTIO_DEVICE_STATUS_ACK 0x01
232 #define VIRTIO_DEVICE_STATUS_DRIVER 0x02
233 #define VIRTIO_DEVICE_STATUS_DRIVER_OK 0x04
234 #define VIRTIO_DEVICE_STATUS_FEATURES_OK 0x08
235 #define VIRTIO_DEVICE_STATUS_DEV_NEED_RESET 0x40
236 #define VIRTIO_DEVICE_STATUS_FAILED 0x80
238 #define VHOST_MAX_VRING 0x100
239 #define VHOST_MAX_QUEUE_PAIRS 0x80
241 /* Declare IOMMU related bits for older kernels */
242 #ifndef VIRTIO_F_IOMMU_PLATFORM
244 #define VIRTIO_F_IOMMU_PLATFORM 33
246 struct vhost_iotlb_msg {
250 #define VHOST_ACCESS_RO 0x1
251 #define VHOST_ACCESS_WO 0x2
252 #define VHOST_ACCESS_RW 0x3
254 #define VHOST_IOTLB_MISS 1
255 #define VHOST_IOTLB_UPDATE 2
256 #define VHOST_IOTLB_INVALIDATE 3
257 #define VHOST_IOTLB_ACCESS_FAIL 4
261 #define VHOST_IOTLB_MSG 0x1
266 struct vhost_iotlb_msg iotlb;
273 * Define virtio 1.0 for older kernels
275 #ifndef VIRTIO_F_VERSION_1
276 #define VIRTIO_F_VERSION_1 32
279 /* Declare packed ring related bits for older kernels */
280 #ifndef VIRTIO_F_RING_PACKED
282 #define VIRTIO_F_RING_PACKED 34
284 struct vring_packed_desc {
291 struct vring_packed_desc_event {
298 * Declare below packed ring defines unconditionally
299 * as Kernel header might use different names.
301 #define VRING_DESC_F_AVAIL (1ULL << 7)
302 #define VRING_DESC_F_USED (1ULL << 15)
304 #define VRING_EVENT_F_ENABLE 0x0
305 #define VRING_EVENT_F_DISABLE 0x1
306 #define VRING_EVENT_F_DESC 0x2
309 * Available and used descs are in same order
311 #ifndef VIRTIO_F_IN_ORDER
312 #define VIRTIO_F_IN_ORDER 35
315 /* Features supported by this builtin vhost-user net driver. */
316 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
317 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
318 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
319 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
320 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
321 (1ULL << VIRTIO_NET_F_MQ) | \
322 (1ULL << VIRTIO_F_VERSION_1) | \
323 (1ULL << VHOST_F_LOG_ALL) | \
324 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
325 (1ULL << VIRTIO_NET_F_GSO) | \
326 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
327 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
328 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
329 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
330 (1ULL << VIRTIO_NET_F_CSUM) | \
331 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
332 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
333 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
334 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
335 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
336 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
337 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
338 (1ULL << VIRTIO_NET_F_MTU) | \
339 (1ULL << VIRTIO_F_IN_ORDER) | \
340 (1ULL << VIRTIO_F_IOMMU_PLATFORM) | \
341 (1ULL << VIRTIO_F_RING_PACKED))
345 uint64_t guest_phys_addr;
346 uint64_t host_phys_addr;
350 struct inflight_mem_info {
357 * Device structure contains all configuration information relating
361 /* Frontend (QEMU) memory and memory region information */
362 struct rte_vhost_memory *mem;
364 uint64_t protocol_features;
368 /* to tell if we need broadcast rarp packet */
369 int16_t broadcast_rarp;
374 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
375 struct inflight_mem_info *inflight_info;
376 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
377 char ifname[IF_NAME_SZ];
381 struct rte_ether_addr mac;
385 struct vhost_device_ops const *notify_ops;
387 uint32_t nr_guest_pages;
388 uint32_t max_guest_pages;
389 struct guest_page *guest_pages;
392 rte_spinlock_t slave_req_lock;
395 int postcopy_listening;
397 struct rte_vdpa_device *vdpa_dev;
399 /* context data for the external message handlers */
401 /* pre and post vhost user message handlers for the device */
402 struct rte_vhost_user_extern_ops extern_ops;
403 } __rte_cache_aligned;
405 static __rte_always_inline bool
406 vq_is_packed(struct virtio_net *dev)
408 return dev->features & (1ull << VIRTIO_F_RING_PACKED);
412 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
414 uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
416 return wrap_counter == !!(flags & VRING_DESC_F_AVAIL) &&
417 wrap_counter != !!(flags & VRING_DESC_F_USED);
421 vq_inc_last_used_packed(struct vhost_virtqueue *vq, uint16_t num)
423 vq->last_used_idx += num;
424 if (vq->last_used_idx >= vq->size) {
425 vq->used_wrap_counter ^= 1;
426 vq->last_used_idx -= vq->size;
431 vq_inc_last_avail_packed(struct vhost_virtqueue *vq, uint16_t num)
433 vq->last_avail_idx += num;
434 if (vq->last_avail_idx >= vq->size) {
435 vq->avail_wrap_counter ^= 1;
436 vq->last_avail_idx -= vq->size;
440 void __vhost_log_cache_write(struct virtio_net *dev,
441 struct vhost_virtqueue *vq,
442 uint64_t addr, uint64_t len);
443 void __vhost_log_cache_write_iova(struct virtio_net *dev,
444 struct vhost_virtqueue *vq,
445 uint64_t iova, uint64_t len);
446 void __vhost_log_cache_sync(struct virtio_net *dev,
447 struct vhost_virtqueue *vq);
448 void __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len);
449 void __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
450 uint64_t iova, uint64_t len);
452 static __rte_always_inline void
453 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
455 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
456 __vhost_log_write(dev, addr, len);
459 static __rte_always_inline void
460 vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
462 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
463 __vhost_log_cache_sync(dev, vq);
466 static __rte_always_inline void
467 vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
468 uint64_t addr, uint64_t len)
470 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL)))
471 __vhost_log_cache_write(dev, vq, addr, len);
474 static __rte_always_inline void
475 vhost_log_cache_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
476 uint64_t offset, uint64_t len)
478 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
479 if (unlikely(vq->log_guest_addr == 0))
481 __vhost_log_cache_write(dev, vq, vq->log_guest_addr + offset,
486 static __rte_always_inline void
487 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
488 uint64_t offset, uint64_t len)
490 if (unlikely(dev->features & (1ULL << VHOST_F_LOG_ALL))) {
491 if (unlikely(vq->log_guest_addr == 0))
493 __vhost_log_write(dev, vq->log_guest_addr + offset, len);
497 static __rte_always_inline void
498 vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
499 uint64_t iova, uint64_t len)
501 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
504 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
505 __vhost_log_cache_write_iova(dev, vq, iova, len);
507 __vhost_log_cache_write(dev, vq, iova, len);
510 static __rte_always_inline void
511 vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
512 uint64_t iova, uint64_t len)
514 if (likely(!(dev->features & (1ULL << VHOST_F_LOG_ALL))))
517 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
518 __vhost_log_write_iova(dev, vq, iova, len);
520 __vhost_log_write(dev, iova, len);
523 extern int vhost_config_log_level;
524 extern int vhost_data_log_level;
526 #define VHOST_LOG_CONFIG(level, fmt, args...) \
527 rte_log(RTE_LOG_ ## level, vhost_config_log_level, \
528 "VHOST_CONFIG: " fmt, ##args)
530 #define VHOST_LOG_DATA(level, fmt, args...) \
531 (void)((RTE_LOG_ ## level <= RTE_LOG_DP_LEVEL) ? \
532 rte_log(RTE_LOG_ ## level, vhost_data_log_level, \
533 "VHOST_DATA : " fmt, ##args) : \
536 #ifdef RTE_LIBRTE_VHOST_DEBUG
537 #define VHOST_MAX_PRINT_BUFF 6072
538 #define PRINT_PACKET(device, addr, size, header) do { \
539 char *pkt_addr = (char *)(addr); \
540 unsigned int index; \
541 char packet[VHOST_MAX_PRINT_BUFF]; \
544 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
546 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
547 for (index = 0; index < (size); index++) { \
548 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
549 "%02hhx ", pkt_addr[index]); \
551 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
553 VHOST_LOG_DATA(DEBUG, "%s", packet); \
556 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
559 #define MAX_VHOST_DEVICE 1024
560 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
562 #define VHOST_BINARY_SEARCH_THRESH 256
564 static __rte_always_inline int guest_page_addrcmp(const void *p1,
567 const struct guest_page *page1 = (const struct guest_page *)p1;
568 const struct guest_page *page2 = (const struct guest_page *)p2;
570 if (page1->guest_phys_addr > page2->guest_phys_addr)
572 if (page1->guest_phys_addr < page2->guest_phys_addr)
578 static __rte_always_inline rte_iova_t
579 gpa_to_first_hpa(struct virtio_net *dev, uint64_t gpa,
580 uint64_t gpa_size, uint64_t *hpa_size)
583 struct guest_page *page;
584 struct guest_page key;
586 *hpa_size = gpa_size;
587 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
588 key.guest_phys_addr = gpa & ~(dev->guest_pages[0].size - 1);
589 page = bsearch(&key, dev->guest_pages, dev->nr_guest_pages,
590 sizeof(struct guest_page), guest_page_addrcmp);
592 if (gpa + gpa_size <=
593 page->guest_phys_addr + page->size) {
594 return gpa - page->guest_phys_addr +
595 page->host_phys_addr;
596 } else if (gpa < page->guest_phys_addr +
598 *hpa_size = page->guest_phys_addr +
600 return gpa - page->guest_phys_addr +
601 page->host_phys_addr;
605 for (i = 0; i < dev->nr_guest_pages; i++) {
606 page = &dev->guest_pages[i];
608 if (gpa >= page->guest_phys_addr) {
609 if (gpa + gpa_size <=
610 page->guest_phys_addr + page->size) {
611 return gpa - page->guest_phys_addr +
612 page->host_phys_addr;
613 } else if (gpa < page->guest_phys_addr +
615 *hpa_size = page->guest_phys_addr +
617 return gpa - page->guest_phys_addr +
618 page->host_phys_addr;
628 /* Convert guest physical address to host physical address */
629 static __rte_always_inline rte_iova_t
630 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
635 hpa = gpa_to_first_hpa(dev, gpa, size, &hpa_size);
636 return hpa_size == size ? hpa : 0;
639 static __rte_always_inline uint64_t
640 hva_to_gpa(struct virtio_net *dev, uint64_t vva, uint64_t len)
642 struct rte_vhost_mem_region *r;
645 if (unlikely(!dev || !dev->mem))
648 for (i = 0; i < dev->mem->nregions; i++) {
649 r = &dev->mem->regions[i];
651 if (vva >= r->host_user_addr &&
652 vva + len < r->host_user_addr + r->size) {
653 return r->guest_phys_addr + vva - r->host_user_addr;
659 static __rte_always_inline struct virtio_net *
662 struct virtio_net *dev = vhost_devices[vid];
664 if (unlikely(!dev)) {
665 VHOST_LOG_CONFIG(ERR,
666 "(%d) device not found.\n", vid);
672 int vhost_new_device(void);
673 void cleanup_device(struct virtio_net *dev, int destroy);
674 void reset_device(struct virtio_net *dev);
675 void vhost_destroy_device(int);
676 void vhost_destroy_device_notify(struct virtio_net *dev);
678 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
679 void cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq);
680 void free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq);
682 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
684 void vhost_attach_vdpa_device(int vid, struct rte_vdpa_device *dev);
686 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
687 void vhost_setup_virtio_net(int vid, bool enable, bool legacy_ol_flags);
688 void vhost_enable_extbuf(int vid);
689 void vhost_enable_linearbuf(int vid);
690 int vhost_enable_guest_notification(struct virtio_net *dev,
691 struct vhost_virtqueue *vq, int enable);
693 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
696 * Backend-specific cleanup.
698 * TODO: fix it; we have one backend now
700 void vhost_backend_cleanup(struct virtio_net *dev);
702 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
703 uint64_t iova, uint64_t *len, uint8_t perm);
704 void *vhost_alloc_copy_ind_table(struct virtio_net *dev,
705 struct vhost_virtqueue *vq,
706 uint64_t desc_addr, uint64_t desc_len);
707 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
708 uint64_t translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
710 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
712 static __rte_always_inline uint64_t
713 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
714 uint64_t iova, uint64_t *len, uint8_t perm)
716 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
717 return rte_vhost_va_from_guest_pa(dev->mem, iova, len);
719 return __vhost_iova_to_vva(dev, vq, iova, len, perm);
722 #define vhost_avail_event(vr) \
723 (*(volatile uint16_t*)&(vr)->used->ring[(vr)->size])
724 #define vhost_used_event(vr) \
725 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
728 * The following is used with VIRTIO_RING_F_EVENT_IDX.
729 * Assuming a given event_idx value from the other size, if we have
730 * just incremented index from old to new_idx, should we trigger an
733 static __rte_always_inline int
734 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
736 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
739 static __rte_always_inline void
740 vhost_vring_call_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
742 /* Flush used->idx update before we read avail->flags. */
743 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
745 /* Don't kick guest if we don't reach index specified by guest. */
746 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
747 uint16_t old = vq->signalled_used;
748 uint16_t new = vq->last_used_idx;
749 bool signalled_used_valid = vq->signalled_used_valid;
751 vq->signalled_used = new;
752 vq->signalled_used_valid = true;
754 VHOST_LOG_DATA(DEBUG, "%s: used_event_idx=%d, old=%d, new=%d\n",
756 vhost_used_event(vq),
759 if ((vhost_need_event(vhost_used_event(vq), new, old) &&
760 (vq->callfd >= 0)) ||
761 unlikely(!signalled_used_valid)) {
762 eventfd_write(vq->callfd, (eventfd_t) 1);
763 if (dev->notify_ops->guest_notified)
764 dev->notify_ops->guest_notified(dev->vid);
767 /* Kick the guest if necessary. */
768 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
769 && (vq->callfd >= 0)) {
770 eventfd_write(vq->callfd, (eventfd_t)1);
771 if (dev->notify_ops->guest_notified)
772 dev->notify_ops->guest_notified(dev->vid);
777 static __rte_always_inline void
778 vhost_vring_call_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
780 uint16_t old, new, off, off_wrap;
781 bool signalled_used_valid, kick = false;
783 /* Flush used desc update. */
784 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
786 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
787 if (vq->driver_event->flags !=
788 VRING_EVENT_F_DISABLE)
793 old = vq->signalled_used;
794 new = vq->last_used_idx;
795 vq->signalled_used = new;
796 signalled_used_valid = vq->signalled_used_valid;
797 vq->signalled_used_valid = true;
799 if (vq->driver_event->flags != VRING_EVENT_F_DESC) {
800 if (vq->driver_event->flags != VRING_EVENT_F_DISABLE)
805 if (unlikely(!signalled_used_valid)) {
810 rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
812 off_wrap = vq->driver_event->off_wrap;
813 off = off_wrap & ~(1 << 15);
818 if (vq->used_wrap_counter != off_wrap >> 15)
821 if (vhost_need_event(off, new, old))
825 eventfd_write(vq->callfd, (eventfd_t)1);
826 if (dev->notify_ops->guest_notified)
827 dev->notify_ops->guest_notified(dev->vid);
831 static __rte_always_inline void
832 free_ind_table(void *idesc)
837 static __rte_always_inline void
838 restore_mbuf(struct rte_mbuf *m)
840 uint32_t mbuf_size, priv_size;
843 priv_size = rte_pktmbuf_priv_size(m->pool);
844 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
845 /* start of buffer is after mbuf structure and priv data */
847 m->buf_addr = (char *)m + mbuf_size;
848 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
853 static __rte_always_inline bool
854 mbuf_is_consumed(struct rte_mbuf *m)
857 if (rte_mbuf_refcnt_read(m) > 1)
865 #endif /* _VHOST_NET_CDEV_H_ */