1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
5 #ifndef _VHOST_NET_CDEV_H_
6 #define _VHOST_NET_CDEV_H_
10 #include <sys/queue.h>
12 #include <linux/vhost.h>
13 #include <linux/virtio_net.h>
14 #include <sys/socket.h>
18 #include <rte_ether.h>
19 #include <rte_rwlock.h>
21 #include "rte_vhost.h"
23 /* Used to indicate that the device is running on a data core */
24 #define VIRTIO_DEV_RUNNING 1
25 /* Used to indicate that the device is ready to operate */
26 #define VIRTIO_DEV_READY 2
28 /* Backend value set by guest. */
29 #define VIRTIO_DEV_STOPPED -1
31 #define BUF_VECTOR_MAX 256
34 * Structure contains buffer address, length and descriptor index
35 * from vring to do scatter RX.
44 * A structure to hold some fields needed in zero copy code path,
45 * mainly for associating an mbuf with the right desc_idx.
48 struct rte_mbuf *mbuf;
52 TAILQ_ENTRY(zcopy_mbuf) next;
54 TAILQ_HEAD(zcopy_mbuf_list, zcopy_mbuf);
57 * Structure contains the info for each batched memory copy.
59 struct batch_copy_elem {
67 * Structure contains variables relevant to RX/TX virtqueues.
69 struct vhost_virtqueue {
70 struct vring_desc *desc;
71 struct vring_avail *avail;
72 struct vring_used *used;
75 uint16_t last_avail_idx;
76 uint16_t last_used_idx;
77 /* Last used index we notify to front end. */
78 uint16_t signalled_used;
79 #define VIRTIO_INVALID_EVENTFD (-1)
80 #define VIRTIO_UNINITIALIZED_EVENTFD (-2)
82 /* Backend value to determine if device should started/stopped */
86 rte_spinlock_t access_lock;
88 /* Used to notify the guest (trigger interrupt) */
90 /* Currently unused as polling mode is enabled */
93 /* Physical address of used ring, for logging */
94 uint64_t log_guest_addr;
98 uint16_t last_zmbuf_idx;
99 struct zcopy_mbuf *zmbufs;
100 struct zcopy_mbuf_list zmbuf_list;
102 struct vring_used_elem *shadow_used_ring;
103 uint16_t shadow_used_idx;
104 struct vhost_vring_addr ring_addrs;
106 struct batch_copy_elem *batch_copy_elems;
107 uint16_t batch_copy_nb_elems;
109 rte_rwlock_t iotlb_lock;
110 rte_rwlock_t iotlb_pending_lock;
111 struct rte_mempool *iotlb_pool;
112 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_list;
114 TAILQ_HEAD(, vhost_iotlb_entry) iotlb_pending_list;
115 } __rte_cache_aligned;
117 /* Old kernels have no such macros defined */
118 #ifndef VIRTIO_NET_F_GUEST_ANNOUNCE
119 #define VIRTIO_NET_F_GUEST_ANNOUNCE 21
122 #ifndef VIRTIO_NET_F_MQ
123 #define VIRTIO_NET_F_MQ 22
126 #define VHOST_MAX_VRING 0x100
127 #define VHOST_MAX_QUEUE_PAIRS 0x80
129 #ifndef VIRTIO_NET_F_MTU
130 #define VIRTIO_NET_F_MTU 3
133 #ifndef VIRTIO_F_ANY_LAYOUT
134 #define VIRTIO_F_ANY_LAYOUT 27
137 /* Declare IOMMU related bits for older kernels */
138 #ifndef VIRTIO_F_IOMMU_PLATFORM
140 #define VIRTIO_F_IOMMU_PLATFORM 33
142 struct vhost_iotlb_msg {
146 #define VHOST_ACCESS_RO 0x1
147 #define VHOST_ACCESS_WO 0x2
148 #define VHOST_ACCESS_RW 0x3
150 #define VHOST_IOTLB_MISS 1
151 #define VHOST_IOTLB_UPDATE 2
152 #define VHOST_IOTLB_INVALIDATE 3
153 #define VHOST_IOTLB_ACCESS_FAIL 4
157 #define VHOST_IOTLB_MSG 0x1
162 struct vhost_iotlb_msg iotlb;
169 * Define virtio 1.0 for older kernels
171 #ifndef VIRTIO_F_VERSION_1
172 #define VIRTIO_F_VERSION_1 32
175 #define VHOST_USER_F_PROTOCOL_FEATURES 30
177 /* Features supported by this builtin vhost-user net driver. */
178 #define VIRTIO_NET_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
179 (1ULL << VIRTIO_F_ANY_LAYOUT) | \
180 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
181 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
182 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
183 (1ULL << VIRTIO_NET_F_MQ) | \
184 (1ULL << VIRTIO_F_VERSION_1) | \
185 (1ULL << VHOST_F_LOG_ALL) | \
186 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
187 (1ULL << VIRTIO_NET_F_GSO) | \
188 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
189 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
190 (1ULL << VIRTIO_NET_F_HOST_UFO) | \
191 (1ULL << VIRTIO_NET_F_HOST_ECN) | \
192 (1ULL << VIRTIO_NET_F_CSUM) | \
193 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
194 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
195 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
196 (1ULL << VIRTIO_NET_F_GUEST_UFO) | \
197 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
198 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \
199 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \
200 (1ULL << VIRTIO_NET_F_MTU) | \
201 (1ULL << VIRTIO_F_IOMMU_PLATFORM))
205 uint64_t guest_phys_addr;
206 uint64_t host_phys_addr;
211 * Device structure contains all configuration information relating
215 /* Frontend (QEMU) memory and memory region information */
216 struct rte_vhost_memory *mem;
218 uint64_t protocol_features;
222 /* to tell if we need broadcast rarp packet */
223 rte_atomic16_t broadcast_rarp;
225 int dequeue_zero_copy;
226 struct vhost_virtqueue *virtqueue[VHOST_MAX_QUEUE_PAIRS * 2];
227 #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ)
228 char ifname[IF_NAME_SZ];
232 struct ether_addr mac;
235 struct vhost_device_ops const *notify_ops;
237 uint32_t nr_guest_pages;
238 uint32_t max_guest_pages;
239 struct guest_page *guest_pages;
242 } __rte_cache_aligned;
245 #define VHOST_LOG_PAGE 4096
248 * Atomically set a bit in memory.
250 static __rte_always_inline void
251 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
253 __sync_fetch_and_or_8(addr, (1U << nr));
256 static __rte_always_inline void
257 vhost_log_page(uint8_t *log_base, uint64_t page)
259 vhost_set_bit(page % 8, &log_base[page / 8]);
262 static __rte_always_inline void
263 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
267 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
268 !dev->log_base || !len))
271 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
274 /* To make sure guest memory updates are committed before logging */
277 page = addr / VHOST_LOG_PAGE;
278 while (page * VHOST_LOG_PAGE < addr + len) {
279 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
284 static __rte_always_inline void
285 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
286 uint64_t offset, uint64_t len)
288 vhost_log_write(dev, vq->log_guest_addr + offset, len);
291 /* Macros for printing using RTE_LOG */
292 #define RTE_LOGTYPE_VHOST_CONFIG RTE_LOGTYPE_USER1
293 #define RTE_LOGTYPE_VHOST_DATA RTE_LOGTYPE_USER1
295 #ifdef RTE_LIBRTE_VHOST_DEBUG
296 #define VHOST_MAX_PRINT_BUFF 6072
297 #define LOG_LEVEL RTE_LOG_DEBUG
298 #define LOG_DEBUG(log_type, fmt, args...) RTE_LOG(DEBUG, log_type, fmt, ##args)
299 #define PRINT_PACKET(device, addr, size, header) do { \
300 char *pkt_addr = (char *)(addr); \
301 unsigned int index; \
302 char packet[VHOST_MAX_PRINT_BUFF]; \
305 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Header size %d: ", (device->vid), (size)); \
307 snprintf(packet, VHOST_MAX_PRINT_BUFF, "(%d) Packet size %d: ", (device->vid), (size)); \
308 for (index = 0; index < (size); index++) { \
309 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), \
310 "%02hhx ", pkt_addr[index]); \
312 snprintf(packet + strnlen(packet, VHOST_MAX_PRINT_BUFF), VHOST_MAX_PRINT_BUFF - strnlen(packet, VHOST_MAX_PRINT_BUFF), "\n"); \
314 LOG_DEBUG(VHOST_DATA, "%s", packet); \
317 #define LOG_LEVEL RTE_LOG_INFO
318 #define LOG_DEBUG(log_type, fmt, args...) do {} while (0)
319 #define PRINT_PACKET(device, addr, size, header) do {} while (0)
322 extern uint64_t VHOST_FEATURES;
323 #define MAX_VHOST_DEVICE 1024
324 extern struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
326 /* Convert guest physical address to host physical address */
327 static __rte_always_inline rte_iova_t
328 gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
331 struct guest_page *page;
333 for (i = 0; i < dev->nr_guest_pages; i++) {
334 page = &dev->guest_pages[i];
336 if (gpa >= page->guest_phys_addr &&
337 gpa + size < page->guest_phys_addr + page->size) {
338 return gpa - page->guest_phys_addr +
339 page->host_phys_addr;
346 struct virtio_net *get_device(int vid);
348 int vhost_new_device(void);
349 void cleanup_device(struct virtio_net *dev, int destroy);
350 void reset_device(struct virtio_net *dev);
351 void vhost_destroy_device(int);
353 void cleanup_vq(struct vhost_virtqueue *vq, int destroy);
354 void free_vq(struct vhost_virtqueue *vq);
356 int alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx);
358 void vhost_set_ifname(int, const char *if_name, unsigned int if_len);
359 void vhost_enable_dequeue_zero_copy(int vid);
361 struct vhost_device_ops const *vhost_driver_callback_get(const char *path);
364 * Backend-specific cleanup.
366 * TODO: fix it; we have one backend now
368 void vhost_backend_cleanup(struct virtio_net *dev);
370 uint64_t __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
371 uint64_t iova, uint64_t size, uint8_t perm);
372 int vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq);
373 void vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq);
375 static __rte_always_inline uint64_t
376 vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
377 uint64_t iova, uint64_t size, uint8_t perm)
379 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
380 return rte_vhost_gpa_to_vva(dev->mem, iova);
382 return __vhost_iova_to_vva(dev, vq, iova, size, perm);
385 #define vhost_used_event(vr) \
386 (*(volatile uint16_t*)&(vr)->avail->ring[(vr)->size])
389 * The following is used with VIRTIO_RING_F_EVENT_IDX.
390 * Assuming a given event_idx value from the other size, if we have
391 * just incremented index from old to new_idx, should we trigger an
394 static __rte_always_inline int
395 vhost_need_event(uint16_t event_idx, uint16_t new_idx, uint16_t old)
397 return (uint16_t)(new_idx - event_idx - 1) < (uint16_t)(new_idx - old);
400 static __rte_always_inline void
401 vhost_vring_call(struct virtio_net *dev, struct vhost_virtqueue *vq)
403 /* Flush used->idx update before we read avail->flags. */
406 /* Don't kick guest if we don't reach index specified by guest. */
407 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
408 uint16_t old = vq->signalled_used;
409 uint16_t new = vq->last_used_idx;
411 LOG_DEBUG(VHOST_DATA, "%s: used_event_idx=%d, old=%d, new=%d\n",
413 vhost_used_event(vq),
415 if (vhost_need_event(vhost_used_event(vq), new, old)
416 && (vq->callfd >= 0)) {
417 vq->signalled_used = vq->last_used_idx;
418 eventfd_write(vq->callfd, (eventfd_t) 1);
421 /* Kick the guest if necessary. */
422 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
423 && (vq->callfd >= 0))
424 eventfd_write(vq->callfd, (eventfd_t)1);
428 #endif /* _VHOST_NET_CDEV_H_ */