1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
10 #include <rte_atomic.h>
11 #include <rte_memory.h>
12 #include <rte_mempool.h>
14 #include "virtio_pci.h"
15 #include "virtio_ring.h"
16 #include "virtio_logs.h"
17 #include "virtio_rxtx.h"
22 * Per virtio_config.h in Linux.
23 * For virtio_pci on SMP, we don't need to order with respect to MMIO
24 * accesses through relaxed memory I/O windows, so smp_mb() et al are
28 #define virtio_mb() rte_smp_mb()
29 #define virtio_rmb() rte_smp_rmb()
30 #define virtio_wmb() rte_smp_wmb()
32 #ifdef RTE_PMD_PACKET_PREFETCH
33 #define rte_packet_prefetch(p) rte_prefetch1(p)
35 #define rte_packet_prefetch(p) do {} while(0)
38 #define VIRTQUEUE_MAX_NAME_SZ 32
40 #ifdef RTE_VIRTIO_USER
42 * Return the physical address (or virtual address in case of
43 * virtio-user) of mbuf data buffer.
45 * The address is firstly casted to the word size (sizeof(uintptr_t))
46 * before casting it to uint64_t. This is to make it work with different
47 * combination of word size (64 bit and 32 bit) and virtio device
48 * (virtio-pci and virtio-user).
50 #define VIRTIO_MBUF_ADDR(mb, vq) \
51 ((uint64_t)(*(uintptr_t *)((uintptr_t)(mb) + (vq)->offset)))
53 #define VIRTIO_MBUF_ADDR(mb, vq) ((mb)->buf_iova)
57 * Return the physical address (or virtual address in case of
58 * virtio-user) of mbuf data buffer, taking care of mbuf data offset
60 #define VIRTIO_MBUF_DATA_DMA_ADDR(mb, vq) \
61 (VIRTIO_MBUF_ADDR(mb, vq) + (mb)->data_off)
63 #define VTNET_SQ_RQ_QUEUE_IDX 0
64 #define VTNET_SQ_TQ_QUEUE_IDX 1
65 #define VTNET_SQ_CQ_QUEUE_IDX 2
67 enum { VTNET_RQ = 0, VTNET_TQ = 1, VTNET_CQ = 2 };
69 * The maximum virtqueue size is 2^15. Use that value as the end of
70 * descriptor chain terminator since it will never be a valid index
71 * in the descriptor table. This is used to verify we are correctly
72 * handling vq_free_cnt.
74 #define VQ_RING_DESC_CHAIN_END 32768
77 * Control the RX mode, ie. promiscuous, allmulti, etc...
78 * All commands require an "out" sg entry containing a 1 byte
79 * state value, zero = disable, non-zero = enable. Commands
80 * 0 and 1 are supported with the VIRTIO_NET_F_CTRL_RX feature.
81 * Commands 2-5 are added with VIRTIO_NET_F_CTRL_RX_EXTRA.
83 #define VIRTIO_NET_CTRL_RX 0
84 #define VIRTIO_NET_CTRL_RX_PROMISC 0
85 #define VIRTIO_NET_CTRL_RX_ALLMULTI 1
86 #define VIRTIO_NET_CTRL_RX_ALLUNI 2
87 #define VIRTIO_NET_CTRL_RX_NOMULTI 3
88 #define VIRTIO_NET_CTRL_RX_NOUNI 4
89 #define VIRTIO_NET_CTRL_RX_NOBCAST 5
94 * The MAC filter table is managed by the hypervisor, the guest should
95 * assume the size is infinite. Filtering should be considered
96 * non-perfect, ie. based on hypervisor resources, the guest may
97 * received packets from sources not specified in the filter list.
99 * In addition to the class/cmd header, the TABLE_SET command requires
100 * two out scatterlists. Each contains a 4 byte count of entries followed
101 * by a concatenated byte stream of the ETH_ALEN MAC addresses. The
102 * first sg list contains unicast addresses, the second is for multicast.
103 * This functionality is present if the VIRTIO_NET_F_CTRL_RX feature
106 * The ADDR_SET command requests one out scatterlist, it contains a
107 * 6 bytes MAC address. This functionality is present if the
108 * VIRTIO_NET_F_CTRL_MAC_ADDR feature is available.
110 struct virtio_net_ctrl_mac {
112 uint8_t macs[][ETHER_ADDR_LEN];
113 } __attribute__((__packed__));
115 #define VIRTIO_NET_CTRL_MAC 1
116 #define VIRTIO_NET_CTRL_MAC_TABLE_SET 0
117 #define VIRTIO_NET_CTRL_MAC_ADDR_SET 1
120 * Control VLAN filtering
122 * The VLAN filter table is controlled via a simple ADD/DEL interface.
123 * VLAN IDs not added may be filtered by the hypervisor. Del is the
124 * opposite of add. Both commands expect an out entry containing a 2
125 * byte VLAN ID. VLAN filtering is available with the
126 * VIRTIO_NET_F_CTRL_VLAN feature bit.
128 #define VIRTIO_NET_CTRL_VLAN 2
129 #define VIRTIO_NET_CTRL_VLAN_ADD 0
130 #define VIRTIO_NET_CTRL_VLAN_DEL 1
133 * Control link announce acknowledgement
135 * The command VIRTIO_NET_CTRL_ANNOUNCE_ACK is used to indicate that
136 * driver has recevied the notification; device would clear the
137 * VIRTIO_NET_S_ANNOUNCE bit in the status field after it receives
140 #define VIRTIO_NET_CTRL_ANNOUNCE 3
141 #define VIRTIO_NET_CTRL_ANNOUNCE_ACK 0
143 struct virtio_net_ctrl_hdr {
146 } __attribute__((packed));
148 typedef uint8_t virtio_net_ctrl_ack;
150 #define VIRTIO_NET_OK 0
151 #define VIRTIO_NET_ERR 1
153 #define VIRTIO_MAX_CTRL_DATA 2048
155 struct virtio_pmd_ctrl {
156 struct virtio_net_ctrl_hdr hdr;
157 virtio_net_ctrl_ack status;
158 uint8_t data[VIRTIO_MAX_CTRL_DATA];
161 struct vq_desc_extra {
168 struct virtio_hw *hw; /**< virtio_hw structure pointer. */
169 struct vring vq_ring; /**< vring keeping desc, used and avail */
170 struct vring_packed ring_packed; /**< vring keeping descs */
171 bool avail_wrap_counter;
172 bool used_wrap_counter;
173 uint16_t event_flags_shadow;
174 uint16_t avail_used_flags;
176 * Last consumed descriptor in the used table,
177 * trails vq_ring.used->idx.
179 uint16_t vq_used_cons_idx;
180 uint16_t vq_nentries; /**< vring desc numbers */
181 uint16_t vq_free_cnt; /**< num of desc available */
182 uint16_t vq_avail_idx; /**< sync until needed */
183 uint16_t vq_free_thresh; /**< free threshold */
185 void *vq_ring_virt_mem; /**< linear address of vring*/
186 unsigned int vq_ring_size;
189 struct virtnet_rx rxq;
190 struct virtnet_tx txq;
191 struct virtnet_ctl cq;
194 rte_iova_t vq_ring_mem; /**< physical address of vring,
195 * or virtual address for virtio_user. */
198 * Head of the free chain in the descriptor table. If
199 * there are no free descriptors, this will be set to
200 * VQ_RING_DESC_CHAIN_END.
202 uint16_t vq_desc_head_idx;
203 uint16_t vq_desc_tail_idx;
204 uint16_t vq_queue_index; /**< PCI queue index */
205 uint16_t offset; /**< relative offset to obtain addr in mbuf */
206 uint16_t *notify_addr;
207 struct rte_mbuf **sw_ring; /**< RX software ring. */
208 struct vq_desc_extra vq_descx[0];
211 /* If multiqueue is provided by host, then we suppport it. */
212 #define VIRTIO_NET_CTRL_MQ 4
213 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET 0
214 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN 1
215 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX 0x8000
218 * This is the first element of the scatter-gather list. If you don't
219 * specify GSO or CSUM features, you can simply ignore the header.
221 struct virtio_net_hdr {
222 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /**< Use csum_start,csum_offset*/
223 #define VIRTIO_NET_HDR_F_DATA_VALID 2 /**< Checksum is valid */
225 #define VIRTIO_NET_HDR_GSO_NONE 0 /**< Not a GSO frame */
226 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /**< GSO frame, IPv4 TCP (TSO) */
227 #define VIRTIO_NET_HDR_GSO_UDP 3 /**< GSO frame, IPv4 UDP (UFO) */
228 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /**< GSO frame, IPv6 TCP */
229 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /**< TCP has ECN set */
231 uint16_t hdr_len; /**< Ethernet + IP + tcp/udp hdrs */
232 uint16_t gso_size; /**< Bytes to append to hdr_len per frame */
233 uint16_t csum_start; /**< Position to start checksumming from */
234 uint16_t csum_offset; /**< Offset after that to place checksum */
238 * This is the version of the header to use when the MRG_RXBUF
239 * feature has been negotiated.
241 struct virtio_net_hdr_mrg_rxbuf {
242 struct virtio_net_hdr hdr;
243 uint16_t num_buffers; /**< Number of merged rx buffers */
246 /* Region reserved to allow for transmit header and indirect ring */
247 #define VIRTIO_MAX_TX_INDIRECT 8
248 struct virtio_tx_region {
249 struct virtio_net_hdr_mrg_rxbuf tx_hdr;
251 struct vring_desc tx_indir[VIRTIO_MAX_TX_INDIRECT]
252 __attribute__((__aligned__(16)));
253 struct vring_packed_desc tx_indir_pq[VIRTIO_MAX_TX_INDIRECT]
254 __attribute__((__aligned__(16)));
259 desc_is_used(struct vring_packed_desc *desc, struct virtqueue *vq)
261 uint16_t used, avail, flags;
264 used = !!(flags & VRING_DESC_F_USED(1));
265 avail = !!(flags & VRING_DESC_F_AVAIL(1));
267 return avail == used && used == vq->used_wrap_counter;
272 vring_desc_init_packed(struct virtqueue *vq, int n)
275 for (i = 0; i < n - 1; i++) {
276 vq->ring_packed.desc_packed[i].id = i;
277 vq->vq_descx[i].next = i + 1;
279 vq->ring_packed.desc_packed[i].id = i;
280 vq->vq_descx[i].next = VQ_RING_DESC_CHAIN_END;
283 /* Chain all the descriptors in the ring with an END */
285 vring_desc_init_split(struct vring_desc *dp, uint16_t n)
289 for (i = 0; i < n - 1; i++)
290 dp[i].next = (uint16_t)(i + 1);
291 dp[i].next = VQ_RING_DESC_CHAIN_END;
295 * Tell the backend not to interrupt us.
298 virtqueue_disable_intr_packed(struct virtqueue *vq)
300 uint16_t *event_flags = &vq->ring_packed.driver_event->desc_event_flags;
302 *event_flags = RING_EVENT_FLAGS_DISABLE;
307 * Tell the backend not to interrupt us.
310 virtqueue_disable_intr(struct virtqueue *vq)
312 if (vtpci_packed_queue(vq->hw))
313 virtqueue_disable_intr_packed(vq);
315 vq->vq_ring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
319 * Tell the backend to interrupt. Implementation for packed virtqueues.
322 virtqueue_enable_intr_packed(struct virtqueue *vq)
324 uint16_t *event_flags = &vq->ring_packed.driver_event->desc_event_flags;
327 if (vq->event_flags_shadow == RING_EVENT_FLAGS_DISABLE) {
329 vq->event_flags_shadow = RING_EVENT_FLAGS_ENABLE;
330 *event_flags = vq->event_flags_shadow;
335 * Tell the backend to interrupt. Implementation for split virtqueues.
338 virtqueue_enable_intr_split(struct virtqueue *vq)
340 vq->vq_ring.avail->flags &= (~VRING_AVAIL_F_NO_INTERRUPT);
344 * Tell the backend to interrupt us.
347 virtqueue_enable_intr(struct virtqueue *vq)
349 if (vtpci_packed_queue(vq->hw))
350 virtqueue_enable_intr_packed(vq);
352 virtqueue_enable_intr_split(vq);
356 * Dump virtqueue internal structures, for debug purpose only.
358 void virtqueue_dump(struct virtqueue *vq);
360 * Get all mbufs to be freed.
362 struct rte_mbuf *virtqueue_detach_unused(struct virtqueue *vq);
364 /* Flush the elements in the used ring. */
365 void virtqueue_rxvq_flush(struct virtqueue *vq);
368 virtqueue_full(const struct virtqueue *vq)
370 return vq->vq_free_cnt == 0;
374 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
376 if (vtpci_queue_idx == hw->max_queue_pairs * 2)
378 else if (vtpci_queue_idx % 2 == 0)
384 #define VIRTQUEUE_NUSED(vq) ((uint16_t)((vq)->vq_ring.used->idx - (vq)->vq_used_cons_idx))
386 void vq_ring_free_chain(struct virtqueue *vq, uint16_t desc_idx);
387 void vq_ring_free_chain_packed(struct virtqueue *vq, uint16_t used_idx);
388 void vq_ring_free_inorder(struct virtqueue *vq, uint16_t desc_idx,
392 vq_update_avail_idx(struct virtqueue *vq)
395 vq->vq_ring.avail->idx = vq->vq_avail_idx;
399 vq_update_avail_ring(struct virtqueue *vq, uint16_t desc_idx)
403 * Place the head of the descriptor chain into the next slot and make
404 * it usable to the host. The chain is made available now rather than
405 * deferring to virtqueue_notify() in the hopes that if the host is
406 * currently running on another CPU, we can keep it processing the new
409 avail_idx = (uint16_t)(vq->vq_avail_idx & (vq->vq_nentries - 1));
410 if (unlikely(vq->vq_ring.avail->ring[avail_idx] != desc_idx))
411 vq->vq_ring.avail->ring[avail_idx] = desc_idx;
416 virtqueue_kick_prepare(struct virtqueue *vq)
419 * Ensure updated avail->idx is visible to vhost before reading
423 return !(vq->vq_ring.used->flags & VRING_USED_F_NO_NOTIFY);
427 virtqueue_kick_prepare_packed(struct virtqueue *vq)
432 * Ensure updated data is visible to vhost before reading the flags.
435 flags = vq->ring_packed.device_event->desc_event_flags;
437 return flags != RING_EVENT_FLAGS_DISABLE;
441 virtqueue_notify(struct virtqueue *vq)
444 * Ensure updated avail->idx is visible to host.
445 * For virtio on IA, the notificaiton is through io port operation
446 * which is a serialization instruction itself.
448 VTPCI_OPS(vq->hw)->notify_queue(vq->hw, vq);
451 #ifdef RTE_LIBRTE_VIRTIO_DEBUG_DUMP
452 #define VIRTQUEUE_DUMP(vq) do { \
453 uint16_t used_idx, nused; \
454 used_idx = (vq)->vq_ring.used->idx; \
455 nused = (uint16_t)(used_idx - (vq)->vq_used_cons_idx); \
456 if (vtpci_packed_queue((vq)->hw)) { \
457 PMD_INIT_LOG(DEBUG, \
458 "VQ: - size=%d; free=%d; used_cons_idx=%d; avail_idx=%d;" \
459 "VQ: - avail_wrap_counter=%d; used_wrap_counter=%d", \
460 (vq)->vq_nentries, (vq)->vq_free_cnt, (vq)->vq_used_cons_idx, \
461 (vq)->vq_avail_idx, (vq)->avail_wrap_counter, \
462 (vq)->used_wrap_counter); \
465 PMD_INIT_LOG(DEBUG, \
466 "VQ: - size=%d; free=%d; used=%d; desc_head_idx=%d;" \
467 " avail.idx=%d; used_cons_idx=%d; used.idx=%d;" \
468 " avail.flags=0x%x; used.flags=0x%x", \
469 (vq)->vq_nentries, (vq)->vq_free_cnt, nused, \
470 (vq)->vq_desc_head_idx, (vq)->vq_ring.avail->idx, \
471 (vq)->vq_used_cons_idx, (vq)->vq_ring.used->idx, \
472 (vq)->vq_ring.avail->flags, (vq)->vq_ring.used->flags); \
475 #define VIRTQUEUE_DUMP(vq) do { } while (0)
478 #endif /* _VIRTQUEUE_H_ */