4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 #include <rte_atomic.h>
41 #include <rte_memory.h>
42 #include <rte_memzone.h>
43 #include <rte_mempool.h>
45 #include "virtio_pci.h"
46 #include "virtio_ring.h"
47 #include "virtio_logs.h"
50 #define wmb() rte_wmb()
51 #define rmb() rte_rmb()
53 #ifdef RTE_PMD_PACKET_PREFETCH
54 #define rte_packet_prefetch(p) rte_prefetch1(p)
56 #define rte_packet_prefetch(p) do {} while(0)
59 #define VIRTQUEUE_MAX_NAME_SZ 32
61 #define RTE_MBUF_DATA_DMA_ADDR(mb) \
62 (uint64_t) ((mb)->buf_physaddr + (uint64_t)((char *)((mb)->pkt.data) - \
63 (char *)(mb)->buf_addr))
65 #define VTNET_SQ_RQ_QUEUE_IDX 0
66 #define VTNET_SQ_TQ_QUEUE_IDX 1
67 #define VTNET_SQ_CQ_QUEUE_IDX 2
69 enum { VTNET_RQ = 0, VTNET_TQ = 1, VTNET_CQ = 2 };
71 * The maximum virtqueue size is 2^15. Use that value as the end of
72 * descriptor chain terminator since it will never be a valid index
73 * in the descriptor table. This is used to verify we are correctly
74 * handling vq_free_cnt.
76 #define VQ_RING_DESC_CHAIN_END 32768
79 * Control the RX mode, ie. promiscuous, allmulti, etc...
80 * All commands require an "out" sg entry containing a 1 byte
81 * state value, zero = disable, non-zero = enable. Commands
82 * 0 and 1 are supported with the VIRTIO_NET_F_CTRL_RX feature.
83 * Commands 2-5 are added with VIRTIO_NET_F_CTRL_RX_EXTRA.
85 #define VIRTIO_NET_CTRL_RX 0
86 #define VIRTIO_NET_CTRL_RX_PROMISC 0
87 #define VIRTIO_NET_CTRL_RX_ALLMULTI 1
88 #define VIRTIO_NET_CTRL_RX_ALLUNI 2
89 #define VIRTIO_NET_CTRL_RX_NOMULTI 3
90 #define VIRTIO_NET_CTRL_RX_NOUNI 4
91 #define VIRTIO_NET_CTRL_RX_NOBCAST 5
94 * Control VLAN filtering
96 * The VLAN filter table is controlled via a simple ADD/DEL interface.
97 * VLAN IDs not added may be filtered by the hypervisor. Del is the
98 * opposite of add. Both commands expect an out entry containing a 2
99 * byte VLAN ID. VLAN filtering is available with the
100 * VIRTIO_NET_F_CTRL_VLAN feature bit.
102 #define VIRTIO_NET_CTRL_VLAN 2
103 #define VIRTIO_NET_CTRL_VLAN_ADD 0
104 #define VIRTIO_NET_CTRL_VLAN_DEL 1
106 struct virtio_net_ctrl_hdr {
109 } __attribute__((packed));
111 typedef uint8_t virtio_net_ctrl_ack;
113 #define VIRTIO_NET_OK 0
114 #define VIRTIO_NET_ERR 1
116 #define VIRTIO_MAX_CTRL_DATA 128
118 struct virtio_pmd_ctrl {
119 struct virtio_net_ctrl_hdr hdr;
120 virtio_net_ctrl_ack status;
121 uint8_t data[VIRTIO_MAX_CTRL_DATA];
125 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
126 struct virtio_hw *hw; /**< virtio_hw structure pointer. */
127 const struct rte_memzone *mz; /**< mem zone to populate RX ring. */
128 const struct rte_memzone *virtio_net_hdr_mz; /**< memzone to populate hdr. */
129 struct rte_mempool *mpool; /**< mempool for mbuf allocation */
130 uint16_t queue_id; /**< DPDK queue index. */
131 uint8_t port_id; /**< Device port identifier. */
133 void *vq_ring_virt_mem; /**< linear address of vring*/
136 phys_addr_t vq_ring_mem; /**< physical address of vring */
138 struct vring vq_ring; /**< vring keeping desc, used and avail */
139 uint16_t vq_free_cnt; /**< num of desc available */
140 uint16_t vq_nentries; /**< vring desc numbers */
141 uint16_t vq_queue_index; /**< PCI queue index */
143 * Head of the free chain in the descriptor table. If
144 * there are no free descriptors, this will be set to
145 * VQ_RING_DESC_CHAIN_END.
147 uint16_t vq_desc_head_idx;
148 uint16_t vq_desc_tail_idx;
150 * Last consumed descriptor in the used table,
151 * trails vq_ring.used->idx.
153 uint16_t vq_used_cons_idx;
154 uint16_t vq_avail_idx;
155 phys_addr_t virtio_net_hdr_mem; /**< hdr for each xmit packet */
157 struct vq_desc_extra {
163 /* If multiqueue is provided by host, then we suppport it. */
164 #ifndef VIRTIO_NET_F_MQ
165 /* Device supports Receive Flow Steering */
166 #define VIRTIO_NET_F_MQ 0x400000
167 #define VIRTIO_NET_CTRL_MQ 4
168 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET 0
169 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN 1
170 #define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX 0x8000
174 * This is the first element of the scatter-gather list. If you don't
175 * specify GSO or CSUM features, you can simply ignore the header.
177 struct virtio_net_hdr {
178 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /**< Use csum_start,csum_offset*/
180 #define VIRTIO_NET_HDR_GSO_NONE 0 /**< Not a GSO frame */
181 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /**< GSO frame, IPv4 TCP (TSO) */
182 #define VIRTIO_NET_HDR_GSO_UDP 3 /**< GSO frame, IPv4 UDP (UFO) */
183 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /**< GSO frame, IPv6 TCP */
184 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /**< TCP has ECN set */
186 uint16_t hdr_len; /**< Ethernet + IP + tcp/udp hdrs */
187 uint16_t gso_size; /**< Bytes to append to hdr_len per frame */
188 uint16_t csum_start; /**< Position to start checksumming from */
189 uint16_t csum_offset; /**< Offset after that to place checksum */
193 * This is the version of the header to use when the MRG_RXBUF
194 * feature has been negotiated.
196 struct virtio_net_hdr_mrg_rxbuf {
197 struct virtio_net_hdr hdr;
198 uint16_t num_buffers; /**< Number of merged rx buffers */
202 * Tell the backend not to interrupt us.
204 void virtqueue_disable_intr(struct virtqueue *vq);
206 * Dump virtqueue internal structures, for debug purpose only.
208 void virtqueue_dump(struct virtqueue *vq);
210 * Get all mbufs to be freed.
212 struct rte_mbuf *virtqueue_detatch_unused(struct virtqueue *vq);
215 virtqueue_full(const struct virtqueue *vq)
217 return vq->vq_free_cnt == 0;
220 #define VIRTQUEUE_NUSED(vq) ((uint16_t)((vq)->vq_ring.used->idx - (vq)->vq_used_cons_idx))
222 static inline void __attribute__((always_inline))
223 vq_update_avail_idx(struct virtqueue *vq)
225 rte_compiler_barrier();
226 vq->vq_ring.avail->idx = vq->vq_avail_idx;
229 static inline void __attribute__((always_inline))
230 vq_update_avail_ring(struct virtqueue *vq, uint16_t desc_idx)
234 * Place the head of the descriptor chain into the next slot and make
235 * it usable to the host. The chain is made available now rather than
236 * deferring to virtqueue_notify() in the hopes that if the host is
237 * currently running on another CPU, we can keep it processing the new
240 avail_idx = (uint16_t)(vq->vq_avail_idx & (vq->vq_nentries - 1));
241 vq->vq_ring.avail->ring[avail_idx] = desc_idx;
245 static inline int __attribute__((always_inline))
246 virtqueue_kick_prepare(struct virtqueue *vq)
248 return !(vq->vq_ring.used->flags & VRING_USED_F_NO_NOTIFY);
251 static inline void __attribute__((always_inline))
252 virtqueue_notify(struct virtqueue *vq)
255 * Ensure updated avail->idx is visible to host. mb() necessary?
256 * For virtio on IA, the notificaiton is through io port operation
257 * which is a serialization instruction itself.
259 VIRTIO_WRITE_REG_2(vq->hw, VIRTIO_PCI_QUEUE_NOTIFY, vq->vq_queue_index);
262 static inline void __attribute__((always_inline))
263 vq_ring_free_chain(struct virtqueue *vq, uint16_t desc_idx)
265 struct vring_desc *dp, *dp_tail;
266 struct vq_desc_extra *dxp;
267 uint16_t desc_idx_last = desc_idx;
269 dp = &vq->vq_ring.desc[desc_idx];
270 dxp = &vq->vq_descx[desc_idx];
271 vq->vq_free_cnt = (uint16_t)(vq->vq_free_cnt + dxp->ndescs);
272 if ((dp->flags & VRING_DESC_F_INDIRECT) == 0) {
273 while (dp->flags & VRING_DESC_F_NEXT) {
274 desc_idx_last = dp->next;
275 dp = &vq->vq_ring.desc[dp->next];
281 * We must append the existing free chain, if any, to the end of
282 * newly freed chain. If the virtqueue was completely used, then
283 * head would be VQ_RING_DESC_CHAIN_END (ASSERTed above).
285 if (vq->vq_desc_tail_idx == VQ_RING_DESC_CHAIN_END) {
286 vq->vq_desc_head_idx = desc_idx;
288 dp_tail = &vq->vq_ring.desc[vq->vq_desc_tail_idx];
289 dp_tail->next = desc_idx;
292 vq->vq_desc_tail_idx = desc_idx_last;
293 dp->next = VQ_RING_DESC_CHAIN_END;
296 static inline int __attribute__((always_inline))
297 virtqueue_enqueue_recv_refill(struct virtqueue *vq, struct rte_mbuf *cookie)
299 struct vq_desc_extra *dxp;
300 struct vring_desc *start_dp;
302 uint16_t head_idx, idx;
304 if (unlikely(vq->vq_free_cnt == 0))
306 if (unlikely(vq->vq_free_cnt < needed))
309 head_idx = vq->vq_desc_head_idx;
310 if (unlikely(head_idx >= vq->vq_nentries))
314 dxp = &vq->vq_descx[idx];
315 dxp->cookie = (void *)cookie;
316 dxp->ndescs = needed;
318 start_dp = vq->vq_ring.desc;
320 (uint64_t) (cookie->buf_physaddr + RTE_PKTMBUF_HEADROOM - sizeof(struct virtio_net_hdr));
321 start_dp[idx].len = cookie->buf_len - RTE_PKTMBUF_HEADROOM + sizeof(struct virtio_net_hdr);
322 start_dp[idx].flags = VRING_DESC_F_WRITE;
323 idx = start_dp[idx].next;
324 vq->vq_desc_head_idx = idx;
325 if (vq->vq_desc_head_idx == VQ_RING_DESC_CHAIN_END)
326 vq->vq_desc_tail_idx = idx;
327 vq->vq_free_cnt = (uint16_t)(vq->vq_free_cnt - needed);
328 vq_update_avail_ring(vq, head_idx);
333 static inline int __attribute__((always_inline))
334 virtqueue_enqueue_xmit(struct virtqueue *txvq, struct rte_mbuf *cookie)
336 struct vq_desc_extra *dxp;
337 struct vring_desc *start_dp;
339 uint16_t head_idx, idx;
341 if (unlikely(txvq->vq_free_cnt == 0))
343 if (unlikely(txvq->vq_free_cnt < needed))
345 head_idx = txvq->vq_desc_head_idx;
346 if (unlikely(head_idx >= txvq->vq_nentries))
350 dxp = &txvq->vq_descx[idx];
351 if (dxp->cookie != NULL)
352 rte_pktmbuf_free_seg(dxp->cookie);
353 dxp->cookie = (void *)cookie;
354 dxp->ndescs = needed;
356 start_dp = txvq->vq_ring.desc;
358 txvq->virtio_net_hdr_mem + idx * sizeof(struct virtio_net_hdr);
359 start_dp[idx].len = sizeof(struct virtio_net_hdr);
360 start_dp[idx].flags = VRING_DESC_F_NEXT;
361 idx = start_dp[idx].next;
362 start_dp[idx].addr = RTE_MBUF_DATA_DMA_ADDR(cookie);
363 start_dp[idx].len = cookie->pkt.data_len;
364 start_dp[idx].flags = 0;
365 idx = start_dp[idx].next;
366 txvq->vq_desc_head_idx = idx;
367 if (txvq->vq_desc_head_idx == VQ_RING_DESC_CHAIN_END)
368 txvq->vq_desc_tail_idx = idx;
369 txvq->vq_free_cnt = (uint16_t)(txvq->vq_free_cnt - needed);
370 vq_update_avail_ring(txvq, head_idx);
375 static inline uint16_t __attribute__((always_inline))
376 virtqueue_dequeue_burst_rx(struct virtqueue *vq, struct rte_mbuf **rx_pkts, uint32_t *len, uint16_t num)
378 struct vring_used_elem *uep;
379 struct rte_mbuf *cookie;
380 uint16_t used_idx, desc_idx;
383 /* Caller does the check */
384 for (i = 0; i < num; i++) {
385 used_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
386 uep = &vq->vq_ring.used->ring[used_idx];
387 desc_idx = (uint16_t) uep->id;
389 cookie = (struct rte_mbuf *)vq->vq_descx[desc_idx].cookie;
391 if (unlikely(cookie == NULL)) {
392 PMD_DRV_LOG(ERR, "vring descriptor with no mbuf cookie at %u\n",
393 vq->vq_used_cons_idx);
397 rte_prefetch0(cookie);
398 rte_packet_prefetch(cookie->pkt.data);
400 vq->vq_used_cons_idx++;
401 vq_ring_free_chain(vq, desc_idx);
402 vq->vq_descx[desc_idx].cookie = NULL;
408 static inline uint16_t __attribute__((always_inline))
409 virtqueue_dequeue_pkt_tx(struct virtqueue *vq)
411 struct vring_used_elem *uep;
412 uint16_t used_idx, desc_idx;
414 used_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
415 uep = &vq->vq_ring.used->ring[used_idx];
416 desc_idx = (uint16_t) uep->id;
417 vq->vq_used_cons_idx++;
418 vq_ring_free_chain(vq, desc_idx);
423 #ifdef RTE_LIBRTE_VIRTIO_DEBUG_DUMP
424 #define VIRTQUEUE_DUMP(vq) do { \
425 uint16_t used_idx, nused; \
426 used_idx = (vq)->vq_ring.used->idx; \
427 nused = (uint16_t)(used_idx - (vq)->vq_used_cons_idx); \
428 PMD_INIT_LOG(DEBUG, \
429 "VQ: %s - size=%d; free=%d; used=%d; desc_head_idx=%d;" \
430 " avail.idx=%d; used_cons_idx=%d; used.idx=%d;" \
431 " avail.flags=0x%x; used.flags=0x%x\n", \
432 (vq)->vq_name, (vq)->vq_nentries, (vq)->vq_free_cnt, nused, \
433 (vq)->vq_desc_head_idx, (vq)->vq_ring.avail->idx, \
434 (vq)->vq_used_cons_idx, (vq)->vq_ring.used->idx, \
435 (vq)->vq_ring.avail->flags, (vq)->vq_ring.used->flags); \
438 #define VIRTQUEUE_DUMP(vq) do { } while (0)
441 #endif /* _VIRTQUEUE_H_ */