ceph/src/dpdk/drivers/net/xenvirt/virtqueue.h

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  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
  16  *       distribution.
  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.
  20  *
  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.
  32  */
  33
  34 #ifndef _VIRTQUEUE_H_
  35 #define _VIRTQUEUE_H_
  36
  37 #include <stdint.h>
  38 #include <linux/virtio_ring.h>
  39 #include <linux/virtio_net.h>
  40
  41 #include <rte_atomic.h>
  42 #include <rte_memory.h>
  43 #include <rte_memzone.h>
  44 #include <rte_mempool.h>
  45
  46 #include "virtio_logs.h"
  47
  48 struct rte_mbuf;
  49
  50 /* The alignment to use between consumer and producer parts of vring. */
  51 #define VIRTIO_PCI_VRING_ALIGN 4096
  52
  53 enum { VTNET_RQ = 0, VTNET_TQ = 1, VTNET_CQ = 2 };
  54
  55 /**
  56  * The maximum virtqueue size is 2^15. Use that value as the end of
  57  * descriptor chain terminator since it will never be a valid index
  58  * in the descriptor table. This is used to verify we are correctly
  59  * handling vq_free_cnt.
  60  */
  61 #define VQ_RING_DESC_CHAIN_END 32768
  62
  63 #define VIRTQUEUE_MAX_NAME_SZ  32
  64
  65 struct pmd_internals {
  66         struct rte_eth_stats eth_stats;
  67         int port_id;
  68         int virtio_idx;
  69 };
  70
  71
  72 struct virtqueue {
  73         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
  74         struct rte_mempool       *mpool;  /**< mempool for mbuf allocation */
  75         uint16_t    queue_id;             /**< DPDK queue index. */
  76         uint16_t    vq_queue_index;       /**< PCI queue index */
  77         uint8_t     port_id;              /**< Device port identifier. */
  78
  79         void        *vq_ring_virt_mem;    /**< virtual address of vring*/
  80         int         vq_alignment;
  81         int         vq_ring_size;
  82
  83         struct vring vq_ring;    /**< vring keeping desc, used and avail */
  84         struct pmd_internals *internals;  /**< virtio device internal info. */
  85         uint16_t    vq_nentries; /**< vring desc numbers */
  86         uint16_t    vq_desc_head_idx;
  87         uint16_t    vq_free_cnt; /**< num of desc available */
  88         uint16_t vq_used_cons_idx; /**< Last consumed desc in used table, trails vq_ring.used->idx*/
  89
  90         struct vq_desc_extra {
  91                 void              *cookie;
  92                 uint16_t          ndescs;
  93         } vq_descx[0] __rte_cache_aligned;
  94 };
  95
  96
  97 #ifdef  RTE_LIBRTE_XENVIRT_DEBUG_DUMP
  98 #define VIRTQUEUE_DUMP(vq) do { \
  99         uint16_t used_idx, nused; \
 100         used_idx = (vq)->vq_ring.used->idx; \
 101         nused = (uint16_t)(used_idx - (vq)->vq_used_cons_idx); \
 102         PMD_INIT_LOG(DEBUG, \
 103           "VQ: %s - size=%d; free=%d; used=%d; desc_head_idx=%d;" \
 104           " avail.idx=%d; used_cons_idx=%d; used.idx=%d;" \
 105           " avail.flags=0x%x; used.flags=0x%x\n", \
 106           (vq)->vq_name, (vq)->vq_nentries, (vq)->vq_free_cnt, nused, \
 107           (vq)->vq_desc_head_idx, (vq)->vq_ring.avail->idx, \
 108           (vq)->vq_used_cons_idx, (vq)->vq_ring.used->idx, \
 109           (vq)->vq_ring.avail->flags, (vq)->vq_ring.used->flags); \
 110 } while (0)
 111 #else
 112 #define VIRTQUEUE_DUMP(vq) do { } while (0)
 113 #endif
 114
 115
 116 /**
 117  *  Dump virtqueue internal structures, for debug purpose only.
 118  */
 119 void virtqueue_dump(struct virtqueue *vq);
 120
 121 /**
 122  *  Get all mbufs to be freed.
 123  */
 124 struct rte_mbuf * virtqueue_detatch_unused(struct virtqueue *vq);
 125
 126 static inline int __attribute__((always_inline))
 127 virtqueue_full(const struct virtqueue *vq)
 128 {
 129         return vq->vq_free_cnt == 0;
 130 }
 131
 132 #define VIRTQUEUE_NUSED(vq) ((uint16_t)((vq)->vq_ring.used->idx - (vq)->vq_used_cons_idx))
 133
 134 static inline void __attribute__((always_inline))
 135 vq_ring_update_avail(struct virtqueue *vq, uint16_t desc_idx)
 136 {
 137         uint16_t avail_idx;
 138         /*
 139          * Place the head of the descriptor chain into the next slot and make
 140          * it usable to the host. The chain is made available now rather than
 141          * deferring to virtqueue_notify() in the hopes that if the host is
 142          * currently running on another CPU, we can keep it processing the new
 143          * descriptor.
 144          */
 145         avail_idx = (uint16_t)(vq->vq_ring.avail->idx & (vq->vq_nentries - 1));
 146         vq->vq_ring.avail->ring[avail_idx] = desc_idx;
 147         rte_smp_wmb();
 148         vq->vq_ring.avail->idx++;
 149 }
 150
 151 static inline void  __attribute__((always_inline))
 152 vq_ring_free_chain(struct virtqueue *vq, uint16_t desc_idx)
 153 {
 154         struct vring_desc *dp;
 155         struct vq_desc_extra *dxp;
 156
 157         dp  = &vq->vq_ring.desc[desc_idx];
 158         dxp = &vq->vq_descx[desc_idx];
 159         vq->vq_free_cnt = (uint16_t)(vq->vq_free_cnt + dxp->ndescs);
 160         while (dp->flags & VRING_DESC_F_NEXT) {
 161                 dp = &vq->vq_ring.desc[dp->next];
 162         }
 163         dxp->ndescs = 0;
 164
 165         /*
 166          * We must append the existing free chain, if any, to the end of
 167          * newly freed chain. If the virtqueue was completely used, then
 168          * head would be VQ_RING_DESC_CHAIN_END (ASSERTed above).
 169          */
 170         dp->next = vq->vq_desc_head_idx;
 171         vq->vq_desc_head_idx = desc_idx;
 172 }
 173
 174 static inline int  __attribute__((always_inline))
 175 virtqueue_enqueue_recv_refill(struct virtqueue *rxvq, struct rte_mbuf *cookie)
 176 {
 177         const uint16_t needed = 1;
 178         const uint16_t head_idx = rxvq->vq_desc_head_idx;
 179         struct vring_desc *start_dp = rxvq->vq_ring.desc;
 180         struct vq_desc_extra *dxp;
 181
 182         if (unlikely(rxvq->vq_free_cnt == 0))
 183                 return -ENOSPC;
 184         if (unlikely(rxvq->vq_free_cnt < needed))
 185                 return -EMSGSIZE;
 186         if (unlikely(head_idx >= rxvq->vq_nentries))
 187                 return -EFAULT;
 188
 189         dxp = &rxvq->vq_descx[head_idx];
 190         dxp->cookie = (void *)cookie;
 191         dxp->ndescs = needed;
 192
 193         start_dp[head_idx].addr  =
 194                 (uint64_t) ((uintptr_t)cookie->buf_addr + RTE_PKTMBUF_HEADROOM - sizeof(struct virtio_net_hdr));
 195         start_dp[head_idx].len   = cookie->buf_len - RTE_PKTMBUF_HEADROOM + sizeof(struct virtio_net_hdr);
 196         start_dp[head_idx].flags = VRING_DESC_F_WRITE;
 197         rxvq->vq_desc_head_idx   = start_dp[head_idx].next;
 198         rxvq->vq_free_cnt        = (uint16_t)(rxvq->vq_free_cnt - needed);
 199         vq_ring_update_avail(rxvq, head_idx);
 200
 201         return 0;
 202 }
 203
 204 static inline int  __attribute__((always_inline))
 205 virtqueue_enqueue_xmit(struct virtqueue *txvq, struct rte_mbuf *cookie)
 206 {
 207
 208         const uint16_t needed = 2;
 209         struct vring_desc *start_dp =  txvq->vq_ring.desc;
 210         uint16_t head_idx = txvq->vq_desc_head_idx;
 211         uint16_t idx      = head_idx;
 212         struct vq_desc_extra *dxp;
 213
 214         if (unlikely(txvq->vq_free_cnt == 0))
 215                 return -ENOSPC;
 216         if (unlikely(txvq->vq_free_cnt < needed))
 217                 return -EMSGSIZE;
 218         if (unlikely(head_idx >= txvq->vq_nentries))
 219                 return -EFAULT;
 220
 221         dxp = &txvq->vq_descx[idx];
 222         dxp->cookie = (void *)cookie;
 223         dxp->ndescs = needed;
 224
 225         start_dp = txvq->vq_ring.desc;
 226         start_dp[idx].addr  = 0;
 227 /*
 228  * TODO: save one desc here?
 229  */
 230         start_dp[idx].len   = sizeof(struct virtio_net_hdr);
 231         start_dp[idx].flags = VRING_DESC_F_NEXT;
 232         start_dp[idx].addr  = (uintptr_t)NULL;
 233         idx = start_dp[idx].next;
 234         start_dp[idx].addr  = (uint64_t)rte_pktmbuf_mtod(cookie, uintptr_t);
 235         start_dp[idx].len   = cookie->data_len;
 236         start_dp[idx].flags = 0;
 237         idx = start_dp[idx].next;
 238         txvq->vq_desc_head_idx = idx;
 239         txvq->vq_free_cnt = (uint16_t)(txvq->vq_free_cnt - needed);
 240         vq_ring_update_avail(txvq, head_idx);
 241
 242         return 0;
 243 }
 244
 245 static inline uint16_t  __attribute__((always_inline))
 246 virtqueue_dequeue_burst(struct virtqueue *vq, struct rte_mbuf **rx_pkts, uint32_t *len, uint16_t num)
 247 {
 248         struct vring_used_elem *uep;
 249         struct rte_mbuf *cookie;
 250         uint16_t used_idx, desc_idx;
 251         uint16_t i;
 252         /*  Caller does the check */
 253         for (i = 0; i < num ; i ++) {
 254                 used_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
 255                 uep = &vq->vq_ring.used->ring[used_idx];
 256                 desc_idx = (uint16_t) uep->id;
 257                 cookie = (struct rte_mbuf *)vq->vq_descx[desc_idx].cookie;
 258                 if (unlikely(cookie == NULL)) {
 259                         PMD_DRV_LOG(ERR, "vring descriptor with no mbuf cookie at %u\n",
 260                                 vq->vq_used_cons_idx);
 261                         RTE_LOG(ERR, PMD, "%s: inconsistent (%u, %u)\n", __func__, used_idx , desc_idx);
 262                         break;
 263                 }
 264                 len[i] = uep->len;
 265                 rx_pkts[i]  = cookie;
 266                 vq->vq_used_cons_idx++;
 267                 vq_ring_free_chain(vq, desc_idx);
 268                 vq->vq_descx[desc_idx].cookie = NULL;
 269         }
 270         return i;
 271 }
 272
 273 #endif /* _VIRTQUEUE_H_ */