net/virtio-user: add virtual device
[dpdk.git] / drivers / net / enic / enic_rx.c
1 /*
2  * Copyright 2008-2014 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * Copyright (c) 2014, Cisco Systems, Inc.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  * notice, this list of conditions and the following disclaimer.
14  *
15  * 2. Redistributions in binary form must reproduce the above copyright
16  * notice, this list of conditions and the following disclaimer in
17  * the documentation and/or other materials provided with the
18  * distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  */
34
35 #include <rte_mbuf.h>
36 #include <rte_ethdev.h>
37 #include <rte_prefetch.h>
38
39 #include "enic_compat.h"
40 #include "rq_enet_desc.h"
41 #include "enic.h"
42
43 #define RTE_PMD_USE_PREFETCH
44
45 #ifdef RTE_PMD_USE_PREFETCH
46 /*
47  * Prefetch a cache line into all cache levels.
48  */
49 #define rte_enic_prefetch(p) rte_prefetch0(p)
50 #else
51 #define rte_enic_prefetch(p) do {} while (0)
52 #endif
53
54 #ifdef RTE_PMD_PACKET_PREFETCH
55 #define rte_packet_prefetch(p) rte_prefetch1(p)
56 #else
57 #define rte_packet_prefetch(p) do {} while (0)
58 #endif
59
60 static inline uint16_t
61 enic_cq_rx_desc_ciflags(struct cq_enet_rq_desc *crd)
62 {
63         return le16_to_cpu(crd->completed_index_flags) & ~CQ_DESC_COMP_NDX_MASK;
64 }
65
66 static inline uint16_t
67 enic_cq_rx_desc_bwflags(struct cq_enet_rq_desc *crd)
68 {
69         return(le16_to_cpu(crd->bytes_written_flags) &
70                 ~CQ_ENET_RQ_DESC_BYTES_WRITTEN_MASK);
71 }
72
73 static inline uint8_t
74 enic_cq_rx_desc_packet_error(uint16_t bwflags)
75 {
76         return((bwflags & CQ_ENET_RQ_DESC_FLAGS_TRUNCATED) ==
77                 CQ_ENET_RQ_DESC_FLAGS_TRUNCATED);
78 }
79
80 static inline uint8_t
81 enic_cq_rx_desc_eop(uint16_t ciflags)
82 {
83         return (ciflags & CQ_ENET_RQ_DESC_FLAGS_EOP)
84                 == CQ_ENET_RQ_DESC_FLAGS_EOP;
85 }
86
87 static inline uint8_t
88 enic_cq_rx_desc_csum_not_calc(struct cq_enet_rq_desc *cqrd)
89 {
90         return ((le16_to_cpu(cqrd->q_number_rss_type_flags) &
91                 CQ_ENET_RQ_DESC_FLAGS_CSUM_NOT_CALC) ==
92                 CQ_ENET_RQ_DESC_FLAGS_CSUM_NOT_CALC);
93 }
94
95 static inline uint8_t
96 enic_cq_rx_desc_ipv4_csum_ok(struct cq_enet_rq_desc *cqrd)
97 {
98         return ((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_IPV4_CSUM_OK) ==
99                 CQ_ENET_RQ_DESC_FLAGS_IPV4_CSUM_OK);
100 }
101
102 static inline uint8_t
103 enic_cq_rx_desc_tcp_udp_csum_ok(struct cq_enet_rq_desc *cqrd)
104 {
105         return((cqrd->flags & CQ_ENET_RQ_DESC_FLAGS_TCP_UDP_CSUM_OK) ==
106                 CQ_ENET_RQ_DESC_FLAGS_TCP_UDP_CSUM_OK);
107 }
108
109 static inline uint8_t
110 enic_cq_rx_desc_rss_type(struct cq_enet_rq_desc *cqrd)
111 {
112         return (uint8_t)((le16_to_cpu(cqrd->q_number_rss_type_flags) >>
113                 CQ_DESC_Q_NUM_BITS) & CQ_ENET_RQ_DESC_RSS_TYPE_MASK);
114 }
115
116 static inline uint32_t
117 enic_cq_rx_desc_rss_hash(struct cq_enet_rq_desc *cqrd)
118 {
119         return le32_to_cpu(cqrd->rss_hash);
120 }
121
122 static inline uint16_t
123 enic_cq_rx_desc_vlan(struct cq_enet_rq_desc *cqrd)
124 {
125         return le16_to_cpu(cqrd->vlan);
126 }
127
128 static inline uint16_t
129 enic_cq_rx_desc_n_bytes(struct cq_desc *cqd)
130 {
131         struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
132         return le16_to_cpu(cqrd->bytes_written_flags) &
133                 CQ_ENET_RQ_DESC_BYTES_WRITTEN_MASK;
134 }
135
136 static inline uint8_t
137 enic_cq_rx_to_pkt_err_flags(struct cq_desc *cqd, uint64_t *pkt_err_flags_out)
138 {
139         struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
140         uint16_t bwflags;
141         int ret = 0;
142         uint64_t pkt_err_flags = 0;
143
144         bwflags = enic_cq_rx_desc_bwflags(cqrd);
145         if (unlikely(enic_cq_rx_desc_packet_error(bwflags))) {
146                 pkt_err_flags = PKT_RX_MAC_ERR;
147                 ret = 1;
148         }
149         *pkt_err_flags_out = pkt_err_flags;
150         return ret;
151 }
152
153 /*
154  * Lookup table to translate RX CQ flags to mbuf flags.
155  */
156 static inline uint32_t
157 enic_cq_rx_flags_to_pkt_type(struct cq_desc *cqd)
158 {
159         struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
160         uint8_t cqrd_flags = cqrd->flags;
161         static const uint32_t cq_type_table[128] __rte_cache_aligned = {
162                 [32] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4,
163                 [34] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4
164                         | RTE_PTYPE_L4_UDP,
165                 [36] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4
166                         | RTE_PTYPE_L4_TCP,
167                 [96] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4
168                         | RTE_PTYPE_L4_FRAG,
169                 [16] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6,
170                 [18] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6
171                         | RTE_PTYPE_L4_UDP,
172                 [20] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6
173                         | RTE_PTYPE_L4_TCP,
174                 [80] =  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6
175                         | RTE_PTYPE_L4_FRAG,
176                 /* All others reserved */
177         };
178         cqrd_flags &= CQ_ENET_RQ_DESC_FLAGS_IPV4_FRAGMENT
179                 | CQ_ENET_RQ_DESC_FLAGS_IPV4 | CQ_ENET_RQ_DESC_FLAGS_IPV6
180                 | CQ_ENET_RQ_DESC_FLAGS_TCP | CQ_ENET_RQ_DESC_FLAGS_UDP;
181         return cq_type_table[cqrd_flags];
182 }
183
184 static inline void
185 enic_cq_rx_to_pkt_flags(struct cq_desc *cqd, struct rte_mbuf *mbuf)
186 {
187         struct cq_enet_rq_desc *cqrd = (struct cq_enet_rq_desc *)cqd;
188         uint16_t ciflags, bwflags, pkt_flags = 0;
189         ciflags = enic_cq_rx_desc_ciflags(cqrd);
190         bwflags = enic_cq_rx_desc_bwflags(cqrd);
191
192         mbuf->ol_flags = 0;
193
194         /* flags are meaningless if !EOP */
195         if (unlikely(!enic_cq_rx_desc_eop(ciflags)))
196                 goto mbuf_flags_done;
197
198         /* VLAN stripping */
199         if (bwflags & CQ_ENET_RQ_DESC_FLAGS_VLAN_STRIPPED) {
200                 pkt_flags |= PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED;
201                 mbuf->vlan_tci = enic_cq_rx_desc_vlan(cqrd);
202         } else {
203                 mbuf->vlan_tci = 0;
204         }
205
206         /* RSS flag */
207         if (enic_cq_rx_desc_rss_type(cqrd)) {
208                 pkt_flags |= PKT_RX_RSS_HASH;
209                 mbuf->hash.rss = enic_cq_rx_desc_rss_hash(cqrd);
210         }
211
212         /* checksum flags */
213         if (!enic_cq_rx_desc_csum_not_calc(cqrd) &&
214                 (mbuf->packet_type & RTE_PTYPE_L3_IPV4)) {
215                 if (unlikely(!enic_cq_rx_desc_ipv4_csum_ok(cqrd)))
216                         pkt_flags |= PKT_RX_IP_CKSUM_BAD;
217                 if (mbuf->packet_type & (RTE_PTYPE_L4_UDP | RTE_PTYPE_L4_TCP)) {
218                         if (unlikely(!enic_cq_rx_desc_tcp_udp_csum_ok(cqrd)))
219                                 pkt_flags |= PKT_RX_L4_CKSUM_BAD;
220                 }
221         }
222
223  mbuf_flags_done:
224         mbuf->ol_flags = pkt_flags;
225 }
226
227 static inline uint32_t
228 enic_ring_add(uint32_t n_descriptors, uint32_t i0, uint32_t i1)
229 {
230         uint32_t d = i0 + i1;
231         RTE_ASSERT(i0 < n_descriptors);
232         RTE_ASSERT(i1 < n_descriptors);
233         d -= (d >= n_descriptors) ? n_descriptors : 0;
234         return d;
235 }
236
237
238 uint16_t
239 enic_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
240                uint16_t nb_pkts)
241 {
242         struct vnic_rq *rq = rx_queue;
243         struct enic *enic = vnic_dev_priv(rq->vdev);
244         unsigned int rx_id;
245         struct rte_mbuf *nmb, *rxmb;
246         uint16_t nb_rx = 0;
247         uint16_t nb_hold;
248         struct vnic_cq *cq;
249         volatile struct cq_desc *cqd_ptr;
250         uint8_t color;
251
252         cq = &enic->cq[enic_cq_rq(enic, rq->index)];
253         rx_id = cq->to_clean;           /* index of cqd, rqd, mbuf_table */
254         cqd_ptr = (struct cq_desc *)(cq->ring.descs) + rx_id;
255
256         nb_hold = rq->rx_nb_hold;       /* mbufs held by software */
257
258         while (nb_rx < nb_pkts) {
259                 volatile struct rq_enet_desc *rqd_ptr;
260                 dma_addr_t dma_addr;
261                 struct cq_desc cqd;
262                 uint64_t ol_err_flags;
263                 uint8_t packet_error;
264
265                 /* Check for pkts available */
266                 color = (cqd_ptr->type_color >> CQ_DESC_COLOR_SHIFT)
267                         & CQ_DESC_COLOR_MASK;
268                 if (color == cq->last_color)
269                         break;
270
271                 /* Get the cq descriptor and rq pointer */
272                 cqd = *cqd_ptr;
273                 rqd_ptr = (struct rq_enet_desc *)(rq->ring.descs) + rx_id;
274
275                 /* allocate a new mbuf */
276                 nmb = rte_mbuf_raw_alloc(rq->mp);
277                 if (nmb == NULL) {
278                         dev_err(enic, "RX mbuf alloc failed port=%u qid=%u",
279                         enic->port_id, (unsigned)rq->index);
280                         rte_eth_devices[enic->port_id].
281                                         data->rx_mbuf_alloc_failed++;
282                         break;
283                 }
284
285                 /* A packet error means descriptor and data are untrusted */
286                 packet_error = enic_cq_rx_to_pkt_err_flags(&cqd, &ol_err_flags);
287
288                 /* Get the mbuf to return and replace with one just allocated */
289                 rxmb = rq->mbuf_ring[rx_id];
290                 rq->mbuf_ring[rx_id] = nmb;
291
292                 /* Increment cqd, rqd, mbuf_table index */
293                 rx_id++;
294                 if (unlikely(rx_id == rq->ring.desc_count)) {
295                         rx_id = 0;
296                         cq->last_color = cq->last_color ? 0 : 1;
297                 }
298
299                 /* Prefetch next mbuf & desc while processing current one */
300                 cqd_ptr = (struct cq_desc *)(cq->ring.descs) + rx_id;
301                 rte_enic_prefetch(cqd_ptr);
302                 rte_enic_prefetch(rq->mbuf_ring[rx_id]);
303                 rte_enic_prefetch((struct rq_enet_desc *)(rq->ring.descs)
304                                  + rx_id);
305
306                 /* Push descriptor for newly allocated mbuf */
307                 dma_addr = (dma_addr_t)(nmb->buf_physaddr
308                            + RTE_PKTMBUF_HEADROOM);
309                 rqd_ptr->address = rte_cpu_to_le_64(dma_addr);
310                 rqd_ptr->length_type = cpu_to_le16(nmb->buf_len
311                                        - RTE_PKTMBUF_HEADROOM);
312
313                 /* Fill in the rest of the mbuf */
314                 rxmb->data_off = RTE_PKTMBUF_HEADROOM;
315                 rxmb->nb_segs = 1;
316                 rxmb->next = NULL;
317                 rxmb->port = enic->port_id;
318                 if (!packet_error) {
319                         rxmb->pkt_len = enic_cq_rx_desc_n_bytes(&cqd);
320                         rxmb->packet_type = enic_cq_rx_flags_to_pkt_type(&cqd);
321                         enic_cq_rx_to_pkt_flags(&cqd, rxmb);
322                 } else {
323                         rxmb->pkt_len = 0;
324                         rxmb->packet_type = 0;
325                         rxmb->ol_flags = 0;
326                 }
327                 rxmb->data_len = rxmb->pkt_len;
328
329                 /* prefetch mbuf data for caller */
330                 rte_packet_prefetch(RTE_PTR_ADD(rxmb->buf_addr,
331                                     RTE_PKTMBUF_HEADROOM));
332
333                 /* store the mbuf address into the next entry of the array */
334                 rx_pkts[nb_rx++] = rxmb;
335         }
336
337         nb_hold += nb_rx;
338         cq->to_clean = rx_id;
339
340         if (nb_hold > rq->rx_free_thresh) {
341                 rq->posted_index = enic_ring_add(rq->ring.desc_count,
342                                 rq->posted_index, nb_hold);
343                 nb_hold = 0;
344                 rte_mb();
345                 iowrite32(rq->posted_index, &rq->ctrl->posted_index);
346         }
347
348         rq->rx_nb_hold = nb_hold;
349
350         return nb_rx;
351 }