4 * Copyright(c) 2013-2015 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.
36 #include <rte_ethdev.h>
37 #include <rte_common.h>
39 #include "base/fm10k_type.h"
41 #ifdef RTE_PMD_PACKET_PREFETCH
42 #define rte_packet_prefetch(p) rte_prefetch1(p)
44 #define rte_packet_prefetch(p) do {} while (0)
47 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
48 static inline void dump_rxd(union fm10k_rx_desc *rxd)
50 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
51 PMD_RX_LOG(DEBUG, "| GLORT | PKT HDR & TYPE |");
52 PMD_RX_LOG(DEBUG, "| 0x%08x | 0x%08x |", rxd->d.glort,
54 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
55 PMD_RX_LOG(DEBUG, "| VLAN & LEN | STATUS |");
56 PMD_RX_LOG(DEBUG, "| 0x%08x | 0x%08x |", rxd->d.vlan_len,
58 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
59 PMD_RX_LOG(DEBUG, "| RESERVED | RSS_HASH |");
60 PMD_RX_LOG(DEBUG, "| 0x%08x | 0x%08x |", 0, rxd->d.rss);
61 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
62 PMD_RX_LOG(DEBUG, "| TIME TAG |");
63 PMD_RX_LOG(DEBUG, "| 0x%016"PRIx64" |", rxd->q.timestamp);
64 PMD_RX_LOG(DEBUG, "+----------------|----------------+");
69 rx_desc_to_ol_flags(struct rte_mbuf *m, const union fm10k_rx_desc *d)
72 static const uint16_t pt_lut[] = { 0,
73 PKT_RX_IPV4_HDR, PKT_RX_IPV4_HDR_EXT,
74 PKT_RX_IPV6_HDR, PKT_RX_IPV6_HDR_EXT,
78 if (d->w.pkt_info & FM10K_RXD_RSSTYPE_MASK)
79 m->ol_flags |= PKT_RX_RSS_HASH;
81 if (unlikely((d->d.staterr &
82 (FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)) ==
83 (FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)))
84 m->ol_flags |= PKT_RX_IP_CKSUM_BAD;
86 if (unlikely((d->d.staterr &
87 (FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)) ==
88 (FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)))
89 m->ol_flags |= PKT_RX_L4_CKSUM_BAD;
91 if (d->d.staterr & FM10K_RXD_STATUS_VEXT)
92 m->ol_flags |= PKT_RX_VLAN_PKT;
94 if (unlikely(d->d.staterr & FM10K_RXD_STATUS_HBO))
95 m->ol_flags |= PKT_RX_HBUF_OVERFLOW;
97 if (unlikely(d->d.staterr & FM10K_RXD_STATUS_RXE))
98 m->ol_flags |= PKT_RX_RECIP_ERR;
100 ptype = (d->d.data & FM10K_RXD_PKTTYPE_MASK_L3) >>
101 FM10K_RXD_PKTTYPE_SHIFT;
102 m->ol_flags |= pt_lut[(uint8_t)ptype];
106 fm10k_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
109 struct rte_mbuf *mbuf;
110 union fm10k_rx_desc desc;
111 struct fm10k_rx_queue *q = rx_queue;
117 next_dd = q->next_dd;
119 nb_pkts = RTE_MIN(nb_pkts, q->alloc_thresh);
120 for (count = 0; count < nb_pkts; ++count) {
121 mbuf = q->sw_ring[next_dd];
122 desc = q->hw_ring[next_dd];
123 if (!(desc.d.staterr & FM10K_RXD_STATUS_DD))
125 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
128 rte_pktmbuf_pkt_len(mbuf) = desc.w.length;
129 rte_pktmbuf_data_len(mbuf) = desc.w.length;
132 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
133 rx_desc_to_ol_flags(mbuf, &desc);
136 mbuf->hash.rss = desc.d.rss;
138 rx_pkts[count] = mbuf;
139 if (++next_dd == q->nb_desc) {
144 /* Prefetch next mbuf while processing current one. */
145 rte_prefetch0(q->sw_ring[next_dd]);
148 * When next RX descriptor is on a cache-line boundary,
149 * prefetch the next 4 RX descriptors and the next 8 pointers
152 if ((next_dd & 0x3) == 0) {
153 rte_prefetch0(&q->hw_ring[next_dd]);
154 rte_prefetch0(&q->sw_ring[next_dd]);
158 q->next_dd = next_dd;
160 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
161 ret = rte_mempool_get_bulk(q->mp,
162 (void **)&q->sw_ring[q->next_alloc],
165 if (unlikely(ret != 0)) {
166 uint8_t port = q->port_id;
167 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
169 * Need to restore next_dd if we cannot allocate new
170 * buffers to replenish the old ones.
172 q->next_dd = (q->next_dd + q->nb_desc - count) %
174 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
178 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
179 mbuf = q->sw_ring[q->next_alloc];
181 /* setup static mbuf fields */
182 fm10k_pktmbuf_reset(mbuf, q->port_id);
184 /* write descriptor */
185 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
186 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
187 q->hw_ring[q->next_alloc] = desc;
189 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
190 q->next_trigger += q->alloc_thresh;
191 if (q->next_trigger >= q->nb_desc) {
192 q->next_trigger = q->alloc_thresh - 1;
201 fm10k_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
204 struct rte_mbuf *mbuf;
205 union fm10k_rx_desc desc;
206 struct fm10k_rx_queue *q = rx_queue;
208 uint16_t nb_rcv, nb_seg;
211 struct rte_mbuf *first_seg = q->pkt_first_seg;
212 struct rte_mbuf *last_seg = q->pkt_last_seg;
215 next_dd = q->next_dd;
218 nb_seg = RTE_MIN(nb_pkts, q->alloc_thresh);
219 for (count = 0; count < nb_seg; count++) {
220 mbuf = q->sw_ring[next_dd];
221 desc = q->hw_ring[next_dd];
222 if (!(desc.d.staterr & FM10K_RXD_STATUS_DD))
224 #ifdef RTE_LIBRTE_FM10K_DEBUG_RX
228 if (++next_dd == q->nb_desc) {
233 /* Prefetch next mbuf while processing current one. */
234 rte_prefetch0(q->sw_ring[next_dd]);
237 * When next RX descriptor is on a cache-line boundary,
238 * prefetch the next 4 RX descriptors and the next 8 pointers
241 if ((next_dd & 0x3) == 0) {
242 rte_prefetch0(&q->hw_ring[next_dd]);
243 rte_prefetch0(&q->sw_ring[next_dd]);
246 /* Fill data length */
247 rte_pktmbuf_data_len(mbuf) = desc.w.length;
250 * If this is the first buffer of the received packet,
251 * set the pointer to the first mbuf of the packet and
252 * initialize its context.
253 * Otherwise, update the total length and the number of segments
254 * of the current scattered packet, and update the pointer to
255 * the last mbuf of the current packet.
259 first_seg->pkt_len = desc.w.length;
262 (uint16_t)(first_seg->pkt_len +
263 rte_pktmbuf_data_len(mbuf));
264 first_seg->nb_segs++;
265 last_seg->next = mbuf;
269 * If this is not the last buffer of the received packet,
270 * update the pointer to the last mbuf of the current scattered
271 * packet and continue to parse the RX ring.
273 if (!(desc.d.staterr & FM10K_RXD_STATUS_EOP)) {
278 first_seg->ol_flags = 0;
279 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
280 rx_desc_to_ol_flags(first_seg, &desc);
282 first_seg->hash.rss = desc.d.rss;
284 /* Prefetch data of first segment, if configured to do so. */
285 rte_packet_prefetch((char *)first_seg->buf_addr +
286 first_seg->data_off);
289 * Store the mbuf address into the next entry of the array
290 * of returned packets.
292 rx_pkts[nb_rcv++] = first_seg;
295 * Setup receipt context for a new packet.
300 q->next_dd = next_dd;
302 if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
303 ret = rte_mempool_get_bulk(q->mp,
304 (void **)&q->sw_ring[q->next_alloc],
307 if (unlikely(ret != 0)) {
308 uint8_t port = q->port_id;
309 PMD_RX_LOG(ERR, "Failed to alloc mbuf");
311 * Need to restore next_dd if we cannot allocate new
312 * buffers to replenish the old ones.
314 q->next_dd = (q->next_dd + q->nb_desc - count) %
316 rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
320 for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
321 mbuf = q->sw_ring[q->next_alloc];
323 /* setup static mbuf fields */
324 fm10k_pktmbuf_reset(mbuf, q->port_id);
326 /* write descriptor */
327 desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
328 desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
329 q->hw_ring[q->next_alloc] = desc;
331 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
332 q->next_trigger += q->alloc_thresh;
333 if (q->next_trigger >= q->nb_desc) {
334 q->next_trigger = q->alloc_thresh - 1;
339 q->pkt_first_seg = first_seg;
340 q->pkt_last_seg = last_seg;
345 static inline void tx_free_descriptors(struct fm10k_tx_queue *q)
347 uint16_t next_rs, count = 0;
349 next_rs = fifo_peek(&q->rs_tracker);
350 if (!(q->hw_ring[next_rs].flags & FM10K_TXD_FLAG_DONE))
353 /* the DONE flag is set on this descriptor so remove the ID
354 * from the RS bit tracker and free the buffers */
355 fifo_remove(&q->rs_tracker);
357 /* wrap around? if so, free buffers from last_free up to but NOT
358 * including nb_desc */
359 if (q->last_free > next_rs) {
360 count = q->nb_desc - q->last_free;
361 while (q->last_free < q->nb_desc) {
362 rte_pktmbuf_free_seg(q->sw_ring[q->last_free]);
363 q->sw_ring[q->last_free] = NULL;
369 /* adjust free descriptor count before the next loop */
370 q->nb_free += count + (next_rs + 1 - q->last_free);
372 /* free buffers from last_free, up to and including next_rs */
373 while (q->last_free <= next_rs) {
374 rte_pktmbuf_free_seg(q->sw_ring[q->last_free]);
375 q->sw_ring[q->last_free] = NULL;
379 if (q->last_free == q->nb_desc)
383 static inline void tx_xmit_pkt(struct fm10k_tx_queue *q, struct rte_mbuf *mb)
388 /* always set the LAST flag on the last descriptor used to
389 * transmit the packet */
390 flags = FM10K_TXD_FLAG_LAST;
391 last_id = q->next_free + mb->nb_segs - 1;
392 if (last_id >= q->nb_desc)
393 last_id = last_id - q->nb_desc;
395 /* but only set the RS flag on the last descriptor if rs_thresh
396 * descriptors will be used since the RS flag was last set */
397 if ((q->nb_used + mb->nb_segs) >= q->rs_thresh) {
398 flags |= FM10K_TXD_FLAG_RS;
399 fifo_insert(&q->rs_tracker, last_id);
402 q->nb_used = q->nb_used + mb->nb_segs;
405 q->nb_free -= mb->nb_segs;
407 q->hw_ring[q->next_free].flags = 0;
408 /* set checksum flags on first descriptor of packet. SCTP checksum
409 * offload is not supported, but we do not explicitly check for this
410 * case in favor of greatly simplified processing. */
411 if (mb->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK))
412 q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_CSUM;
414 /* set vlan if requested */
415 if (mb->ol_flags & PKT_TX_VLAN_PKT)
416 q->hw_ring[q->next_free].vlan = mb->vlan_tci;
418 q->sw_ring[q->next_free] = mb;
419 q->hw_ring[q->next_free].buffer_addr =
420 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
421 q->hw_ring[q->next_free].buflen =
422 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
423 if (++q->next_free == q->nb_desc)
426 /* fill up the rings */
427 for (mb = mb->next; mb != NULL; mb = mb->next) {
428 q->sw_ring[q->next_free] = mb;
429 q->hw_ring[q->next_free].buffer_addr =
430 rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
431 q->hw_ring[q->next_free].buflen =
432 rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
433 q->hw_ring[q->next_free].flags = 0;
434 if (++q->next_free == q->nb_desc)
438 q->hw_ring[last_id].flags = flags;
442 fm10k_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
445 struct fm10k_tx_queue *q = tx_queue;
449 for (count = 0; count < nb_pkts; ++count) {
452 /* running low on descriptors? try to free some... */
453 if (q->nb_free < q->free_trigger)
454 tx_free_descriptors(q);
456 /* make sure there are enough free descriptors to transmit the
457 * entire packet before doing anything */
458 if (q->nb_free < mb->nb_segs)
461 /* sanity check to make sure the mbuf is valid */
462 if ((mb->nb_segs == 0) ||
463 ((mb->nb_segs > 1) && (mb->next == NULL)))
466 /* process the packet */
470 /* update the tail pointer if any packets were processed */
471 if (likely(count > 0))
472 FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_free);