xref: /dpdk/drivers/net/fm10k/fm10k_rxtx.c (revision daa02b5c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2013-2016 Intel Corporation
3  */
4 
5 #include <inttypes.h>
6 
7 #include <ethdev_driver.h>
8 #include <rte_common.h>
9 #include <rte_net.h>
10 #include "fm10k.h"
11 #include "base/fm10k_type.h"
12 
13 #ifdef RTE_PMD_PACKET_PREFETCH
14 #define rte_packet_prefetch(p)  rte_prefetch1(p)
15 #else
16 #define rte_packet_prefetch(p)  do {} while (0)
17 #endif
18 
19 #ifdef RTE_ETHDEV_DEBUG_RX
dump_rxd(union fm10k_rx_desc * rxd)20 static inline void dump_rxd(union fm10k_rx_desc *rxd)
21 {
22 	PMD_RX_LOG(DEBUG, "+----------------|----------------+");
23 	PMD_RX_LOG(DEBUG, "|     GLORT      | PKT HDR & TYPE |");
24 	PMD_RX_LOG(DEBUG, "|   0x%08x   |   0x%08x   |", rxd->d.glort,
25 			rxd->d.data);
26 	PMD_RX_LOG(DEBUG, "+----------------|----------------+");
27 	PMD_RX_LOG(DEBUG, "|   VLAN & LEN   |     STATUS     |");
28 	PMD_RX_LOG(DEBUG, "|   0x%08x   |   0x%08x   |", rxd->d.vlan_len,
29 			rxd->d.staterr);
30 	PMD_RX_LOG(DEBUG, "+----------------|----------------+");
31 	PMD_RX_LOG(DEBUG, "|    RESERVED    |    RSS_HASH    |");
32 	PMD_RX_LOG(DEBUG, "|   0x%08x   |   0x%08x   |", 0, rxd->d.rss);
33 	PMD_RX_LOG(DEBUG, "+----------------|----------------+");
34 	PMD_RX_LOG(DEBUG, "|            TIME TAG             |");
35 	PMD_RX_LOG(DEBUG, "|       0x%016"PRIx64"        |", rxd->q.timestamp);
36 	PMD_RX_LOG(DEBUG, "+----------------|----------------+");
37 }
38 #endif
39 
40 #define FM10K_TX_OFFLOAD_MASK (RTE_MBUF_F_TX_VLAN |        \
41 		RTE_MBUF_F_TX_IPV6 |            \
42 		RTE_MBUF_F_TX_IPV4 |            \
43 		RTE_MBUF_F_TX_IP_CKSUM |        \
44 		RTE_MBUF_F_TX_L4_MASK |         \
45 		RTE_MBUF_F_TX_TCP_SEG)
46 
47 #define FM10K_TX_OFFLOAD_NOTSUP_MASK \
48 		(RTE_MBUF_F_TX_OFFLOAD_MASK ^ FM10K_TX_OFFLOAD_MASK)
49 
50 /* @note: When this function is changed, make corresponding change to
51  * fm10k_dev_supported_ptypes_get()
52  */
53 static inline void
rx_desc_to_ol_flags(struct rte_mbuf * m,const union fm10k_rx_desc * d)54 rx_desc_to_ol_flags(struct rte_mbuf *m, const union fm10k_rx_desc *d)
55 {
56 	static const uint32_t
57 		ptype_table[FM10K_RXD_PKTTYPE_MASK >> FM10K_RXD_PKTTYPE_SHIFT]
58 			__rte_cache_aligned = {
59 		[FM10K_PKTTYPE_OTHER] = RTE_PTYPE_L2_ETHER,
60 		[FM10K_PKTTYPE_IPV4] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4,
61 		[FM10K_PKTTYPE_IPV4_EX] = RTE_PTYPE_L2_ETHER |
62 			RTE_PTYPE_L3_IPV4_EXT,
63 		[FM10K_PKTTYPE_IPV6] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6,
64 		[FM10K_PKTTYPE_IPV6_EX] = RTE_PTYPE_L2_ETHER |
65 			RTE_PTYPE_L3_IPV6_EXT,
66 		[FM10K_PKTTYPE_IPV4 | FM10K_PKTTYPE_TCP] = RTE_PTYPE_L2_ETHER |
67 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
68 		[FM10K_PKTTYPE_IPV6 | FM10K_PKTTYPE_TCP] = RTE_PTYPE_L2_ETHER |
69 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
70 		[FM10K_PKTTYPE_IPV4 | FM10K_PKTTYPE_UDP] = RTE_PTYPE_L2_ETHER |
71 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
72 		[FM10K_PKTTYPE_IPV6 | FM10K_PKTTYPE_UDP] = RTE_PTYPE_L2_ETHER |
73 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
74 	};
75 
76 	m->packet_type = ptype_table[(d->w.pkt_info & FM10K_RXD_PKTTYPE_MASK)
77 						>> FM10K_RXD_PKTTYPE_SHIFT];
78 
79 	if (d->w.pkt_info & FM10K_RXD_RSSTYPE_MASK)
80 		m->ol_flags |= RTE_MBUF_F_RX_RSS_HASH;
81 
82 	if (unlikely((d->d.staterr &
83 		(FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)) ==
84 		(FM10K_RXD_STATUS_IPCS | FM10K_RXD_STATUS_IPE)))
85 		m->ol_flags |= RTE_MBUF_F_RX_IP_CKSUM_BAD;
86 	else
87 		m->ol_flags |= RTE_MBUF_F_RX_IP_CKSUM_GOOD;
88 
89 	if (unlikely((d->d.staterr &
90 		(FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)) ==
91 		(FM10K_RXD_STATUS_L4CS | FM10K_RXD_STATUS_L4E)))
92 		m->ol_flags |= RTE_MBUF_F_RX_L4_CKSUM_BAD;
93 	else
94 		m->ol_flags |= RTE_MBUF_F_RX_L4_CKSUM_GOOD;
95 }
96 
97 uint16_t
fm10k_recv_pkts(void * rx_queue,struct rte_mbuf ** rx_pkts,uint16_t nb_pkts)98 fm10k_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
99 	uint16_t nb_pkts)
100 {
101 	struct rte_mbuf *mbuf;
102 	union fm10k_rx_desc desc;
103 	struct fm10k_rx_queue *q = rx_queue;
104 	uint16_t count = 0;
105 	int alloc = 0;
106 	uint16_t next_dd;
107 	int ret;
108 
109 	next_dd = q->next_dd;
110 
111 	nb_pkts = RTE_MIN(nb_pkts, q->alloc_thresh);
112 	for (count = 0; count < nb_pkts; ++count) {
113 		if (!(q->hw_ring[next_dd].d.staterr & FM10K_RXD_STATUS_DD))
114 			break;
115 		mbuf = q->sw_ring[next_dd];
116 		desc = q->hw_ring[next_dd];
117 #ifdef RTE_ETHDEV_DEBUG_RX
118 		dump_rxd(&desc);
119 #endif
120 		rte_pktmbuf_pkt_len(mbuf) = desc.w.length;
121 		rte_pktmbuf_data_len(mbuf) = desc.w.length;
122 
123 		mbuf->ol_flags = 0;
124 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
125 		rx_desc_to_ol_flags(mbuf, &desc);
126 #endif
127 
128 		mbuf->hash.rss = desc.d.rss;
129 		/**
130 		 * Packets in fm10k device always carry at least one VLAN tag.
131 		 * For those packets coming in without VLAN tag,
132 		 * the port default VLAN tag will be used.
133 		 * So, always RTE_MBUF_F_RX_VLAN flag is set and vlan_tci
134 		 * is valid for each RX packet's mbuf.
135 		 */
136 		mbuf->ol_flags |= RTE_MBUF_F_RX_VLAN | RTE_MBUF_F_RX_VLAN_STRIPPED;
137 		mbuf->vlan_tci = desc.w.vlan;
138 		/**
139 		 * mbuf->vlan_tci_outer is an idle field in fm10k driver,
140 		 * so it can be selected to store sglort value.
141 		 */
142 		if (q->rx_ftag_en)
143 			mbuf->vlan_tci_outer = rte_le_to_cpu_16(desc.w.sglort);
144 
145 		rx_pkts[count] = mbuf;
146 		if (++next_dd == q->nb_desc) {
147 			next_dd = 0;
148 			alloc = 1;
149 		}
150 
151 		/* Prefetch next mbuf while processing current one. */
152 		rte_prefetch0(q->sw_ring[next_dd]);
153 
154 		/*
155 		 * When next RX descriptor is on a cache-line boundary,
156 		 * prefetch the next 4 RX descriptors and the next 8 pointers
157 		 * to mbufs.
158 		 */
159 		if ((next_dd & 0x3) == 0) {
160 			rte_prefetch0(&q->hw_ring[next_dd]);
161 			rte_prefetch0(&q->sw_ring[next_dd]);
162 		}
163 	}
164 
165 	q->next_dd = next_dd;
166 
167 	if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
168 		ret = rte_mempool_get_bulk(q->mp,
169 					(void **)&q->sw_ring[q->next_alloc],
170 					q->alloc_thresh);
171 
172 		if (unlikely(ret != 0)) {
173 			uint16_t port = q->port_id;
174 			PMD_RX_LOG(ERR, "Failed to alloc mbuf");
175 			/*
176 			 * Need to restore next_dd if we cannot allocate new
177 			 * buffers to replenish the old ones.
178 			 */
179 			q->next_dd = (q->next_dd + q->nb_desc - count) %
180 								q->nb_desc;
181 			rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
182 			return 0;
183 		}
184 
185 		for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
186 			mbuf = q->sw_ring[q->next_alloc];
187 
188 			/* setup static mbuf fields */
189 			fm10k_pktmbuf_reset(mbuf, q->port_id);
190 
191 			/* write descriptor */
192 			desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
193 			desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
194 			q->hw_ring[q->next_alloc] = desc;
195 		}
196 		FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
197 		q->next_trigger += q->alloc_thresh;
198 		if (q->next_trigger >= q->nb_desc) {
199 			q->next_trigger = q->alloc_thresh - 1;
200 			q->next_alloc = 0;
201 		}
202 	}
203 
204 	return count;
205 }
206 
207 uint16_t
fm10k_recv_scattered_pkts(void * rx_queue,struct rte_mbuf ** rx_pkts,uint16_t nb_pkts)208 fm10k_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
209 				uint16_t nb_pkts)
210 {
211 	struct rte_mbuf *mbuf;
212 	union fm10k_rx_desc desc;
213 	struct fm10k_rx_queue *q = rx_queue;
214 	uint16_t count = 0;
215 	uint16_t nb_rcv, nb_seg;
216 	int alloc = 0;
217 	uint16_t next_dd;
218 	struct rte_mbuf *first_seg = q->pkt_first_seg;
219 	struct rte_mbuf *last_seg = q->pkt_last_seg;
220 	int ret;
221 
222 	next_dd = q->next_dd;
223 	nb_rcv = 0;
224 
225 	nb_seg = RTE_MIN(nb_pkts, q->alloc_thresh);
226 	for (count = 0; count < nb_seg; count++) {
227 		if (!(q->hw_ring[next_dd].d.staterr & FM10K_RXD_STATUS_DD))
228 			break;
229 		mbuf = q->sw_ring[next_dd];
230 		desc = q->hw_ring[next_dd];
231 #ifdef RTE_ETHDEV_DEBUG_RX
232 		dump_rxd(&desc);
233 #endif
234 
235 		if (++next_dd == q->nb_desc) {
236 			next_dd = 0;
237 			alloc = 1;
238 		}
239 
240 		/* Prefetch next mbuf while processing current one. */
241 		rte_prefetch0(q->sw_ring[next_dd]);
242 
243 		/*
244 		 * When next RX descriptor is on a cache-line boundary,
245 		 * prefetch the next 4 RX descriptors and the next 8 pointers
246 		 * to mbufs.
247 		 */
248 		if ((next_dd & 0x3) == 0) {
249 			rte_prefetch0(&q->hw_ring[next_dd]);
250 			rte_prefetch0(&q->sw_ring[next_dd]);
251 		}
252 
253 		/* Fill data length */
254 		rte_pktmbuf_data_len(mbuf) = desc.w.length;
255 
256 		/*
257 		 * If this is the first buffer of the received packet,
258 		 * set the pointer to the first mbuf of the packet and
259 		 * initialize its context.
260 		 * Otherwise, update the total length and the number of segments
261 		 * of the current scattered packet, and update the pointer to
262 		 * the last mbuf of the current packet.
263 		 */
264 		if (!first_seg) {
265 			first_seg = mbuf;
266 			first_seg->pkt_len = desc.w.length;
267 		} else {
268 			first_seg->pkt_len =
269 					(uint16_t)(first_seg->pkt_len +
270 					rte_pktmbuf_data_len(mbuf));
271 			first_seg->nb_segs++;
272 			last_seg->next = mbuf;
273 		}
274 
275 		/*
276 		 * If this is not the last buffer of the received packet,
277 		 * update the pointer to the last mbuf of the current scattered
278 		 * packet and continue to parse the RX ring.
279 		 */
280 		if (!(desc.d.staterr & FM10K_RXD_STATUS_EOP)) {
281 			last_seg = mbuf;
282 			continue;
283 		}
284 
285 		first_seg->ol_flags = 0;
286 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
287 		rx_desc_to_ol_flags(first_seg, &desc);
288 #endif
289 		first_seg->hash.rss = desc.d.rss;
290 		/**
291 		 * Packets in fm10k device always carry at least one VLAN tag.
292 		 * For those packets coming in without VLAN tag,
293 		 * the port default VLAN tag will be used.
294 		 * So, always RTE_MBUF_F_RX_VLAN flag is set and vlan_tci
295 		 * is valid for each RX packet's mbuf.
296 		 */
297 		first_seg->ol_flags |= RTE_MBUF_F_RX_VLAN | RTE_MBUF_F_RX_VLAN_STRIPPED;
298 		first_seg->vlan_tci = desc.w.vlan;
299 		/**
300 		 * mbuf->vlan_tci_outer is an idle field in fm10k driver,
301 		 * so it can be selected to store sglort value.
302 		 */
303 		if (q->rx_ftag_en)
304 			first_seg->vlan_tci_outer =
305 				rte_le_to_cpu_16(desc.w.sglort);
306 
307 		/* Prefetch data of first segment, if configured to do so. */
308 		rte_packet_prefetch((char *)first_seg->buf_addr +
309 			first_seg->data_off);
310 
311 		/*
312 		 * Store the mbuf address into the next entry of the array
313 		 * of returned packets.
314 		 */
315 		rx_pkts[nb_rcv++] = first_seg;
316 
317 		/*
318 		 * Setup receipt context for a new packet.
319 		 */
320 		first_seg = NULL;
321 	}
322 
323 	q->next_dd = next_dd;
324 
325 	if ((q->next_dd > q->next_trigger) || (alloc == 1)) {
326 		ret = rte_mempool_get_bulk(q->mp,
327 					(void **)&q->sw_ring[q->next_alloc],
328 					q->alloc_thresh);
329 
330 		if (unlikely(ret != 0)) {
331 			uint16_t port = q->port_id;
332 			PMD_RX_LOG(ERR, "Failed to alloc mbuf");
333 			/*
334 			 * Need to restore next_dd if we cannot allocate new
335 			 * buffers to replenish the old ones.
336 			 */
337 			q->next_dd = (q->next_dd + q->nb_desc - count) %
338 								q->nb_desc;
339 			rte_eth_devices[port].data->rx_mbuf_alloc_failed++;
340 			return 0;
341 		}
342 
343 		for (; q->next_alloc <= q->next_trigger; ++q->next_alloc) {
344 			mbuf = q->sw_ring[q->next_alloc];
345 
346 			/* setup static mbuf fields */
347 			fm10k_pktmbuf_reset(mbuf, q->port_id);
348 
349 			/* write descriptor */
350 			desc.q.pkt_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
351 			desc.q.hdr_addr = MBUF_DMA_ADDR_DEFAULT(mbuf);
352 			q->hw_ring[q->next_alloc] = desc;
353 		}
354 		FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_trigger);
355 		q->next_trigger += q->alloc_thresh;
356 		if (q->next_trigger >= q->nb_desc) {
357 			q->next_trigger = q->alloc_thresh - 1;
358 			q->next_alloc = 0;
359 		}
360 	}
361 
362 	q->pkt_first_seg = first_seg;
363 	q->pkt_last_seg = last_seg;
364 
365 	return nb_rcv;
366 }
367 
368 uint32_t
fm10k_dev_rx_queue_count(void * rx_queue)369 fm10k_dev_rx_queue_count(void *rx_queue)
370 {
371 #define FM10K_RXQ_SCAN_INTERVAL 4
372 	volatile union fm10k_rx_desc *rxdp;
373 	struct fm10k_rx_queue *rxq;
374 	uint16_t desc = 0;
375 
376 	rxq = rx_queue;
377 	rxdp = &rxq->hw_ring[rxq->next_dd];
378 	while ((desc < rxq->nb_desc) &&
379 		rxdp->w.status & rte_cpu_to_le_16(FM10K_RXD_STATUS_DD)) {
380 		/**
381 		 * Check the DD bit of a rx descriptor of each group of 4 desc,
382 		 * to avoid checking too frequently and downgrading performance
383 		 * too much.
384 		 */
385 		desc += FM10K_RXQ_SCAN_INTERVAL;
386 		rxdp += FM10K_RXQ_SCAN_INTERVAL;
387 		if (rxq->next_dd + desc >= rxq->nb_desc)
388 			rxdp = &rxq->hw_ring[rxq->next_dd + desc -
389 				rxq->nb_desc];
390 	}
391 
392 	return desc;
393 }
394 
395 int
fm10k_dev_rx_descriptor_status(void * rx_queue,uint16_t offset)396 fm10k_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
397 {
398 	volatile union fm10k_rx_desc *rxdp;
399 	struct fm10k_rx_queue *rxq = rx_queue;
400 	uint16_t nb_hold, trigger_last;
401 	uint16_t desc;
402 	int ret;
403 
404 	if (unlikely(offset >= rxq->nb_desc)) {
405 		PMD_DRV_LOG(ERR, "Invalid RX descriptor offset %u", offset);
406 		return 0;
407 	}
408 
409 	if (rxq->next_trigger < rxq->alloc_thresh)
410 		trigger_last = rxq->next_trigger +
411 					rxq->nb_desc - rxq->alloc_thresh;
412 	else
413 		trigger_last = rxq->next_trigger - rxq->alloc_thresh;
414 
415 	if (rxq->next_dd < trigger_last)
416 		nb_hold = rxq->next_dd + rxq->nb_desc - trigger_last;
417 	else
418 		nb_hold = rxq->next_dd - trigger_last;
419 
420 	if (offset >= rxq->nb_desc - nb_hold)
421 		return RTE_ETH_RX_DESC_UNAVAIL;
422 
423 	desc = rxq->next_dd + offset;
424 	if (desc >= rxq->nb_desc)
425 		desc -= rxq->nb_desc;
426 
427 	rxdp = &rxq->hw_ring[desc];
428 
429 	ret = !!(rxdp->w.status &
430 			rte_cpu_to_le_16(FM10K_RXD_STATUS_DD));
431 
432 	return ret;
433 }
434 
435 int
fm10k_dev_tx_descriptor_status(void * tx_queue,uint16_t offset)436 fm10k_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
437 {
438 	volatile struct fm10k_tx_desc *txdp;
439 	struct fm10k_tx_queue *txq = tx_queue;
440 	uint16_t desc;
441 	uint16_t next_rs = txq->nb_desc;
442 	struct fifo rs_tracker = txq->rs_tracker;
443 	struct fifo *r = &rs_tracker;
444 
445 	if (unlikely(offset >= txq->nb_desc))
446 		return -EINVAL;
447 
448 	desc = txq->next_free + offset;
449 	/* go to next desc that has the RS bit */
450 	desc = (desc / txq->rs_thresh + 1) *
451 		txq->rs_thresh - 1;
452 
453 	if (desc >= txq->nb_desc) {
454 		desc -= txq->nb_desc;
455 		if (desc >= txq->nb_desc)
456 			desc -= txq->nb_desc;
457 	}
458 
459 	r->head = r->list;
460 	for ( ; r->head != r->endp; ) {
461 		if (*r->head >= desc && *r->head < next_rs)
462 			next_rs = *r->head;
463 		++r->head;
464 	}
465 
466 	txdp = &txq->hw_ring[next_rs];
467 	if (txdp->flags & FM10K_TXD_FLAG_DONE)
468 		return RTE_ETH_TX_DESC_DONE;
469 
470 	return RTE_ETH_TX_DESC_FULL;
471 }
472 
473 /*
474  * Free multiple TX mbuf at a time if they are in the same pool
475  *
476  * @txep: software desc ring index that starts to free
477  * @num: number of descs to free
478  *
479  */
tx_free_bulk_mbuf(struct rte_mbuf ** txep,int num)480 static inline void tx_free_bulk_mbuf(struct rte_mbuf **txep, int num)
481 {
482 	struct rte_mbuf *m, *free[RTE_FM10K_TX_MAX_FREE_BUF_SZ];
483 	int i;
484 	int nb_free = 0;
485 
486 	if (unlikely(num == 0))
487 		return;
488 
489 	m = rte_pktmbuf_prefree_seg(txep[0]);
490 	if (likely(m != NULL)) {
491 		free[0] = m;
492 		nb_free = 1;
493 		for (i = 1; i < num; i++) {
494 			m = rte_pktmbuf_prefree_seg(txep[i]);
495 			if (likely(m != NULL)) {
496 				if (likely(m->pool == free[0]->pool))
497 					free[nb_free++] = m;
498 				else {
499 					rte_mempool_put_bulk(free[0]->pool,
500 							(void *)free, nb_free);
501 					free[0] = m;
502 					nb_free = 1;
503 				}
504 			}
505 			txep[i] = NULL;
506 		}
507 		rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
508 	} else {
509 		for (i = 1; i < num; i++) {
510 			m = rte_pktmbuf_prefree_seg(txep[i]);
511 			if (m != NULL)
512 				rte_mempool_put(m->pool, m);
513 			txep[i] = NULL;
514 		}
515 	}
516 }
517 
tx_free_descriptors(struct fm10k_tx_queue * q)518 static inline void tx_free_descriptors(struct fm10k_tx_queue *q)
519 {
520 	uint16_t next_rs, count = 0;
521 
522 	next_rs = fifo_peek(&q->rs_tracker);
523 	if (!(q->hw_ring[next_rs].flags & FM10K_TXD_FLAG_DONE))
524 		return;
525 
526 	/* the DONE flag is set on this descriptor so remove the ID
527 	 * from the RS bit tracker and free the buffers */
528 	fifo_remove(&q->rs_tracker);
529 
530 	/* wrap around? if so, free buffers from last_free up to but NOT
531 	 * including nb_desc */
532 	if (q->last_free > next_rs) {
533 		count = q->nb_desc - q->last_free;
534 		tx_free_bulk_mbuf(&q->sw_ring[q->last_free], count);
535 		q->last_free = 0;
536 	}
537 
538 	/* adjust free descriptor count before the next loop */
539 	q->nb_free += count + (next_rs + 1 - q->last_free);
540 
541 	/* free buffers from last_free, up to and including next_rs */
542 	if (q->last_free <= next_rs) {
543 		count = next_rs - q->last_free + 1;
544 		tx_free_bulk_mbuf(&q->sw_ring[q->last_free], count);
545 		q->last_free += count;
546 	}
547 
548 	if (q->last_free == q->nb_desc)
549 		q->last_free = 0;
550 }
551 
tx_xmit_pkt(struct fm10k_tx_queue * q,struct rte_mbuf * mb)552 static inline void tx_xmit_pkt(struct fm10k_tx_queue *q, struct rte_mbuf *mb)
553 {
554 	uint16_t last_id;
555 	uint8_t flags, hdrlen;
556 
557 	/* always set the LAST flag on the last descriptor used to
558 	 * transmit the packet */
559 	flags = FM10K_TXD_FLAG_LAST;
560 	last_id = q->next_free + mb->nb_segs - 1;
561 	if (last_id >= q->nb_desc)
562 		last_id = last_id - q->nb_desc;
563 
564 	/* but only set the RS flag on the last descriptor if rs_thresh
565 	 * descriptors will be used since the RS flag was last set */
566 	if ((q->nb_used + mb->nb_segs) >= q->rs_thresh) {
567 		flags |= FM10K_TXD_FLAG_RS;
568 		fifo_insert(&q->rs_tracker, last_id);
569 		q->nb_used = 0;
570 	} else {
571 		q->nb_used = q->nb_used + mb->nb_segs;
572 	}
573 
574 	q->nb_free -= mb->nb_segs;
575 
576 	q->hw_ring[q->next_free].flags = 0;
577 	if (q->tx_ftag_en)
578 		q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_FTAG;
579 	/* set checksum flags on first descriptor of packet. SCTP checksum
580 	 * offload is not supported, but we do not explicitly check for this
581 	 * case in favor of greatly simplified processing. */
582 	if (mb->ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_L4_MASK | RTE_MBUF_F_TX_TCP_SEG))
583 		q->hw_ring[q->next_free].flags |= FM10K_TXD_FLAG_CSUM;
584 
585 	/* set vlan if requested */
586 	if (mb->ol_flags & RTE_MBUF_F_TX_VLAN)
587 		q->hw_ring[q->next_free].vlan = mb->vlan_tci;
588 	else
589 		q->hw_ring[q->next_free].vlan = 0;
590 
591 	q->sw_ring[q->next_free] = mb;
592 	q->hw_ring[q->next_free].buffer_addr =
593 			rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
594 	q->hw_ring[q->next_free].buflen =
595 			rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
596 
597 	if (mb->ol_flags & RTE_MBUF_F_TX_TCP_SEG) {
598 		hdrlen = mb->l2_len + mb->l3_len + mb->l4_len;
599 		hdrlen += (mb->ol_flags & RTE_MBUF_F_TX_TUNNEL_MASK) ?
600 			  mb->outer_l2_len + mb->outer_l3_len : 0;
601 		if (q->hw_ring[q->next_free].flags & FM10K_TXD_FLAG_FTAG)
602 			hdrlen += sizeof(struct fm10k_ftag);
603 
604 		if (likely((hdrlen >= FM10K_TSO_MIN_HEADERLEN) &&
605 				(hdrlen <= FM10K_TSO_MAX_HEADERLEN) &&
606 				(mb->tso_segsz >= FM10K_TSO_MINMSS))) {
607 			q->hw_ring[q->next_free].mss = mb->tso_segsz;
608 			q->hw_ring[q->next_free].hdrlen = hdrlen;
609 		}
610 	}
611 
612 	if (++q->next_free == q->nb_desc)
613 		q->next_free = 0;
614 
615 	/* fill up the rings */
616 	for (mb = mb->next; mb != NULL; mb = mb->next) {
617 		q->sw_ring[q->next_free] = mb;
618 		q->hw_ring[q->next_free].buffer_addr =
619 				rte_cpu_to_le_64(MBUF_DMA_ADDR(mb));
620 		q->hw_ring[q->next_free].buflen =
621 				rte_cpu_to_le_16(rte_pktmbuf_data_len(mb));
622 		q->hw_ring[q->next_free].flags = 0;
623 		if (++q->next_free == q->nb_desc)
624 			q->next_free = 0;
625 	}
626 
627 	q->hw_ring[last_id].flags |= flags;
628 }
629 
630 uint16_t
fm10k_xmit_pkts(void * tx_queue,struct rte_mbuf ** tx_pkts,uint16_t nb_pkts)631 fm10k_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
632 	uint16_t nb_pkts)
633 {
634 	struct fm10k_tx_queue *q = tx_queue;
635 	struct rte_mbuf *mb;
636 	uint16_t count;
637 
638 	for (count = 0; count < nb_pkts; ++count) {
639 		mb = tx_pkts[count];
640 
641 		/* running low on descriptors? try to free some... */
642 		if (q->nb_free < q->free_thresh)
643 			tx_free_descriptors(q);
644 
645 		/* make sure there are enough free descriptors to transmit the
646 		 * entire packet before doing anything */
647 		if (q->nb_free < mb->nb_segs)
648 			break;
649 
650 		/* sanity check to make sure the mbuf is valid */
651 		if ((mb->nb_segs == 0) ||
652 		    ((mb->nb_segs > 1) && (mb->next == NULL)))
653 			break;
654 
655 		/* process the packet */
656 		tx_xmit_pkt(q, mb);
657 	}
658 
659 	/* update the tail pointer if any packets were processed */
660 	if (likely(count > 0))
661 		FM10K_PCI_REG_WRITE(q->tail_ptr, q->next_free);
662 
663 	return count;
664 }
665 
666 uint16_t
fm10k_prep_pkts(__rte_unused void * tx_queue,struct rte_mbuf ** tx_pkts,uint16_t nb_pkts)667 fm10k_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
668 		uint16_t nb_pkts)
669 {
670 	int i, ret;
671 	struct rte_mbuf *m;
672 
673 	for (i = 0; i < nb_pkts; i++) {
674 		m = tx_pkts[i];
675 
676 		if ((m->ol_flags & RTE_MBUF_F_TX_TCP_SEG) &&
677 				(m->tso_segsz < FM10K_TSO_MINMSS)) {
678 			rte_errno = EINVAL;
679 			return i;
680 		}
681 
682 		if (m->ol_flags & FM10K_TX_OFFLOAD_NOTSUP_MASK) {
683 			rte_errno = ENOTSUP;
684 			return i;
685 		}
686 
687 #ifdef RTE_ETHDEV_DEBUG_TX
688 		ret = rte_validate_tx_offload(m);
689 		if (ret != 0) {
690 			rte_errno = -ret;
691 			return i;
692 		}
693 #endif
694 		ret = rte_net_intel_cksum_prepare(m);
695 		if (ret != 0) {
696 			rte_errno = -ret;
697 			return i;
698 		}
699 	}
700 
701 	return i;
702 }
703