xref: /dpdk/drivers/net/dpaa2/dpaa2_rxtx.c (revision 2a7bb4fdf61e9edfb7adbaecb50e728b82da9e23)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  *
3  *   Copyright (c) 2016 Freescale Semiconductor, Inc. All rights reserved.
4  *   Copyright 2016-2018 NXP
5  *
6  */
7 
8 #include <time.h>
9 #include <net/if.h>
10 
11 #include <rte_mbuf.h>
12 #include <rte_ethdev_driver.h>
13 #include <rte_malloc.h>
14 #include <rte_memcpy.h>
15 #include <rte_string_fns.h>
16 #include <rte_dev.h>
17 
18 #include <rte_fslmc.h>
19 #include <fslmc_vfio.h>
20 #include <dpaa2_hw_pvt.h>
21 #include <dpaa2_hw_dpio.h>
22 #include <dpaa2_hw_mempool.h>
23 
24 #include "dpaa2_pmd_logs.h"
25 #include "dpaa2_ethdev.h"
26 #include "base/dpaa2_hw_dpni_annot.h"
27 
28 static inline uint32_t __attribute__((hot))
29 dpaa2_dev_rx_parse_slow(struct rte_mbuf *mbuf,
30 			struct dpaa2_annot_hdr *annotation);
31 
32 #define DPAA2_MBUF_TO_CONTIG_FD(_mbuf, _fd, _bpid)  do { \
33 	DPAA2_SET_FD_ADDR(_fd, DPAA2_MBUF_VADDR_TO_IOVA(_mbuf)); \
34 	DPAA2_SET_FD_LEN(_fd, _mbuf->data_len); \
35 	DPAA2_SET_ONLY_FD_BPID(_fd, _bpid); \
36 	DPAA2_SET_FD_OFFSET(_fd, _mbuf->data_off); \
37 	DPAA2_SET_FD_FRC(_fd, 0);		\
38 	DPAA2_RESET_FD_CTRL(_fd);		\
39 	DPAA2_RESET_FD_FLC(_fd);		\
40 } while (0)
41 
42 static inline void __attribute__((hot))
43 dpaa2_dev_rx_parse_new(struct rte_mbuf *m, const struct qbman_fd *fd)
44 {
45 	struct dpaa2_annot_hdr *annotation;
46 	uint16_t frc = DPAA2_GET_FD_FRC_PARSE_SUM(fd);
47 
48 	m->packet_type = RTE_PTYPE_UNKNOWN;
49 	switch (frc) {
50 	case DPAA2_PKT_TYPE_ETHER:
51 		m->packet_type = RTE_PTYPE_L2_ETHER;
52 		break;
53 	case DPAA2_PKT_TYPE_IPV4:
54 		m->packet_type = RTE_PTYPE_L2_ETHER |
55 			RTE_PTYPE_L3_IPV4;
56 		break;
57 	case DPAA2_PKT_TYPE_IPV6:
58 		m->packet_type = RTE_PTYPE_L2_ETHER |
59 			RTE_PTYPE_L3_IPV6;
60 		break;
61 	case DPAA2_PKT_TYPE_IPV4_EXT:
62 		m->packet_type = RTE_PTYPE_L2_ETHER |
63 			RTE_PTYPE_L3_IPV4_EXT;
64 		break;
65 	case DPAA2_PKT_TYPE_IPV6_EXT:
66 		m->packet_type = RTE_PTYPE_L2_ETHER |
67 			RTE_PTYPE_L3_IPV6_EXT;
68 		break;
69 	case DPAA2_PKT_TYPE_IPV4_TCP:
70 		m->packet_type = RTE_PTYPE_L2_ETHER |
71 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP;
72 		break;
73 	case DPAA2_PKT_TYPE_IPV6_TCP:
74 		m->packet_type = RTE_PTYPE_L2_ETHER |
75 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
76 		break;
77 	case DPAA2_PKT_TYPE_IPV4_UDP:
78 		m->packet_type = RTE_PTYPE_L2_ETHER |
79 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP;
80 		break;
81 	case DPAA2_PKT_TYPE_IPV6_UDP:
82 		m->packet_type = RTE_PTYPE_L2_ETHER |
83 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
84 		break;
85 	case DPAA2_PKT_TYPE_IPV4_SCTP:
86 		m->packet_type = RTE_PTYPE_L2_ETHER |
87 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP;
88 		break;
89 	case DPAA2_PKT_TYPE_IPV6_SCTP:
90 		m->packet_type = RTE_PTYPE_L2_ETHER |
91 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
92 		break;
93 	case DPAA2_PKT_TYPE_IPV4_ICMP:
94 		m->packet_type = RTE_PTYPE_L2_ETHER |
95 			RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_ICMP;
96 		break;
97 	case DPAA2_PKT_TYPE_IPV6_ICMP:
98 		m->packet_type = RTE_PTYPE_L2_ETHER |
99 			RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_ICMP;
100 		break;
101 	default:
102 		m->packet_type = dpaa2_dev_rx_parse_slow(m,
103 		  (void *)((size_t)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd))
104 			 + DPAA2_FD_PTA_SIZE));
105 	}
106 	m->hash.rss = fd->simple.flc_hi;
107 	m->ol_flags |= PKT_RX_RSS_HASH;
108 
109 	if (dpaa2_enable_ts == PMD_DPAA2_ENABLE_TS) {
110 		annotation = (struct dpaa2_annot_hdr *)
111 			((size_t)DPAA2_IOVA_TO_VADDR(
112 			DPAA2_GET_FD_ADDR(fd)) + DPAA2_FD_PTA_SIZE);
113 		m->timestamp = annotation->word2;
114 		m->ol_flags |= PKT_RX_TIMESTAMP;
115 		DPAA2_PMD_DP_DEBUG("pkt timestamp:0x%" PRIx64 "", m->timestamp);
116 	}
117 
118 	DPAA2_PMD_DP_DEBUG("HW frc = 0x%x\t packet type =0x%x "
119 		"ol_flags =0x%" PRIx64 "",
120 		frc, m->packet_type, m->ol_flags);
121 }
122 
123 static inline uint32_t __attribute__((hot))
124 dpaa2_dev_rx_parse_slow(struct rte_mbuf *mbuf,
125 			struct dpaa2_annot_hdr *annotation)
126 {
127 	uint32_t pkt_type = RTE_PTYPE_UNKNOWN;
128 	uint16_t *vlan_tci;
129 
130 	DPAA2_PMD_DP_DEBUG("(slow parse)annotation(3)=0x%" PRIx64 "\t"
131 			"(4)=0x%" PRIx64 "\t",
132 			annotation->word3, annotation->word4);
133 
134 	if (BIT_ISSET_AT_POS(annotation->word3, L2_VLAN_1_PRESENT)) {
135 		vlan_tci = rte_pktmbuf_mtod_offset(mbuf, uint16_t *,
136 			(VLAN_TCI_OFFSET_1(annotation->word5) >> 16));
137 		mbuf->vlan_tci = rte_be_to_cpu_16(*vlan_tci);
138 		mbuf->ol_flags |= PKT_RX_VLAN;
139 		pkt_type |= RTE_PTYPE_L2_ETHER_VLAN;
140 	} else if (BIT_ISSET_AT_POS(annotation->word3, L2_VLAN_N_PRESENT)) {
141 		vlan_tci = rte_pktmbuf_mtod_offset(mbuf, uint16_t *,
142 			(VLAN_TCI_OFFSET_1(annotation->word5) >> 16));
143 		mbuf->vlan_tci = rte_be_to_cpu_16(*vlan_tci);
144 		mbuf->ol_flags |= PKT_RX_VLAN | PKT_RX_QINQ;
145 		pkt_type |= RTE_PTYPE_L2_ETHER_QINQ;
146 	}
147 
148 	if (BIT_ISSET_AT_POS(annotation->word3, L2_ARP_PRESENT)) {
149 		pkt_type |= RTE_PTYPE_L2_ETHER_ARP;
150 		goto parse_done;
151 	} else if (BIT_ISSET_AT_POS(annotation->word3, L2_ETH_MAC_PRESENT)) {
152 		pkt_type |= RTE_PTYPE_L2_ETHER;
153 	} else {
154 		goto parse_done;
155 	}
156 
157 	if (BIT_ISSET_AT_POS(annotation->word4, L3_IPV4_1_PRESENT |
158 			     L3_IPV4_N_PRESENT)) {
159 		pkt_type |= RTE_PTYPE_L3_IPV4;
160 		if (BIT_ISSET_AT_POS(annotation->word4, L3_IP_1_OPT_PRESENT |
161 			L3_IP_N_OPT_PRESENT))
162 			pkt_type |= RTE_PTYPE_L3_IPV4_EXT;
163 
164 	} else if (BIT_ISSET_AT_POS(annotation->word4, L3_IPV6_1_PRESENT |
165 		  L3_IPV6_N_PRESENT)) {
166 		pkt_type |= RTE_PTYPE_L3_IPV6;
167 		if (BIT_ISSET_AT_POS(annotation->word4, L3_IP_1_OPT_PRESENT |
168 		    L3_IP_N_OPT_PRESENT))
169 			pkt_type |= RTE_PTYPE_L3_IPV6_EXT;
170 	} else {
171 		goto parse_done;
172 	}
173 
174 	if (BIT_ISSET_AT_POS(annotation->word8, DPAA2_ETH_FAS_L3CE))
175 		mbuf->ol_flags |= PKT_RX_IP_CKSUM_BAD;
176 	else if (BIT_ISSET_AT_POS(annotation->word8, DPAA2_ETH_FAS_L4CE))
177 		mbuf->ol_flags |= PKT_RX_L4_CKSUM_BAD;
178 
179 	if (BIT_ISSET_AT_POS(annotation->word4, L3_IP_1_FIRST_FRAGMENT |
180 	    L3_IP_1_MORE_FRAGMENT |
181 	    L3_IP_N_FIRST_FRAGMENT |
182 	    L3_IP_N_MORE_FRAGMENT)) {
183 		pkt_type |= RTE_PTYPE_L4_FRAG;
184 		goto parse_done;
185 	} else {
186 		pkt_type |= RTE_PTYPE_L4_NONFRAG;
187 	}
188 
189 	if (BIT_ISSET_AT_POS(annotation->word4, L3_PROTO_UDP_PRESENT))
190 		pkt_type |= RTE_PTYPE_L4_UDP;
191 
192 	else if (BIT_ISSET_AT_POS(annotation->word4, L3_PROTO_TCP_PRESENT))
193 		pkt_type |= RTE_PTYPE_L4_TCP;
194 
195 	else if (BIT_ISSET_AT_POS(annotation->word4, L3_PROTO_SCTP_PRESENT))
196 		pkt_type |= RTE_PTYPE_L4_SCTP;
197 
198 	else if (BIT_ISSET_AT_POS(annotation->word4, L3_PROTO_ICMP_PRESENT))
199 		pkt_type |= RTE_PTYPE_L4_ICMP;
200 
201 	else if (BIT_ISSET_AT_POS(annotation->word4, L3_IP_UNKNOWN_PROTOCOL))
202 		pkt_type |= RTE_PTYPE_UNKNOWN;
203 
204 parse_done:
205 	return pkt_type;
206 }
207 
208 static inline uint32_t __attribute__((hot))
209 dpaa2_dev_rx_parse(struct rte_mbuf *mbuf, void *hw_annot_addr)
210 {
211 	struct dpaa2_annot_hdr *annotation =
212 			(struct dpaa2_annot_hdr *)hw_annot_addr;
213 
214 	DPAA2_PMD_DP_DEBUG("(fast parse) Annotation = 0x%" PRIx64 "\t",
215 			   annotation->word4);
216 
217 	if (BIT_ISSET_AT_POS(annotation->word8, DPAA2_ETH_FAS_L3CE))
218 		mbuf->ol_flags |= PKT_RX_IP_CKSUM_BAD;
219 	else if (BIT_ISSET_AT_POS(annotation->word8, DPAA2_ETH_FAS_L4CE))
220 		mbuf->ol_flags |= PKT_RX_L4_CKSUM_BAD;
221 
222 	mbuf->ol_flags |= PKT_RX_TIMESTAMP;
223 	mbuf->timestamp = annotation->word2;
224 	DPAA2_PMD_DP_DEBUG("pkt timestamp: 0x%" PRIx64 "", mbuf->timestamp);
225 
226 	/* Check detailed parsing requirement */
227 	if (annotation->word3 & 0x7FFFFC3FFFF)
228 		return dpaa2_dev_rx_parse_slow(mbuf, annotation);
229 
230 	/* Return some common types from parse processing */
231 	switch (annotation->word4) {
232 	case DPAA2_L3_IPv4:
233 		return RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4;
234 	case DPAA2_L3_IPv6:
235 		return  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6;
236 	case DPAA2_L3_IPv4_TCP:
237 		return  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4 |
238 				RTE_PTYPE_L4_TCP;
239 	case DPAA2_L3_IPv4_UDP:
240 		return  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4 |
241 				RTE_PTYPE_L4_UDP;
242 	case DPAA2_L3_IPv6_TCP:
243 		return  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6 |
244 				RTE_PTYPE_L4_TCP;
245 	case DPAA2_L3_IPv6_UDP:
246 		return  RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6 |
247 				RTE_PTYPE_L4_UDP;
248 	default:
249 		break;
250 	}
251 
252 	return dpaa2_dev_rx_parse_slow(mbuf, annotation);
253 }
254 
255 static inline struct rte_mbuf *__attribute__((hot))
256 eth_sg_fd_to_mbuf(const struct qbman_fd *fd)
257 {
258 	struct qbman_sge *sgt, *sge;
259 	size_t sg_addr, fd_addr;
260 	int i = 0;
261 	struct rte_mbuf *first_seg, *next_seg, *cur_seg, *temp;
262 
263 	fd_addr = (size_t)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd));
264 
265 	/* Get Scatter gather table address */
266 	sgt = (struct qbman_sge *)(fd_addr + DPAA2_GET_FD_OFFSET(fd));
267 
268 	sge = &sgt[i++];
269 	sg_addr = (size_t)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FLE_ADDR(sge));
270 
271 	/* First Scatter gather entry */
272 	first_seg = DPAA2_INLINE_MBUF_FROM_BUF(sg_addr,
273 		rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size);
274 	/* Prepare all the metadata for first segment */
275 	first_seg->buf_addr = (uint8_t *)sg_addr;
276 	first_seg->ol_flags = 0;
277 	first_seg->data_off = DPAA2_GET_FLE_OFFSET(sge);
278 	first_seg->data_len = sge->length  & 0x1FFFF;
279 	first_seg->pkt_len = DPAA2_GET_FD_LEN(fd);
280 	first_seg->nb_segs = 1;
281 	first_seg->next = NULL;
282 	if (dpaa2_svr_family == SVR_LX2160A)
283 		dpaa2_dev_rx_parse_new(first_seg, fd);
284 	else
285 		first_seg->packet_type = dpaa2_dev_rx_parse(first_seg,
286 			(void *)((size_t)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd))
287 			 + DPAA2_FD_PTA_SIZE));
288 
289 	rte_mbuf_refcnt_set(first_seg, 1);
290 	cur_seg = first_seg;
291 	while (!DPAA2_SG_IS_FINAL(sge)) {
292 		sge = &sgt[i++];
293 		sg_addr = (size_t)DPAA2_IOVA_TO_VADDR(
294 				DPAA2_GET_FLE_ADDR(sge));
295 		next_seg = DPAA2_INLINE_MBUF_FROM_BUF(sg_addr,
296 			rte_dpaa2_bpid_info[DPAA2_GET_FLE_BPID(sge)].meta_data_size);
297 		next_seg->buf_addr  = (uint8_t *)sg_addr;
298 		next_seg->data_off  = DPAA2_GET_FLE_OFFSET(sge);
299 		next_seg->data_len  = sge->length  & 0x1FFFF;
300 		first_seg->nb_segs += 1;
301 		rte_mbuf_refcnt_set(next_seg, 1);
302 		cur_seg->next = next_seg;
303 		next_seg->next = NULL;
304 		cur_seg = next_seg;
305 	}
306 	temp = DPAA2_INLINE_MBUF_FROM_BUF(fd_addr,
307 		rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size);
308 	rte_mbuf_refcnt_set(temp, 1);
309 	rte_pktmbuf_free_seg(temp);
310 
311 	return (void *)first_seg;
312 }
313 
314 static inline struct rte_mbuf *__attribute__((hot))
315 eth_fd_to_mbuf(const struct qbman_fd *fd)
316 {
317 	struct rte_mbuf *mbuf = DPAA2_INLINE_MBUF_FROM_BUF(
318 		DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd)),
319 		     rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size);
320 
321 	/* need to repopulated some of the fields,
322 	 * as they may have changed in last transmission
323 	 */
324 	mbuf->nb_segs = 1;
325 	mbuf->ol_flags = 0;
326 	mbuf->data_off = DPAA2_GET_FD_OFFSET(fd);
327 	mbuf->data_len = DPAA2_GET_FD_LEN(fd);
328 	mbuf->pkt_len = mbuf->data_len;
329 	mbuf->next = NULL;
330 	rte_mbuf_refcnt_set(mbuf, 1);
331 
332 	/* Parse the packet */
333 	/* parse results for LX2 are there in FRC field of FD.
334 	 * For other DPAA2 platforms , parse results are after
335 	 * the private - sw annotation area
336 	 */
337 
338 	if (dpaa2_svr_family == SVR_LX2160A)
339 		dpaa2_dev_rx_parse_new(mbuf, fd);
340 	else
341 		mbuf->packet_type = dpaa2_dev_rx_parse(mbuf,
342 			(void *)((size_t)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd))
343 			 + DPAA2_FD_PTA_SIZE));
344 
345 	DPAA2_PMD_DP_DEBUG("to mbuf - mbuf =%p, mbuf->buf_addr =%p, off = %d,"
346 		"fd_off=%d fd =%" PRIx64 ", meta = %d  bpid =%d, len=%d\n",
347 		mbuf, mbuf->buf_addr, mbuf->data_off,
348 		DPAA2_GET_FD_OFFSET(fd), DPAA2_GET_FD_ADDR(fd),
349 		rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size,
350 		DPAA2_GET_FD_BPID(fd), DPAA2_GET_FD_LEN(fd));
351 
352 	return mbuf;
353 }
354 
355 static int __attribute__ ((noinline)) __attribute__((hot))
356 eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
357 		  struct qbman_fd *fd, uint16_t bpid)
358 {
359 	struct rte_mbuf *cur_seg = mbuf, *prev_seg, *mi, *temp;
360 	struct qbman_sge *sgt, *sge = NULL;
361 	int i;
362 
363 	temp = rte_pktmbuf_alloc(mbuf->pool);
364 	if (temp == NULL) {
365 		DPAA2_PMD_DP_DEBUG("No memory to allocate S/G table\n");
366 		return -ENOMEM;
367 	}
368 
369 	DPAA2_SET_FD_ADDR(fd, DPAA2_MBUF_VADDR_TO_IOVA(temp));
370 	DPAA2_SET_FD_LEN(fd, mbuf->pkt_len);
371 	DPAA2_SET_ONLY_FD_BPID(fd, bpid);
372 	DPAA2_SET_FD_OFFSET(fd, temp->data_off);
373 	DPAA2_SET_FD_ASAL(fd, DPAA2_ASAL_VAL);
374 	DPAA2_FD_SET_FORMAT(fd, qbman_fd_sg);
375 	/*Set Scatter gather table and Scatter gather entries*/
376 	sgt = (struct qbman_sge *)(
377 			(size_t)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd))
378 			+ DPAA2_GET_FD_OFFSET(fd));
379 
380 	for (i = 0; i < mbuf->nb_segs; i++) {
381 		sge = &sgt[i];
382 		/*Resetting the buffer pool id and offset field*/
383 		sge->fin_bpid_offset = 0;
384 		DPAA2_SET_FLE_ADDR(sge, DPAA2_MBUF_VADDR_TO_IOVA(cur_seg));
385 		DPAA2_SET_FLE_OFFSET(sge, cur_seg->data_off);
386 		sge->length = cur_seg->data_len;
387 		if (RTE_MBUF_DIRECT(cur_seg)) {
388 			if (rte_mbuf_refcnt_read(cur_seg) > 1) {
389 				/* If refcnt > 1, invalid bpid is set to ensure
390 				 * buffer is not freed by HW
391 				 */
392 				DPAA2_SET_FLE_IVP(sge);
393 				rte_mbuf_refcnt_update(cur_seg, -1);
394 			} else
395 				DPAA2_SET_FLE_BPID(sge,
396 						mempool_to_bpid(cur_seg->pool));
397 			cur_seg = cur_seg->next;
398 		} else {
399 			/* Get owner MBUF from indirect buffer */
400 			mi = rte_mbuf_from_indirect(cur_seg);
401 			if (rte_mbuf_refcnt_read(mi) > 1) {
402 				/* If refcnt > 1, invalid bpid is set to ensure
403 				 * owner buffer is not freed by HW
404 				 */
405 				DPAA2_SET_FLE_IVP(sge);
406 			} else {
407 				DPAA2_SET_FLE_BPID(sge,
408 						   mempool_to_bpid(mi->pool));
409 				rte_mbuf_refcnt_update(mi, 1);
410 			}
411 			prev_seg = cur_seg;
412 			cur_seg = cur_seg->next;
413 			prev_seg->next = NULL;
414 			rte_pktmbuf_free(prev_seg);
415 		}
416 	}
417 	DPAA2_SG_SET_FINAL(sge, true);
418 	return 0;
419 }
420 
421 static void
422 eth_mbuf_to_fd(struct rte_mbuf *mbuf,
423 	       struct qbman_fd *fd, uint16_t bpid) __attribute__((unused));
424 
425 static void __attribute__ ((noinline)) __attribute__((hot))
426 eth_mbuf_to_fd(struct rte_mbuf *mbuf,
427 	       struct qbman_fd *fd, uint16_t bpid)
428 {
429 	DPAA2_MBUF_TO_CONTIG_FD(mbuf, fd, bpid);
430 
431 	DPAA2_PMD_DP_DEBUG("mbuf =%p, mbuf->buf_addr =%p, off = %d,"
432 		"fd_off=%d fd =%" PRIx64 ", meta = %d  bpid =%d, len=%d\n",
433 		mbuf, mbuf->buf_addr, mbuf->data_off,
434 		DPAA2_GET_FD_OFFSET(fd), DPAA2_GET_FD_ADDR(fd),
435 		rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size,
436 		DPAA2_GET_FD_BPID(fd), DPAA2_GET_FD_LEN(fd));
437 	if (RTE_MBUF_DIRECT(mbuf)) {
438 		if (rte_mbuf_refcnt_read(mbuf) > 1) {
439 			DPAA2_SET_FD_IVP(fd);
440 			rte_mbuf_refcnt_update(mbuf, -1);
441 		}
442 	} else {
443 		struct rte_mbuf *mi;
444 
445 		mi = rte_mbuf_from_indirect(mbuf);
446 		if (rte_mbuf_refcnt_read(mi) > 1)
447 			DPAA2_SET_FD_IVP(fd);
448 		else
449 			rte_mbuf_refcnt_update(mi, 1);
450 		rte_pktmbuf_free(mbuf);
451 	}
452 }
453 
454 static inline int __attribute__((hot))
455 eth_copy_mbuf_to_fd(struct rte_mbuf *mbuf,
456 		    struct qbman_fd *fd, uint16_t bpid)
457 {
458 	struct rte_mbuf *m;
459 	void *mb = NULL;
460 
461 	if (rte_dpaa2_mbuf_alloc_bulk(
462 		rte_dpaa2_bpid_info[bpid].bp_list->mp, &mb, 1)) {
463 		DPAA2_PMD_DP_DEBUG("Unable to allocated DPAA2 buffer\n");
464 		return -1;
465 	}
466 	m = (struct rte_mbuf *)mb;
467 	memcpy((char *)m->buf_addr + mbuf->data_off,
468 	       (void *)((char *)mbuf->buf_addr + mbuf->data_off),
469 		mbuf->pkt_len);
470 
471 	/* Copy required fields */
472 	m->data_off = mbuf->data_off;
473 	m->ol_flags = mbuf->ol_flags;
474 	m->packet_type = mbuf->packet_type;
475 	m->tx_offload = mbuf->tx_offload;
476 
477 	DPAA2_MBUF_TO_CONTIG_FD(m, fd, bpid);
478 
479 	DPAA2_PMD_DP_DEBUG(
480 		"mbuf: %p, BMAN buf addr: %p, fdaddr: %" PRIx64 ", bpid: %d,"
481 		" meta: %d, off: %d, len: %d\n",
482 		(void *)mbuf,
483 		mbuf->buf_addr,
484 		DPAA2_GET_FD_ADDR(fd),
485 		DPAA2_GET_FD_BPID(fd),
486 		rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size,
487 		DPAA2_GET_FD_OFFSET(fd),
488 		DPAA2_GET_FD_LEN(fd));
489 
490 return 0;
491 }
492 
493 /* This function assumes that caller will be keep the same value for nb_pkts
494  * across calls per queue, if that is not the case, better use non-prefetch
495  * version of rx call.
496  * It will return the packets as requested in previous call without honoring
497  * the current nb_pkts or bufs space.
498  */
499 uint16_t
500 dpaa2_dev_prefetch_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
501 {
502 	/* Function receive frames for a given device and VQ*/
503 	struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)queue;
504 	struct qbman_result *dq_storage, *dq_storage1 = NULL;
505 	uint32_t fqid = dpaa2_q->fqid;
506 	int ret, num_rx = 0, pull_size;
507 	uint8_t pending, status;
508 	struct qbman_swp *swp;
509 	const struct qbman_fd *fd, *next_fd;
510 	struct qbman_pull_desc pulldesc;
511 	struct queue_storage_info_t *q_storage = dpaa2_q->q_storage;
512 	struct rte_eth_dev_data *eth_data = dpaa2_q->eth_data;
513 
514 	if (unlikely(!DPAA2_PER_LCORE_ETHRX_DPIO)) {
515 		ret = dpaa2_affine_qbman_ethrx_swp();
516 		if (ret) {
517 			DPAA2_PMD_ERR("Failure in affining portal");
518 			return 0;
519 		}
520 	}
521 
522 	if (unlikely(!rte_dpaa2_bpid_info &&
523 		     rte_eal_process_type() == RTE_PROC_SECONDARY))
524 		rte_dpaa2_bpid_info = dpaa2_q->bp_array;
525 
526 	swp = DPAA2_PER_LCORE_ETHRX_PORTAL;
527 	pull_size = (nb_pkts > dpaa2_dqrr_size) ? dpaa2_dqrr_size : nb_pkts;
528 	if (unlikely(!q_storage->active_dqs)) {
529 		q_storage->toggle = 0;
530 		dq_storage = q_storage->dq_storage[q_storage->toggle];
531 		q_storage->last_num_pkts = pull_size;
532 		qbman_pull_desc_clear(&pulldesc);
533 		qbman_pull_desc_set_numframes(&pulldesc,
534 					      q_storage->last_num_pkts);
535 		qbman_pull_desc_set_fq(&pulldesc, fqid);
536 		qbman_pull_desc_set_storage(&pulldesc, dq_storage,
537 			(uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
538 		if (check_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)) {
539 			while (!qbman_check_command_complete(
540 			       get_swp_active_dqs(
541 			       DPAA2_PER_LCORE_ETHRX_DPIO->index)))
542 				;
543 			clear_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index);
544 		}
545 		while (1) {
546 			if (qbman_swp_pull(swp, &pulldesc)) {
547 				DPAA2_PMD_DP_DEBUG("VDQ command is not issued."
548 						  " QBMAN is busy (1)\n");
549 				/* Portal was busy, try again */
550 				continue;
551 			}
552 			break;
553 		}
554 		q_storage->active_dqs = dq_storage;
555 		q_storage->active_dpio_id = DPAA2_PER_LCORE_ETHRX_DPIO->index;
556 		set_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index,
557 				   dq_storage);
558 	}
559 
560 	dq_storage = q_storage->active_dqs;
561 	rte_prefetch0((void *)(size_t)(dq_storage));
562 	rte_prefetch0((void *)(size_t)(dq_storage + 1));
563 
564 	/* Prepare next pull descriptor. This will give space for the
565 	 * prefething done on DQRR entries
566 	 */
567 	q_storage->toggle ^= 1;
568 	dq_storage1 = q_storage->dq_storage[q_storage->toggle];
569 	qbman_pull_desc_clear(&pulldesc);
570 	qbman_pull_desc_set_numframes(&pulldesc, pull_size);
571 	qbman_pull_desc_set_fq(&pulldesc, fqid);
572 	qbman_pull_desc_set_storage(&pulldesc, dq_storage1,
573 		(uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage1)), 1);
574 
575 	/* Check if the previous issued command is completed.
576 	 * Also seems like the SWP is shared between the Ethernet Driver
577 	 * and the SEC driver.
578 	 */
579 	while (!qbman_check_command_complete(dq_storage))
580 		;
581 	if (dq_storage == get_swp_active_dqs(q_storage->active_dpio_id))
582 		clear_swp_active_dqs(q_storage->active_dpio_id);
583 
584 	pending = 1;
585 
586 	do {
587 		/* Loop until the dq_storage is updated with
588 		 * new token by QBMAN
589 		 */
590 		while (!qbman_check_new_result(dq_storage))
591 			;
592 		rte_prefetch0((void *)((size_t)(dq_storage + 2)));
593 		/* Check whether Last Pull command is Expired and
594 		 * setting Condition for Loop termination
595 		 */
596 		if (qbman_result_DQ_is_pull_complete(dq_storage)) {
597 			pending = 0;
598 			/* Check for valid frame. */
599 			status = qbman_result_DQ_flags(dq_storage);
600 			if (unlikely((status & QBMAN_DQ_STAT_VALIDFRAME) == 0))
601 				continue;
602 		}
603 		fd = qbman_result_DQ_fd(dq_storage);
604 
605 		if (dpaa2_svr_family != SVR_LX2160A) {
606 			next_fd = qbman_result_DQ_fd(dq_storage + 1);
607 			/* Prefetch Annotation address for the parse results */
608 			rte_prefetch0((void *)(size_t)(DPAA2_GET_FD_ADDR(
609 				      next_fd) + DPAA2_FD_PTA_SIZE + 16));
610 		}
611 
612 		if (unlikely(DPAA2_FD_GET_FORMAT(fd) == qbman_fd_sg))
613 			bufs[num_rx] = eth_sg_fd_to_mbuf(fd);
614 		else
615 			bufs[num_rx] = eth_fd_to_mbuf(fd);
616 		bufs[num_rx]->port = eth_data->port_id;
617 
618 		if (eth_data->dev_conf.rxmode.offloads &
619 				DEV_RX_OFFLOAD_VLAN_STRIP)
620 			rte_vlan_strip(bufs[num_rx]);
621 
622 		dq_storage++;
623 		num_rx++;
624 	} while (pending);
625 
626 	if (check_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)) {
627 		while (!qbman_check_command_complete(
628 		       get_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)))
629 			;
630 		clear_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index);
631 	}
632 	/* issue a volatile dequeue command for next pull */
633 	while (1) {
634 		if (qbman_swp_pull(swp, &pulldesc)) {
635 			DPAA2_PMD_DP_DEBUG("VDQ command is not issued."
636 					  "QBMAN is busy (2)\n");
637 			continue;
638 		}
639 		break;
640 	}
641 	q_storage->active_dqs = dq_storage1;
642 	q_storage->active_dpio_id = DPAA2_PER_LCORE_ETHRX_DPIO->index;
643 	set_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index, dq_storage1);
644 
645 	dpaa2_q->rx_pkts += num_rx;
646 
647 	return num_rx;
648 }
649 
650 void __attribute__((hot))
651 dpaa2_dev_process_parallel_event(struct qbman_swp *swp,
652 				 const struct qbman_fd *fd,
653 				 const struct qbman_result *dq,
654 				 struct dpaa2_queue *rxq,
655 				 struct rte_event *ev)
656 {
657 	rte_prefetch0((void *)(size_t)(DPAA2_GET_FD_ADDR(fd) +
658 		DPAA2_FD_PTA_SIZE + 16));
659 
660 	ev->flow_id = rxq->ev.flow_id;
661 	ev->sub_event_type = rxq->ev.sub_event_type;
662 	ev->event_type = RTE_EVENT_TYPE_ETHDEV;
663 	ev->op = RTE_EVENT_OP_NEW;
664 	ev->sched_type = rxq->ev.sched_type;
665 	ev->queue_id = rxq->ev.queue_id;
666 	ev->priority = rxq->ev.priority;
667 
668 	ev->mbuf = eth_fd_to_mbuf(fd);
669 
670 	qbman_swp_dqrr_consume(swp, dq);
671 }
672 
673 void __attribute__((hot))
674 dpaa2_dev_process_atomic_event(struct qbman_swp *swp __attribute__((unused)),
675 			       const struct qbman_fd *fd,
676 			       const struct qbman_result *dq,
677 			       struct dpaa2_queue *rxq,
678 			       struct rte_event *ev)
679 {
680 	uint8_t dqrr_index;
681 
682 	rte_prefetch0((void *)(size_t)(DPAA2_GET_FD_ADDR(fd) +
683 		DPAA2_FD_PTA_SIZE + 16));
684 
685 	ev->flow_id = rxq->ev.flow_id;
686 	ev->sub_event_type = rxq->ev.sub_event_type;
687 	ev->event_type = RTE_EVENT_TYPE_ETHDEV;
688 	ev->op = RTE_EVENT_OP_NEW;
689 	ev->sched_type = rxq->ev.sched_type;
690 	ev->queue_id = rxq->ev.queue_id;
691 	ev->priority = rxq->ev.priority;
692 
693 	ev->mbuf = eth_fd_to_mbuf(fd);
694 
695 	dqrr_index = qbman_get_dqrr_idx(dq);
696 	ev->mbuf->seqn = dqrr_index + 1;
697 	DPAA2_PER_LCORE_DQRR_SIZE++;
698 	DPAA2_PER_LCORE_DQRR_HELD |= 1 << dqrr_index;
699 	DPAA2_PER_LCORE_DQRR_MBUF(dqrr_index) = ev->mbuf;
700 }
701 
702 /*
703  * Callback to handle sending packets through WRIOP based interface
704  */
705 uint16_t
706 dpaa2_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
707 {
708 	/* Function to transmit the frames to given device and VQ*/
709 	uint32_t loop, retry_count;
710 	int32_t ret;
711 	struct qbman_fd fd_arr[MAX_TX_RING_SLOTS];
712 	struct rte_mbuf *mi;
713 	uint32_t frames_to_send;
714 	struct rte_mempool *mp;
715 	struct qbman_eq_desc eqdesc;
716 	struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)queue;
717 	struct qbman_swp *swp;
718 	uint16_t num_tx = 0;
719 	uint16_t bpid;
720 	struct rte_eth_dev_data *eth_data = dpaa2_q->eth_data;
721 	struct dpaa2_dev_priv *priv = eth_data->dev_private;
722 	uint32_t flags[MAX_TX_RING_SLOTS] = {0};
723 
724 	if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
725 		ret = dpaa2_affine_qbman_swp();
726 		if (ret) {
727 			DPAA2_PMD_ERR("Failure in affining portal");
728 			return 0;
729 		}
730 	}
731 	swp = DPAA2_PER_LCORE_PORTAL;
732 
733 	DPAA2_PMD_DP_DEBUG("===> eth_data =%p, fqid =%d\n",
734 			eth_data, dpaa2_q->fqid);
735 
736 	/*Prepare enqueue descriptor*/
737 	qbman_eq_desc_clear(&eqdesc);
738 	qbman_eq_desc_set_no_orp(&eqdesc, DPAA2_EQ_RESP_ERR_FQ);
739 	qbman_eq_desc_set_qd(&eqdesc, priv->qdid,
740 			     dpaa2_q->flow_id, dpaa2_q->tc_index);
741 	/*Clear the unused FD fields before sending*/
742 	while (nb_pkts) {
743 		/*Check if the queue is congested*/
744 		retry_count = 0;
745 		while (qbman_result_SCN_state(dpaa2_q->cscn)) {
746 			retry_count++;
747 			/* Retry for some time before giving up */
748 			if (retry_count > CONG_RETRY_COUNT)
749 				goto skip_tx;
750 		}
751 
752 		frames_to_send = (nb_pkts > dpaa2_eqcr_size) ?
753 			dpaa2_eqcr_size : nb_pkts;
754 
755 		for (loop = 0; loop < frames_to_send; loop++) {
756 			if ((*bufs)->seqn) {
757 				uint8_t dqrr_index = (*bufs)->seqn - 1;
758 
759 				flags[loop] = QBMAN_ENQUEUE_FLAG_DCA |
760 						dqrr_index;
761 				DPAA2_PER_LCORE_DQRR_SIZE--;
762 				DPAA2_PER_LCORE_DQRR_HELD &= ~(1 << dqrr_index);
763 				(*bufs)->seqn = DPAA2_INVALID_MBUF_SEQN;
764 			}
765 
766 			if (likely(RTE_MBUF_DIRECT(*bufs))) {
767 				mp = (*bufs)->pool;
768 				/* Check the basic scenario and set
769 				 * the FD appropriately here itself.
770 				 */
771 				if (likely(mp && mp->ops_index ==
772 				    priv->bp_list->dpaa2_ops_index &&
773 				    (*bufs)->nb_segs == 1 &&
774 				    rte_mbuf_refcnt_read((*bufs)) == 1)) {
775 					if (unlikely(((*bufs)->ol_flags
776 						& PKT_TX_VLAN_PKT) ||
777 						(eth_data->dev_conf.txmode.offloads
778 						& DEV_TX_OFFLOAD_VLAN_INSERT))) {
779 						ret = rte_vlan_insert(bufs);
780 						if (ret)
781 							goto send_n_return;
782 					}
783 					DPAA2_MBUF_TO_CONTIG_FD((*bufs),
784 					&fd_arr[loop], mempool_to_bpid(mp));
785 					bufs++;
786 					continue;
787 				}
788 			} else {
789 				mi = rte_mbuf_from_indirect(*bufs);
790 				mp = mi->pool;
791 			}
792 			/* Not a hw_pkt pool allocated frame */
793 			if (unlikely(!mp || !priv->bp_list)) {
794 				DPAA2_PMD_ERR("Err: No buffer pool attached");
795 				goto send_n_return;
796 			}
797 
798 			if (unlikely(((*bufs)->ol_flags & PKT_TX_VLAN_PKT) ||
799 				(eth_data->dev_conf.txmode.offloads
800 				& DEV_TX_OFFLOAD_VLAN_INSERT))) {
801 				int ret = rte_vlan_insert(bufs);
802 				if (ret)
803 					goto send_n_return;
804 			}
805 			if (mp->ops_index != priv->bp_list->dpaa2_ops_index) {
806 				DPAA2_PMD_WARN("Non DPAA2 buffer pool");
807 				/* alloc should be from the default buffer pool
808 				 * attached to this interface
809 				 */
810 				bpid = priv->bp_list->buf_pool.bpid;
811 
812 				if (unlikely((*bufs)->nb_segs > 1)) {
813 					DPAA2_PMD_ERR("S/G support not added"
814 						" for non hw offload buffer");
815 					goto send_n_return;
816 				}
817 				if (eth_copy_mbuf_to_fd(*bufs,
818 							&fd_arr[loop], bpid)) {
819 					goto send_n_return;
820 				}
821 				/* free the original packet */
822 				rte_pktmbuf_free(*bufs);
823 			} else {
824 				bpid = mempool_to_bpid(mp);
825 				if (unlikely((*bufs)->nb_segs > 1)) {
826 					if (eth_mbuf_to_sg_fd(*bufs,
827 							&fd_arr[loop], bpid))
828 						goto send_n_return;
829 				} else {
830 					eth_mbuf_to_fd(*bufs,
831 						       &fd_arr[loop], bpid);
832 				}
833 			}
834 			bufs++;
835 		}
836 		loop = 0;
837 		while (loop < frames_to_send) {
838 			loop += qbman_swp_enqueue_multiple(swp, &eqdesc,
839 					&fd_arr[loop], &flags[loop],
840 					frames_to_send - loop);
841 		}
842 
843 		num_tx += frames_to_send;
844 		nb_pkts -= frames_to_send;
845 	}
846 	dpaa2_q->tx_pkts += num_tx;
847 	return num_tx;
848 
849 send_n_return:
850 	/* send any already prepared fd */
851 	if (loop) {
852 		unsigned int i = 0;
853 
854 		while (i < loop) {
855 			i += qbman_swp_enqueue_multiple(swp, &eqdesc,
856 							&fd_arr[i],
857 							&flags[loop],
858 							loop - i);
859 		}
860 		num_tx += loop;
861 	}
862 skip_tx:
863 	dpaa2_q->tx_pkts += num_tx;
864 	return num_tx;
865 }
866 
867 /**
868  * Dummy DPDK callback for TX.
869  *
870  * This function is used to temporarily replace the real callback during
871  * unsafe control operations on the queue, or in case of error.
872  *
873  * @param dpdk_txq
874  *   Generic pointer to TX queue structure.
875  * @param[in] pkts
876  *   Packets to transmit.
877  * @param pkts_n
878  *   Number of packets in array.
879  *
880  * @return
881  *   Number of packets successfully transmitted (<= pkts_n).
882  */
883 uint16_t
884 dummy_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
885 {
886 	(void)queue;
887 	(void)bufs;
888 	(void)nb_pkts;
889 	return 0;
890 }
891