xref: /dpdk/drivers/net/dpaa2/dpaa2_rxtx.c (revision 089e5ed727a15da2729cfee9b63533dd120bd04c)
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_FD_SET_FORMAT(fd, qbman_fd_sg);
374 	DPAA2_RESET_FD_FRC(fd);
375 	DPAA2_RESET_FD_CTRL(fd);
376 	/*Set Scatter gather table and Scatter gather entries*/
377 	sgt = (struct qbman_sge *)(
378 			(size_t)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd))
379 			+ DPAA2_GET_FD_OFFSET(fd));
380 
381 	for (i = 0; i < mbuf->nb_segs; i++) {
382 		sge = &sgt[i];
383 		/*Resetting the buffer pool id and offset field*/
384 		sge->fin_bpid_offset = 0;
385 		DPAA2_SET_FLE_ADDR(sge, DPAA2_MBUF_VADDR_TO_IOVA(cur_seg));
386 		DPAA2_SET_FLE_OFFSET(sge, cur_seg->data_off);
387 		sge->length = cur_seg->data_len;
388 		if (RTE_MBUF_DIRECT(cur_seg)) {
389 			if (rte_mbuf_refcnt_read(cur_seg) > 1) {
390 				/* If refcnt > 1, invalid bpid is set to ensure
391 				 * buffer is not freed by HW
392 				 */
393 				DPAA2_SET_FLE_IVP(sge);
394 				rte_mbuf_refcnt_update(cur_seg, -1);
395 			} else
396 				DPAA2_SET_FLE_BPID(sge,
397 						mempool_to_bpid(cur_seg->pool));
398 			cur_seg = cur_seg->next;
399 		} else {
400 			/* Get owner MBUF from indirect buffer */
401 			mi = rte_mbuf_from_indirect(cur_seg);
402 			if (rte_mbuf_refcnt_read(mi) > 1) {
403 				/* If refcnt > 1, invalid bpid is set to ensure
404 				 * owner buffer is not freed by HW
405 				 */
406 				DPAA2_SET_FLE_IVP(sge);
407 			} else {
408 				DPAA2_SET_FLE_BPID(sge,
409 						   mempool_to_bpid(mi->pool));
410 				rte_mbuf_refcnt_update(mi, 1);
411 			}
412 			prev_seg = cur_seg;
413 			cur_seg = cur_seg->next;
414 			prev_seg->next = NULL;
415 			rte_pktmbuf_free(prev_seg);
416 		}
417 	}
418 	DPAA2_SG_SET_FINAL(sge, true);
419 	return 0;
420 }
421 
422 static void
423 eth_mbuf_to_fd(struct rte_mbuf *mbuf,
424 	       struct qbman_fd *fd, uint16_t bpid) __attribute__((unused));
425 
426 static void __attribute__ ((noinline)) __attribute__((hot))
427 eth_mbuf_to_fd(struct rte_mbuf *mbuf,
428 	       struct qbman_fd *fd, uint16_t bpid)
429 {
430 	DPAA2_MBUF_TO_CONTIG_FD(mbuf, fd, bpid);
431 
432 	DPAA2_PMD_DP_DEBUG("mbuf =%p, mbuf->buf_addr =%p, off = %d,"
433 		"fd_off=%d fd =%" PRIx64 ", meta = %d  bpid =%d, len=%d\n",
434 		mbuf, mbuf->buf_addr, mbuf->data_off,
435 		DPAA2_GET_FD_OFFSET(fd), DPAA2_GET_FD_ADDR(fd),
436 		rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size,
437 		DPAA2_GET_FD_BPID(fd), DPAA2_GET_FD_LEN(fd));
438 	if (RTE_MBUF_DIRECT(mbuf)) {
439 		if (rte_mbuf_refcnt_read(mbuf) > 1) {
440 			DPAA2_SET_FD_IVP(fd);
441 			rte_mbuf_refcnt_update(mbuf, -1);
442 		}
443 	} else {
444 		struct rte_mbuf *mi;
445 
446 		mi = rte_mbuf_from_indirect(mbuf);
447 		if (rte_mbuf_refcnt_read(mi) > 1)
448 			DPAA2_SET_FD_IVP(fd);
449 		else
450 			rte_mbuf_refcnt_update(mi, 1);
451 		rte_pktmbuf_free(mbuf);
452 	}
453 }
454 
455 static inline int __attribute__((hot))
456 eth_copy_mbuf_to_fd(struct rte_mbuf *mbuf,
457 		    struct qbman_fd *fd, uint16_t bpid)
458 {
459 	struct rte_mbuf *m;
460 	void *mb = NULL;
461 
462 	if (rte_dpaa2_mbuf_alloc_bulk(
463 		rte_dpaa2_bpid_info[bpid].bp_list->mp, &mb, 1)) {
464 		DPAA2_PMD_DP_DEBUG("Unable to allocated DPAA2 buffer\n");
465 		return -1;
466 	}
467 	m = (struct rte_mbuf *)mb;
468 	memcpy((char *)m->buf_addr + mbuf->data_off,
469 	       (void *)((char *)mbuf->buf_addr + mbuf->data_off),
470 		mbuf->pkt_len);
471 
472 	/* Copy required fields */
473 	m->data_off = mbuf->data_off;
474 	m->ol_flags = mbuf->ol_flags;
475 	m->packet_type = mbuf->packet_type;
476 	m->tx_offload = mbuf->tx_offload;
477 
478 	DPAA2_MBUF_TO_CONTIG_FD(m, fd, bpid);
479 
480 	DPAA2_PMD_DP_DEBUG(
481 		"mbuf: %p, BMAN buf addr: %p, fdaddr: %" PRIx64 ", bpid: %d,"
482 		" meta: %d, off: %d, len: %d\n",
483 		(void *)mbuf,
484 		mbuf->buf_addr,
485 		DPAA2_GET_FD_ADDR(fd),
486 		DPAA2_GET_FD_BPID(fd),
487 		rte_dpaa2_bpid_info[DPAA2_GET_FD_BPID(fd)].meta_data_size,
488 		DPAA2_GET_FD_OFFSET(fd),
489 		DPAA2_GET_FD_LEN(fd));
490 
491 return 0;
492 }
493 
494 /* This function assumes that caller will be keep the same value for nb_pkts
495  * across calls per queue, if that is not the case, better use non-prefetch
496  * version of rx call.
497  * It will return the packets as requested in previous call without honoring
498  * the current nb_pkts or bufs space.
499  */
500 uint16_t
501 dpaa2_dev_prefetch_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
502 {
503 	/* Function receive frames for a given device and VQ*/
504 	struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)queue;
505 	struct qbman_result *dq_storage, *dq_storage1 = NULL;
506 	uint32_t fqid = dpaa2_q->fqid;
507 	int ret, num_rx = 0, pull_size;
508 	uint8_t pending, status;
509 	struct qbman_swp *swp;
510 	const struct qbman_fd *fd, *next_fd;
511 	struct qbman_pull_desc pulldesc;
512 	struct queue_storage_info_t *q_storage = dpaa2_q->q_storage;
513 	struct rte_eth_dev_data *eth_data = dpaa2_q->eth_data;
514 
515 	if (unlikely(!DPAA2_PER_LCORE_ETHRX_DPIO)) {
516 		ret = dpaa2_affine_qbman_ethrx_swp();
517 		if (ret) {
518 			DPAA2_PMD_ERR("Failure in affining portal");
519 			return 0;
520 		}
521 	}
522 
523 	if (unlikely(!rte_dpaa2_bpid_info &&
524 		     rte_eal_process_type() == RTE_PROC_SECONDARY))
525 		rte_dpaa2_bpid_info = dpaa2_q->bp_array;
526 
527 	swp = DPAA2_PER_LCORE_ETHRX_PORTAL;
528 	pull_size = (nb_pkts > dpaa2_dqrr_size) ? dpaa2_dqrr_size : nb_pkts;
529 	if (unlikely(!q_storage->active_dqs)) {
530 		q_storage->toggle = 0;
531 		dq_storage = q_storage->dq_storage[q_storage->toggle];
532 		q_storage->last_num_pkts = pull_size;
533 		qbman_pull_desc_clear(&pulldesc);
534 		qbman_pull_desc_set_numframes(&pulldesc,
535 					      q_storage->last_num_pkts);
536 		qbman_pull_desc_set_fq(&pulldesc, fqid);
537 		qbman_pull_desc_set_storage(&pulldesc, dq_storage,
538 			(uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
539 		if (check_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)) {
540 			while (!qbman_check_command_complete(
541 			       get_swp_active_dqs(
542 			       DPAA2_PER_LCORE_ETHRX_DPIO->index)))
543 				;
544 			clear_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index);
545 		}
546 		while (1) {
547 			if (qbman_swp_pull(swp, &pulldesc)) {
548 				DPAA2_PMD_DP_DEBUG("VDQ command is not issued."
549 						  " QBMAN is busy (1)\n");
550 				/* Portal was busy, try again */
551 				continue;
552 			}
553 			break;
554 		}
555 		q_storage->active_dqs = dq_storage;
556 		q_storage->active_dpio_id = DPAA2_PER_LCORE_ETHRX_DPIO->index;
557 		set_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index,
558 				   dq_storage);
559 	}
560 
561 	dq_storage = q_storage->active_dqs;
562 	rte_prefetch0((void *)(size_t)(dq_storage));
563 	rte_prefetch0((void *)(size_t)(dq_storage + 1));
564 
565 	/* Prepare next pull descriptor. This will give space for the
566 	 * prefething done on DQRR entries
567 	 */
568 	q_storage->toggle ^= 1;
569 	dq_storage1 = q_storage->dq_storage[q_storage->toggle];
570 	qbman_pull_desc_clear(&pulldesc);
571 	qbman_pull_desc_set_numframes(&pulldesc, pull_size);
572 	qbman_pull_desc_set_fq(&pulldesc, fqid);
573 	qbman_pull_desc_set_storage(&pulldesc, dq_storage1,
574 		(uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage1)), 1);
575 
576 	/* Check if the previous issued command is completed.
577 	 * Also seems like the SWP is shared between the Ethernet Driver
578 	 * and the SEC driver.
579 	 */
580 	while (!qbman_check_command_complete(dq_storage))
581 		;
582 	if (dq_storage == get_swp_active_dqs(q_storage->active_dpio_id))
583 		clear_swp_active_dqs(q_storage->active_dpio_id);
584 
585 	pending = 1;
586 
587 	do {
588 		/* Loop until the dq_storage is updated with
589 		 * new token by QBMAN
590 		 */
591 		while (!qbman_check_new_result(dq_storage))
592 			;
593 		rte_prefetch0((void *)((size_t)(dq_storage + 2)));
594 		/* Check whether Last Pull command is Expired and
595 		 * setting Condition for Loop termination
596 		 */
597 		if (qbman_result_DQ_is_pull_complete(dq_storage)) {
598 			pending = 0;
599 			/* Check for valid frame. */
600 			status = qbman_result_DQ_flags(dq_storage);
601 			if (unlikely((status & QBMAN_DQ_STAT_VALIDFRAME) == 0))
602 				continue;
603 		}
604 		fd = qbman_result_DQ_fd(dq_storage);
605 
606 		if (dpaa2_svr_family != SVR_LX2160A) {
607 			next_fd = qbman_result_DQ_fd(dq_storage + 1);
608 			/* Prefetch Annotation address for the parse results */
609 			rte_prefetch0((void *)(size_t)(DPAA2_GET_FD_ADDR(
610 				      next_fd) + DPAA2_FD_PTA_SIZE + 16));
611 		}
612 
613 		if (unlikely(DPAA2_FD_GET_FORMAT(fd) == qbman_fd_sg))
614 			bufs[num_rx] = eth_sg_fd_to_mbuf(fd);
615 		else
616 			bufs[num_rx] = eth_fd_to_mbuf(fd);
617 		bufs[num_rx]->port = eth_data->port_id;
618 
619 		if (eth_data->dev_conf.rxmode.offloads &
620 				DEV_RX_OFFLOAD_VLAN_STRIP)
621 			rte_vlan_strip(bufs[num_rx]);
622 
623 		dq_storage++;
624 		num_rx++;
625 	} while (pending);
626 
627 	if (check_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)) {
628 		while (!qbman_check_command_complete(
629 		       get_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)))
630 			;
631 		clear_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index);
632 	}
633 	/* issue a volatile dequeue command for next pull */
634 	while (1) {
635 		if (qbman_swp_pull(swp, &pulldesc)) {
636 			DPAA2_PMD_DP_DEBUG("VDQ command is not issued."
637 					  "QBMAN is busy (2)\n");
638 			continue;
639 		}
640 		break;
641 	}
642 	q_storage->active_dqs = dq_storage1;
643 	q_storage->active_dpio_id = DPAA2_PER_LCORE_ETHRX_DPIO->index;
644 	set_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index, dq_storage1);
645 
646 	dpaa2_q->rx_pkts += num_rx;
647 
648 	return num_rx;
649 }
650 
651 void __attribute__((hot))
652 dpaa2_dev_process_parallel_event(struct qbman_swp *swp,
653 				 const struct qbman_fd *fd,
654 				 const struct qbman_result *dq,
655 				 struct dpaa2_queue *rxq,
656 				 struct rte_event *ev)
657 {
658 	rte_prefetch0((void *)(size_t)(DPAA2_GET_FD_ADDR(fd) +
659 		DPAA2_FD_PTA_SIZE + 16));
660 
661 	ev->flow_id = rxq->ev.flow_id;
662 	ev->sub_event_type = rxq->ev.sub_event_type;
663 	ev->event_type = RTE_EVENT_TYPE_ETHDEV;
664 	ev->op = RTE_EVENT_OP_NEW;
665 	ev->sched_type = rxq->ev.sched_type;
666 	ev->queue_id = rxq->ev.queue_id;
667 	ev->priority = rxq->ev.priority;
668 
669 	ev->mbuf = eth_fd_to_mbuf(fd);
670 
671 	qbman_swp_dqrr_consume(swp, dq);
672 }
673 
674 void __attribute__((hot))
675 dpaa2_dev_process_atomic_event(struct qbman_swp *swp __attribute__((unused)),
676 			       const struct qbman_fd *fd,
677 			       const struct qbman_result *dq,
678 			       struct dpaa2_queue *rxq,
679 			       struct rte_event *ev)
680 {
681 	uint8_t dqrr_index;
682 
683 	rte_prefetch0((void *)(size_t)(DPAA2_GET_FD_ADDR(fd) +
684 		DPAA2_FD_PTA_SIZE + 16));
685 
686 	ev->flow_id = rxq->ev.flow_id;
687 	ev->sub_event_type = rxq->ev.sub_event_type;
688 	ev->event_type = RTE_EVENT_TYPE_ETHDEV;
689 	ev->op = RTE_EVENT_OP_NEW;
690 	ev->sched_type = rxq->ev.sched_type;
691 	ev->queue_id = rxq->ev.queue_id;
692 	ev->priority = rxq->ev.priority;
693 
694 	ev->mbuf = eth_fd_to_mbuf(fd);
695 
696 	dqrr_index = qbman_get_dqrr_idx(dq);
697 	ev->mbuf->seqn = dqrr_index + 1;
698 	DPAA2_PER_LCORE_DQRR_SIZE++;
699 	DPAA2_PER_LCORE_DQRR_HELD |= 1 << dqrr_index;
700 	DPAA2_PER_LCORE_DQRR_MBUF(dqrr_index) = ev->mbuf;
701 }
702 
703 void __attribute__((hot))
704 dpaa2_dev_process_ordered_event(struct qbman_swp *swp,
705 				const struct qbman_fd *fd,
706 				const struct qbman_result *dq,
707 				struct dpaa2_queue *rxq,
708 				struct rte_event *ev)
709 {
710 	rte_prefetch0((void *)(size_t)(DPAA2_GET_FD_ADDR(fd) +
711 		DPAA2_FD_PTA_SIZE + 16));
712 
713 	ev->flow_id = rxq->ev.flow_id;
714 	ev->sub_event_type = rxq->ev.sub_event_type;
715 	ev->event_type = RTE_EVENT_TYPE_ETHDEV;
716 	ev->op = RTE_EVENT_OP_NEW;
717 	ev->sched_type = rxq->ev.sched_type;
718 	ev->queue_id = rxq->ev.queue_id;
719 	ev->priority = rxq->ev.priority;
720 
721 	ev->mbuf = eth_fd_to_mbuf(fd);
722 
723 	ev->mbuf->seqn = DPAA2_ENQUEUE_FLAG_ORP;
724 	ev->mbuf->seqn |= qbman_result_DQ_odpid(dq) << DPAA2_EQCR_OPRID_SHIFT;
725 	ev->mbuf->seqn |= qbman_result_DQ_seqnum(dq) << DPAA2_EQCR_SEQNUM_SHIFT;
726 
727 	qbman_swp_dqrr_consume(swp, dq);
728 }
729 
730 /*
731  * Callback to handle sending packets through WRIOP based interface
732  */
733 uint16_t
734 dpaa2_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
735 {
736 	/* Function to transmit the frames to given device and VQ*/
737 	uint32_t loop, retry_count;
738 	int32_t ret;
739 	struct qbman_fd fd_arr[MAX_TX_RING_SLOTS];
740 	struct rte_mbuf *mi;
741 	uint32_t frames_to_send;
742 	struct rte_mempool *mp;
743 	struct qbman_eq_desc eqdesc;
744 	struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)queue;
745 	struct qbman_swp *swp;
746 	uint16_t num_tx = 0;
747 	uint16_t bpid;
748 	struct rte_eth_dev_data *eth_data = dpaa2_q->eth_data;
749 	struct dpaa2_dev_priv *priv = eth_data->dev_private;
750 	uint32_t flags[MAX_TX_RING_SLOTS] = {0};
751 
752 	if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
753 		ret = dpaa2_affine_qbman_swp();
754 		if (ret) {
755 			DPAA2_PMD_ERR("Failure in affining portal");
756 			return 0;
757 		}
758 	}
759 	swp = DPAA2_PER_LCORE_PORTAL;
760 
761 	DPAA2_PMD_DP_DEBUG("===> eth_data =%p, fqid =%d\n",
762 			eth_data, dpaa2_q->fqid);
763 
764 	/*Prepare enqueue descriptor*/
765 	qbman_eq_desc_clear(&eqdesc);
766 	qbman_eq_desc_set_no_orp(&eqdesc, DPAA2_EQ_RESP_ERR_FQ);
767 	qbman_eq_desc_set_qd(&eqdesc, priv->qdid,
768 			     dpaa2_q->flow_id, dpaa2_q->tc_index);
769 	/*Clear the unused FD fields before sending*/
770 	while (nb_pkts) {
771 		/*Check if the queue is congested*/
772 		retry_count = 0;
773 		while (qbman_result_SCN_state(dpaa2_q->cscn)) {
774 			retry_count++;
775 			/* Retry for some time before giving up */
776 			if (retry_count > CONG_RETRY_COUNT)
777 				goto skip_tx;
778 		}
779 
780 		frames_to_send = (nb_pkts > dpaa2_eqcr_size) ?
781 			dpaa2_eqcr_size : nb_pkts;
782 
783 		for (loop = 0; loop < frames_to_send; loop++) {
784 			if ((*bufs)->seqn) {
785 				uint8_t dqrr_index = (*bufs)->seqn - 1;
786 
787 				flags[loop] = QBMAN_ENQUEUE_FLAG_DCA |
788 						dqrr_index;
789 				DPAA2_PER_LCORE_DQRR_SIZE--;
790 				DPAA2_PER_LCORE_DQRR_HELD &= ~(1 << dqrr_index);
791 				(*bufs)->seqn = DPAA2_INVALID_MBUF_SEQN;
792 			}
793 
794 			if (likely(RTE_MBUF_DIRECT(*bufs))) {
795 				mp = (*bufs)->pool;
796 				/* Check the basic scenario and set
797 				 * the FD appropriately here itself.
798 				 */
799 				if (likely(mp && mp->ops_index ==
800 				    priv->bp_list->dpaa2_ops_index &&
801 				    (*bufs)->nb_segs == 1 &&
802 				    rte_mbuf_refcnt_read((*bufs)) == 1)) {
803 					if (unlikely(((*bufs)->ol_flags
804 						& PKT_TX_VLAN_PKT) ||
805 						(eth_data->dev_conf.txmode.offloads
806 						& DEV_TX_OFFLOAD_VLAN_INSERT))) {
807 						ret = rte_vlan_insert(bufs);
808 						if (ret)
809 							goto send_n_return;
810 					}
811 					DPAA2_MBUF_TO_CONTIG_FD((*bufs),
812 					&fd_arr[loop], mempool_to_bpid(mp));
813 					bufs++;
814 					continue;
815 				}
816 			} else {
817 				mi = rte_mbuf_from_indirect(*bufs);
818 				mp = mi->pool;
819 			}
820 			/* Not a hw_pkt pool allocated frame */
821 			if (unlikely(!mp || !priv->bp_list)) {
822 				DPAA2_PMD_ERR("Err: No buffer pool attached");
823 				goto send_n_return;
824 			}
825 
826 			if (unlikely(((*bufs)->ol_flags & PKT_TX_VLAN_PKT) ||
827 				(eth_data->dev_conf.txmode.offloads
828 				& DEV_TX_OFFLOAD_VLAN_INSERT))) {
829 				int ret = rte_vlan_insert(bufs);
830 				if (ret)
831 					goto send_n_return;
832 			}
833 			if (mp->ops_index != priv->bp_list->dpaa2_ops_index) {
834 				DPAA2_PMD_WARN("Non DPAA2 buffer pool");
835 				/* alloc should be from the default buffer pool
836 				 * attached to this interface
837 				 */
838 				bpid = priv->bp_list->buf_pool.bpid;
839 
840 				if (unlikely((*bufs)->nb_segs > 1)) {
841 					DPAA2_PMD_ERR("S/G support not added"
842 						" for non hw offload buffer");
843 					goto send_n_return;
844 				}
845 				if (eth_copy_mbuf_to_fd(*bufs,
846 							&fd_arr[loop], bpid)) {
847 					goto send_n_return;
848 				}
849 				/* free the original packet */
850 				rte_pktmbuf_free(*bufs);
851 			} else {
852 				bpid = mempool_to_bpid(mp);
853 				if (unlikely((*bufs)->nb_segs > 1)) {
854 					if (eth_mbuf_to_sg_fd(*bufs,
855 							&fd_arr[loop], bpid))
856 						goto send_n_return;
857 				} else {
858 					eth_mbuf_to_fd(*bufs,
859 						       &fd_arr[loop], bpid);
860 				}
861 			}
862 			bufs++;
863 		}
864 		loop = 0;
865 		while (loop < frames_to_send) {
866 			loop += qbman_swp_enqueue_multiple(swp, &eqdesc,
867 					&fd_arr[loop], &flags[loop],
868 					frames_to_send - loop);
869 		}
870 
871 		num_tx += frames_to_send;
872 		nb_pkts -= frames_to_send;
873 	}
874 	dpaa2_q->tx_pkts += num_tx;
875 	return num_tx;
876 
877 send_n_return:
878 	/* send any already prepared fd */
879 	if (loop) {
880 		unsigned int i = 0;
881 
882 		while (i < loop) {
883 			i += qbman_swp_enqueue_multiple(swp, &eqdesc,
884 							&fd_arr[i],
885 							&flags[loop],
886 							loop - i);
887 		}
888 		num_tx += loop;
889 	}
890 skip_tx:
891 	dpaa2_q->tx_pkts += num_tx;
892 	return num_tx;
893 }
894 
895 void
896 dpaa2_dev_free_eqresp_buf(uint16_t eqresp_ci)
897 {
898 	struct dpaa2_dpio_dev *dpio_dev = DPAA2_PER_LCORE_DPIO;
899 	struct qbman_fd *fd;
900 	struct rte_mbuf *m;
901 
902 	fd = qbman_result_eqresp_fd(&dpio_dev->eqresp[eqresp_ci]);
903 	m = eth_fd_to_mbuf(fd);
904 	rte_pktmbuf_free(m);
905 }
906 
907 static void
908 dpaa2_set_enqueue_descriptor(struct dpaa2_queue *dpaa2_q,
909 			     struct rte_mbuf *m,
910 			     struct qbman_eq_desc *eqdesc)
911 {
912 	struct rte_eth_dev_data *eth_data = dpaa2_q->eth_data;
913 	struct dpaa2_dev_priv *priv = eth_data->dev_private;
914 	struct dpaa2_dpio_dev *dpio_dev = DPAA2_PER_LCORE_DPIO;
915 	struct eqresp_metadata *eqresp_meta;
916 	uint16_t orpid, seqnum;
917 	uint8_t dq_idx;
918 
919 	qbman_eq_desc_set_qd(eqdesc, priv->qdid, dpaa2_q->flow_id,
920 			     dpaa2_q->tc_index);
921 
922 	if (m->seqn & DPAA2_ENQUEUE_FLAG_ORP) {
923 		orpid = (m->seqn & DPAA2_EQCR_OPRID_MASK) >>
924 			DPAA2_EQCR_OPRID_SHIFT;
925 		seqnum = (m->seqn & DPAA2_EQCR_SEQNUM_MASK) >>
926 			DPAA2_EQCR_SEQNUM_SHIFT;
927 
928 		if (!priv->en_loose_ordered) {
929 			qbman_eq_desc_set_orp(eqdesc, 1, orpid, seqnum, 0);
930 			qbman_eq_desc_set_response(eqdesc, (uint64_t)
931 				DPAA2_VADDR_TO_IOVA(&dpio_dev->eqresp[
932 				dpio_dev->eqresp_pi]), 1);
933 			qbman_eq_desc_set_token(eqdesc, 1);
934 
935 			eqresp_meta = &dpio_dev->eqresp_meta[
936 				dpio_dev->eqresp_pi];
937 			eqresp_meta->dpaa2_q = dpaa2_q;
938 			eqresp_meta->mp = m->pool;
939 
940 			dpio_dev->eqresp_pi + 1 < MAX_EQ_RESP_ENTRIES ?
941 				dpio_dev->eqresp_pi++ :
942 				(dpio_dev->eqresp_pi = 0);
943 		} else {
944 			qbman_eq_desc_set_orp(eqdesc, 0, orpid, seqnum, 0);
945 		}
946 	} else {
947 		dq_idx = m->seqn - 1;
948 		qbman_eq_desc_set_dca(eqdesc, 1, dq_idx, 0);
949 		DPAA2_PER_LCORE_DQRR_SIZE--;
950 		DPAA2_PER_LCORE_DQRR_HELD &= ~(1 << dq_idx);
951 	}
952 	m->seqn = DPAA2_INVALID_MBUF_SEQN;
953 }
954 
955 /* Callback to handle sending ordered packets through WRIOP based interface */
956 uint16_t
957 dpaa2_dev_tx_ordered(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
958 {
959 	/* Function to transmit the frames to given device and VQ*/
960 	struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)queue;
961 	struct rte_eth_dev_data *eth_data = dpaa2_q->eth_data;
962 	struct dpaa2_dev_priv *priv = eth_data->dev_private;
963 	struct dpaa2_queue *order_sendq = (struct dpaa2_queue *)priv->tx_vq[0];
964 	struct qbman_fd fd_arr[MAX_TX_RING_SLOTS];
965 	struct rte_mbuf *mi;
966 	struct rte_mempool *mp;
967 	struct qbman_eq_desc eqdesc[MAX_TX_RING_SLOTS];
968 	struct qbman_swp *swp;
969 	uint32_t frames_to_send, num_free_eq_desc;
970 	uint32_t loop, retry_count;
971 	int32_t ret;
972 	uint16_t num_tx = 0;
973 	uint16_t bpid;
974 
975 	if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
976 		ret = dpaa2_affine_qbman_swp();
977 		if (ret) {
978 			DPAA2_PMD_ERR("Failure in affining portal");
979 			return 0;
980 		}
981 	}
982 	swp = DPAA2_PER_LCORE_PORTAL;
983 
984 	DPAA2_PMD_DP_DEBUG("===> eth_data =%p, fqid =%d\n",
985 			   eth_data, dpaa2_q->fqid);
986 
987 	/* This would also handle normal and atomic queues as any type
988 	 * of packet can be enqueued when ordered queues are being used.
989 	 */
990 	while (nb_pkts) {
991 		/*Check if the queue is congested*/
992 		retry_count = 0;
993 		while (qbman_result_SCN_state(dpaa2_q->cscn)) {
994 			retry_count++;
995 			/* Retry for some time before giving up */
996 			if (retry_count > CONG_RETRY_COUNT)
997 				goto skip_tx;
998 		}
999 
1000 		frames_to_send = (nb_pkts > dpaa2_eqcr_size) ?
1001 			dpaa2_eqcr_size : nb_pkts;
1002 
1003 		if (!priv->en_loose_ordered) {
1004 			if ((*bufs)->seqn & DPAA2_ENQUEUE_FLAG_ORP) {
1005 				num_free_eq_desc = dpaa2_free_eq_descriptors();
1006 				if (num_free_eq_desc < frames_to_send)
1007 					frames_to_send = num_free_eq_desc;
1008 			}
1009 		}
1010 
1011 		for (loop = 0; loop < frames_to_send; loop++) {
1012 			/*Prepare enqueue descriptor*/
1013 			qbman_eq_desc_clear(&eqdesc[loop]);
1014 
1015 			if ((*bufs)->seqn) {
1016 				/* Use only queue 0 for Tx in case of atomic/
1017 				 * ordered packets as packets can get unordered
1018 				 * when being tranmitted out from the interface
1019 				 */
1020 				dpaa2_set_enqueue_descriptor(order_sendq,
1021 							     (*bufs),
1022 							     &eqdesc[loop]);
1023 			} else {
1024 				qbman_eq_desc_set_no_orp(&eqdesc[loop],
1025 							 DPAA2_EQ_RESP_ERR_FQ);
1026 				qbman_eq_desc_set_qd(&eqdesc[loop], priv->qdid,
1027 						     dpaa2_q->flow_id,
1028 						     dpaa2_q->tc_index);
1029 			}
1030 
1031 			if (likely(RTE_MBUF_DIRECT(*bufs))) {
1032 				mp = (*bufs)->pool;
1033 				/* Check the basic scenario and set
1034 				 * the FD appropriately here itself.
1035 				 */
1036 				if (likely(mp && mp->ops_index ==
1037 				    priv->bp_list->dpaa2_ops_index &&
1038 				    (*bufs)->nb_segs == 1 &&
1039 				    rte_mbuf_refcnt_read((*bufs)) == 1)) {
1040 					if (unlikely((*bufs)->ol_flags
1041 						& PKT_TX_VLAN_PKT)) {
1042 					  ret = rte_vlan_insert(bufs);
1043 					  if (ret)
1044 						goto send_n_return;
1045 					}
1046 					DPAA2_MBUF_TO_CONTIG_FD((*bufs),
1047 						&fd_arr[loop],
1048 						mempool_to_bpid(mp));
1049 					bufs++;
1050 					continue;
1051 				}
1052 			} else {
1053 				mi = rte_mbuf_from_indirect(*bufs);
1054 				mp = mi->pool;
1055 			}
1056 			/* Not a hw_pkt pool allocated frame */
1057 			if (unlikely(!mp || !priv->bp_list)) {
1058 				DPAA2_PMD_ERR("Err: No buffer pool attached");
1059 				goto send_n_return;
1060 			}
1061 
1062 			if (mp->ops_index != priv->bp_list->dpaa2_ops_index) {
1063 				DPAA2_PMD_WARN("Non DPAA2 buffer pool");
1064 				/* alloc should be from the default buffer pool
1065 				 * attached to this interface
1066 				 */
1067 				bpid = priv->bp_list->buf_pool.bpid;
1068 
1069 				if (unlikely((*bufs)->nb_segs > 1)) {
1070 					DPAA2_PMD_ERR(
1071 						"S/G not supp for non hw offload buffer");
1072 					goto send_n_return;
1073 				}
1074 				if (eth_copy_mbuf_to_fd(*bufs,
1075 							&fd_arr[loop], bpid)) {
1076 					goto send_n_return;
1077 				}
1078 				/* free the original packet */
1079 				rte_pktmbuf_free(*bufs);
1080 			} else {
1081 				bpid = mempool_to_bpid(mp);
1082 				if (unlikely((*bufs)->nb_segs > 1)) {
1083 					if (eth_mbuf_to_sg_fd(*bufs,
1084 							      &fd_arr[loop],
1085 							      bpid))
1086 						goto send_n_return;
1087 				} else {
1088 					eth_mbuf_to_fd(*bufs,
1089 						       &fd_arr[loop], bpid);
1090 				}
1091 			}
1092 			bufs++;
1093 		}
1094 		loop = 0;
1095 		while (loop < frames_to_send) {
1096 			loop += qbman_swp_enqueue_multiple_desc(swp,
1097 					&eqdesc[loop], &fd_arr[loop],
1098 					frames_to_send - loop);
1099 		}
1100 
1101 		num_tx += frames_to_send;
1102 		nb_pkts -= frames_to_send;
1103 	}
1104 	dpaa2_q->tx_pkts += num_tx;
1105 	return num_tx;
1106 
1107 send_n_return:
1108 	/* send any already prepared fd */
1109 	if (loop) {
1110 		unsigned int i = 0;
1111 
1112 		while (i < loop) {
1113 			i += qbman_swp_enqueue_multiple_desc(swp, &eqdesc[loop],
1114 							&fd_arr[i], loop - i);
1115 		}
1116 		num_tx += loop;
1117 	}
1118 skip_tx:
1119 	dpaa2_q->tx_pkts += num_tx;
1120 	return num_tx;
1121 }
1122 
1123 /**
1124  * Dummy DPDK callback for TX.
1125  *
1126  * This function is used to temporarily replace the real callback during
1127  * unsafe control operations on the queue, or in case of error.
1128  *
1129  * @param dpdk_txq
1130  *   Generic pointer to TX queue structure.
1131  * @param[in] pkts
1132  *   Packets to transmit.
1133  * @param pkts_n
1134  *   Number of packets in array.
1135  *
1136  * @return
1137  *   Number of packets successfully transmitted (<= pkts_n).
1138  */
1139 uint16_t
1140 dummy_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
1141 {
1142 	(void)queue;
1143 	(void)bufs;
1144 	(void)nb_pkts;
1145 	return 0;
1146 }
1147 
1148 #if defined(RTE_TOOLCHAIN_GCC)
1149 #pragma GCC diagnostic push
1150 #pragma GCC diagnostic ignored "-Wcast-qual"
1151 #elif defined(RTE_TOOLCHAIN_CLANG)
1152 #pragma clang diagnostic push
1153 #pragma clang diagnostic ignored "-Wcast-qual"
1154 #endif
1155 
1156 /* This function loopbacks all the received packets.*/
1157 uint16_t
1158 dpaa2_dev_loopback_rx(void *queue,
1159 		      struct rte_mbuf **bufs __rte_unused,
1160 		      uint16_t nb_pkts)
1161 {
1162 	/* Function receive frames for a given device and VQ*/
1163 	struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)queue;
1164 	struct qbman_result *dq_storage, *dq_storage1 = NULL;
1165 	uint32_t fqid = dpaa2_q->fqid;
1166 	int ret, num_rx = 0, num_tx = 0, pull_size;
1167 	uint8_t pending, status;
1168 	struct qbman_swp *swp;
1169 	struct qbman_fd *fd[DPAA2_LX2_DQRR_RING_SIZE];
1170 	struct qbman_pull_desc pulldesc;
1171 	struct qbman_eq_desc eqdesc;
1172 	struct queue_storage_info_t *q_storage = dpaa2_q->q_storage;
1173 	struct rte_eth_dev_data *eth_data = dpaa2_q->eth_data;
1174 	struct dpaa2_dev_priv *priv = eth_data->dev_private;
1175 	struct dpaa2_queue *tx_q = priv->tx_vq[0];
1176 	/* todo - currently we are using 1st TX queue only for loopback*/
1177 
1178 	if (unlikely(!DPAA2_PER_LCORE_ETHRX_DPIO)) {
1179 		ret = dpaa2_affine_qbman_ethrx_swp();
1180 		if (ret) {
1181 			DPAA2_PMD_ERR("Failure in affining portal");
1182 			return 0;
1183 		}
1184 	}
1185 	swp = DPAA2_PER_LCORE_ETHRX_PORTAL;
1186 	pull_size = (nb_pkts > dpaa2_dqrr_size) ? dpaa2_dqrr_size : nb_pkts;
1187 	if (unlikely(!q_storage->active_dqs)) {
1188 		q_storage->toggle = 0;
1189 		dq_storage = q_storage->dq_storage[q_storage->toggle];
1190 		q_storage->last_num_pkts = pull_size;
1191 		qbman_pull_desc_clear(&pulldesc);
1192 		qbman_pull_desc_set_numframes(&pulldesc,
1193 					      q_storage->last_num_pkts);
1194 		qbman_pull_desc_set_fq(&pulldesc, fqid);
1195 		qbman_pull_desc_set_storage(&pulldesc, dq_storage,
1196 			(size_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
1197 		if (check_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)) {
1198 			while (!qbman_check_command_complete(
1199 			       get_swp_active_dqs(
1200 			       DPAA2_PER_LCORE_ETHRX_DPIO->index)))
1201 				;
1202 			clear_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index);
1203 		}
1204 		while (1) {
1205 			if (qbman_swp_pull(swp, &pulldesc)) {
1206 				DPAA2_PMD_DP_DEBUG(
1207 					"VDQ command not issued.QBMAN busy\n");
1208 				/* Portal was busy, try again */
1209 				continue;
1210 			}
1211 			break;
1212 		}
1213 		q_storage->active_dqs = dq_storage;
1214 		q_storage->active_dpio_id = DPAA2_PER_LCORE_ETHRX_DPIO->index;
1215 		set_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index,
1216 				   dq_storage);
1217 	}
1218 
1219 	dq_storage = q_storage->active_dqs;
1220 	rte_prefetch0((void *)(size_t)(dq_storage));
1221 	rte_prefetch0((void *)(size_t)(dq_storage + 1));
1222 
1223 	/* Prepare next pull descriptor. This will give space for the
1224 	 * prefething done on DQRR entries
1225 	 */
1226 	q_storage->toggle ^= 1;
1227 	dq_storage1 = q_storage->dq_storage[q_storage->toggle];
1228 	qbman_pull_desc_clear(&pulldesc);
1229 	qbman_pull_desc_set_numframes(&pulldesc, pull_size);
1230 	qbman_pull_desc_set_fq(&pulldesc, fqid);
1231 	qbman_pull_desc_set_storage(&pulldesc, dq_storage1,
1232 		(size_t)(DPAA2_VADDR_TO_IOVA(dq_storage1)), 1);
1233 
1234 	/*Prepare enqueue descriptor*/
1235 	qbman_eq_desc_clear(&eqdesc);
1236 	qbman_eq_desc_set_no_orp(&eqdesc, DPAA2_EQ_RESP_ERR_FQ);
1237 	qbman_eq_desc_set_response(&eqdesc, 0, 0);
1238 	qbman_eq_desc_set_fq(&eqdesc, tx_q->fqid);
1239 
1240 	/* Check if the previous issued command is completed.
1241 	 * Also seems like the SWP is shared between the Ethernet Driver
1242 	 * and the SEC driver.
1243 	 */
1244 	while (!qbman_check_command_complete(dq_storage))
1245 		;
1246 	if (dq_storage == get_swp_active_dqs(q_storage->active_dpio_id))
1247 		clear_swp_active_dqs(q_storage->active_dpio_id);
1248 
1249 	pending = 1;
1250 
1251 	do {
1252 		/* Loop until the dq_storage is updated with
1253 		 * new token by QBMAN
1254 		 */
1255 		while (!qbman_check_new_result(dq_storage))
1256 			;
1257 		rte_prefetch0((void *)((size_t)(dq_storage + 2)));
1258 		/* Check whether Last Pull command is Expired and
1259 		 * setting Condition for Loop termination
1260 		 */
1261 		if (qbman_result_DQ_is_pull_complete(dq_storage)) {
1262 			pending = 0;
1263 			/* Check for valid frame. */
1264 			status = qbman_result_DQ_flags(dq_storage);
1265 			if (unlikely((status & QBMAN_DQ_STAT_VALIDFRAME) == 0))
1266 				continue;
1267 		}
1268 		fd[num_rx] = (struct qbman_fd *)qbman_result_DQ_fd(dq_storage);
1269 
1270 		dq_storage++;
1271 		num_rx++;
1272 	} while (pending);
1273 
1274 	while (num_tx < num_rx) {
1275 		num_tx += qbman_swp_enqueue_multiple_fd(swp, &eqdesc,
1276 				&fd[num_tx], 0, num_rx - num_tx);
1277 	}
1278 
1279 	if (check_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)) {
1280 		while (!qbman_check_command_complete(
1281 		       get_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index)))
1282 			;
1283 		clear_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index);
1284 	}
1285 	/* issue a volatile dequeue command for next pull */
1286 	while (1) {
1287 		if (qbman_swp_pull(swp, &pulldesc)) {
1288 			DPAA2_PMD_DP_DEBUG("VDQ command is not issued."
1289 					  "QBMAN is busy (2)\n");
1290 			continue;
1291 		}
1292 		break;
1293 	}
1294 	q_storage->active_dqs = dq_storage1;
1295 	q_storage->active_dpio_id = DPAA2_PER_LCORE_ETHRX_DPIO->index;
1296 	set_swp_active_dqs(DPAA2_PER_LCORE_ETHRX_DPIO->index, dq_storage1);
1297 
1298 	dpaa2_q->rx_pkts += num_rx;
1299 	dpaa2_q->tx_pkts += num_tx;
1300 
1301 	return 0;
1302 }
1303 #if defined(RTE_TOOLCHAIN_GCC)
1304 #pragma GCC diagnostic pop
1305 #elif defined(RTE_TOOLCHAIN_CLANG)
1306 #pragma clang diagnostic pop
1307 #endif
1308