xref: /f-stack/dpdk/drivers/net/ena/ena_ethdev.c (revision 2bfe3f2e)
1 /*-
2 * BSD LICENSE
3 *
4 * Copyright (c) 2015-2016 Amazon.com, Inc. or its affiliates.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * * Neither the name of copyright holder nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33 
34 #include <rte_ether.h>
35 #include <rte_ethdev.h>
36 #include <rte_ethdev_pci.h>
37 #include <rte_tcp.h>
38 #include <rte_atomic.h>
39 #include <rte_dev.h>
40 #include <rte_errno.h>
41 #include <rte_version.h>
42 #include <rte_eal_memconfig.h>
43 #include <rte_net.h>
44 
45 #include "ena_ethdev.h"
46 #include "ena_logs.h"
47 #include "ena_platform.h"
48 #include "ena_com.h"
49 #include "ena_eth_com.h"
50 
51 #include <ena_common_defs.h>
52 #include <ena_regs_defs.h>
53 #include <ena_admin_defs.h>
54 #include <ena_eth_io_defs.h>
55 
56 #define DRV_MODULE_VER_MAJOR	1
57 #define DRV_MODULE_VER_MINOR	0
58 #define DRV_MODULE_VER_SUBMINOR	0
59 
60 #define ENA_IO_TXQ_IDX(q)	(2 * (q))
61 #define ENA_IO_RXQ_IDX(q)	(2 * (q) + 1)
62 /*reverse version of ENA_IO_RXQ_IDX*/
63 #define ENA_IO_RXQ_IDX_REV(q)	((q - 1) / 2)
64 
65 /* While processing submitted and completed descriptors (rx and tx path
66  * respectively) in a loop it is desired to:
67  *  - perform batch submissions while populating sumbissmion queue
68  *  - avoid blocking transmission of other packets during cleanup phase
69  * Hence the utilization ratio of 1/8 of a queue size.
70  */
71 #define ENA_RING_DESCS_RATIO(ring_size)	(ring_size / 8)
72 
73 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
74 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
75 
76 #define GET_L4_HDR_LEN(mbuf)					\
77 	((rte_pktmbuf_mtod_offset(mbuf,	struct tcp_hdr *,	\
78 		mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
79 
80 #define ENA_RX_RSS_TABLE_LOG_SIZE  7
81 #define ENA_RX_RSS_TABLE_SIZE	(1 << ENA_RX_RSS_TABLE_LOG_SIZE)
82 #define ENA_HASH_KEY_SIZE	40
83 #define ENA_ETH_SS_STATS	0xFF
84 #define ETH_GSTRING_LEN	32
85 
86 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
87 
88 enum ethtool_stringset {
89 	ETH_SS_TEST             = 0,
90 	ETH_SS_STATS,
91 };
92 
93 struct ena_stats {
94 	char name[ETH_GSTRING_LEN];
95 	int stat_offset;
96 };
97 
98 #define ENA_STAT_ENA_COM_ENTRY(stat) { \
99 	.name = #stat, \
100 	.stat_offset = offsetof(struct ena_com_stats_admin, stat) \
101 }
102 
103 #define ENA_STAT_ENTRY(stat, stat_type) { \
104 	.name = #stat, \
105 	.stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
106 }
107 
108 #define ENA_STAT_RX_ENTRY(stat) \
109 	ENA_STAT_ENTRY(stat, rx)
110 
111 #define ENA_STAT_TX_ENTRY(stat) \
112 	ENA_STAT_ENTRY(stat, tx)
113 
114 #define ENA_STAT_GLOBAL_ENTRY(stat) \
115 	ENA_STAT_ENTRY(stat, dev)
116 
117 static const struct ena_stats ena_stats_global_strings[] = {
118 	ENA_STAT_GLOBAL_ENTRY(tx_timeout),
119 	ENA_STAT_GLOBAL_ENTRY(io_suspend),
120 	ENA_STAT_GLOBAL_ENTRY(io_resume),
121 	ENA_STAT_GLOBAL_ENTRY(wd_expired),
122 	ENA_STAT_GLOBAL_ENTRY(interface_up),
123 	ENA_STAT_GLOBAL_ENTRY(interface_down),
124 	ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
125 };
126 
127 static const struct ena_stats ena_stats_tx_strings[] = {
128 	ENA_STAT_TX_ENTRY(cnt),
129 	ENA_STAT_TX_ENTRY(bytes),
130 	ENA_STAT_TX_ENTRY(queue_stop),
131 	ENA_STAT_TX_ENTRY(queue_wakeup),
132 	ENA_STAT_TX_ENTRY(dma_mapping_err),
133 	ENA_STAT_TX_ENTRY(linearize),
134 	ENA_STAT_TX_ENTRY(linearize_failed),
135 	ENA_STAT_TX_ENTRY(tx_poll),
136 	ENA_STAT_TX_ENTRY(doorbells),
137 	ENA_STAT_TX_ENTRY(prepare_ctx_err),
138 	ENA_STAT_TX_ENTRY(missing_tx_comp),
139 	ENA_STAT_TX_ENTRY(bad_req_id),
140 };
141 
142 static const struct ena_stats ena_stats_rx_strings[] = {
143 	ENA_STAT_RX_ENTRY(cnt),
144 	ENA_STAT_RX_ENTRY(bytes),
145 	ENA_STAT_RX_ENTRY(refil_partial),
146 	ENA_STAT_RX_ENTRY(bad_csum),
147 	ENA_STAT_RX_ENTRY(page_alloc_fail),
148 	ENA_STAT_RX_ENTRY(skb_alloc_fail),
149 	ENA_STAT_RX_ENTRY(dma_mapping_err),
150 	ENA_STAT_RX_ENTRY(bad_desc_num),
151 	ENA_STAT_RX_ENTRY(small_copy_len_pkt),
152 };
153 
154 static const struct ena_stats ena_stats_ena_com_strings[] = {
155 	ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
156 	ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
157 	ENA_STAT_ENA_COM_ENTRY(completed_cmd),
158 	ENA_STAT_ENA_COM_ENTRY(out_of_space),
159 	ENA_STAT_ENA_COM_ENTRY(no_completion),
160 };
161 
162 #define ENA_STATS_ARRAY_GLOBAL	ARRAY_SIZE(ena_stats_global_strings)
163 #define ENA_STATS_ARRAY_TX	ARRAY_SIZE(ena_stats_tx_strings)
164 #define ENA_STATS_ARRAY_RX	ARRAY_SIZE(ena_stats_rx_strings)
165 #define ENA_STATS_ARRAY_ENA_COM	ARRAY_SIZE(ena_stats_ena_com_strings)
166 
167 /** Vendor ID used by Amazon devices */
168 #define PCI_VENDOR_ID_AMAZON 0x1D0F
169 /** Amazon devices */
170 #define PCI_DEVICE_ID_ENA_VF	0xEC20
171 #define PCI_DEVICE_ID_ENA_LLQ_VF	0xEC21
172 
173 #define	ENA_TX_OFFLOAD_MASK	(\
174 	PKT_TX_L4_MASK |         \
175 	PKT_TX_IP_CKSUM |        \
176 	PKT_TX_TCP_SEG)
177 
178 #define	ENA_TX_OFFLOAD_NOTSUP_MASK	\
179 	(PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
180 
181 static const struct rte_pci_id pci_id_ena_map[] = {
182 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
183 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
184 	{ .device_id = 0 },
185 };
186 
187 static int ena_device_init(struct ena_com_dev *ena_dev,
188 			   struct ena_com_dev_get_features_ctx *get_feat_ctx);
189 static int ena_dev_configure(struct rte_eth_dev *dev);
190 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
191 				  uint16_t nb_pkts);
192 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
193 		uint16_t nb_pkts);
194 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
195 			      uint16_t nb_desc, unsigned int socket_id,
196 			      const struct rte_eth_txconf *tx_conf);
197 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
198 			      uint16_t nb_desc, unsigned int socket_id,
199 			      const struct rte_eth_rxconf *rx_conf,
200 			      struct rte_mempool *mp);
201 static uint16_t eth_ena_recv_pkts(void *rx_queue,
202 				  struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
203 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
204 static void ena_init_rings(struct ena_adapter *adapter);
205 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
206 static int ena_start(struct rte_eth_dev *dev);
207 static void ena_close(struct rte_eth_dev *dev);
208 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
209 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
210 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
211 static void ena_rx_queue_release(void *queue);
212 static void ena_tx_queue_release(void *queue);
213 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
214 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
215 static int ena_link_update(struct rte_eth_dev *dev,
216 			   int wait_to_complete);
217 static int ena_queue_restart(struct ena_ring *ring);
218 static int ena_queue_restart_all(struct rte_eth_dev *dev,
219 				 enum ena_ring_type ring_type);
220 static void ena_stats_restart(struct rte_eth_dev *dev);
221 static void ena_infos_get(struct rte_eth_dev *dev,
222 			  struct rte_eth_dev_info *dev_info);
223 static int ena_rss_reta_update(struct rte_eth_dev *dev,
224 			       struct rte_eth_rss_reta_entry64 *reta_conf,
225 			       uint16_t reta_size);
226 static int ena_rss_reta_query(struct rte_eth_dev *dev,
227 			      struct rte_eth_rss_reta_entry64 *reta_conf,
228 			      uint16_t reta_size);
229 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
230 
231 static const struct eth_dev_ops ena_dev_ops = {
232 	.dev_configure        = ena_dev_configure,
233 	.dev_infos_get        = ena_infos_get,
234 	.rx_queue_setup       = ena_rx_queue_setup,
235 	.tx_queue_setup       = ena_tx_queue_setup,
236 	.dev_start            = ena_start,
237 	.link_update          = ena_link_update,
238 	.stats_get            = ena_stats_get,
239 	.mtu_set              = ena_mtu_set,
240 	.rx_queue_release     = ena_rx_queue_release,
241 	.tx_queue_release     = ena_tx_queue_release,
242 	.dev_close            = ena_close,
243 	.reta_update          = ena_rss_reta_update,
244 	.reta_query           = ena_rss_reta_query,
245 };
246 
247 #define NUMA_NO_NODE	SOCKET_ID_ANY
248 
249 static inline int ena_cpu_to_node(int cpu)
250 {
251 	struct rte_config *config = rte_eal_get_configuration();
252 
253 	if (likely(cpu < RTE_MAX_MEMZONE))
254 		return config->mem_config->memzone[cpu].socket_id;
255 
256 	return NUMA_NO_NODE;
257 }
258 
259 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
260 				       struct ena_com_rx_ctx *ena_rx_ctx)
261 {
262 	uint64_t ol_flags = 0;
263 	uint32_t packet_type = 0;
264 
265 	if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
266 		packet_type |= RTE_PTYPE_L4_TCP;
267 	else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
268 		packet_type |= RTE_PTYPE_L4_UDP;
269 
270 	if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
271 		packet_type |= RTE_PTYPE_L3_IPV4;
272 	else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
273 		packet_type |= RTE_PTYPE_L3_IPV6;
274 
275 	if (unlikely(ena_rx_ctx->l4_csum_err))
276 		ol_flags |= PKT_RX_L4_CKSUM_BAD;
277 	if (unlikely(ena_rx_ctx->l3_csum_err))
278 		ol_flags |= PKT_RX_IP_CKSUM_BAD;
279 
280 	mbuf->ol_flags = ol_flags;
281 	mbuf->packet_type = packet_type;
282 }
283 
284 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
285 				       struct ena_com_tx_ctx *ena_tx_ctx)
286 {
287 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
288 
289 	if (mbuf->ol_flags &
290 	    (PKT_TX_L4_MASK | PKT_TX_IP_CKSUM | PKT_TX_TCP_SEG)) {
291 		/* check if TSO is required */
292 		if (mbuf->ol_flags & PKT_TX_TCP_SEG) {
293 			ena_tx_ctx->tso_enable = true;
294 
295 			ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
296 		}
297 
298 		/* check if L3 checksum is needed */
299 		if (mbuf->ol_flags & PKT_TX_IP_CKSUM)
300 			ena_tx_ctx->l3_csum_enable = true;
301 
302 		if (mbuf->ol_flags & PKT_TX_IPV6) {
303 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
304 		} else {
305 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
306 
307 			/* set don't fragment (DF) flag */
308 			if (mbuf->packet_type &
309 				(RTE_PTYPE_L4_NONFRAG
310 				 | RTE_PTYPE_INNER_L4_NONFRAG))
311 				ena_tx_ctx->df = true;
312 		}
313 
314 		/* check if L4 checksum is needed */
315 		switch (mbuf->ol_flags & PKT_TX_L4_MASK) {
316 		case PKT_TX_TCP_CKSUM:
317 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
318 			ena_tx_ctx->l4_csum_enable = true;
319 			break;
320 		case PKT_TX_UDP_CKSUM:
321 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
322 			ena_tx_ctx->l4_csum_enable = true;
323 			break;
324 		default:
325 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
326 			ena_tx_ctx->l4_csum_enable = false;
327 			break;
328 		}
329 
330 		ena_meta->mss = mbuf->tso_segsz;
331 		ena_meta->l3_hdr_len = mbuf->l3_len;
332 		ena_meta->l3_hdr_offset = mbuf->l2_len;
333 		/* this param needed only for TSO */
334 		ena_meta->l3_outer_hdr_len = 0;
335 		ena_meta->l3_outer_hdr_offset = 0;
336 
337 		ena_tx_ctx->meta_valid = true;
338 	} else {
339 		ena_tx_ctx->meta_valid = false;
340 	}
341 }
342 
343 static void ena_config_host_info(struct ena_com_dev *ena_dev)
344 {
345 	struct ena_admin_host_info *host_info;
346 	int rc;
347 
348 	/* Allocate only the host info */
349 	rc = ena_com_allocate_host_info(ena_dev);
350 	if (rc) {
351 		RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
352 		return;
353 	}
354 
355 	host_info = ena_dev->host_attr.host_info;
356 
357 	host_info->os_type = ENA_ADMIN_OS_DPDK;
358 	host_info->kernel_ver = RTE_VERSION;
359 	snprintf((char *)host_info->kernel_ver_str,
360 		 sizeof(host_info->kernel_ver_str),
361 		 "%s", rte_version());
362 	host_info->os_dist = RTE_VERSION;
363 	snprintf((char *)host_info->os_dist_str,
364 		 sizeof(host_info->os_dist_str),
365 		 "%s", rte_version());
366 	host_info->driver_version =
367 		(DRV_MODULE_VER_MAJOR) |
368 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
369 		(DRV_MODULE_VER_SUBMINOR <<
370 			ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
371 
372 	rc = ena_com_set_host_attributes(ena_dev);
373 	if (rc) {
374 		RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
375 		if (rc != -EPERM)
376 			goto err;
377 	}
378 
379 	return;
380 
381 err:
382 	ena_com_delete_host_info(ena_dev);
383 }
384 
385 static int
386 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
387 {
388 	if (sset != ETH_SS_STATS)
389 		return -EOPNOTSUPP;
390 
391 	 /* Workaround for clang:
392 	 * touch internal structures to prevent
393 	 * compiler error
394 	 */
395 	ENA_TOUCH(ena_stats_global_strings);
396 	ENA_TOUCH(ena_stats_tx_strings);
397 	ENA_TOUCH(ena_stats_rx_strings);
398 	ENA_TOUCH(ena_stats_ena_com_strings);
399 
400 	return  dev->data->nb_tx_queues *
401 		(ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
402 		ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
403 }
404 
405 static void ena_config_debug_area(struct ena_adapter *adapter)
406 {
407 	u32 debug_area_size;
408 	int rc, ss_count;
409 
410 	ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
411 	if (ss_count <= 0) {
412 		RTE_LOG(ERR, PMD, "SS count is negative\n");
413 		return;
414 	}
415 
416 	/* allocate 32 bytes for each string and 64bit for the value */
417 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
418 
419 	rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
420 	if (rc) {
421 		RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
422 		return;
423 	}
424 
425 	rc = ena_com_set_host_attributes(&adapter->ena_dev);
426 	if (rc) {
427 		RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
428 		if (rc != -EPERM)
429 			goto err;
430 	}
431 
432 	return;
433 err:
434 	ena_com_delete_debug_area(&adapter->ena_dev);
435 }
436 
437 static void ena_close(struct rte_eth_dev *dev)
438 {
439 	struct ena_adapter *adapter =
440 		(struct ena_adapter *)(dev->data->dev_private);
441 
442 	adapter->state = ENA_ADAPTER_STATE_STOPPED;
443 
444 	ena_rx_queue_release_all(dev);
445 	ena_tx_queue_release_all(dev);
446 }
447 
448 static int ena_rss_reta_update(struct rte_eth_dev *dev,
449 			       struct rte_eth_rss_reta_entry64 *reta_conf,
450 			       uint16_t reta_size)
451 {
452 	struct ena_adapter *adapter =
453 		(struct ena_adapter *)(dev->data->dev_private);
454 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
455 	int ret, i;
456 	u16 entry_value;
457 	int conf_idx;
458 	int idx;
459 
460 	if ((reta_size == 0) || (reta_conf == NULL))
461 		return -EINVAL;
462 
463 	if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
464 		RTE_LOG(WARNING, PMD,
465 			"indirection table %d is bigger than supported (%d)\n",
466 			reta_size, ENA_RX_RSS_TABLE_SIZE);
467 		ret = -EINVAL;
468 		goto err;
469 	}
470 
471 	for (i = 0 ; i < reta_size ; i++) {
472 		/* each reta_conf is for 64 entries.
473 		 * to support 128 we use 2 conf of 64
474 		 */
475 		conf_idx = i / RTE_RETA_GROUP_SIZE;
476 		idx = i % RTE_RETA_GROUP_SIZE;
477 		if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
478 			entry_value =
479 				ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
480 			ret = ena_com_indirect_table_fill_entry(ena_dev,
481 								i,
482 								entry_value);
483 			if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
484 				RTE_LOG(ERR, PMD,
485 					"Cannot fill indirect table\n");
486 				ret = -ENOTSUP;
487 				goto err;
488 			}
489 		}
490 	}
491 
492 	ret = ena_com_indirect_table_set(ena_dev);
493 	if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
494 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
495 		ret = -ENOTSUP;
496 		goto err;
497 	}
498 
499 	RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries  for port %d\n",
500 		__func__, reta_size, adapter->rte_dev->data->port_id);
501 err:
502 	return ret;
503 }
504 
505 /* Query redirection table. */
506 static int ena_rss_reta_query(struct rte_eth_dev *dev,
507 			      struct rte_eth_rss_reta_entry64 *reta_conf,
508 			      uint16_t reta_size)
509 {
510 	struct ena_adapter *adapter =
511 		(struct ena_adapter *)(dev->data->dev_private);
512 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
513 	int ret;
514 	int i;
515 	u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
516 	int reta_conf_idx;
517 	int reta_idx;
518 
519 	if (reta_size == 0 || reta_conf == NULL ||
520 	    (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
521 		return -EINVAL;
522 
523 	ret = ena_com_indirect_table_get(ena_dev, indirect_table);
524 	if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
525 		RTE_LOG(ERR, PMD, "cannot get indirect table\n");
526 		ret = -ENOTSUP;
527 		goto err;
528 	}
529 
530 	for (i = 0 ; i < reta_size ; i++) {
531 		reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
532 		reta_idx = i % RTE_RETA_GROUP_SIZE;
533 		if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
534 			reta_conf[reta_conf_idx].reta[reta_idx] =
535 				ENA_IO_RXQ_IDX_REV(indirect_table[i]);
536 	}
537 err:
538 	return ret;
539 }
540 
541 static int ena_rss_init_default(struct ena_adapter *adapter)
542 {
543 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
544 	uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
545 	int rc, i;
546 	u32 val;
547 
548 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
549 	if (unlikely(rc)) {
550 		RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
551 		goto err_rss_init;
552 	}
553 
554 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
555 		val = i % nb_rx_queues;
556 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
557 						       ENA_IO_RXQ_IDX(val));
558 		if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
559 			RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
560 			goto err_fill_indir;
561 		}
562 	}
563 
564 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
565 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
566 	if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
567 		RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
568 		goto err_fill_indir;
569 	}
570 
571 	rc = ena_com_set_default_hash_ctrl(ena_dev);
572 	if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
573 		RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
574 		goto err_fill_indir;
575 	}
576 
577 	rc = ena_com_indirect_table_set(ena_dev);
578 	if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
579 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
580 		goto err_fill_indir;
581 	}
582 	RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
583 		adapter->rte_dev->data->port_id);
584 
585 	return 0;
586 
587 err_fill_indir:
588 	ena_com_rss_destroy(ena_dev);
589 err_rss_init:
590 
591 	return rc;
592 }
593 
594 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
595 {
596 	struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
597 	int nb_queues = dev->data->nb_rx_queues;
598 	int i;
599 
600 	for (i = 0; i < nb_queues; i++)
601 		ena_rx_queue_release(queues[i]);
602 }
603 
604 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
605 {
606 	struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
607 	int nb_queues = dev->data->nb_tx_queues;
608 	int i;
609 
610 	for (i = 0; i < nb_queues; i++)
611 		ena_tx_queue_release(queues[i]);
612 }
613 
614 static void ena_rx_queue_release(void *queue)
615 {
616 	struct ena_ring *ring = (struct ena_ring *)queue;
617 	struct ena_adapter *adapter = ring->adapter;
618 	int ena_qid;
619 
620 	ena_assert_msg(ring->configured,
621 		       "API violation - releasing not configured queue");
622 	ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
623 		       "API violation");
624 
625 	/* Destroy HW queue */
626 	ena_qid = ENA_IO_RXQ_IDX(ring->id);
627 	ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
628 
629 	/* Free all bufs */
630 	ena_rx_queue_release_bufs(ring);
631 
632 	/* Free ring resources */
633 	if (ring->rx_buffer_info)
634 		rte_free(ring->rx_buffer_info);
635 	ring->rx_buffer_info = NULL;
636 
637 	ring->configured = 0;
638 
639 	RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
640 		ring->port_id, ring->id);
641 }
642 
643 static void ena_tx_queue_release(void *queue)
644 {
645 	struct ena_ring *ring = (struct ena_ring *)queue;
646 	struct ena_adapter *adapter = ring->adapter;
647 	int ena_qid;
648 
649 	ena_assert_msg(ring->configured,
650 		       "API violation. Releasing not configured queue");
651 	ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
652 		       "API violation");
653 
654 	/* Destroy HW queue */
655 	ena_qid = ENA_IO_TXQ_IDX(ring->id);
656 	ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
657 
658 	/* Free all bufs */
659 	ena_tx_queue_release_bufs(ring);
660 
661 	/* Free ring resources */
662 	if (ring->tx_buffer_info)
663 		rte_free(ring->tx_buffer_info);
664 
665 	if (ring->empty_tx_reqs)
666 		rte_free(ring->empty_tx_reqs);
667 
668 	ring->empty_tx_reqs = NULL;
669 	ring->tx_buffer_info = NULL;
670 
671 	ring->configured = 0;
672 
673 	RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
674 		ring->port_id, ring->id);
675 }
676 
677 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
678 {
679 	unsigned int ring_mask = ring->ring_size - 1;
680 
681 	while (ring->next_to_clean != ring->next_to_use) {
682 		struct rte_mbuf *m =
683 			ring->rx_buffer_info[ring->next_to_clean & ring_mask];
684 
685 		if (m)
686 			rte_mbuf_raw_free(m);
687 
688 		ring->next_to_clean++;
689 	}
690 }
691 
692 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
693 {
694 	unsigned int i;
695 
696 	for (i = 0; i < ring->ring_size; ++i) {
697 		struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
698 
699 		if (tx_buf->mbuf)
700 			rte_pktmbuf_free(tx_buf->mbuf);
701 
702 		ring->next_to_clean++;
703 	}
704 }
705 
706 static int ena_link_update(struct rte_eth_dev *dev,
707 			   __rte_unused int wait_to_complete)
708 {
709 	struct rte_eth_link *link = &dev->data->dev_link;
710 
711 	link->link_status = 1;
712 	link->link_speed = ETH_SPEED_NUM_10G;
713 	link->link_duplex = ETH_LINK_FULL_DUPLEX;
714 
715 	return 0;
716 }
717 
718 static int ena_queue_restart_all(struct rte_eth_dev *dev,
719 				 enum ena_ring_type ring_type)
720 {
721 	struct ena_adapter *adapter =
722 		(struct ena_adapter *)(dev->data->dev_private);
723 	struct ena_ring *queues = NULL;
724 	int i = 0;
725 	int rc = 0;
726 
727 	queues = (ring_type == ENA_RING_TYPE_RX) ?
728 		adapter->rx_ring : adapter->tx_ring;
729 
730 	for (i = 0; i < adapter->num_queues; i++) {
731 		if (queues[i].configured) {
732 			if (ring_type == ENA_RING_TYPE_RX) {
733 				ena_assert_msg(
734 					dev->data->rx_queues[i] == &queues[i],
735 					"Inconsistent state of rx queues\n");
736 			} else {
737 				ena_assert_msg(
738 					dev->data->tx_queues[i] == &queues[i],
739 					"Inconsistent state of tx queues\n");
740 			}
741 
742 			rc = ena_queue_restart(&queues[i]);
743 
744 			if (rc) {
745 				PMD_INIT_LOG(ERR,
746 					     "failed to restart queue %d type(%d)",
747 					     i, ring_type);
748 				return -1;
749 			}
750 		}
751 	}
752 
753 	return 0;
754 }
755 
756 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
757 {
758 	uint32_t max_frame_len = adapter->max_mtu;
759 
760 	if (adapter->rte_eth_dev_data->dev_conf.rxmode.jumbo_frame == 1)
761 		max_frame_len =
762 			adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
763 
764 	return max_frame_len;
765 }
766 
767 static int ena_check_valid_conf(struct ena_adapter *adapter)
768 {
769 	uint32_t max_frame_len = ena_get_mtu_conf(adapter);
770 
771 	if (max_frame_len > adapter->max_mtu) {
772 		PMD_INIT_LOG(ERR, "Unsupported MTU of %d", max_frame_len);
773 		return -1;
774 	}
775 
776 	return 0;
777 }
778 
779 static int
780 ena_calc_queue_size(struct ena_com_dev *ena_dev,
781 		    struct ena_com_dev_get_features_ctx *get_feat_ctx)
782 {
783 	uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
784 
785 	queue_size = RTE_MIN(queue_size,
786 			     get_feat_ctx->max_queues.max_cq_depth);
787 	queue_size = RTE_MIN(queue_size,
788 			     get_feat_ctx->max_queues.max_sq_depth);
789 
790 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
791 		queue_size = RTE_MIN(queue_size,
792 				     get_feat_ctx->max_queues.max_llq_depth);
793 
794 	/* Round down to power of 2 */
795 	if (!rte_is_power_of_2(queue_size))
796 		queue_size = rte_align32pow2(queue_size >> 1);
797 
798 	if (queue_size == 0) {
799 		PMD_INIT_LOG(ERR, "Invalid queue size");
800 		return -EFAULT;
801 	}
802 
803 	return queue_size;
804 }
805 
806 static void ena_stats_restart(struct rte_eth_dev *dev)
807 {
808 	struct ena_adapter *adapter =
809 		(struct ena_adapter *)(dev->data->dev_private);
810 
811 	rte_atomic64_init(&adapter->drv_stats->ierrors);
812 	rte_atomic64_init(&adapter->drv_stats->oerrors);
813 	rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
814 }
815 
816 static int ena_stats_get(struct rte_eth_dev *dev,
817 			  struct rte_eth_stats *stats)
818 {
819 	struct ena_admin_basic_stats ena_stats;
820 	struct ena_adapter *adapter =
821 		(struct ena_adapter *)(dev->data->dev_private);
822 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
823 	int rc;
824 
825 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
826 		return -ENOTSUP;
827 
828 	memset(&ena_stats, 0, sizeof(ena_stats));
829 	rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
830 	if (unlikely(rc)) {
831 		RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
832 		return rc;
833 	}
834 
835 	/* Set of basic statistics from ENA */
836 	stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
837 					  ena_stats.rx_pkts_low);
838 	stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
839 					  ena_stats.tx_pkts_low);
840 	stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
841 					ena_stats.rx_bytes_low);
842 	stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
843 					ena_stats.tx_bytes_low);
844 	stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
845 					 ena_stats.rx_drops_low);
846 
847 	/* Driver related stats */
848 	stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
849 	stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
850 	stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
851 	return 0;
852 }
853 
854 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
855 {
856 	struct ena_adapter *adapter;
857 	struct ena_com_dev *ena_dev;
858 	int rc = 0;
859 
860 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
861 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
862 	adapter = (struct ena_adapter *)(dev->data->dev_private);
863 
864 	ena_dev = &adapter->ena_dev;
865 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
866 
867 	if (mtu > ena_get_mtu_conf(adapter)) {
868 		RTE_LOG(ERR, PMD,
869 			"Given MTU (%d) exceeds maximum MTU supported (%d)\n",
870 			mtu, ena_get_mtu_conf(adapter));
871 		rc = -EINVAL;
872 		goto err;
873 	}
874 
875 	rc = ena_com_set_dev_mtu(ena_dev, mtu);
876 	if (rc)
877 		RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
878 	else
879 		RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
880 
881 err:
882 	return rc;
883 }
884 
885 static int ena_start(struct rte_eth_dev *dev)
886 {
887 	struct ena_adapter *adapter =
888 		(struct ena_adapter *)(dev->data->dev_private);
889 	int rc = 0;
890 
891 	if (!(adapter->state == ENA_ADAPTER_STATE_CONFIG ||
892 	      adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
893 		PMD_INIT_LOG(ERR, "API violation");
894 		return -1;
895 	}
896 
897 	rc = ena_check_valid_conf(adapter);
898 	if (rc)
899 		return rc;
900 
901 	rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
902 	if (rc)
903 		return rc;
904 
905 	rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
906 	if (rc)
907 		return rc;
908 
909 	if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
910 	    ETH_MQ_RX_RSS_FLAG) {
911 		rc = ena_rss_init_default(adapter);
912 		if (rc)
913 			return rc;
914 	}
915 
916 	ena_stats_restart(dev);
917 
918 	adapter->state = ENA_ADAPTER_STATE_RUNNING;
919 
920 	return 0;
921 }
922 
923 static int ena_queue_restart(struct ena_ring *ring)
924 {
925 	int rc, bufs_num;
926 
927 	ena_assert_msg(ring->configured == 1,
928 		       "Trying to restart unconfigured queue\n");
929 
930 	ring->next_to_clean = 0;
931 	ring->next_to_use = 0;
932 
933 	if (ring->type == ENA_RING_TYPE_TX)
934 		return 0;
935 
936 	bufs_num = ring->ring_size - 1;
937 	rc = ena_populate_rx_queue(ring, bufs_num);
938 	if (rc != bufs_num) {
939 		PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
940 		return (-1);
941 	}
942 
943 	return 0;
944 }
945 
946 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
947 			      uint16_t queue_idx,
948 			      uint16_t nb_desc,
949 			      __rte_unused unsigned int socket_id,
950 			      __rte_unused const struct rte_eth_txconf *tx_conf)
951 {
952 	struct ena_com_create_io_ctx ctx =
953 		/* policy set to _HOST just to satisfy icc compiler */
954 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
955 		  ENA_COM_IO_QUEUE_DIRECTION_TX, 0, 0, 0, 0 };
956 	struct ena_ring *txq = NULL;
957 	struct ena_adapter *adapter =
958 		(struct ena_adapter *)(dev->data->dev_private);
959 	unsigned int i;
960 	int ena_qid;
961 	int rc;
962 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
963 
964 	txq = &adapter->tx_ring[queue_idx];
965 
966 	if (txq->configured) {
967 		RTE_LOG(CRIT, PMD,
968 			"API violation. Queue %d is already configured\n",
969 			queue_idx);
970 		return -1;
971 	}
972 
973 	if (!rte_is_power_of_2(nb_desc)) {
974 		RTE_LOG(ERR, PMD,
975 			"Unsupported size of RX queue: %d is not a power of 2.",
976 			nb_desc);
977 		return -EINVAL;
978 	}
979 
980 	if (nb_desc > adapter->tx_ring_size) {
981 		RTE_LOG(ERR, PMD,
982 			"Unsupported size of TX queue (max size: %d)\n",
983 			adapter->tx_ring_size);
984 		return -EINVAL;
985 	}
986 
987 	ena_qid = ENA_IO_TXQ_IDX(queue_idx);
988 
989 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
990 	ctx.qid = ena_qid;
991 	ctx.msix_vector = -1; /* admin interrupts not used */
992 	ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
993 	ctx.queue_size = adapter->tx_ring_size;
994 	ctx.numa_node = ena_cpu_to_node(queue_idx);
995 
996 	rc = ena_com_create_io_queue(ena_dev, &ctx);
997 	if (rc) {
998 		RTE_LOG(ERR, PMD,
999 			"failed to create io TX queue #%d (qid:%d) rc: %d\n",
1000 			queue_idx, ena_qid, rc);
1001 	}
1002 	txq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1003 	txq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1004 
1005 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1006 				     &txq->ena_com_io_sq,
1007 				     &txq->ena_com_io_cq);
1008 	if (rc) {
1009 		RTE_LOG(ERR, PMD,
1010 			"Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1011 			queue_idx, rc);
1012 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1013 		goto err;
1014 	}
1015 
1016 	txq->port_id = dev->data->port_id;
1017 	txq->next_to_clean = 0;
1018 	txq->next_to_use = 0;
1019 	txq->ring_size = nb_desc;
1020 
1021 	txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1022 					  sizeof(struct ena_tx_buffer) *
1023 					  txq->ring_size,
1024 					  RTE_CACHE_LINE_SIZE);
1025 	if (!txq->tx_buffer_info) {
1026 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1027 		return -ENOMEM;
1028 	}
1029 
1030 	txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1031 					 sizeof(u16) * txq->ring_size,
1032 					 RTE_CACHE_LINE_SIZE);
1033 	if (!txq->empty_tx_reqs) {
1034 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1035 		rte_free(txq->tx_buffer_info);
1036 		return -ENOMEM;
1037 	}
1038 	for (i = 0; i < txq->ring_size; i++)
1039 		txq->empty_tx_reqs[i] = i;
1040 
1041 	/* Store pointer to this queue in upper layer */
1042 	txq->configured = 1;
1043 	dev->data->tx_queues[queue_idx] = txq;
1044 err:
1045 	return rc;
1046 }
1047 
1048 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1049 			      uint16_t queue_idx,
1050 			      uint16_t nb_desc,
1051 			      __rte_unused unsigned int socket_id,
1052 			      __rte_unused const struct rte_eth_rxconf *rx_conf,
1053 			      struct rte_mempool *mp)
1054 {
1055 	struct ena_com_create_io_ctx ctx =
1056 		/* policy set to _HOST just to satisfy icc compiler */
1057 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
1058 		  ENA_COM_IO_QUEUE_DIRECTION_RX, 0, 0, 0, 0 };
1059 	struct ena_adapter *adapter =
1060 		(struct ena_adapter *)(dev->data->dev_private);
1061 	struct ena_ring *rxq = NULL;
1062 	uint16_t ena_qid = 0;
1063 	int rc = 0;
1064 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1065 
1066 	rxq = &adapter->rx_ring[queue_idx];
1067 	if (rxq->configured) {
1068 		RTE_LOG(CRIT, PMD,
1069 			"API violation. Queue %d is already configured\n",
1070 			queue_idx);
1071 		return -1;
1072 	}
1073 
1074 	if (!rte_is_power_of_2(nb_desc)) {
1075 		RTE_LOG(ERR, PMD,
1076 			"Unsupported size of TX queue: %d is not a power of 2.",
1077 			nb_desc);
1078 		return -EINVAL;
1079 	}
1080 
1081 	if (nb_desc > adapter->rx_ring_size) {
1082 		RTE_LOG(ERR, PMD,
1083 			"Unsupported size of RX queue (max size: %d)\n",
1084 			adapter->rx_ring_size);
1085 		return -EINVAL;
1086 	}
1087 
1088 	ena_qid = ENA_IO_RXQ_IDX(queue_idx);
1089 
1090 	ctx.qid = ena_qid;
1091 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1092 	ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1093 	ctx.msix_vector = -1; /* admin interrupts not used */
1094 	ctx.queue_size = adapter->rx_ring_size;
1095 	ctx.numa_node = ena_cpu_to_node(queue_idx);
1096 
1097 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1098 	if (rc)
1099 		RTE_LOG(ERR, PMD, "failed to create io RX queue #%d rc: %d\n",
1100 			queue_idx, rc);
1101 
1102 	rxq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1103 	rxq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1104 
1105 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1106 				     &rxq->ena_com_io_sq,
1107 				     &rxq->ena_com_io_cq);
1108 	if (rc) {
1109 		RTE_LOG(ERR, PMD,
1110 			"Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1111 			queue_idx, rc);
1112 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1113 	}
1114 
1115 	rxq->port_id = dev->data->port_id;
1116 	rxq->next_to_clean = 0;
1117 	rxq->next_to_use = 0;
1118 	rxq->ring_size = nb_desc;
1119 	rxq->mb_pool = mp;
1120 
1121 	rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1122 					  sizeof(struct rte_mbuf *) * nb_desc,
1123 					  RTE_CACHE_LINE_SIZE);
1124 	if (!rxq->rx_buffer_info) {
1125 		RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1126 		return -ENOMEM;
1127 	}
1128 
1129 	/* Store pointer to this queue in upper layer */
1130 	rxq->configured = 1;
1131 	dev->data->rx_queues[queue_idx] = rxq;
1132 
1133 	return rc;
1134 }
1135 
1136 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1137 {
1138 	unsigned int i;
1139 	int rc;
1140 	uint16_t ring_size = rxq->ring_size;
1141 	uint16_t ring_mask = ring_size - 1;
1142 	uint16_t next_to_use = rxq->next_to_use;
1143 	uint16_t in_use;
1144 	struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1145 
1146 	if (unlikely(!count))
1147 		return 0;
1148 
1149 	in_use = rxq->next_to_use - rxq->next_to_clean;
1150 	ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1151 
1152 	count = RTE_MIN(count,
1153 			(uint16_t)(ring_size - (next_to_use & ring_mask)));
1154 
1155 	/* get resources for incoming packets */
1156 	rc = rte_mempool_get_bulk(rxq->mb_pool,
1157 				  (void **)(&mbufs[next_to_use & ring_mask]),
1158 				  count);
1159 	if (unlikely(rc < 0)) {
1160 		rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1161 		PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1162 		return 0;
1163 	}
1164 
1165 	for (i = 0; i < count; i++) {
1166 		uint16_t next_to_use_masked = next_to_use & ring_mask;
1167 		struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
1168 		struct ena_com_buf ebuf;
1169 
1170 		rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1171 		/* prepare physical address for DMA transaction */
1172 		ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1173 		ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1174 		/* pass resource to device */
1175 		rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1176 						&ebuf, next_to_use_masked);
1177 		if (unlikely(rc)) {
1178 			rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbuf),
1179 					     count - i);
1180 			RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1181 			break;
1182 		}
1183 		next_to_use++;
1184 	}
1185 
1186 	/* When we submitted free recources to device... */
1187 	if (i > 0) {
1188 		/* ...let HW know that it can fill buffers with data */
1189 		rte_wmb();
1190 		ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1191 
1192 		rxq->next_to_use = next_to_use;
1193 	}
1194 
1195 	return i;
1196 }
1197 
1198 static int ena_device_init(struct ena_com_dev *ena_dev,
1199 			   struct ena_com_dev_get_features_ctx *get_feat_ctx)
1200 {
1201 	int rc;
1202 	bool readless_supported;
1203 
1204 	/* Initialize mmio registers */
1205 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
1206 	if (rc) {
1207 		RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1208 		return rc;
1209 	}
1210 
1211 	/* The PCIe configuration space revision id indicate if mmio reg
1212 	 * read is disabled.
1213 	 */
1214 	readless_supported =
1215 		!(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1216 			       & ENA_MMIO_DISABLE_REG_READ);
1217 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1218 
1219 	/* reset device */
1220 	rc = ena_com_dev_reset(ena_dev);
1221 	if (rc) {
1222 		RTE_LOG(ERR, PMD, "cannot reset device\n");
1223 		goto err_mmio_read_less;
1224 	}
1225 
1226 	/* check FW version */
1227 	rc = ena_com_validate_version(ena_dev);
1228 	if (rc) {
1229 		RTE_LOG(ERR, PMD, "device version is too low\n");
1230 		goto err_mmio_read_less;
1231 	}
1232 
1233 	ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1234 
1235 	/* ENA device administration layer init */
1236 	rc = ena_com_admin_init(ena_dev, NULL, true);
1237 	if (rc) {
1238 		RTE_LOG(ERR, PMD,
1239 			"cannot initialize ena admin queue with device\n");
1240 		goto err_mmio_read_less;
1241 	}
1242 
1243 	/* To enable the msix interrupts the driver needs to know the number
1244 	 * of queues. So the driver uses polling mode to retrieve this
1245 	 * information.
1246 	 */
1247 	ena_com_set_admin_polling_mode(ena_dev, true);
1248 
1249 	ena_config_host_info(ena_dev);
1250 
1251 	/* Get Device Attributes and features */
1252 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1253 	if (rc) {
1254 		RTE_LOG(ERR, PMD,
1255 			"cannot get attribute for ena device rc= %d\n", rc);
1256 		goto err_admin_init;
1257 	}
1258 
1259 	return 0;
1260 
1261 err_admin_init:
1262 	ena_com_admin_destroy(ena_dev);
1263 
1264 err_mmio_read_less:
1265 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1266 
1267 	return rc;
1268 }
1269 
1270 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1271 {
1272 	struct rte_pci_device *pci_dev;
1273 	struct ena_adapter *adapter =
1274 		(struct ena_adapter *)(eth_dev->data->dev_private);
1275 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1276 	struct ena_com_dev_get_features_ctx get_feat_ctx;
1277 	int queue_size, rc;
1278 
1279 	static int adapters_found;
1280 
1281 	memset(adapter, 0, sizeof(struct ena_adapter));
1282 	ena_dev = &adapter->ena_dev;
1283 
1284 	eth_dev->dev_ops = &ena_dev_ops;
1285 	eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1286 	eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1287 	eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1288 	adapter->rte_eth_dev_data = eth_dev->data;
1289 	adapter->rte_dev = eth_dev;
1290 
1291 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1292 		return 0;
1293 
1294 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1295 	adapter->pdev = pci_dev;
1296 
1297 	PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1298 		     pci_dev->addr.domain,
1299 		     pci_dev->addr.bus,
1300 		     pci_dev->addr.devid,
1301 		     pci_dev->addr.function);
1302 
1303 	adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1304 	adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1305 
1306 	/* Present ENA_MEM_BAR indicates available LLQ mode.
1307 	 * Use corresponding policy
1308 	 */
1309 	if (adapter->dev_mem_base)
1310 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
1311 	else if (adapter->regs)
1312 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1313 	else
1314 		PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1315 			     ENA_REGS_BAR);
1316 
1317 	ena_dev->reg_bar = adapter->regs;
1318 	ena_dev->dmadev = adapter->pdev;
1319 
1320 	adapter->id_number = adapters_found;
1321 
1322 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1323 		 adapter->id_number);
1324 
1325 	/* device specific initialization routine */
1326 	rc = ena_device_init(ena_dev, &get_feat_ctx);
1327 	if (rc) {
1328 		PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1329 		return -1;
1330 	}
1331 
1332 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1333 		if (get_feat_ctx.max_queues.max_llq_num == 0) {
1334 			PMD_INIT_LOG(ERR,
1335 				     "Trying to use LLQ but llq_num is 0.\n"
1336 				     "Fall back into regular queues.");
1337 			ena_dev->tx_mem_queue_type =
1338 				ENA_ADMIN_PLACEMENT_POLICY_HOST;
1339 			adapter->num_queues =
1340 				get_feat_ctx.max_queues.max_sq_num;
1341 		} else {
1342 			adapter->num_queues =
1343 				get_feat_ctx.max_queues.max_llq_num;
1344 		}
1345 	} else {
1346 		adapter->num_queues = get_feat_ctx.max_queues.max_sq_num;
1347 	}
1348 
1349 	queue_size = ena_calc_queue_size(ena_dev, &get_feat_ctx);
1350 	if ((queue_size <= 0) || (adapter->num_queues <= 0))
1351 		return -EFAULT;
1352 
1353 	adapter->tx_ring_size = queue_size;
1354 	adapter->rx_ring_size = queue_size;
1355 
1356 	/* prepare ring structures */
1357 	ena_init_rings(adapter);
1358 
1359 	ena_config_debug_area(adapter);
1360 
1361 	/* Set max MTU for this device */
1362 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1363 
1364 	/* set device support for TSO */
1365 	adapter->tso4_supported = get_feat_ctx.offload.tx &
1366 				  ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1367 
1368 	/* Copy MAC address and point DPDK to it */
1369 	eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1370 	ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1371 			(struct ether_addr *)adapter->mac_addr);
1372 
1373 	adapter->drv_stats = rte_zmalloc("adapter stats",
1374 					 sizeof(*adapter->drv_stats),
1375 					 RTE_CACHE_LINE_SIZE);
1376 	if (!adapter->drv_stats) {
1377 		RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1378 		return -ENOMEM;
1379 	}
1380 
1381 	adapters_found++;
1382 	adapter->state = ENA_ADAPTER_STATE_INIT;
1383 
1384 	return 0;
1385 }
1386 
1387 static int ena_dev_configure(struct rte_eth_dev *dev)
1388 {
1389 	struct ena_adapter *adapter =
1390 		(struct ena_adapter *)(dev->data->dev_private);
1391 
1392 	if (!(adapter->state == ENA_ADAPTER_STATE_INIT ||
1393 	      adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
1394 		PMD_INIT_LOG(ERR, "Illegal adapter state: %d",
1395 			     adapter->state);
1396 		return -1;
1397 	}
1398 
1399 	switch (adapter->state) {
1400 	case ENA_ADAPTER_STATE_INIT:
1401 	case ENA_ADAPTER_STATE_STOPPED:
1402 		adapter->state = ENA_ADAPTER_STATE_CONFIG;
1403 		break;
1404 	case ENA_ADAPTER_STATE_CONFIG:
1405 		RTE_LOG(WARNING, PMD,
1406 			"Ivalid driver state while trying to configure device\n");
1407 		break;
1408 	default:
1409 		break;
1410 	}
1411 
1412 	return 0;
1413 }
1414 
1415 static void ena_init_rings(struct ena_adapter *adapter)
1416 {
1417 	int i;
1418 
1419 	for (i = 0; i < adapter->num_queues; i++) {
1420 		struct ena_ring *ring = &adapter->tx_ring[i];
1421 
1422 		ring->configured = 0;
1423 		ring->type = ENA_RING_TYPE_TX;
1424 		ring->adapter = adapter;
1425 		ring->id = i;
1426 		ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1427 		ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1428 	}
1429 
1430 	for (i = 0; i < adapter->num_queues; i++) {
1431 		struct ena_ring *ring = &adapter->rx_ring[i];
1432 
1433 		ring->configured = 0;
1434 		ring->type = ENA_RING_TYPE_RX;
1435 		ring->adapter = adapter;
1436 		ring->id = i;
1437 	}
1438 }
1439 
1440 static void ena_infos_get(struct rte_eth_dev *dev,
1441 			  struct rte_eth_dev_info *dev_info)
1442 {
1443 	struct ena_adapter *adapter;
1444 	struct ena_com_dev *ena_dev;
1445 	struct ena_com_dev_get_features_ctx feat;
1446 	uint32_t rx_feat = 0, tx_feat = 0;
1447 	int rc = 0;
1448 
1449 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
1450 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1451 	adapter = (struct ena_adapter *)(dev->data->dev_private);
1452 
1453 	ena_dev = &adapter->ena_dev;
1454 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1455 
1456 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1457 
1458 	dev_info->speed_capa =
1459 			ETH_LINK_SPEED_1G   |
1460 			ETH_LINK_SPEED_2_5G |
1461 			ETH_LINK_SPEED_5G   |
1462 			ETH_LINK_SPEED_10G  |
1463 			ETH_LINK_SPEED_25G  |
1464 			ETH_LINK_SPEED_40G  |
1465 			ETH_LINK_SPEED_50G  |
1466 			ETH_LINK_SPEED_100G;
1467 
1468 	/* Get supported features from HW */
1469 	rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1470 	if (unlikely(rc)) {
1471 		RTE_LOG(ERR, PMD,
1472 			"Cannot get attribute for ena device rc= %d\n", rc);
1473 		return;
1474 	}
1475 
1476 	/* Set Tx & Rx features available for device */
1477 	if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1478 		tx_feat	|= DEV_TX_OFFLOAD_TCP_TSO;
1479 
1480 	if (feat.offload.tx &
1481 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1482 		tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1483 			DEV_TX_OFFLOAD_UDP_CKSUM |
1484 			DEV_TX_OFFLOAD_TCP_CKSUM;
1485 
1486 	if (feat.offload.rx_supported &
1487 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1488 		rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1489 			DEV_RX_OFFLOAD_UDP_CKSUM  |
1490 			DEV_RX_OFFLOAD_TCP_CKSUM;
1491 
1492 	/* Inform framework about available features */
1493 	dev_info->rx_offload_capa = rx_feat;
1494 	dev_info->tx_offload_capa = tx_feat;
1495 
1496 	dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1497 	dev_info->max_rx_pktlen  = adapter->max_mtu;
1498 	dev_info->max_mac_addrs = 1;
1499 
1500 	dev_info->max_rx_queues = adapter->num_queues;
1501 	dev_info->max_tx_queues = adapter->num_queues;
1502 	dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1503 }
1504 
1505 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1506 				  uint16_t nb_pkts)
1507 {
1508 	struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1509 	unsigned int ring_size = rx_ring->ring_size;
1510 	unsigned int ring_mask = ring_size - 1;
1511 	uint16_t next_to_clean = rx_ring->next_to_clean;
1512 	uint16_t desc_in_use = 0;
1513 	unsigned int recv_idx = 0;
1514 	struct rte_mbuf *mbuf = NULL;
1515 	struct rte_mbuf *mbuf_head = NULL;
1516 	struct rte_mbuf *mbuf_prev = NULL;
1517 	struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1518 	unsigned int completed;
1519 
1520 	struct ena_com_rx_ctx ena_rx_ctx;
1521 	int rc = 0;
1522 
1523 	/* Check adapter state */
1524 	if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1525 		RTE_LOG(ALERT, PMD,
1526 			"Trying to receive pkts while device is NOT running\n");
1527 		return 0;
1528 	}
1529 
1530 	desc_in_use = rx_ring->next_to_use - next_to_clean;
1531 	if (unlikely(nb_pkts > desc_in_use))
1532 		nb_pkts = desc_in_use;
1533 
1534 	for (completed = 0; completed < nb_pkts; completed++) {
1535 		int segments = 0;
1536 
1537 		ena_rx_ctx.max_bufs = rx_ring->ring_size;
1538 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1539 		ena_rx_ctx.descs = 0;
1540 		/* receive packet context */
1541 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1542 				    rx_ring->ena_com_io_sq,
1543 				    &ena_rx_ctx);
1544 		if (unlikely(rc)) {
1545 			RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1546 			return 0;
1547 		}
1548 
1549 		if (unlikely(ena_rx_ctx.descs == 0))
1550 			break;
1551 
1552 		while (segments < ena_rx_ctx.descs) {
1553 			mbuf = rx_buff_info[next_to_clean & ring_mask];
1554 			mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1555 			mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1556 			mbuf->refcnt = 1;
1557 			mbuf->next = NULL;
1558 			if (segments == 0) {
1559 				mbuf->nb_segs = ena_rx_ctx.descs;
1560 				mbuf->port = rx_ring->port_id;
1561 				mbuf->pkt_len = 0;
1562 				mbuf_head = mbuf;
1563 			} else {
1564 				/* for multi-segment pkts create mbuf chain */
1565 				mbuf_prev->next = mbuf;
1566 			}
1567 			mbuf_head->pkt_len += mbuf->data_len;
1568 
1569 			mbuf_prev = mbuf;
1570 			segments++;
1571 			next_to_clean++;
1572 		}
1573 
1574 		/* fill mbuf attributes if any */
1575 		ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1576 		mbuf_head->hash.rss = (uint32_t)rx_ring->id;
1577 
1578 		/* pass to DPDK application head mbuf */
1579 		rx_pkts[recv_idx] = mbuf_head;
1580 		recv_idx++;
1581 	}
1582 
1583 	rx_ring->next_to_clean = next_to_clean;
1584 
1585 	desc_in_use = desc_in_use - completed + 1;
1586 	/* Burst refill to save doorbells, memory barriers, const interval */
1587 	if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1588 		ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1589 
1590 	return recv_idx;
1591 }
1592 
1593 static uint16_t
1594 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1595 		uint16_t nb_pkts)
1596 {
1597 	int32_t ret;
1598 	uint32_t i;
1599 	struct rte_mbuf *m;
1600 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1601 	struct ipv4_hdr *ip_hdr;
1602 	uint64_t ol_flags;
1603 	uint16_t frag_field;
1604 
1605 	for (i = 0; i != nb_pkts; i++) {
1606 		m = tx_pkts[i];
1607 		ol_flags = m->ol_flags;
1608 
1609 		if (!(ol_flags & PKT_TX_IPV4))
1610 			continue;
1611 
1612 		/* If there was not L2 header length specified, assume it is
1613 		 * length of the ethernet header.
1614 		 */
1615 		if (unlikely(m->l2_len == 0))
1616 			m->l2_len = sizeof(struct ether_hdr);
1617 
1618 		ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1619 						 m->l2_len);
1620 		frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1621 
1622 		if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1623 			m->packet_type |= RTE_PTYPE_L4_NONFRAG;
1624 
1625 			/* If IPv4 header has DF flag enabled and TSO support is
1626 			 * disabled, partial chcecksum should not be calculated.
1627 			 */
1628 			if (!tx_ring->adapter->tso4_supported)
1629 				continue;
1630 		}
1631 
1632 		if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
1633 				(ol_flags & PKT_TX_L4_MASK) ==
1634 				PKT_TX_SCTP_CKSUM) {
1635 			rte_errno = -ENOTSUP;
1636 			return i;
1637 		}
1638 
1639 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1640 		ret = rte_validate_tx_offload(m);
1641 		if (ret != 0) {
1642 			rte_errno = ret;
1643 			return i;
1644 		}
1645 #endif
1646 
1647 		/* In case we are supposed to TSO and have DF not set (DF=0)
1648 		 * hardware must be provided with partial checksum, otherwise
1649 		 * it will take care of necessary calculations.
1650 		 */
1651 
1652 		ret = rte_net_intel_cksum_flags_prepare(m,
1653 			ol_flags & ~PKT_TX_TCP_SEG);
1654 		if (ret != 0) {
1655 			rte_errno = ret;
1656 			return i;
1657 		}
1658 	}
1659 
1660 	return i;
1661 }
1662 
1663 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1664 				  uint16_t nb_pkts)
1665 {
1666 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1667 	uint16_t next_to_use = tx_ring->next_to_use;
1668 	uint16_t next_to_clean = tx_ring->next_to_clean;
1669 	struct rte_mbuf *mbuf;
1670 	unsigned int ring_size = tx_ring->ring_size;
1671 	unsigned int ring_mask = ring_size - 1;
1672 	struct ena_com_tx_ctx ena_tx_ctx;
1673 	struct ena_tx_buffer *tx_info;
1674 	struct ena_com_buf *ebuf;
1675 	uint16_t rc, req_id, total_tx_descs = 0;
1676 	uint16_t sent_idx = 0, empty_tx_reqs;
1677 	int nb_hw_desc;
1678 
1679 	/* Check adapter state */
1680 	if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1681 		RTE_LOG(ALERT, PMD,
1682 			"Trying to xmit pkts while device is NOT running\n");
1683 		return 0;
1684 	}
1685 
1686 	empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
1687 	if (nb_pkts > empty_tx_reqs)
1688 		nb_pkts = empty_tx_reqs;
1689 
1690 	for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
1691 		mbuf = tx_pkts[sent_idx];
1692 
1693 		req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
1694 		tx_info = &tx_ring->tx_buffer_info[req_id];
1695 		tx_info->mbuf = mbuf;
1696 		tx_info->num_of_bufs = 0;
1697 		ebuf = tx_info->bufs;
1698 
1699 		/* Prepare TX context */
1700 		memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1701 		memset(&ena_tx_ctx.ena_meta, 0x0,
1702 		       sizeof(struct ena_com_tx_meta));
1703 		ena_tx_ctx.ena_bufs = ebuf;
1704 		ena_tx_ctx.req_id = req_id;
1705 		if (tx_ring->tx_mem_queue_type ==
1706 				ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1707 			/* prepare the push buffer with
1708 			 * virtual address of the data
1709 			 */
1710 			ena_tx_ctx.header_len =
1711 				RTE_MIN(mbuf->data_len,
1712 					tx_ring->tx_max_header_size);
1713 			ena_tx_ctx.push_header =
1714 				(void *)((char *)mbuf->buf_addr +
1715 					 mbuf->data_off);
1716 		} /* there's no else as we take advantage of memset zeroing */
1717 
1718 		/* Set TX offloads flags, if applicable */
1719 		ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx);
1720 
1721 		if (unlikely(mbuf->ol_flags &
1722 			     (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
1723 			rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
1724 
1725 		rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
1726 
1727 		/* Process first segment taking into
1728 		 * consideration pushed header
1729 		 */
1730 		if (mbuf->data_len > ena_tx_ctx.header_len) {
1731 			ebuf->paddr = mbuf->buf_iova +
1732 				      mbuf->data_off +
1733 				      ena_tx_ctx.header_len;
1734 			ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
1735 			ebuf++;
1736 			tx_info->num_of_bufs++;
1737 		}
1738 
1739 		while ((mbuf = mbuf->next) != NULL) {
1740 			ebuf->paddr = mbuf->buf_iova + mbuf->data_off;
1741 			ebuf->len = mbuf->data_len;
1742 			ebuf++;
1743 			tx_info->num_of_bufs++;
1744 		}
1745 
1746 		ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1747 
1748 		/* Write data to device */
1749 		rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
1750 					&ena_tx_ctx, &nb_hw_desc);
1751 		if (unlikely(rc))
1752 			break;
1753 
1754 		tx_info->tx_descs = nb_hw_desc;
1755 
1756 		next_to_use++;
1757 	}
1758 
1759 	/* If there are ready packets to be xmitted... */
1760 	if (sent_idx > 0) {
1761 		/* ...let HW do its best :-) */
1762 		rte_wmb();
1763 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
1764 
1765 		tx_ring->next_to_use = next_to_use;
1766 	}
1767 
1768 	/* Clear complete packets  */
1769 	while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
1770 		/* Get Tx info & store how many descs were processed  */
1771 		tx_info = &tx_ring->tx_buffer_info[req_id];
1772 		total_tx_descs += tx_info->tx_descs;
1773 
1774 		/* Free whole mbuf chain  */
1775 		mbuf = tx_info->mbuf;
1776 		rte_pktmbuf_free(mbuf);
1777 		tx_info->mbuf = NULL;
1778 
1779 		/* Put back descriptor to the ring for reuse */
1780 		tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
1781 		next_to_clean++;
1782 
1783 		/* If too many descs to clean, leave it for another run */
1784 		if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
1785 			break;
1786 	}
1787 
1788 	if (total_tx_descs > 0) {
1789 		/* acknowledge completion of sent packets */
1790 		ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
1791 		tx_ring->next_to_clean = next_to_clean;
1792 	}
1793 
1794 	return sent_idx;
1795 }
1796 
1797 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1798 	struct rte_pci_device *pci_dev)
1799 {
1800 	return rte_eth_dev_pci_generic_probe(pci_dev,
1801 		sizeof(struct ena_adapter), eth_ena_dev_init);
1802 }
1803 
1804 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
1805 {
1806 	return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1807 }
1808 
1809 static struct rte_pci_driver rte_ena_pmd = {
1810 	.id_table = pci_id_ena_map,
1811 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1812 	.probe = eth_ena_pci_probe,
1813 	.remove = eth_ena_pci_remove,
1814 };
1815 
1816 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
1817 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
1818 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
1819