xref: /dpdk/drivers/net/ena/ena_ethdev.c (revision b79e4c00af0e7cfb8601ab0208659d226b82bd10)
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 void 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 
264 	if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
265 		ol_flags |= PKT_TX_TCP_CKSUM;
266 	else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
267 		ol_flags |= PKT_TX_UDP_CKSUM;
268 
269 	if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
270 		ol_flags |= PKT_TX_IPV4;
271 	else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
272 		ol_flags |= PKT_TX_IPV6;
273 
274 	if (unlikely(ena_rx_ctx->l4_csum_err))
275 		ol_flags |= PKT_RX_L4_CKSUM_BAD;
276 	if (unlikely(ena_rx_ctx->l3_csum_err))
277 		ol_flags |= PKT_RX_IP_CKSUM_BAD;
278 
279 	mbuf->ol_flags = ol_flags;
280 }
281 
282 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
283 				       struct ena_com_tx_ctx *ena_tx_ctx)
284 {
285 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
286 
287 	if (mbuf->ol_flags &
288 	    (PKT_TX_L4_MASK | PKT_TX_IP_CKSUM | PKT_TX_TCP_SEG)) {
289 		/* check if TSO is required */
290 		if (mbuf->ol_flags & PKT_TX_TCP_SEG) {
291 			ena_tx_ctx->tso_enable = true;
292 
293 			ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
294 		}
295 
296 		/* check if L3 checksum is needed */
297 		if (mbuf->ol_flags & PKT_TX_IP_CKSUM)
298 			ena_tx_ctx->l3_csum_enable = true;
299 
300 		if (mbuf->ol_flags & PKT_TX_IPV6) {
301 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
302 		} else {
303 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
304 
305 			/* set don't fragment (DF) flag */
306 			if (mbuf->packet_type &
307 				(RTE_PTYPE_L4_NONFRAG
308 				 | RTE_PTYPE_INNER_L4_NONFRAG))
309 				ena_tx_ctx->df = true;
310 		}
311 
312 		/* check if L4 checksum is needed */
313 		switch (mbuf->ol_flags & PKT_TX_L4_MASK) {
314 		case PKT_TX_TCP_CKSUM:
315 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
316 			ena_tx_ctx->l4_csum_enable = true;
317 			break;
318 		case PKT_TX_UDP_CKSUM:
319 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
320 			ena_tx_ctx->l4_csum_enable = true;
321 			break;
322 		default:
323 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
324 			ena_tx_ctx->l4_csum_enable = false;
325 			break;
326 		}
327 
328 		ena_meta->mss = mbuf->tso_segsz;
329 		ena_meta->l3_hdr_len = mbuf->l3_len;
330 		ena_meta->l3_hdr_offset = mbuf->l2_len;
331 		/* this param needed only for TSO */
332 		ena_meta->l3_outer_hdr_len = 0;
333 		ena_meta->l3_outer_hdr_offset = 0;
334 
335 		ena_tx_ctx->meta_valid = true;
336 	} else {
337 		ena_tx_ctx->meta_valid = false;
338 	}
339 }
340 
341 static void ena_config_host_info(struct ena_com_dev *ena_dev)
342 {
343 	struct ena_admin_host_info *host_info;
344 	int rc;
345 
346 	/* Allocate only the host info */
347 	rc = ena_com_allocate_host_info(ena_dev);
348 	if (rc) {
349 		RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
350 		return;
351 	}
352 
353 	host_info = ena_dev->host_attr.host_info;
354 
355 	host_info->os_type = ENA_ADMIN_OS_DPDK;
356 	host_info->kernel_ver = RTE_VERSION;
357 	snprintf((char *)host_info->kernel_ver_str,
358 		 sizeof(host_info->kernel_ver_str),
359 		 "%s", rte_version());
360 	host_info->os_dist = RTE_VERSION;
361 	snprintf((char *)host_info->os_dist_str,
362 		 sizeof(host_info->os_dist_str),
363 		 "%s", rte_version());
364 	host_info->driver_version =
365 		(DRV_MODULE_VER_MAJOR) |
366 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
367 		(DRV_MODULE_VER_SUBMINOR <<
368 			ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
369 
370 	rc = ena_com_set_host_attributes(ena_dev);
371 	if (rc) {
372 		RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
373 		if (rc != -EPERM)
374 			goto err;
375 	}
376 
377 	return;
378 
379 err:
380 	ena_com_delete_host_info(ena_dev);
381 }
382 
383 static int
384 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
385 {
386 	if (sset != ETH_SS_STATS)
387 		return -EOPNOTSUPP;
388 
389 	 /* Workaround for clang:
390 	 * touch internal structures to prevent
391 	 * compiler error
392 	 */
393 	ENA_TOUCH(ena_stats_global_strings);
394 	ENA_TOUCH(ena_stats_tx_strings);
395 	ENA_TOUCH(ena_stats_rx_strings);
396 	ENA_TOUCH(ena_stats_ena_com_strings);
397 
398 	return  dev->data->nb_tx_queues *
399 		(ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
400 		ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
401 }
402 
403 static void ena_config_debug_area(struct ena_adapter *adapter)
404 {
405 	u32 debug_area_size;
406 	int rc, ss_count;
407 
408 	ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
409 	if (ss_count <= 0) {
410 		RTE_LOG(ERR, PMD, "SS count is negative\n");
411 		return;
412 	}
413 
414 	/* allocate 32 bytes for each string and 64bit for the value */
415 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
416 
417 	rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
418 	if (rc) {
419 		RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
420 		return;
421 	}
422 
423 	rc = ena_com_set_host_attributes(&adapter->ena_dev);
424 	if (rc) {
425 		RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
426 		if (rc != -EPERM)
427 			goto err;
428 	}
429 
430 	return;
431 err:
432 	ena_com_delete_debug_area(&adapter->ena_dev);
433 }
434 
435 static void ena_close(struct rte_eth_dev *dev)
436 {
437 	struct ena_adapter *adapter =
438 		(struct ena_adapter *)(dev->data->dev_private);
439 
440 	adapter->state = ENA_ADAPTER_STATE_STOPPED;
441 
442 	ena_rx_queue_release_all(dev);
443 	ena_tx_queue_release_all(dev);
444 }
445 
446 static int ena_rss_reta_update(struct rte_eth_dev *dev,
447 			       struct rte_eth_rss_reta_entry64 *reta_conf,
448 			       uint16_t reta_size)
449 {
450 	struct ena_adapter *adapter =
451 		(struct ena_adapter *)(dev->data->dev_private);
452 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
453 	int ret, i;
454 	u16 entry_value;
455 	int conf_idx;
456 	int idx;
457 
458 	if ((reta_size == 0) || (reta_conf == NULL))
459 		return -EINVAL;
460 
461 	if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
462 		RTE_LOG(WARNING, PMD,
463 			"indirection table %d is bigger than supported (%d)\n",
464 			reta_size, ENA_RX_RSS_TABLE_SIZE);
465 		ret = -EINVAL;
466 		goto err;
467 	}
468 
469 	for (i = 0 ; i < reta_size ; i++) {
470 		/* each reta_conf is for 64 entries.
471 		 * to support 128 we use 2 conf of 64
472 		 */
473 		conf_idx = i / RTE_RETA_GROUP_SIZE;
474 		idx = i % RTE_RETA_GROUP_SIZE;
475 		if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
476 			entry_value =
477 				ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
478 			ret = ena_com_indirect_table_fill_entry(ena_dev,
479 								i,
480 								entry_value);
481 			if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
482 				RTE_LOG(ERR, PMD,
483 					"Cannot fill indirect table\n");
484 				ret = -ENOTSUP;
485 				goto err;
486 			}
487 		}
488 	}
489 
490 	ret = ena_com_indirect_table_set(ena_dev);
491 	if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
492 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
493 		ret = -ENOTSUP;
494 		goto err;
495 	}
496 
497 	RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries  for port %d\n",
498 		__func__, reta_size, adapter->rte_dev->data->port_id);
499 err:
500 	return ret;
501 }
502 
503 /* Query redirection table. */
504 static int ena_rss_reta_query(struct rte_eth_dev *dev,
505 			      struct rte_eth_rss_reta_entry64 *reta_conf,
506 			      uint16_t reta_size)
507 {
508 	struct ena_adapter *adapter =
509 		(struct ena_adapter *)(dev->data->dev_private);
510 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
511 	int ret;
512 	int i;
513 	u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
514 	int reta_conf_idx;
515 	int reta_idx;
516 
517 	if (reta_size == 0 || reta_conf == NULL ||
518 	    (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
519 		return -EINVAL;
520 
521 	ret = ena_com_indirect_table_get(ena_dev, indirect_table);
522 	if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
523 		RTE_LOG(ERR, PMD, "cannot get indirect table\n");
524 		ret = -ENOTSUP;
525 		goto err;
526 	}
527 
528 	for (i = 0 ; i < reta_size ; i++) {
529 		reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
530 		reta_idx = i % RTE_RETA_GROUP_SIZE;
531 		if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
532 			reta_conf[reta_conf_idx].reta[reta_idx] =
533 				ENA_IO_RXQ_IDX_REV(indirect_table[i]);
534 	}
535 err:
536 	return ret;
537 }
538 
539 static int ena_rss_init_default(struct ena_adapter *adapter)
540 {
541 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
542 	uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
543 	int rc, i;
544 	u32 val;
545 
546 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
547 	if (unlikely(rc)) {
548 		RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
549 		goto err_rss_init;
550 	}
551 
552 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
553 		val = i % nb_rx_queues;
554 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
555 						       ENA_IO_RXQ_IDX(val));
556 		if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
557 			RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
558 			goto err_fill_indir;
559 		}
560 	}
561 
562 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
563 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
564 	if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
565 		RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
566 		goto err_fill_indir;
567 	}
568 
569 	rc = ena_com_set_default_hash_ctrl(ena_dev);
570 	if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
571 		RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
572 		goto err_fill_indir;
573 	}
574 
575 	rc = ena_com_indirect_table_set(ena_dev);
576 	if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
577 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
578 		goto err_fill_indir;
579 	}
580 	RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
581 		adapter->rte_dev->data->port_id);
582 
583 	return 0;
584 
585 err_fill_indir:
586 	ena_com_rss_destroy(ena_dev);
587 err_rss_init:
588 
589 	return rc;
590 }
591 
592 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
593 {
594 	struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
595 	int nb_queues = dev->data->nb_rx_queues;
596 	int i;
597 
598 	for (i = 0; i < nb_queues; i++)
599 		ena_rx_queue_release(queues[i]);
600 }
601 
602 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
603 {
604 	struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
605 	int nb_queues = dev->data->nb_tx_queues;
606 	int i;
607 
608 	for (i = 0; i < nb_queues; i++)
609 		ena_tx_queue_release(queues[i]);
610 }
611 
612 static void ena_rx_queue_release(void *queue)
613 {
614 	struct ena_ring *ring = (struct ena_ring *)queue;
615 	struct ena_adapter *adapter = ring->adapter;
616 	int ena_qid;
617 
618 	ena_assert_msg(ring->configured,
619 		       "API violation - releasing not configured queue");
620 	ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
621 		       "API violation");
622 
623 	/* Destroy HW queue */
624 	ena_qid = ENA_IO_RXQ_IDX(ring->id);
625 	ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
626 
627 	/* Free all bufs */
628 	ena_rx_queue_release_bufs(ring);
629 
630 	/* Free ring resources */
631 	if (ring->rx_buffer_info)
632 		rte_free(ring->rx_buffer_info);
633 	ring->rx_buffer_info = NULL;
634 
635 	ring->configured = 0;
636 
637 	RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
638 		ring->port_id, ring->id);
639 }
640 
641 static void ena_tx_queue_release(void *queue)
642 {
643 	struct ena_ring *ring = (struct ena_ring *)queue;
644 	struct ena_adapter *adapter = ring->adapter;
645 	int ena_qid;
646 
647 	ena_assert_msg(ring->configured,
648 		       "API violation. Releasing not configured queue");
649 	ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
650 		       "API violation");
651 
652 	/* Destroy HW queue */
653 	ena_qid = ENA_IO_TXQ_IDX(ring->id);
654 	ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
655 
656 	/* Free all bufs */
657 	ena_tx_queue_release_bufs(ring);
658 
659 	/* Free ring resources */
660 	if (ring->tx_buffer_info)
661 		rte_free(ring->tx_buffer_info);
662 
663 	if (ring->empty_tx_reqs)
664 		rte_free(ring->empty_tx_reqs);
665 
666 	ring->empty_tx_reqs = NULL;
667 	ring->tx_buffer_info = NULL;
668 
669 	ring->configured = 0;
670 
671 	RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
672 		ring->port_id, ring->id);
673 }
674 
675 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
676 {
677 	unsigned int ring_mask = ring->ring_size - 1;
678 
679 	while (ring->next_to_clean != ring->next_to_use) {
680 		struct rte_mbuf *m =
681 			ring->rx_buffer_info[ring->next_to_clean & ring_mask];
682 
683 		if (m)
684 			rte_mbuf_raw_free(m);
685 
686 		ring->next_to_clean++;
687 	}
688 }
689 
690 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
691 {
692 	unsigned int i;
693 
694 	for (i = 0; i < ring->ring_size; ++i) {
695 		struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
696 
697 		if (tx_buf->mbuf)
698 			rte_pktmbuf_free(tx_buf->mbuf);
699 
700 		ring->next_to_clean++;
701 	}
702 }
703 
704 static int ena_link_update(struct rte_eth_dev *dev,
705 			   __rte_unused int wait_to_complete)
706 {
707 	struct rte_eth_link *link = &dev->data->dev_link;
708 
709 	link->link_status = 1;
710 	link->link_speed = ETH_SPEED_NUM_10G;
711 	link->link_duplex = ETH_LINK_FULL_DUPLEX;
712 
713 	return 0;
714 }
715 
716 static int ena_queue_restart_all(struct rte_eth_dev *dev,
717 				 enum ena_ring_type ring_type)
718 {
719 	struct ena_adapter *adapter =
720 		(struct ena_adapter *)(dev->data->dev_private);
721 	struct ena_ring *queues = NULL;
722 	int i = 0;
723 	int rc = 0;
724 
725 	queues = (ring_type == ENA_RING_TYPE_RX) ?
726 		adapter->rx_ring : adapter->tx_ring;
727 
728 	for (i = 0; i < adapter->num_queues; i++) {
729 		if (queues[i].configured) {
730 			if (ring_type == ENA_RING_TYPE_RX) {
731 				ena_assert_msg(
732 					dev->data->rx_queues[i] == &queues[i],
733 					"Inconsistent state of rx queues\n");
734 			} else {
735 				ena_assert_msg(
736 					dev->data->tx_queues[i] == &queues[i],
737 					"Inconsistent state of tx queues\n");
738 			}
739 
740 			rc = ena_queue_restart(&queues[i]);
741 
742 			if (rc) {
743 				PMD_INIT_LOG(ERR,
744 					     "failed to restart queue %d type(%d)",
745 					     i, ring_type);
746 				return -1;
747 			}
748 		}
749 	}
750 
751 	return 0;
752 }
753 
754 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
755 {
756 	uint32_t max_frame_len = adapter->max_mtu;
757 
758 	if (adapter->rte_eth_dev_data->dev_conf.rxmode.jumbo_frame == 1)
759 		max_frame_len =
760 			adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
761 
762 	return max_frame_len;
763 }
764 
765 static int ena_check_valid_conf(struct ena_adapter *adapter)
766 {
767 	uint32_t max_frame_len = ena_get_mtu_conf(adapter);
768 
769 	if (max_frame_len > adapter->max_mtu) {
770 		PMD_INIT_LOG(ERR, "Unsupported MTU of %d", max_frame_len);
771 		return -1;
772 	}
773 
774 	return 0;
775 }
776 
777 static int
778 ena_calc_queue_size(struct ena_com_dev *ena_dev,
779 		    struct ena_com_dev_get_features_ctx *get_feat_ctx)
780 {
781 	uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
782 
783 	queue_size = RTE_MIN(queue_size,
784 			     get_feat_ctx->max_queues.max_cq_depth);
785 	queue_size = RTE_MIN(queue_size,
786 			     get_feat_ctx->max_queues.max_sq_depth);
787 
788 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
789 		queue_size = RTE_MIN(queue_size,
790 				     get_feat_ctx->max_queues.max_llq_depth);
791 
792 	/* Round down to power of 2 */
793 	if (!rte_is_power_of_2(queue_size))
794 		queue_size = rte_align32pow2(queue_size >> 1);
795 
796 	if (queue_size == 0) {
797 		PMD_INIT_LOG(ERR, "Invalid queue size");
798 		return -EFAULT;
799 	}
800 
801 	return queue_size;
802 }
803 
804 static void ena_stats_restart(struct rte_eth_dev *dev)
805 {
806 	struct ena_adapter *adapter =
807 		(struct ena_adapter *)(dev->data->dev_private);
808 
809 	rte_atomic64_init(&adapter->drv_stats->ierrors);
810 	rte_atomic64_init(&adapter->drv_stats->oerrors);
811 	rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
812 }
813 
814 static void ena_stats_get(struct rte_eth_dev *dev,
815 			  struct rte_eth_stats *stats)
816 {
817 	struct ena_admin_basic_stats ena_stats;
818 	struct ena_adapter *adapter =
819 		(struct ena_adapter *)(dev->data->dev_private);
820 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
821 	int rc;
822 
823 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
824 		return;
825 
826 	memset(&ena_stats, 0, sizeof(ena_stats));
827 	rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
828 	if (unlikely(rc)) {
829 		RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
830 		return;
831 	}
832 
833 	/* Set of basic statistics from ENA */
834 	stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
835 					  ena_stats.rx_pkts_low);
836 	stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
837 					  ena_stats.tx_pkts_low);
838 	stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
839 					ena_stats.rx_bytes_low);
840 	stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
841 					ena_stats.tx_bytes_low);
842 	stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
843 					 ena_stats.rx_drops_low);
844 
845 	/* Driver related stats */
846 	stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
847 	stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
848 	stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
849 }
850 
851 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
852 {
853 	struct ena_adapter *adapter;
854 	struct ena_com_dev *ena_dev;
855 	int rc = 0;
856 
857 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
858 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
859 	adapter = (struct ena_adapter *)(dev->data->dev_private);
860 
861 	ena_dev = &adapter->ena_dev;
862 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
863 
864 	if (mtu > ena_get_mtu_conf(adapter)) {
865 		RTE_LOG(ERR, PMD,
866 			"Given MTU (%d) exceeds maximum MTU supported (%d)\n",
867 			mtu, ena_get_mtu_conf(adapter));
868 		rc = -EINVAL;
869 		goto err;
870 	}
871 
872 	rc = ena_com_set_dev_mtu(ena_dev, mtu);
873 	if (rc)
874 		RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
875 	else
876 		RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
877 
878 err:
879 	return rc;
880 }
881 
882 static int ena_start(struct rte_eth_dev *dev)
883 {
884 	struct ena_adapter *adapter =
885 		(struct ena_adapter *)(dev->data->dev_private);
886 	int rc = 0;
887 
888 	if (!(adapter->state == ENA_ADAPTER_STATE_CONFIG ||
889 	      adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
890 		PMD_INIT_LOG(ERR, "API violation");
891 		return -1;
892 	}
893 
894 	rc = ena_check_valid_conf(adapter);
895 	if (rc)
896 		return rc;
897 
898 	rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
899 	if (rc)
900 		return rc;
901 
902 	rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
903 	if (rc)
904 		return rc;
905 
906 	if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
907 	    ETH_MQ_RX_RSS_FLAG) {
908 		rc = ena_rss_init_default(adapter);
909 		if (rc)
910 			return rc;
911 	}
912 
913 	ena_stats_restart(dev);
914 
915 	adapter->state = ENA_ADAPTER_STATE_RUNNING;
916 
917 	return 0;
918 }
919 
920 static int ena_queue_restart(struct ena_ring *ring)
921 {
922 	int rc, bufs_num;
923 
924 	ena_assert_msg(ring->configured == 1,
925 		       "Trying to restart unconfigured queue\n");
926 
927 	ring->next_to_clean = 0;
928 	ring->next_to_use = 0;
929 
930 	if (ring->type == ENA_RING_TYPE_TX)
931 		return 0;
932 
933 	bufs_num = ring->ring_size - 1;
934 	rc = ena_populate_rx_queue(ring, bufs_num);
935 	if (rc != bufs_num) {
936 		PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
937 		return (-1);
938 	}
939 
940 	return 0;
941 }
942 
943 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
944 			      uint16_t queue_idx,
945 			      uint16_t nb_desc,
946 			      __rte_unused unsigned int socket_id,
947 			      __rte_unused const struct rte_eth_txconf *tx_conf)
948 {
949 	struct ena_com_create_io_ctx ctx =
950 		/* policy set to _HOST just to satisfy icc compiler */
951 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
952 		  ENA_COM_IO_QUEUE_DIRECTION_TX, 0, 0, 0, 0 };
953 	struct ena_ring *txq = NULL;
954 	struct ena_adapter *adapter =
955 		(struct ena_adapter *)(dev->data->dev_private);
956 	unsigned int i;
957 	int ena_qid;
958 	int rc;
959 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
960 
961 	txq = &adapter->tx_ring[queue_idx];
962 
963 	if (txq->configured) {
964 		RTE_LOG(CRIT, PMD,
965 			"API violation. Queue %d is already configured\n",
966 			queue_idx);
967 		return -1;
968 	}
969 
970 	if (!rte_is_power_of_2(nb_desc)) {
971 		RTE_LOG(ERR, PMD,
972 			"Unsupported size of RX queue: %d is not a power of 2.",
973 			nb_desc);
974 		return -EINVAL;
975 	}
976 
977 	if (nb_desc > adapter->tx_ring_size) {
978 		RTE_LOG(ERR, PMD,
979 			"Unsupported size of TX queue (max size: %d)\n",
980 			adapter->tx_ring_size);
981 		return -EINVAL;
982 	}
983 
984 	ena_qid = ENA_IO_TXQ_IDX(queue_idx);
985 
986 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
987 	ctx.qid = ena_qid;
988 	ctx.msix_vector = -1; /* admin interrupts not used */
989 	ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
990 	ctx.queue_size = adapter->tx_ring_size;
991 	ctx.numa_node = ena_cpu_to_node(queue_idx);
992 
993 	rc = ena_com_create_io_queue(ena_dev, &ctx);
994 	if (rc) {
995 		RTE_LOG(ERR, PMD,
996 			"failed to create io TX queue #%d (qid:%d) rc: %d\n",
997 			queue_idx, ena_qid, rc);
998 	}
999 	txq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1000 	txq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1001 
1002 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1003 				     &txq->ena_com_io_sq,
1004 				     &txq->ena_com_io_cq);
1005 	if (rc) {
1006 		RTE_LOG(ERR, PMD,
1007 			"Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1008 			queue_idx, rc);
1009 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1010 		goto err;
1011 	}
1012 
1013 	txq->port_id = dev->data->port_id;
1014 	txq->next_to_clean = 0;
1015 	txq->next_to_use = 0;
1016 	txq->ring_size = nb_desc;
1017 
1018 	txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1019 					  sizeof(struct ena_tx_buffer) *
1020 					  txq->ring_size,
1021 					  RTE_CACHE_LINE_SIZE);
1022 	if (!txq->tx_buffer_info) {
1023 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1024 		return -ENOMEM;
1025 	}
1026 
1027 	txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1028 					 sizeof(u16) * txq->ring_size,
1029 					 RTE_CACHE_LINE_SIZE);
1030 	if (!txq->empty_tx_reqs) {
1031 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1032 		rte_free(txq->tx_buffer_info);
1033 		return -ENOMEM;
1034 	}
1035 	for (i = 0; i < txq->ring_size; i++)
1036 		txq->empty_tx_reqs[i] = i;
1037 
1038 	/* Store pointer to this queue in upper layer */
1039 	txq->configured = 1;
1040 	dev->data->tx_queues[queue_idx] = txq;
1041 err:
1042 	return rc;
1043 }
1044 
1045 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1046 			      uint16_t queue_idx,
1047 			      uint16_t nb_desc,
1048 			      __rte_unused unsigned int socket_id,
1049 			      __rte_unused const struct rte_eth_rxconf *rx_conf,
1050 			      struct rte_mempool *mp)
1051 {
1052 	struct ena_com_create_io_ctx ctx =
1053 		/* policy set to _HOST just to satisfy icc compiler */
1054 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
1055 		  ENA_COM_IO_QUEUE_DIRECTION_RX, 0, 0, 0, 0 };
1056 	struct ena_adapter *adapter =
1057 		(struct ena_adapter *)(dev->data->dev_private);
1058 	struct ena_ring *rxq = NULL;
1059 	uint16_t ena_qid = 0;
1060 	int rc = 0;
1061 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1062 
1063 	rxq = &adapter->rx_ring[queue_idx];
1064 	if (rxq->configured) {
1065 		RTE_LOG(CRIT, PMD,
1066 			"API violation. Queue %d is already configured\n",
1067 			queue_idx);
1068 		return -1;
1069 	}
1070 
1071 	if (!rte_is_power_of_2(nb_desc)) {
1072 		RTE_LOG(ERR, PMD,
1073 			"Unsupported size of TX queue: %d is not a power of 2.",
1074 			nb_desc);
1075 		return -EINVAL;
1076 	}
1077 
1078 	if (nb_desc > adapter->rx_ring_size) {
1079 		RTE_LOG(ERR, PMD,
1080 			"Unsupported size of RX queue (max size: %d)\n",
1081 			adapter->rx_ring_size);
1082 		return -EINVAL;
1083 	}
1084 
1085 	ena_qid = ENA_IO_RXQ_IDX(queue_idx);
1086 
1087 	ctx.qid = ena_qid;
1088 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1089 	ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1090 	ctx.msix_vector = -1; /* admin interrupts not used */
1091 	ctx.queue_size = adapter->rx_ring_size;
1092 	ctx.numa_node = ena_cpu_to_node(queue_idx);
1093 
1094 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1095 	if (rc)
1096 		RTE_LOG(ERR, PMD, "failed to create io RX queue #%d rc: %d\n",
1097 			queue_idx, rc);
1098 
1099 	rxq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1100 	rxq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1101 
1102 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1103 				     &rxq->ena_com_io_sq,
1104 				     &rxq->ena_com_io_cq);
1105 	if (rc) {
1106 		RTE_LOG(ERR, PMD,
1107 			"Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1108 			queue_idx, rc);
1109 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1110 	}
1111 
1112 	rxq->port_id = dev->data->port_id;
1113 	rxq->next_to_clean = 0;
1114 	rxq->next_to_use = 0;
1115 	rxq->ring_size = nb_desc;
1116 	rxq->mb_pool = mp;
1117 
1118 	rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1119 					  sizeof(struct rte_mbuf *) * nb_desc,
1120 					  RTE_CACHE_LINE_SIZE);
1121 	if (!rxq->rx_buffer_info) {
1122 		RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1123 		return -ENOMEM;
1124 	}
1125 
1126 	/* Store pointer to this queue in upper layer */
1127 	rxq->configured = 1;
1128 	dev->data->rx_queues[queue_idx] = rxq;
1129 
1130 	return rc;
1131 }
1132 
1133 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1134 {
1135 	unsigned int i;
1136 	int rc;
1137 	uint16_t ring_size = rxq->ring_size;
1138 	uint16_t ring_mask = ring_size - 1;
1139 	uint16_t next_to_use = rxq->next_to_use;
1140 	uint16_t in_use;
1141 	struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1142 
1143 	if (unlikely(!count))
1144 		return 0;
1145 
1146 	in_use = rxq->next_to_use - rxq->next_to_clean;
1147 	ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1148 
1149 	count = RTE_MIN(count,
1150 			(uint16_t)(ring_size - (next_to_use & ring_mask)));
1151 
1152 	/* get resources for incoming packets */
1153 	rc = rte_mempool_get_bulk(rxq->mb_pool,
1154 				  (void **)(&mbufs[next_to_use & ring_mask]),
1155 				  count);
1156 	if (unlikely(rc < 0)) {
1157 		rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1158 		PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1159 		return 0;
1160 	}
1161 
1162 	for (i = 0; i < count; i++) {
1163 		uint16_t next_to_use_masked = next_to_use & ring_mask;
1164 		struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
1165 		struct ena_com_buf ebuf;
1166 
1167 		rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1168 		/* prepare physical address for DMA transaction */
1169 		ebuf.paddr = mbuf->buf_physaddr + RTE_PKTMBUF_HEADROOM;
1170 		ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1171 		/* pass resource to device */
1172 		rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1173 						&ebuf, next_to_use_masked);
1174 		if (unlikely(rc)) {
1175 			rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbuf),
1176 					     count - i);
1177 			RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1178 			break;
1179 		}
1180 		next_to_use++;
1181 	}
1182 
1183 	/* When we submitted free recources to device... */
1184 	if (i > 0) {
1185 		/* ...let HW know that it can fill buffers with data */
1186 		rte_wmb();
1187 		ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1188 
1189 		rxq->next_to_use = next_to_use;
1190 	}
1191 
1192 	return i;
1193 }
1194 
1195 static int ena_device_init(struct ena_com_dev *ena_dev,
1196 			   struct ena_com_dev_get_features_ctx *get_feat_ctx)
1197 {
1198 	int rc;
1199 	bool readless_supported;
1200 
1201 	/* Initialize mmio registers */
1202 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
1203 	if (rc) {
1204 		RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1205 		return rc;
1206 	}
1207 
1208 	/* The PCIe configuration space revision id indicate if mmio reg
1209 	 * read is disabled.
1210 	 */
1211 	readless_supported =
1212 		!(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1213 			       & ENA_MMIO_DISABLE_REG_READ);
1214 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1215 
1216 	/* reset device */
1217 	rc = ena_com_dev_reset(ena_dev);
1218 	if (rc) {
1219 		RTE_LOG(ERR, PMD, "cannot reset device\n");
1220 		goto err_mmio_read_less;
1221 	}
1222 
1223 	/* check FW version */
1224 	rc = ena_com_validate_version(ena_dev);
1225 	if (rc) {
1226 		RTE_LOG(ERR, PMD, "device version is too low\n");
1227 		goto err_mmio_read_less;
1228 	}
1229 
1230 	ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1231 
1232 	/* ENA device administration layer init */
1233 	rc = ena_com_admin_init(ena_dev, NULL, true);
1234 	if (rc) {
1235 		RTE_LOG(ERR, PMD,
1236 			"cannot initialize ena admin queue with device\n");
1237 		goto err_mmio_read_less;
1238 	}
1239 
1240 	/* To enable the msix interrupts the driver needs to know the number
1241 	 * of queues. So the driver uses polling mode to retrieve this
1242 	 * information.
1243 	 */
1244 	ena_com_set_admin_polling_mode(ena_dev, true);
1245 
1246 	ena_config_host_info(ena_dev);
1247 
1248 	/* Get Device Attributes and features */
1249 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1250 	if (rc) {
1251 		RTE_LOG(ERR, PMD,
1252 			"cannot get attribute for ena device rc= %d\n", rc);
1253 		goto err_admin_init;
1254 	}
1255 
1256 	return 0;
1257 
1258 err_admin_init:
1259 	ena_com_admin_destroy(ena_dev);
1260 
1261 err_mmio_read_less:
1262 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1263 
1264 	return rc;
1265 }
1266 
1267 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1268 {
1269 	struct rte_pci_device *pci_dev;
1270 	struct ena_adapter *adapter =
1271 		(struct ena_adapter *)(eth_dev->data->dev_private);
1272 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1273 	struct ena_com_dev_get_features_ctx get_feat_ctx;
1274 	int queue_size, rc;
1275 
1276 	static int adapters_found;
1277 
1278 	memset(adapter, 0, sizeof(struct ena_adapter));
1279 	ena_dev = &adapter->ena_dev;
1280 
1281 	eth_dev->dev_ops = &ena_dev_ops;
1282 	eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1283 	eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1284 	eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1285 	adapter->rte_eth_dev_data = eth_dev->data;
1286 	adapter->rte_dev = eth_dev;
1287 
1288 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1289 		return 0;
1290 
1291 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1292 	adapter->pdev = pci_dev;
1293 
1294 	PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1295 		     pci_dev->addr.domain,
1296 		     pci_dev->addr.bus,
1297 		     pci_dev->addr.devid,
1298 		     pci_dev->addr.function);
1299 
1300 	adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1301 	adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1302 
1303 	/* Present ENA_MEM_BAR indicates available LLQ mode.
1304 	 * Use corresponding policy
1305 	 */
1306 	if (adapter->dev_mem_base)
1307 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
1308 	else if (adapter->regs)
1309 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1310 	else
1311 		PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1312 			     ENA_REGS_BAR);
1313 
1314 	ena_dev->reg_bar = adapter->regs;
1315 	ena_dev->dmadev = adapter->pdev;
1316 
1317 	adapter->id_number = adapters_found;
1318 
1319 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1320 		 adapter->id_number);
1321 
1322 	/* device specific initialization routine */
1323 	rc = ena_device_init(ena_dev, &get_feat_ctx);
1324 	if (rc) {
1325 		PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1326 		return -1;
1327 	}
1328 
1329 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1330 		if (get_feat_ctx.max_queues.max_llq_num == 0) {
1331 			PMD_INIT_LOG(ERR,
1332 				     "Trying to use LLQ but llq_num is 0.\n"
1333 				     "Fall back into regular queues.");
1334 			ena_dev->tx_mem_queue_type =
1335 				ENA_ADMIN_PLACEMENT_POLICY_HOST;
1336 			adapter->num_queues =
1337 				get_feat_ctx.max_queues.max_sq_num;
1338 		} else {
1339 			adapter->num_queues =
1340 				get_feat_ctx.max_queues.max_llq_num;
1341 		}
1342 	} else {
1343 		adapter->num_queues = get_feat_ctx.max_queues.max_sq_num;
1344 	}
1345 
1346 	queue_size = ena_calc_queue_size(ena_dev, &get_feat_ctx);
1347 	if ((queue_size <= 0) || (adapter->num_queues <= 0))
1348 		return -EFAULT;
1349 
1350 	adapter->tx_ring_size = queue_size;
1351 	adapter->rx_ring_size = queue_size;
1352 
1353 	/* prepare ring structures */
1354 	ena_init_rings(adapter);
1355 
1356 	ena_config_debug_area(adapter);
1357 
1358 	/* Set max MTU for this device */
1359 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1360 
1361 	/* set device support for TSO */
1362 	adapter->tso4_supported = get_feat_ctx.offload.tx &
1363 				  ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1364 
1365 	/* Copy MAC address and point DPDK to it */
1366 	eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1367 	ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1368 			(struct ether_addr *)adapter->mac_addr);
1369 
1370 	adapter->drv_stats = rte_zmalloc("adapter stats",
1371 					 sizeof(*adapter->drv_stats),
1372 					 RTE_CACHE_LINE_SIZE);
1373 	if (!adapter->drv_stats) {
1374 		RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1375 		return -ENOMEM;
1376 	}
1377 
1378 	adapters_found++;
1379 	adapter->state = ENA_ADAPTER_STATE_INIT;
1380 
1381 	return 0;
1382 }
1383 
1384 static int ena_dev_configure(struct rte_eth_dev *dev)
1385 {
1386 	struct ena_adapter *adapter =
1387 		(struct ena_adapter *)(dev->data->dev_private);
1388 
1389 	if (!(adapter->state == ENA_ADAPTER_STATE_INIT ||
1390 	      adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
1391 		PMD_INIT_LOG(ERR, "Illegal adapter state: %d",
1392 			     adapter->state);
1393 		return -1;
1394 	}
1395 
1396 	switch (adapter->state) {
1397 	case ENA_ADAPTER_STATE_INIT:
1398 	case ENA_ADAPTER_STATE_STOPPED:
1399 		adapter->state = ENA_ADAPTER_STATE_CONFIG;
1400 		break;
1401 	case ENA_ADAPTER_STATE_CONFIG:
1402 		RTE_LOG(WARNING, PMD,
1403 			"Ivalid driver state while trying to configure device\n");
1404 		break;
1405 	default:
1406 		break;
1407 	}
1408 
1409 	return 0;
1410 }
1411 
1412 static void ena_init_rings(struct ena_adapter *adapter)
1413 {
1414 	int i;
1415 
1416 	for (i = 0; i < adapter->num_queues; i++) {
1417 		struct ena_ring *ring = &adapter->tx_ring[i];
1418 
1419 		ring->configured = 0;
1420 		ring->type = ENA_RING_TYPE_TX;
1421 		ring->adapter = adapter;
1422 		ring->id = i;
1423 		ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1424 		ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1425 	}
1426 
1427 	for (i = 0; i < adapter->num_queues; i++) {
1428 		struct ena_ring *ring = &adapter->rx_ring[i];
1429 
1430 		ring->configured = 0;
1431 		ring->type = ENA_RING_TYPE_RX;
1432 		ring->adapter = adapter;
1433 		ring->id = i;
1434 	}
1435 }
1436 
1437 static void ena_infos_get(struct rte_eth_dev *dev,
1438 			  struct rte_eth_dev_info *dev_info)
1439 {
1440 	struct ena_adapter *adapter;
1441 	struct ena_com_dev *ena_dev;
1442 	struct ena_com_dev_get_features_ctx feat;
1443 	uint32_t rx_feat = 0, tx_feat = 0;
1444 	int rc = 0;
1445 
1446 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
1447 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1448 	adapter = (struct ena_adapter *)(dev->data->dev_private);
1449 
1450 	ena_dev = &adapter->ena_dev;
1451 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1452 
1453 	dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1454 
1455 	dev_info->speed_capa =
1456 			ETH_LINK_SPEED_1G   |
1457 			ETH_LINK_SPEED_2_5G |
1458 			ETH_LINK_SPEED_5G   |
1459 			ETH_LINK_SPEED_10G  |
1460 			ETH_LINK_SPEED_25G  |
1461 			ETH_LINK_SPEED_40G  |
1462 			ETH_LINK_SPEED_50G  |
1463 			ETH_LINK_SPEED_100G;
1464 
1465 	/* Get supported features from HW */
1466 	rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1467 	if (unlikely(rc)) {
1468 		RTE_LOG(ERR, PMD,
1469 			"Cannot get attribute for ena device rc= %d\n", rc);
1470 		return;
1471 	}
1472 
1473 	/* Set Tx & Rx features available for device */
1474 	if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1475 		tx_feat	|= DEV_TX_OFFLOAD_TCP_TSO;
1476 
1477 	if (feat.offload.tx &
1478 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1479 		tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1480 			DEV_TX_OFFLOAD_UDP_CKSUM |
1481 			DEV_TX_OFFLOAD_TCP_CKSUM;
1482 
1483 	if (feat.offload.rx_supported &
1484 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1485 		rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1486 			DEV_RX_OFFLOAD_UDP_CKSUM  |
1487 			DEV_RX_OFFLOAD_TCP_CKSUM;
1488 
1489 	/* Inform framework about available features */
1490 	dev_info->rx_offload_capa = rx_feat;
1491 	dev_info->tx_offload_capa = tx_feat;
1492 
1493 	dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1494 	dev_info->max_rx_pktlen  = adapter->max_mtu;
1495 	dev_info->max_mac_addrs = 1;
1496 
1497 	dev_info->max_rx_queues = adapter->num_queues;
1498 	dev_info->max_tx_queues = adapter->num_queues;
1499 	dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1500 }
1501 
1502 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1503 				  uint16_t nb_pkts)
1504 {
1505 	struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1506 	unsigned int ring_size = rx_ring->ring_size;
1507 	unsigned int ring_mask = ring_size - 1;
1508 	uint16_t next_to_clean = rx_ring->next_to_clean;
1509 	uint16_t desc_in_use = 0;
1510 	unsigned int recv_idx = 0;
1511 	struct rte_mbuf *mbuf = NULL;
1512 	struct rte_mbuf *mbuf_head = NULL;
1513 	struct rte_mbuf *mbuf_prev = NULL;
1514 	struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1515 	unsigned int completed;
1516 
1517 	struct ena_com_rx_ctx ena_rx_ctx;
1518 	int rc = 0;
1519 
1520 	/* Check adapter state */
1521 	if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1522 		RTE_LOG(ALERT, PMD,
1523 			"Trying to receive pkts while device is NOT running\n");
1524 		return 0;
1525 	}
1526 
1527 	desc_in_use = rx_ring->next_to_use - next_to_clean;
1528 	if (unlikely(nb_pkts > desc_in_use))
1529 		nb_pkts = desc_in_use;
1530 
1531 	for (completed = 0; completed < nb_pkts; completed++) {
1532 		int segments = 0;
1533 
1534 		ena_rx_ctx.max_bufs = rx_ring->ring_size;
1535 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1536 		ena_rx_ctx.descs = 0;
1537 		/* receive packet context */
1538 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1539 				    rx_ring->ena_com_io_sq,
1540 				    &ena_rx_ctx);
1541 		if (unlikely(rc)) {
1542 			RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1543 			return 0;
1544 		}
1545 
1546 		if (unlikely(ena_rx_ctx.descs == 0))
1547 			break;
1548 
1549 		while (segments < ena_rx_ctx.descs) {
1550 			mbuf = rx_buff_info[next_to_clean & ring_mask];
1551 			mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1552 			mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1553 			mbuf->refcnt = 1;
1554 			mbuf->next = NULL;
1555 			if (segments == 0) {
1556 				mbuf->nb_segs = ena_rx_ctx.descs;
1557 				mbuf->port = rx_ring->port_id;
1558 				mbuf->pkt_len = 0;
1559 				mbuf_head = mbuf;
1560 			} else {
1561 				/* for multi-segment pkts create mbuf chain */
1562 				mbuf_prev->next = mbuf;
1563 			}
1564 			mbuf_head->pkt_len += mbuf->data_len;
1565 
1566 			mbuf_prev = mbuf;
1567 			segments++;
1568 			next_to_clean++;
1569 		}
1570 
1571 		/* fill mbuf attributes if any */
1572 		ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1573 		mbuf_head->hash.rss = (uint32_t)rx_ring->id;
1574 
1575 		/* pass to DPDK application head mbuf */
1576 		rx_pkts[recv_idx] = mbuf_head;
1577 		recv_idx++;
1578 	}
1579 
1580 	rx_ring->next_to_clean = next_to_clean;
1581 
1582 	desc_in_use = desc_in_use - completed + 1;
1583 	/* Burst refill to save doorbells, memory barriers, const interval */
1584 	if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1585 		ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1586 
1587 	return recv_idx;
1588 }
1589 
1590 static uint16_t
1591 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1592 		uint16_t nb_pkts)
1593 {
1594 	int32_t ret;
1595 	uint32_t i;
1596 	struct rte_mbuf *m;
1597 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1598 	struct ipv4_hdr *ip_hdr;
1599 	uint64_t ol_flags;
1600 	uint16_t frag_field;
1601 
1602 	for (i = 0; i != nb_pkts; i++) {
1603 		m = tx_pkts[i];
1604 		ol_flags = m->ol_flags;
1605 
1606 		if (!(ol_flags & PKT_TX_IPV4))
1607 			continue;
1608 
1609 		/* If there was not L2 header length specified, assume it is
1610 		 * length of the ethernet header.
1611 		 */
1612 		if (unlikely(m->l2_len == 0))
1613 			m->l2_len = sizeof(struct ether_hdr);
1614 
1615 		ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1616 						 m->l2_len);
1617 		frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1618 
1619 		if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1620 			m->packet_type |= RTE_PTYPE_L4_NONFRAG;
1621 
1622 			/* If IPv4 header has DF flag enabled and TSO support is
1623 			 * disabled, partial chcecksum should not be calculated.
1624 			 */
1625 			if (!tx_ring->adapter->tso4_supported)
1626 				continue;
1627 		}
1628 
1629 		if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
1630 				(ol_flags & PKT_TX_L4_MASK) ==
1631 				PKT_TX_SCTP_CKSUM) {
1632 			rte_errno = -ENOTSUP;
1633 			return i;
1634 		}
1635 
1636 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1637 		ret = rte_validate_tx_offload(m);
1638 		if (ret != 0) {
1639 			rte_errno = ret;
1640 			return i;
1641 		}
1642 #endif
1643 
1644 		/* In case we are supposed to TSO and have DF not set (DF=0)
1645 		 * hardware must be provided with partial checksum, otherwise
1646 		 * it will take care of necessary calculations.
1647 		 */
1648 
1649 		ret = rte_net_intel_cksum_flags_prepare(m,
1650 			ol_flags & ~PKT_TX_TCP_SEG);
1651 		if (ret != 0) {
1652 			rte_errno = ret;
1653 			return i;
1654 		}
1655 	}
1656 
1657 	return i;
1658 }
1659 
1660 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1661 				  uint16_t nb_pkts)
1662 {
1663 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1664 	uint16_t next_to_use = tx_ring->next_to_use;
1665 	uint16_t next_to_clean = tx_ring->next_to_clean;
1666 	struct rte_mbuf *mbuf;
1667 	unsigned int ring_size = tx_ring->ring_size;
1668 	unsigned int ring_mask = ring_size - 1;
1669 	struct ena_com_tx_ctx ena_tx_ctx;
1670 	struct ena_tx_buffer *tx_info;
1671 	struct ena_com_buf *ebuf;
1672 	uint16_t rc, req_id, total_tx_descs = 0;
1673 	uint16_t sent_idx = 0, empty_tx_reqs;
1674 	int nb_hw_desc;
1675 
1676 	/* Check adapter state */
1677 	if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1678 		RTE_LOG(ALERT, PMD,
1679 			"Trying to xmit pkts while device is NOT running\n");
1680 		return 0;
1681 	}
1682 
1683 	empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
1684 	if (nb_pkts > empty_tx_reqs)
1685 		nb_pkts = empty_tx_reqs;
1686 
1687 	for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
1688 		mbuf = tx_pkts[sent_idx];
1689 
1690 		req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
1691 		tx_info = &tx_ring->tx_buffer_info[req_id];
1692 		tx_info->mbuf = mbuf;
1693 		tx_info->num_of_bufs = 0;
1694 		ebuf = tx_info->bufs;
1695 
1696 		/* Prepare TX context */
1697 		memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1698 		memset(&ena_tx_ctx.ena_meta, 0x0,
1699 		       sizeof(struct ena_com_tx_meta));
1700 		ena_tx_ctx.ena_bufs = ebuf;
1701 		ena_tx_ctx.req_id = req_id;
1702 		if (tx_ring->tx_mem_queue_type ==
1703 				ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1704 			/* prepare the push buffer with
1705 			 * virtual address of the data
1706 			 */
1707 			ena_tx_ctx.header_len =
1708 				RTE_MIN(mbuf->data_len,
1709 					tx_ring->tx_max_header_size);
1710 			ena_tx_ctx.push_header =
1711 				(void *)((char *)mbuf->buf_addr +
1712 					 mbuf->data_off);
1713 		} /* there's no else as we take advantage of memset zeroing */
1714 
1715 		/* Set TX offloads flags, if applicable */
1716 		ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx);
1717 
1718 		if (unlikely(mbuf->ol_flags &
1719 			     (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
1720 			rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
1721 
1722 		rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
1723 
1724 		/* Process first segment taking into
1725 		 * consideration pushed header
1726 		 */
1727 		if (mbuf->data_len > ena_tx_ctx.header_len) {
1728 			ebuf->paddr = mbuf->buf_physaddr +
1729 				      mbuf->data_off +
1730 				      ena_tx_ctx.header_len;
1731 			ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
1732 			ebuf++;
1733 			tx_info->num_of_bufs++;
1734 		}
1735 
1736 		while ((mbuf = mbuf->next) != NULL) {
1737 			ebuf->paddr = mbuf->buf_physaddr + mbuf->data_off;
1738 			ebuf->len = mbuf->data_len;
1739 			ebuf++;
1740 			tx_info->num_of_bufs++;
1741 		}
1742 
1743 		ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1744 
1745 		/* Write data to device */
1746 		rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
1747 					&ena_tx_ctx, &nb_hw_desc);
1748 		if (unlikely(rc))
1749 			break;
1750 
1751 		tx_info->tx_descs = nb_hw_desc;
1752 
1753 		next_to_use++;
1754 	}
1755 
1756 	/* If there are ready packets to be xmitted... */
1757 	if (sent_idx > 0) {
1758 		/* ...let HW do its best :-) */
1759 		rte_wmb();
1760 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
1761 
1762 		tx_ring->next_to_use = next_to_use;
1763 	}
1764 
1765 	/* Clear complete packets  */
1766 	while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
1767 		/* Get Tx info & store how many descs were processed  */
1768 		tx_info = &tx_ring->tx_buffer_info[req_id];
1769 		total_tx_descs += tx_info->tx_descs;
1770 
1771 		/* Free whole mbuf chain  */
1772 		mbuf = tx_info->mbuf;
1773 		rte_pktmbuf_free(mbuf);
1774 		tx_info->mbuf = NULL;
1775 
1776 		/* Put back descriptor to the ring for reuse */
1777 		tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
1778 		next_to_clean++;
1779 
1780 		/* If too many descs to clean, leave it for another run */
1781 		if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
1782 			break;
1783 	}
1784 
1785 	if (total_tx_descs > 0) {
1786 		/* acknowledge completion of sent packets */
1787 		ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
1788 		tx_ring->next_to_clean = next_to_clean;
1789 	}
1790 
1791 	return sent_idx;
1792 }
1793 
1794 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1795 	struct rte_pci_device *pci_dev)
1796 {
1797 	return rte_eth_dev_pci_generic_probe(pci_dev,
1798 		sizeof(struct ena_adapter), eth_ena_dev_init);
1799 }
1800 
1801 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
1802 {
1803 	return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1804 }
1805 
1806 static struct rte_pci_driver rte_ena_pmd = {
1807 	.id_table = pci_id_ena_map,
1808 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1809 	.probe = eth_ena_pci_probe,
1810 	.remove = eth_ena_pci_remove,
1811 };
1812 
1813 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
1814 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
1815 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
1816