xref: /f-stack/dpdk/drivers/net/ena/ena_ethdev.c (revision 04b1440d)
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_driver.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	1
58 #define DRV_MODULE_VER_SUBMINOR	1
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 #define ENA_MAX_RING_DESC	ENA_DEFAULT_RING_SIZE
89 #define ENA_MIN_RING_DESC	128
90 
91 enum ethtool_stringset {
92 	ETH_SS_TEST             = 0,
93 	ETH_SS_STATS,
94 };
95 
96 struct ena_stats {
97 	char name[ETH_GSTRING_LEN];
98 	int stat_offset;
99 };
100 
101 #define ENA_STAT_ENA_COM_ENTRY(stat) { \
102 	.name = #stat, \
103 	.stat_offset = offsetof(struct ena_com_stats_admin, stat) \
104 }
105 
106 #define ENA_STAT_ENTRY(stat, stat_type) { \
107 	.name = #stat, \
108 	.stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
109 }
110 
111 #define ENA_STAT_RX_ENTRY(stat) \
112 	ENA_STAT_ENTRY(stat, rx)
113 
114 #define ENA_STAT_TX_ENTRY(stat) \
115 	ENA_STAT_ENTRY(stat, tx)
116 
117 #define ENA_STAT_GLOBAL_ENTRY(stat) \
118 	ENA_STAT_ENTRY(stat, dev)
119 
120 /*
121  * Each rte_memzone should have unique name.
122  * To satisfy it, count number of allocation and add it to name.
123  */
124 uint32_t ena_alloc_cnt;
125 
126 static const struct ena_stats ena_stats_global_strings[] = {
127 	ENA_STAT_GLOBAL_ENTRY(tx_timeout),
128 	ENA_STAT_GLOBAL_ENTRY(io_suspend),
129 	ENA_STAT_GLOBAL_ENTRY(io_resume),
130 	ENA_STAT_GLOBAL_ENTRY(wd_expired),
131 	ENA_STAT_GLOBAL_ENTRY(interface_up),
132 	ENA_STAT_GLOBAL_ENTRY(interface_down),
133 	ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
134 };
135 
136 static const struct ena_stats ena_stats_tx_strings[] = {
137 	ENA_STAT_TX_ENTRY(cnt),
138 	ENA_STAT_TX_ENTRY(bytes),
139 	ENA_STAT_TX_ENTRY(queue_stop),
140 	ENA_STAT_TX_ENTRY(queue_wakeup),
141 	ENA_STAT_TX_ENTRY(dma_mapping_err),
142 	ENA_STAT_TX_ENTRY(linearize),
143 	ENA_STAT_TX_ENTRY(linearize_failed),
144 	ENA_STAT_TX_ENTRY(tx_poll),
145 	ENA_STAT_TX_ENTRY(doorbells),
146 	ENA_STAT_TX_ENTRY(prepare_ctx_err),
147 	ENA_STAT_TX_ENTRY(missing_tx_comp),
148 	ENA_STAT_TX_ENTRY(bad_req_id),
149 };
150 
151 static const struct ena_stats ena_stats_rx_strings[] = {
152 	ENA_STAT_RX_ENTRY(cnt),
153 	ENA_STAT_RX_ENTRY(bytes),
154 	ENA_STAT_RX_ENTRY(refil_partial),
155 	ENA_STAT_RX_ENTRY(bad_csum),
156 	ENA_STAT_RX_ENTRY(page_alloc_fail),
157 	ENA_STAT_RX_ENTRY(skb_alloc_fail),
158 	ENA_STAT_RX_ENTRY(dma_mapping_err),
159 	ENA_STAT_RX_ENTRY(bad_desc_num),
160 	ENA_STAT_RX_ENTRY(small_copy_len_pkt),
161 };
162 
163 static const struct ena_stats ena_stats_ena_com_strings[] = {
164 	ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
165 	ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
166 	ENA_STAT_ENA_COM_ENTRY(completed_cmd),
167 	ENA_STAT_ENA_COM_ENTRY(out_of_space),
168 	ENA_STAT_ENA_COM_ENTRY(no_completion),
169 };
170 
171 #define ENA_STATS_ARRAY_GLOBAL	ARRAY_SIZE(ena_stats_global_strings)
172 #define ENA_STATS_ARRAY_TX	ARRAY_SIZE(ena_stats_tx_strings)
173 #define ENA_STATS_ARRAY_RX	ARRAY_SIZE(ena_stats_rx_strings)
174 #define ENA_STATS_ARRAY_ENA_COM	ARRAY_SIZE(ena_stats_ena_com_strings)
175 
176 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
177 			DEV_TX_OFFLOAD_UDP_CKSUM |\
178 			DEV_TX_OFFLOAD_IPV4_CKSUM |\
179 			DEV_TX_OFFLOAD_TCP_TSO)
180 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
181 		       PKT_TX_IP_CKSUM |\
182 		       PKT_TX_TCP_SEG)
183 
184 /** Vendor ID used by Amazon devices */
185 #define PCI_VENDOR_ID_AMAZON 0x1D0F
186 /** Amazon devices */
187 #define PCI_DEVICE_ID_ENA_VF	0xEC20
188 #define PCI_DEVICE_ID_ENA_LLQ_VF	0xEC21
189 
190 #define	ENA_TX_OFFLOAD_MASK	(\
191 	PKT_TX_L4_MASK |         \
192 	PKT_TX_IPV6 |            \
193 	PKT_TX_IPV4 |            \
194 	PKT_TX_IP_CKSUM |        \
195 	PKT_TX_TCP_SEG)
196 
197 #define	ENA_TX_OFFLOAD_NOTSUP_MASK	\
198 	(PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
199 
200 int ena_logtype_init;
201 int ena_logtype_driver;
202 
203 static const struct rte_pci_id pci_id_ena_map[] = {
204 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
205 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
206 	{ .device_id = 0 },
207 };
208 
209 static struct ena_aenq_handlers aenq_handlers;
210 
211 static int ena_device_init(struct ena_com_dev *ena_dev,
212 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
213 			   bool *wd_state);
214 static int ena_dev_configure(struct rte_eth_dev *dev);
215 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
216 				  uint16_t nb_pkts);
217 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
218 		uint16_t nb_pkts);
219 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
220 			      uint16_t nb_desc, unsigned int socket_id,
221 			      const struct rte_eth_txconf *tx_conf);
222 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
223 			      uint16_t nb_desc, unsigned int socket_id,
224 			      const struct rte_eth_rxconf *rx_conf,
225 			      struct rte_mempool *mp);
226 static uint16_t eth_ena_recv_pkts(void *rx_queue,
227 				  struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
228 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
229 static void ena_init_rings(struct ena_adapter *adapter);
230 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
231 static int ena_start(struct rte_eth_dev *dev);
232 static void ena_stop(struct rte_eth_dev *dev);
233 static void ena_close(struct rte_eth_dev *dev);
234 static int ena_dev_reset(struct rte_eth_dev *dev);
235 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
236 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
237 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
238 static void ena_rx_queue_release(void *queue);
239 static void ena_tx_queue_release(void *queue);
240 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
241 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
242 static int ena_link_update(struct rte_eth_dev *dev,
243 			   int wait_to_complete);
244 static int ena_create_io_queue(struct ena_ring *ring);
245 static void ena_queue_stop(struct ena_ring *ring);
246 static void ena_queue_stop_all(struct rte_eth_dev *dev,
247 			      enum ena_ring_type ring_type);
248 static int ena_queue_start(struct ena_ring *ring);
249 static int ena_queue_start_all(struct rte_eth_dev *dev,
250 			       enum ena_ring_type ring_type);
251 static void ena_stats_restart(struct rte_eth_dev *dev);
252 static void ena_infos_get(struct rte_eth_dev *dev,
253 			  struct rte_eth_dev_info *dev_info);
254 static int ena_rss_reta_update(struct rte_eth_dev *dev,
255 			       struct rte_eth_rss_reta_entry64 *reta_conf,
256 			       uint16_t reta_size);
257 static int ena_rss_reta_query(struct rte_eth_dev *dev,
258 			      struct rte_eth_rss_reta_entry64 *reta_conf,
259 			      uint16_t reta_size);
260 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
261 static void ena_interrupt_handler_rte(void *cb_arg);
262 static void ena_timer_wd_callback(struct rte_timer *timer, void *arg);
263 static void ena_destroy_device(struct rte_eth_dev *eth_dev);
264 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev);
265 
266 static const struct eth_dev_ops ena_dev_ops = {
267 	.dev_configure        = ena_dev_configure,
268 	.dev_infos_get        = ena_infos_get,
269 	.rx_queue_setup       = ena_rx_queue_setup,
270 	.tx_queue_setup       = ena_tx_queue_setup,
271 	.dev_start            = ena_start,
272 	.dev_stop             = ena_stop,
273 	.link_update          = ena_link_update,
274 	.stats_get            = ena_stats_get,
275 	.mtu_set              = ena_mtu_set,
276 	.rx_queue_release     = ena_rx_queue_release,
277 	.tx_queue_release     = ena_tx_queue_release,
278 	.dev_close            = ena_close,
279 	.dev_reset            = ena_dev_reset,
280 	.reta_update          = ena_rss_reta_update,
281 	.reta_query           = ena_rss_reta_query,
282 };
283 
284 #define NUMA_NO_NODE	SOCKET_ID_ANY
285 
286 static inline int ena_cpu_to_node(int cpu)
287 {
288 	struct rte_config *config = rte_eal_get_configuration();
289 	struct rte_fbarray *arr = &config->mem_config->memzones;
290 	const struct rte_memzone *mz;
291 
292 	if (unlikely(cpu >= RTE_MAX_MEMZONE))
293 		return NUMA_NO_NODE;
294 
295 	mz = rte_fbarray_get(arr, cpu);
296 
297 	return mz->socket_id;
298 }
299 
300 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
301 				       struct ena_com_rx_ctx *ena_rx_ctx)
302 {
303 	uint64_t ol_flags = 0;
304 	uint32_t packet_type = 0;
305 
306 	if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
307 		packet_type |= RTE_PTYPE_L4_TCP;
308 	else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
309 		packet_type |= RTE_PTYPE_L4_UDP;
310 
311 	if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
312 		packet_type |= RTE_PTYPE_L3_IPV4;
313 	else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
314 		packet_type |= RTE_PTYPE_L3_IPV6;
315 
316 	if (unlikely(ena_rx_ctx->l4_csum_err))
317 		ol_flags |= PKT_RX_L4_CKSUM_BAD;
318 	if (unlikely(ena_rx_ctx->l3_csum_err))
319 		ol_flags |= PKT_RX_IP_CKSUM_BAD;
320 
321 	mbuf->ol_flags = ol_flags;
322 	mbuf->packet_type = packet_type;
323 }
324 
325 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
326 				       struct ena_com_tx_ctx *ena_tx_ctx,
327 				       uint64_t queue_offloads)
328 {
329 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
330 
331 	if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
332 	    (queue_offloads & QUEUE_OFFLOADS)) {
333 		/* check if TSO is required */
334 		if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
335 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
336 			ena_tx_ctx->tso_enable = true;
337 
338 			ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
339 		}
340 
341 		/* check if L3 checksum is needed */
342 		if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
343 		    (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
344 			ena_tx_ctx->l3_csum_enable = true;
345 
346 		if (mbuf->ol_flags & PKT_TX_IPV6) {
347 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
348 		} else {
349 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
350 
351 			/* set don't fragment (DF) flag */
352 			if (mbuf->packet_type &
353 				(RTE_PTYPE_L4_NONFRAG
354 				 | RTE_PTYPE_INNER_L4_NONFRAG))
355 				ena_tx_ctx->df = true;
356 		}
357 
358 		/* check if L4 checksum is needed */
359 		if ((mbuf->ol_flags & PKT_TX_TCP_CKSUM) &&
360 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
361 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
362 			ena_tx_ctx->l4_csum_enable = true;
363 		} else if ((mbuf->ol_flags & PKT_TX_UDP_CKSUM) &&
364 			   (queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
365 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
366 			ena_tx_ctx->l4_csum_enable = true;
367 		} else {
368 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
369 			ena_tx_ctx->l4_csum_enable = false;
370 		}
371 
372 		ena_meta->mss = mbuf->tso_segsz;
373 		ena_meta->l3_hdr_len = mbuf->l3_len;
374 		ena_meta->l3_hdr_offset = mbuf->l2_len;
375 
376 		ena_tx_ctx->meta_valid = true;
377 	} else {
378 		ena_tx_ctx->meta_valid = false;
379 	}
380 }
381 
382 static inline int validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
383 {
384 	if (likely(req_id < rx_ring->ring_size))
385 		return 0;
386 
387 	RTE_LOG(ERR, PMD, "Invalid rx req_id: %hu\n", req_id);
388 
389 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
390 	rx_ring->adapter->trigger_reset = true;
391 
392 	return -EFAULT;
393 }
394 
395 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
396 {
397 	struct ena_tx_buffer *tx_info = NULL;
398 
399 	if (likely(req_id < tx_ring->ring_size)) {
400 		tx_info = &tx_ring->tx_buffer_info[req_id];
401 		if (likely(tx_info->mbuf))
402 			return 0;
403 	}
404 
405 	if (tx_info)
406 		RTE_LOG(ERR, PMD, "tx_info doesn't have valid mbuf\n");
407 	else
408 		RTE_LOG(ERR, PMD, "Invalid req_id: %hu\n", req_id);
409 
410 	/* Trigger device reset */
411 	tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
412 	tx_ring->adapter->trigger_reset	= true;
413 	return -EFAULT;
414 }
415 
416 static void ena_config_host_info(struct ena_com_dev *ena_dev)
417 {
418 	struct ena_admin_host_info *host_info;
419 	int rc;
420 
421 	/* Allocate only the host info */
422 	rc = ena_com_allocate_host_info(ena_dev);
423 	if (rc) {
424 		RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
425 		return;
426 	}
427 
428 	host_info = ena_dev->host_attr.host_info;
429 
430 	host_info->os_type = ENA_ADMIN_OS_DPDK;
431 	host_info->kernel_ver = RTE_VERSION;
432 	snprintf((char *)host_info->kernel_ver_str,
433 		 sizeof(host_info->kernel_ver_str),
434 		 "%s", rte_version());
435 	host_info->os_dist = RTE_VERSION;
436 	snprintf((char *)host_info->os_dist_str,
437 		 sizeof(host_info->os_dist_str),
438 		 "%s", rte_version());
439 	host_info->driver_version =
440 		(DRV_MODULE_VER_MAJOR) |
441 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
442 		(DRV_MODULE_VER_SUBMINOR <<
443 			ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
444 
445 	rc = ena_com_set_host_attributes(ena_dev);
446 	if (rc) {
447 		if (rc == -ENA_COM_UNSUPPORTED)
448 			RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
449 		else
450 			RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
451 
452 		goto err;
453 	}
454 
455 	return;
456 
457 err:
458 	ena_com_delete_host_info(ena_dev);
459 }
460 
461 static int
462 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
463 {
464 	if (sset != ETH_SS_STATS)
465 		return -EOPNOTSUPP;
466 
467 	 /* Workaround for clang:
468 	 * touch internal structures to prevent
469 	 * compiler error
470 	 */
471 	ENA_TOUCH(ena_stats_global_strings);
472 	ENA_TOUCH(ena_stats_tx_strings);
473 	ENA_TOUCH(ena_stats_rx_strings);
474 	ENA_TOUCH(ena_stats_ena_com_strings);
475 
476 	return  dev->data->nb_tx_queues *
477 		(ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
478 		ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
479 }
480 
481 static void ena_config_debug_area(struct ena_adapter *adapter)
482 {
483 	u32 debug_area_size;
484 	int rc, ss_count;
485 
486 	ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
487 	if (ss_count <= 0) {
488 		RTE_LOG(ERR, PMD, "SS count is negative\n");
489 		return;
490 	}
491 
492 	/* allocate 32 bytes for each string and 64bit for the value */
493 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
494 
495 	rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
496 	if (rc) {
497 		RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
498 		return;
499 	}
500 
501 	rc = ena_com_set_host_attributes(&adapter->ena_dev);
502 	if (rc) {
503 		if (rc == -ENA_COM_UNSUPPORTED)
504 			RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
505 		else
506 			RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
507 
508 		goto err;
509 	}
510 
511 	return;
512 err:
513 	ena_com_delete_debug_area(&adapter->ena_dev);
514 }
515 
516 static void ena_close(struct rte_eth_dev *dev)
517 {
518 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
519 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
520 	struct ena_adapter *adapter =
521 		(struct ena_adapter *)(dev->data->dev_private);
522 
523 	if (adapter->state == ENA_ADAPTER_STATE_RUNNING)
524 		ena_stop(dev);
525 	adapter->state = ENA_ADAPTER_STATE_CLOSED;
526 
527 	ena_rx_queue_release_all(dev);
528 	ena_tx_queue_release_all(dev);
529 
530 	rte_free(adapter->drv_stats);
531 	adapter->drv_stats = NULL;
532 
533 	rte_intr_disable(intr_handle);
534 	rte_intr_callback_unregister(intr_handle,
535 				     ena_interrupt_handler_rte,
536 				     adapter);
537 
538 	/*
539 	 * MAC is not allocated dynamically. Setting NULL should prevent from
540 	 * release of the resource in the rte_eth_dev_release_port().
541 	 */
542 	dev->data->mac_addrs = NULL;
543 }
544 
545 static int
546 ena_dev_reset(struct rte_eth_dev *dev)
547 {
548 	int rc = 0;
549 
550 	ena_destroy_device(dev);
551 	rc = eth_ena_dev_init(dev);
552 	if (rc)
553 		PMD_INIT_LOG(CRIT, "Cannot initialize device\n");
554 
555 	return rc;
556 }
557 
558 static int ena_rss_reta_update(struct rte_eth_dev *dev,
559 			       struct rte_eth_rss_reta_entry64 *reta_conf,
560 			       uint16_t reta_size)
561 {
562 	struct ena_adapter *adapter =
563 		(struct ena_adapter *)(dev->data->dev_private);
564 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
565 	int rc, i;
566 	u16 entry_value;
567 	int conf_idx;
568 	int idx;
569 
570 	if ((reta_size == 0) || (reta_conf == NULL))
571 		return -EINVAL;
572 
573 	if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
574 		RTE_LOG(WARNING, PMD,
575 			"indirection table %d is bigger than supported (%d)\n",
576 			reta_size, ENA_RX_RSS_TABLE_SIZE);
577 		return -EINVAL;
578 	}
579 
580 	for (i = 0 ; i < reta_size ; i++) {
581 		/* each reta_conf is for 64 entries.
582 		 * to support 128 we use 2 conf of 64
583 		 */
584 		conf_idx = i / RTE_RETA_GROUP_SIZE;
585 		idx = i % RTE_RETA_GROUP_SIZE;
586 		if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
587 			entry_value =
588 				ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
589 
590 			rc = ena_com_indirect_table_fill_entry(ena_dev,
591 							       i,
592 							       entry_value);
593 			if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
594 				RTE_LOG(ERR, PMD,
595 					"Cannot fill indirect table\n");
596 				return rc;
597 			}
598 		}
599 	}
600 
601 	rc = ena_com_indirect_table_set(ena_dev);
602 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
603 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
604 		return rc;
605 	}
606 
607 	RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries  for port %d\n",
608 		__func__, reta_size, adapter->rte_dev->data->port_id);
609 
610 	return 0;
611 }
612 
613 /* Query redirection table. */
614 static int ena_rss_reta_query(struct rte_eth_dev *dev,
615 			      struct rte_eth_rss_reta_entry64 *reta_conf,
616 			      uint16_t reta_size)
617 {
618 	struct ena_adapter *adapter =
619 		(struct ena_adapter *)(dev->data->dev_private);
620 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
621 	int rc;
622 	int i;
623 	u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
624 	int reta_conf_idx;
625 	int reta_idx;
626 
627 	if (reta_size == 0 || reta_conf == NULL ||
628 	    (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
629 		return -EINVAL;
630 
631 	rc = ena_com_indirect_table_get(ena_dev, indirect_table);
632 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
633 		RTE_LOG(ERR, PMD, "cannot get indirect table\n");
634 		return -ENOTSUP;
635 	}
636 
637 	for (i = 0 ; i < reta_size ; i++) {
638 		reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
639 		reta_idx = i % RTE_RETA_GROUP_SIZE;
640 		if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
641 			reta_conf[reta_conf_idx].reta[reta_idx] =
642 				ENA_IO_RXQ_IDX_REV(indirect_table[i]);
643 	}
644 
645 	return 0;
646 }
647 
648 static int ena_rss_init_default(struct ena_adapter *adapter)
649 {
650 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
651 	uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
652 	int rc, i;
653 	u32 val;
654 
655 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
656 	if (unlikely(rc)) {
657 		RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
658 		goto err_rss_init;
659 	}
660 
661 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
662 		val = i % nb_rx_queues;
663 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
664 						       ENA_IO_RXQ_IDX(val));
665 		if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
666 			RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
667 			goto err_fill_indir;
668 		}
669 	}
670 
671 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
672 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
673 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
674 		RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
675 		goto err_fill_indir;
676 	}
677 
678 	rc = ena_com_set_default_hash_ctrl(ena_dev);
679 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
680 		RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
681 		goto err_fill_indir;
682 	}
683 
684 	rc = ena_com_indirect_table_set(ena_dev);
685 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
686 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
687 		goto err_fill_indir;
688 	}
689 	RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
690 		adapter->rte_dev->data->port_id);
691 
692 	return 0;
693 
694 err_fill_indir:
695 	ena_com_rss_destroy(ena_dev);
696 err_rss_init:
697 
698 	return rc;
699 }
700 
701 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
702 {
703 	struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
704 	int nb_queues = dev->data->nb_rx_queues;
705 	int i;
706 
707 	for (i = 0; i < nb_queues; i++)
708 		ena_rx_queue_release(queues[i]);
709 }
710 
711 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
712 {
713 	struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
714 	int nb_queues = dev->data->nb_tx_queues;
715 	int i;
716 
717 	for (i = 0; i < nb_queues; i++)
718 		ena_tx_queue_release(queues[i]);
719 }
720 
721 static void ena_rx_queue_release(void *queue)
722 {
723 	struct ena_ring *ring = (struct ena_ring *)queue;
724 
725 	/* Free ring resources */
726 	if (ring->rx_buffer_info)
727 		rte_free(ring->rx_buffer_info);
728 	ring->rx_buffer_info = NULL;
729 
730 	if (ring->rx_refill_buffer)
731 		rte_free(ring->rx_refill_buffer);
732 	ring->rx_refill_buffer = NULL;
733 
734 	if (ring->empty_rx_reqs)
735 		rte_free(ring->empty_rx_reqs);
736 	ring->empty_rx_reqs = NULL;
737 
738 	ring->configured = 0;
739 
740 	RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
741 		ring->port_id, ring->id);
742 }
743 
744 static void ena_tx_queue_release(void *queue)
745 {
746 	struct ena_ring *ring = (struct ena_ring *)queue;
747 
748 	/* Free ring resources */
749 	if (ring->tx_buffer_info)
750 		rte_free(ring->tx_buffer_info);
751 
752 	if (ring->empty_tx_reqs)
753 		rte_free(ring->empty_tx_reqs);
754 
755 	ring->empty_tx_reqs = NULL;
756 	ring->tx_buffer_info = NULL;
757 
758 	ring->configured = 0;
759 
760 	RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
761 		ring->port_id, ring->id);
762 }
763 
764 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
765 {
766 	unsigned int i;
767 
768 	for (i = 0; i < ring->ring_size; ++i)
769 		if (ring->rx_buffer_info[i]) {
770 			rte_mbuf_raw_free(ring->rx_buffer_info[i]);
771 			ring->rx_buffer_info[i] = NULL;
772 		}
773 }
774 
775 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
776 {
777 	unsigned int i;
778 
779 	for (i = 0; i < ring->ring_size; ++i) {
780 		struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
781 
782 		if (tx_buf->mbuf)
783 			rte_pktmbuf_free(tx_buf->mbuf);
784 	}
785 }
786 
787 static int ena_link_update(struct rte_eth_dev *dev,
788 			   __rte_unused int wait_to_complete)
789 {
790 	struct rte_eth_link *link = &dev->data->dev_link;
791 	struct ena_adapter *adapter;
792 
793 	adapter = (struct ena_adapter *)(dev->data->dev_private);
794 
795 	link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
796 	link->link_speed = ETH_SPEED_NUM_NONE;
797 	link->link_duplex = ETH_LINK_FULL_DUPLEX;
798 
799 	return 0;
800 }
801 
802 static int ena_queue_start_all(struct rte_eth_dev *dev,
803 			       enum ena_ring_type ring_type)
804 {
805 	struct ena_adapter *adapter =
806 		(struct ena_adapter *)(dev->data->dev_private);
807 	struct ena_ring *queues = NULL;
808 	int nb_queues;
809 	int i = 0;
810 	int rc = 0;
811 
812 	if (ring_type == ENA_RING_TYPE_RX) {
813 		queues = adapter->rx_ring;
814 		nb_queues = dev->data->nb_rx_queues;
815 	} else {
816 		queues = adapter->tx_ring;
817 		nb_queues = dev->data->nb_tx_queues;
818 	}
819 	for (i = 0; i < nb_queues; i++) {
820 		if (queues[i].configured) {
821 			if (ring_type == ENA_RING_TYPE_RX) {
822 				ena_assert_msg(
823 					dev->data->rx_queues[i] == &queues[i],
824 					"Inconsistent state of rx queues\n");
825 			} else {
826 				ena_assert_msg(
827 					dev->data->tx_queues[i] == &queues[i],
828 					"Inconsistent state of tx queues\n");
829 			}
830 
831 			rc = ena_queue_start(&queues[i]);
832 
833 			if (rc) {
834 				PMD_INIT_LOG(ERR,
835 					     "failed to start queue %d type(%d)",
836 					     i, ring_type);
837 				goto err;
838 			}
839 		}
840 	}
841 
842 	return 0;
843 
844 err:
845 	while (i--)
846 		if (queues[i].configured)
847 			ena_queue_stop(&queues[i]);
848 
849 	return rc;
850 }
851 
852 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
853 {
854 	uint32_t max_frame_len = adapter->max_mtu;
855 
856 	if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
857 	    DEV_RX_OFFLOAD_JUMBO_FRAME)
858 		max_frame_len =
859 			adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
860 
861 	return max_frame_len;
862 }
863 
864 static int ena_check_valid_conf(struct ena_adapter *adapter)
865 {
866 	uint32_t max_frame_len = ena_get_mtu_conf(adapter);
867 
868 	if (max_frame_len > adapter->max_mtu || max_frame_len < ENA_MIN_MTU) {
869 		PMD_INIT_LOG(ERR, "Unsupported MTU of %d. "
870 				  "max mtu: %d, min mtu: %d\n",
871 			     max_frame_len, adapter->max_mtu, ENA_MIN_MTU);
872 		return ENA_COM_UNSUPPORTED;
873 	}
874 
875 	return 0;
876 }
877 
878 static int
879 ena_calc_queue_size(struct ena_com_dev *ena_dev,
880 		    u16 *max_tx_sgl_size,
881 		    struct ena_com_dev_get_features_ctx *get_feat_ctx)
882 {
883 	uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
884 
885 	queue_size = RTE_MIN(queue_size,
886 			     get_feat_ctx->max_queues.max_cq_depth);
887 	queue_size = RTE_MIN(queue_size,
888 			     get_feat_ctx->max_queues.max_sq_depth);
889 
890 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
891 		queue_size = RTE_MIN(queue_size,
892 				     get_feat_ctx->max_queues.max_llq_depth);
893 
894 	/* Round down to power of 2 */
895 	if (!rte_is_power_of_2(queue_size))
896 		queue_size = rte_align32pow2(queue_size >> 1);
897 
898 	if (unlikely(queue_size == 0)) {
899 		PMD_INIT_LOG(ERR, "Invalid queue size");
900 		return -EFAULT;
901 	}
902 
903 	*max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
904 		get_feat_ctx->max_queues.max_packet_tx_descs);
905 
906 	return queue_size;
907 }
908 
909 static void ena_stats_restart(struct rte_eth_dev *dev)
910 {
911 	struct ena_adapter *adapter =
912 		(struct ena_adapter *)(dev->data->dev_private);
913 
914 	rte_atomic64_init(&adapter->drv_stats->ierrors);
915 	rte_atomic64_init(&adapter->drv_stats->oerrors);
916 	rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
917 }
918 
919 static int ena_stats_get(struct rte_eth_dev *dev,
920 			  struct rte_eth_stats *stats)
921 {
922 	struct ena_admin_basic_stats ena_stats;
923 	struct ena_adapter *adapter =
924 		(struct ena_adapter *)(dev->data->dev_private);
925 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
926 	int rc;
927 
928 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
929 		return -ENOTSUP;
930 
931 	memset(&ena_stats, 0, sizeof(ena_stats));
932 	rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
933 	if (unlikely(rc)) {
934 		RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
935 		return rc;
936 	}
937 
938 	/* Set of basic statistics from ENA */
939 	stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
940 					  ena_stats.rx_pkts_low);
941 	stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
942 					  ena_stats.tx_pkts_low);
943 	stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
944 					ena_stats.rx_bytes_low);
945 	stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
946 					ena_stats.tx_bytes_low);
947 	stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
948 					 ena_stats.rx_drops_low);
949 
950 	/* Driver related stats */
951 	stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
952 	stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
953 	stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
954 	return 0;
955 }
956 
957 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
958 {
959 	struct ena_adapter *adapter;
960 	struct ena_com_dev *ena_dev;
961 	int rc = 0;
962 
963 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
964 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
965 	adapter = (struct ena_adapter *)(dev->data->dev_private);
966 
967 	ena_dev = &adapter->ena_dev;
968 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
969 
970 	if (mtu > ena_get_mtu_conf(adapter) || mtu < ENA_MIN_MTU) {
971 		RTE_LOG(ERR, PMD,
972 			"Invalid MTU setting. new_mtu: %d "
973 			"max mtu: %d min mtu: %d\n",
974 			mtu, ena_get_mtu_conf(adapter), ENA_MIN_MTU);
975 		return -EINVAL;
976 	}
977 
978 	rc = ena_com_set_dev_mtu(ena_dev, mtu);
979 	if (rc)
980 		RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
981 	else
982 		RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
983 
984 	return rc;
985 }
986 
987 static int ena_start(struct rte_eth_dev *dev)
988 {
989 	struct ena_adapter *adapter =
990 		(struct ena_adapter *)(dev->data->dev_private);
991 	uint64_t ticks;
992 	int rc = 0;
993 
994 	rc = ena_check_valid_conf(adapter);
995 	if (rc)
996 		return rc;
997 
998 	rc = ena_queue_start_all(dev, ENA_RING_TYPE_RX);
999 	if (rc)
1000 		return rc;
1001 
1002 	rc = ena_queue_start_all(dev, ENA_RING_TYPE_TX);
1003 	if (rc)
1004 		goto err_start_tx;
1005 
1006 	if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
1007 	    ETH_MQ_RX_RSS_FLAG && adapter->rte_dev->data->nb_rx_queues > 0) {
1008 		rc = ena_rss_init_default(adapter);
1009 		if (rc)
1010 			goto err_rss_init;
1011 	}
1012 
1013 	ena_stats_restart(dev);
1014 
1015 	adapter->timestamp_wd = rte_get_timer_cycles();
1016 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
1017 
1018 	ticks = rte_get_timer_hz();
1019 	rte_timer_reset(&adapter->timer_wd, ticks, PERIODICAL, rte_lcore_id(),
1020 			ena_timer_wd_callback, adapter);
1021 
1022 	adapter->state = ENA_ADAPTER_STATE_RUNNING;
1023 
1024 	return 0;
1025 
1026 err_rss_init:
1027 	ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1028 err_start_tx:
1029 	ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1030 	return rc;
1031 }
1032 
1033 static void ena_stop(struct rte_eth_dev *dev)
1034 {
1035 	struct ena_adapter *adapter =
1036 		(struct ena_adapter *)(dev->data->dev_private);
1037 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1038 	int rc;
1039 
1040 	rte_timer_stop_sync(&adapter->timer_wd);
1041 	ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1042 	ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1043 
1044 	if (adapter->trigger_reset) {
1045 		rc = ena_com_dev_reset(ena_dev, adapter->reset_reason);
1046 		if (rc)
1047 			RTE_LOG(ERR, PMD, "Device reset failed rc=%d\n", rc);
1048 	}
1049 
1050 	adapter->state = ENA_ADAPTER_STATE_STOPPED;
1051 }
1052 
1053 static int ena_create_io_queue(struct ena_ring *ring)
1054 {
1055 	struct ena_adapter *adapter;
1056 	struct ena_com_dev *ena_dev;
1057 	struct ena_com_create_io_ctx ctx =
1058 		/* policy set to _HOST just to satisfy icc compiler */
1059 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
1060 		  0, 0, 0, 0, 0 };
1061 	uint16_t ena_qid;
1062 	unsigned int i;
1063 	int rc;
1064 
1065 	adapter = ring->adapter;
1066 	ena_dev = &adapter->ena_dev;
1067 
1068 	if (ring->type == ENA_RING_TYPE_TX) {
1069 		ena_qid = ENA_IO_TXQ_IDX(ring->id);
1070 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1071 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1072 		ctx.queue_size = adapter->tx_ring_size;
1073 		for (i = 0; i < ring->ring_size; i++)
1074 			ring->empty_tx_reqs[i] = i;
1075 	} else {
1076 		ena_qid = ENA_IO_RXQ_IDX(ring->id);
1077 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1078 		ctx.queue_size = adapter->rx_ring_size;
1079 		for (i = 0; i < ring->ring_size; i++)
1080 			ring->empty_rx_reqs[i] = i;
1081 	}
1082 	ctx.qid = ena_qid;
1083 	ctx.msix_vector = -1; /* interrupts not used */
1084 	ctx.numa_node = ena_cpu_to_node(ring->id);
1085 
1086 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1087 	if (rc) {
1088 		RTE_LOG(ERR, PMD,
1089 			"failed to create io queue #%d (qid:%d) rc: %d\n",
1090 			ring->id, ena_qid, rc);
1091 		return rc;
1092 	}
1093 
1094 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1095 				     &ring->ena_com_io_sq,
1096 				     &ring->ena_com_io_cq);
1097 	if (rc) {
1098 		RTE_LOG(ERR, PMD,
1099 			"Failed to get io queue handlers. queue num %d rc: %d\n",
1100 			ring->id, rc);
1101 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1102 		return rc;
1103 	}
1104 
1105 	if (ring->type == ENA_RING_TYPE_TX)
1106 		ena_com_update_numa_node(ring->ena_com_io_cq, ctx.numa_node);
1107 
1108 	return 0;
1109 }
1110 
1111 static void ena_queue_stop(struct ena_ring *ring)
1112 {
1113 	struct ena_com_dev *ena_dev = &ring->adapter->ena_dev;
1114 
1115 	if (ring->type == ENA_RING_TYPE_RX) {
1116 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(ring->id));
1117 		ena_rx_queue_release_bufs(ring);
1118 	} else {
1119 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(ring->id));
1120 		ena_tx_queue_release_bufs(ring);
1121 	}
1122 }
1123 
1124 static void ena_queue_stop_all(struct rte_eth_dev *dev,
1125 			      enum ena_ring_type ring_type)
1126 {
1127 	struct ena_adapter *adapter =
1128 		(struct ena_adapter *)(dev->data->dev_private);
1129 	struct ena_ring *queues = NULL;
1130 	uint16_t nb_queues, i;
1131 
1132 	if (ring_type == ENA_RING_TYPE_RX) {
1133 		queues = adapter->rx_ring;
1134 		nb_queues = dev->data->nb_rx_queues;
1135 	} else {
1136 		queues = adapter->tx_ring;
1137 		nb_queues = dev->data->nb_tx_queues;
1138 	}
1139 
1140 	for (i = 0; i < nb_queues; ++i)
1141 		if (queues[i].configured)
1142 			ena_queue_stop(&queues[i]);
1143 }
1144 
1145 static int ena_queue_start(struct ena_ring *ring)
1146 {
1147 	int rc, bufs_num;
1148 
1149 	ena_assert_msg(ring->configured == 1,
1150 		       "Trying to start unconfigured queue\n");
1151 
1152 	rc = ena_create_io_queue(ring);
1153 	if (rc) {
1154 		PMD_INIT_LOG(ERR, "Failed to create IO queue!\n");
1155 		return rc;
1156 	}
1157 
1158 	ring->next_to_clean = 0;
1159 	ring->next_to_use = 0;
1160 
1161 	if (ring->type == ENA_RING_TYPE_TX)
1162 		return 0;
1163 
1164 	bufs_num = ring->ring_size - 1;
1165 	rc = ena_populate_rx_queue(ring, bufs_num);
1166 	if (rc != bufs_num) {
1167 		ena_com_destroy_io_queue(&ring->adapter->ena_dev,
1168 					 ENA_IO_RXQ_IDX(ring->id));
1169 		PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
1170 		return ENA_COM_FAULT;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
1177 			      uint16_t queue_idx,
1178 			      uint16_t nb_desc,
1179 			      __rte_unused unsigned int socket_id,
1180 			      const struct rte_eth_txconf *tx_conf)
1181 {
1182 	struct ena_ring *txq = NULL;
1183 	struct ena_adapter *adapter =
1184 		(struct ena_adapter *)(dev->data->dev_private);
1185 	unsigned int i;
1186 
1187 	txq = &adapter->tx_ring[queue_idx];
1188 
1189 	if (txq->configured) {
1190 		RTE_LOG(CRIT, PMD,
1191 			"API violation. Queue %d is already configured\n",
1192 			queue_idx);
1193 		return ENA_COM_FAULT;
1194 	}
1195 
1196 	if (!rte_is_power_of_2(nb_desc)) {
1197 		RTE_LOG(ERR, PMD,
1198 			"Unsupported size of TX queue: %d is not a power of 2.",
1199 			nb_desc);
1200 		return -EINVAL;
1201 	}
1202 
1203 	if (nb_desc > adapter->tx_ring_size) {
1204 		RTE_LOG(ERR, PMD,
1205 			"Unsupported size of TX queue (max size: %d)\n",
1206 			adapter->tx_ring_size);
1207 		return -EINVAL;
1208 	}
1209 
1210 	txq->port_id = dev->data->port_id;
1211 	txq->next_to_clean = 0;
1212 	txq->next_to_use = 0;
1213 	txq->ring_size = nb_desc;
1214 
1215 	txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1216 					  sizeof(struct ena_tx_buffer) *
1217 					  txq->ring_size,
1218 					  RTE_CACHE_LINE_SIZE);
1219 	if (!txq->tx_buffer_info) {
1220 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1221 		return -ENOMEM;
1222 	}
1223 
1224 	txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1225 					 sizeof(u16) * txq->ring_size,
1226 					 RTE_CACHE_LINE_SIZE);
1227 	if (!txq->empty_tx_reqs) {
1228 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1229 		rte_free(txq->tx_buffer_info);
1230 		return -ENOMEM;
1231 	}
1232 
1233 	for (i = 0; i < txq->ring_size; i++)
1234 		txq->empty_tx_reqs[i] = i;
1235 
1236 	if (tx_conf != NULL) {
1237 		txq->offloads =
1238 			tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1239 	}
1240 
1241 	/* Store pointer to this queue in upper layer */
1242 	txq->configured = 1;
1243 	dev->data->tx_queues[queue_idx] = txq;
1244 
1245 	return 0;
1246 }
1247 
1248 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1249 			      uint16_t queue_idx,
1250 			      uint16_t nb_desc,
1251 			      __rte_unused unsigned int socket_id,
1252 			      __rte_unused const struct rte_eth_rxconf *rx_conf,
1253 			      struct rte_mempool *mp)
1254 {
1255 	struct ena_adapter *adapter =
1256 		(struct ena_adapter *)(dev->data->dev_private);
1257 	struct ena_ring *rxq = NULL;
1258 	int i;
1259 
1260 	rxq = &adapter->rx_ring[queue_idx];
1261 	if (rxq->configured) {
1262 		RTE_LOG(CRIT, PMD,
1263 			"API violation. Queue %d is already configured\n",
1264 			queue_idx);
1265 		return ENA_COM_FAULT;
1266 	}
1267 
1268 	if (!rte_is_power_of_2(nb_desc)) {
1269 		RTE_LOG(ERR, PMD,
1270 			"Unsupported size of RX queue: %d is not a power of 2.",
1271 			nb_desc);
1272 		return -EINVAL;
1273 	}
1274 
1275 	if (nb_desc > adapter->rx_ring_size) {
1276 		RTE_LOG(ERR, PMD,
1277 			"Unsupported size of RX queue (max size: %d)\n",
1278 			adapter->rx_ring_size);
1279 		return -EINVAL;
1280 	}
1281 
1282 	rxq->port_id = dev->data->port_id;
1283 	rxq->next_to_clean = 0;
1284 	rxq->next_to_use = 0;
1285 	rxq->ring_size = nb_desc;
1286 	rxq->mb_pool = mp;
1287 
1288 	rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1289 					  sizeof(struct rte_mbuf *) * nb_desc,
1290 					  RTE_CACHE_LINE_SIZE);
1291 	if (!rxq->rx_buffer_info) {
1292 		RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1293 		return -ENOMEM;
1294 	}
1295 
1296 	rxq->rx_refill_buffer = rte_zmalloc("rxq->rx_refill_buffer",
1297 					    sizeof(struct rte_mbuf *) * nb_desc,
1298 					    RTE_CACHE_LINE_SIZE);
1299 
1300 	if (!rxq->rx_refill_buffer) {
1301 		RTE_LOG(ERR, PMD, "failed to alloc mem for rx refill buffer\n");
1302 		rte_free(rxq->rx_buffer_info);
1303 		rxq->rx_buffer_info = NULL;
1304 		return -ENOMEM;
1305 	}
1306 
1307 	rxq->empty_rx_reqs = rte_zmalloc("rxq->empty_rx_reqs",
1308 					 sizeof(uint16_t) * nb_desc,
1309 					 RTE_CACHE_LINE_SIZE);
1310 	if (!rxq->empty_rx_reqs) {
1311 		RTE_LOG(ERR, PMD, "failed to alloc mem for empty rx reqs\n");
1312 		rte_free(rxq->rx_buffer_info);
1313 		rxq->rx_buffer_info = NULL;
1314 		rte_free(rxq->rx_refill_buffer);
1315 		rxq->rx_refill_buffer = NULL;
1316 		return -ENOMEM;
1317 	}
1318 
1319 	for (i = 0; i < nb_desc; i++)
1320 		rxq->empty_rx_reqs[i] = i;
1321 
1322 	/* Store pointer to this queue in upper layer */
1323 	rxq->configured = 1;
1324 	dev->data->rx_queues[queue_idx] = rxq;
1325 
1326 	return 0;
1327 }
1328 
1329 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1330 {
1331 	unsigned int i;
1332 	int rc;
1333 	uint16_t ring_size = rxq->ring_size;
1334 	uint16_t ring_mask = ring_size - 1;
1335 	uint16_t next_to_use = rxq->next_to_use;
1336 	uint16_t in_use, req_id;
1337 	struct rte_mbuf **mbufs = rxq->rx_refill_buffer;
1338 
1339 	if (unlikely(!count))
1340 		return 0;
1341 
1342 	in_use = rxq->next_to_use - rxq->next_to_clean;
1343 	ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1344 
1345 	/* get resources for incoming packets */
1346 	rc = rte_mempool_get_bulk(rxq->mb_pool, (void **)mbufs, count);
1347 	if (unlikely(rc < 0)) {
1348 		rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1349 		PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1350 		return 0;
1351 	}
1352 
1353 	for (i = 0; i < count; i++) {
1354 		uint16_t next_to_use_masked = next_to_use & ring_mask;
1355 		struct rte_mbuf *mbuf = mbufs[i];
1356 		struct ena_com_buf ebuf;
1357 
1358 		if (likely((i + 4) < count))
1359 			rte_prefetch0(mbufs[i + 4]);
1360 
1361 		req_id = rxq->empty_rx_reqs[next_to_use_masked];
1362 		rc = validate_rx_req_id(rxq, req_id);
1363 		if (unlikely(rc < 0))
1364 			break;
1365 		rxq->rx_buffer_info[req_id] = mbuf;
1366 
1367 		/* prepare physical address for DMA transaction */
1368 		ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1369 		ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1370 		/* pass resource to device */
1371 		rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1372 						&ebuf, req_id);
1373 		if (unlikely(rc)) {
1374 			RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1375 			rxq->rx_buffer_info[req_id] = NULL;
1376 			break;
1377 		}
1378 		next_to_use++;
1379 	}
1380 
1381 	if (unlikely(i < count)) {
1382 		RTE_LOG(WARNING, PMD, "refilled rx qid %d with only %d "
1383 			"buffers (from %d)\n", rxq->id, i, count);
1384 		rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbufs[i]),
1385 				     count - i);
1386 	}
1387 
1388 	/* When we submitted free recources to device... */
1389 	if (likely(i > 0)) {
1390 		/* ...let HW know that it can fill buffers with data
1391 		 *
1392 		 * Add memory barrier to make sure the desc were written before
1393 		 * issue a doorbell
1394 		 */
1395 		rte_wmb();
1396 		ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1397 
1398 		rxq->next_to_use = next_to_use;
1399 	}
1400 
1401 	return i;
1402 }
1403 
1404 static int ena_device_init(struct ena_com_dev *ena_dev,
1405 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
1406 			   bool *wd_state)
1407 {
1408 	uint32_t aenq_groups;
1409 	int rc;
1410 	bool readless_supported;
1411 
1412 	/* Initialize mmio registers */
1413 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
1414 	if (rc) {
1415 		RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1416 		return rc;
1417 	}
1418 
1419 	/* The PCIe configuration space revision id indicate if mmio reg
1420 	 * read is disabled.
1421 	 */
1422 	readless_supported =
1423 		!(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1424 			       & ENA_MMIO_DISABLE_REG_READ);
1425 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1426 
1427 	/* reset device */
1428 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1429 	if (rc) {
1430 		RTE_LOG(ERR, PMD, "cannot reset device\n");
1431 		goto err_mmio_read_less;
1432 	}
1433 
1434 	/* check FW version */
1435 	rc = ena_com_validate_version(ena_dev);
1436 	if (rc) {
1437 		RTE_LOG(ERR, PMD, "device version is too low\n");
1438 		goto err_mmio_read_less;
1439 	}
1440 
1441 	ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1442 
1443 	/* ENA device administration layer init */
1444 	rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
1445 	if (rc) {
1446 		RTE_LOG(ERR, PMD,
1447 			"cannot initialize ena admin queue with device\n");
1448 		goto err_mmio_read_less;
1449 	}
1450 
1451 	/* To enable the msix interrupts the driver needs to know the number
1452 	 * of queues. So the driver uses polling mode to retrieve this
1453 	 * information.
1454 	 */
1455 	ena_com_set_admin_polling_mode(ena_dev, true);
1456 
1457 	ena_config_host_info(ena_dev);
1458 
1459 	/* Get Device Attributes and features */
1460 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1461 	if (rc) {
1462 		RTE_LOG(ERR, PMD,
1463 			"cannot get attribute for ena device rc= %d\n", rc);
1464 		goto err_admin_init;
1465 	}
1466 
1467 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1468 		      BIT(ENA_ADMIN_NOTIFICATION) |
1469 		      BIT(ENA_ADMIN_KEEP_ALIVE) |
1470 		      BIT(ENA_ADMIN_FATAL_ERROR) |
1471 		      BIT(ENA_ADMIN_WARNING);
1472 
1473 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
1474 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1475 	if (rc) {
1476 		RTE_LOG(ERR, PMD, "Cannot configure aenq groups rc: %d\n", rc);
1477 		goto err_admin_init;
1478 	}
1479 
1480 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
1481 
1482 	return 0;
1483 
1484 err_admin_init:
1485 	ena_com_admin_destroy(ena_dev);
1486 
1487 err_mmio_read_less:
1488 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1489 
1490 	return rc;
1491 }
1492 
1493 static void ena_interrupt_handler_rte(void *cb_arg)
1494 {
1495 	struct ena_adapter *adapter = (struct ena_adapter *)cb_arg;
1496 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1497 
1498 	ena_com_admin_q_comp_intr_handler(ena_dev);
1499 	if (likely(adapter->state != ENA_ADAPTER_STATE_CLOSED))
1500 		ena_com_aenq_intr_handler(ena_dev, adapter);
1501 }
1502 
1503 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
1504 {
1505 	if (!adapter->wd_state)
1506 		return;
1507 
1508 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
1509 		return;
1510 
1511 	if (unlikely((rte_get_timer_cycles() - adapter->timestamp_wd) >=
1512 	    adapter->keep_alive_timeout)) {
1513 		RTE_LOG(ERR, PMD, "Keep alive timeout\n");
1514 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
1515 		adapter->trigger_reset = true;
1516 	}
1517 }
1518 
1519 /* Check if admin queue is enabled */
1520 static void check_for_admin_com_state(struct ena_adapter *adapter)
1521 {
1522 	if (unlikely(!ena_com_get_admin_running_state(&adapter->ena_dev))) {
1523 		RTE_LOG(ERR, PMD, "ENA admin queue is not in running state!\n");
1524 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
1525 		adapter->trigger_reset = true;
1526 	}
1527 }
1528 
1529 static void ena_timer_wd_callback(__rte_unused struct rte_timer *timer,
1530 				  void *arg)
1531 {
1532 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1533 	struct rte_eth_dev *dev = adapter->rte_dev;
1534 
1535 	check_for_missing_keep_alive(adapter);
1536 	check_for_admin_com_state(adapter);
1537 
1538 	if (unlikely(adapter->trigger_reset)) {
1539 		RTE_LOG(ERR, PMD, "Trigger reset is on\n");
1540 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1541 			NULL);
1542 	}
1543 }
1544 
1545 static int ena_calc_io_queue_num(__rte_unused struct ena_com_dev *ena_dev,
1546 				 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1547 {
1548 	int io_sq_num, io_cq_num, io_queue_num;
1549 
1550 	io_sq_num = get_feat_ctx->max_queues.max_sq_num;
1551 	io_cq_num = get_feat_ctx->max_queues.max_cq_num;
1552 
1553 	io_queue_num = RTE_MIN(io_sq_num, io_cq_num);
1554 
1555 	if (unlikely(io_queue_num == 0)) {
1556 		RTE_LOG(ERR, PMD, "Number of IO queues should not be 0\n");
1557 		return -EFAULT;
1558 	}
1559 
1560 	return io_queue_num;
1561 }
1562 
1563 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1564 {
1565 	struct rte_pci_device *pci_dev;
1566 	struct rte_intr_handle *intr_handle;
1567 	struct ena_adapter *adapter =
1568 		(struct ena_adapter *)(eth_dev->data->dev_private);
1569 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1570 	struct ena_com_dev_get_features_ctx get_feat_ctx;
1571 	int queue_size, rc;
1572 	u16 tx_sgl_size = 0;
1573 
1574 	static int adapters_found;
1575 	bool wd_state;
1576 
1577 	eth_dev->dev_ops = &ena_dev_ops;
1578 	eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1579 	eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1580 	eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1581 
1582 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1583 		return 0;
1584 
1585 	memset(adapter, 0, sizeof(struct ena_adapter));
1586 	ena_dev = &adapter->ena_dev;
1587 
1588 	adapter->rte_eth_dev_data = eth_dev->data;
1589 	adapter->rte_dev = eth_dev;
1590 
1591 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1592 	adapter->pdev = pci_dev;
1593 
1594 	PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1595 		     pci_dev->addr.domain,
1596 		     pci_dev->addr.bus,
1597 		     pci_dev->addr.devid,
1598 		     pci_dev->addr.function);
1599 
1600 	intr_handle = &pci_dev->intr_handle;
1601 
1602 	adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1603 	adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1604 
1605 	if (!adapter->regs) {
1606 		PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1607 			     ENA_REGS_BAR);
1608 		return -ENXIO;
1609 	}
1610 
1611 	ena_dev->reg_bar = adapter->regs;
1612 	ena_dev->dmadev = adapter->pdev;
1613 
1614 	adapter->id_number = adapters_found;
1615 
1616 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1617 		 adapter->id_number);
1618 
1619 	/* device specific initialization routine */
1620 	rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
1621 	if (rc) {
1622 		PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1623 		goto err;
1624 	}
1625 	adapter->wd_state = wd_state;
1626 
1627 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1628 	adapter->num_queues = ena_calc_io_queue_num(ena_dev,
1629 						    &get_feat_ctx);
1630 
1631 	queue_size = ena_calc_queue_size(ena_dev, &tx_sgl_size, &get_feat_ctx);
1632 	if (queue_size <= 0 || adapter->num_queues <= 0) {
1633 		rc = -EFAULT;
1634 		goto err_device_destroy;
1635 	}
1636 
1637 	adapter->tx_ring_size = queue_size;
1638 	adapter->rx_ring_size = queue_size;
1639 
1640 	adapter->max_tx_sgl_size = tx_sgl_size;
1641 
1642 	/* prepare ring structures */
1643 	ena_init_rings(adapter);
1644 
1645 	ena_config_debug_area(adapter);
1646 
1647 	/* Set max MTU for this device */
1648 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1649 
1650 	/* set device support for TSO */
1651 	adapter->tso4_supported = get_feat_ctx.offload.tx &
1652 				  ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1653 
1654 	/* Copy MAC address and point DPDK to it */
1655 	eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1656 	ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1657 			(struct ether_addr *)adapter->mac_addr);
1658 
1659 	/*
1660 	 * Pass the information to the rte_eth_dev_close() that it should also
1661 	 * release the private port resources.
1662 	 */
1663 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1664 
1665 	adapter->drv_stats = rte_zmalloc("adapter stats",
1666 					 sizeof(*adapter->drv_stats),
1667 					 RTE_CACHE_LINE_SIZE);
1668 	if (!adapter->drv_stats) {
1669 		RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1670 		rc = -ENOMEM;
1671 		goto err_delete_debug_area;
1672 	}
1673 
1674 	rte_intr_callback_register(intr_handle,
1675 				   ena_interrupt_handler_rte,
1676 				   adapter);
1677 	rte_intr_enable(intr_handle);
1678 	ena_com_set_admin_polling_mode(ena_dev, false);
1679 	ena_com_admin_aenq_enable(ena_dev);
1680 
1681 	if (adapters_found == 0)
1682 		rte_timer_subsystem_init();
1683 	rte_timer_init(&adapter->timer_wd);
1684 
1685 	adapters_found++;
1686 	adapter->state = ENA_ADAPTER_STATE_INIT;
1687 
1688 	return 0;
1689 
1690 err_delete_debug_area:
1691 	ena_com_delete_debug_area(ena_dev);
1692 
1693 err_device_destroy:
1694 	ena_com_delete_host_info(ena_dev);
1695 	ena_com_admin_destroy(ena_dev);
1696 
1697 err:
1698 	return rc;
1699 }
1700 
1701 static void ena_destroy_device(struct rte_eth_dev *eth_dev)
1702 {
1703 	struct ena_adapter *adapter =
1704 		(struct ena_adapter *)(eth_dev->data->dev_private);
1705 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1706 
1707 	if (adapter->state == ENA_ADAPTER_STATE_FREE)
1708 		return;
1709 
1710 	ena_com_set_admin_running_state(ena_dev, false);
1711 
1712 	if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1713 		ena_close(eth_dev);
1714 
1715 	ena_com_delete_debug_area(ena_dev);
1716 	ena_com_delete_host_info(ena_dev);
1717 
1718 	ena_com_abort_admin_commands(ena_dev);
1719 	ena_com_wait_for_abort_completion(ena_dev);
1720 	ena_com_admin_destroy(ena_dev);
1721 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1722 
1723 	adapter->state = ENA_ADAPTER_STATE_FREE;
1724 }
1725 
1726 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1727 {
1728 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1729 		return 0;
1730 
1731 	ena_destroy_device(eth_dev);
1732 
1733 	eth_dev->dev_ops = NULL;
1734 	eth_dev->rx_pkt_burst = NULL;
1735 	eth_dev->tx_pkt_burst = NULL;
1736 	eth_dev->tx_pkt_prepare = NULL;
1737 
1738 	return 0;
1739 }
1740 
1741 static int ena_dev_configure(struct rte_eth_dev *dev)
1742 {
1743 	struct ena_adapter *adapter =
1744 		(struct ena_adapter *)(dev->data->dev_private);
1745 
1746 	adapter->state = ENA_ADAPTER_STATE_CONFIG;
1747 
1748 	adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1749 	adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1750 	return 0;
1751 }
1752 
1753 static void ena_init_rings(struct ena_adapter *adapter)
1754 {
1755 	int i;
1756 
1757 	for (i = 0; i < adapter->num_queues; i++) {
1758 		struct ena_ring *ring = &adapter->tx_ring[i];
1759 
1760 		ring->configured = 0;
1761 		ring->type = ENA_RING_TYPE_TX;
1762 		ring->adapter = adapter;
1763 		ring->id = i;
1764 		ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1765 		ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1766 		ring->sgl_size = adapter->max_tx_sgl_size;
1767 	}
1768 
1769 	for (i = 0; i < adapter->num_queues; i++) {
1770 		struct ena_ring *ring = &adapter->rx_ring[i];
1771 
1772 		ring->configured = 0;
1773 		ring->type = ENA_RING_TYPE_RX;
1774 		ring->adapter = adapter;
1775 		ring->id = i;
1776 	}
1777 }
1778 
1779 static void ena_infos_get(struct rte_eth_dev *dev,
1780 			  struct rte_eth_dev_info *dev_info)
1781 {
1782 	struct ena_adapter *adapter;
1783 	struct ena_com_dev *ena_dev;
1784 	struct ena_com_dev_get_features_ctx feat;
1785 	uint64_t rx_feat = 0, tx_feat = 0;
1786 	int rc = 0;
1787 
1788 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
1789 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1790 	adapter = (struct ena_adapter *)(dev->data->dev_private);
1791 
1792 	ena_dev = &adapter->ena_dev;
1793 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1794 
1795 	dev_info->speed_capa =
1796 			ETH_LINK_SPEED_1G   |
1797 			ETH_LINK_SPEED_2_5G |
1798 			ETH_LINK_SPEED_5G   |
1799 			ETH_LINK_SPEED_10G  |
1800 			ETH_LINK_SPEED_25G  |
1801 			ETH_LINK_SPEED_40G  |
1802 			ETH_LINK_SPEED_50G  |
1803 			ETH_LINK_SPEED_100G;
1804 
1805 	/* Get supported features from HW */
1806 	rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1807 	if (unlikely(rc)) {
1808 		RTE_LOG(ERR, PMD,
1809 			"Cannot get attribute for ena device rc= %d\n", rc);
1810 		return;
1811 	}
1812 
1813 	/* Set Tx & Rx features available for device */
1814 	if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1815 		tx_feat	|= DEV_TX_OFFLOAD_TCP_TSO;
1816 
1817 	if (feat.offload.tx &
1818 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1819 		tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1820 			DEV_TX_OFFLOAD_UDP_CKSUM |
1821 			DEV_TX_OFFLOAD_TCP_CKSUM;
1822 
1823 	if (feat.offload.rx_supported &
1824 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1825 		rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1826 			DEV_RX_OFFLOAD_UDP_CKSUM  |
1827 			DEV_RX_OFFLOAD_TCP_CKSUM;
1828 
1829 	rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1830 
1831 	/* Inform framework about available features */
1832 	dev_info->rx_offload_capa = rx_feat;
1833 	dev_info->rx_queue_offload_capa = rx_feat;
1834 	dev_info->tx_offload_capa = tx_feat;
1835 	dev_info->tx_queue_offload_capa = tx_feat;
1836 
1837 	dev_info->flow_type_rss_offloads = ETH_RSS_IP | ETH_RSS_TCP |
1838 					   ETH_RSS_UDP;
1839 
1840 	dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1841 	dev_info->max_rx_pktlen  = adapter->max_mtu;
1842 	dev_info->max_mac_addrs = 1;
1843 
1844 	dev_info->max_rx_queues = adapter->num_queues;
1845 	dev_info->max_tx_queues = adapter->num_queues;
1846 	dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1847 
1848 	adapter->tx_supported_offloads = tx_feat;
1849 	adapter->rx_supported_offloads = rx_feat;
1850 
1851 	dev_info->rx_desc_lim.nb_max = ENA_MAX_RING_DESC;
1852 	dev_info->rx_desc_lim.nb_min = ENA_MIN_RING_DESC;
1853 
1854 	dev_info->tx_desc_lim.nb_max = ENA_MAX_RING_DESC;
1855 	dev_info->tx_desc_lim.nb_min = ENA_MIN_RING_DESC;
1856 	dev_info->tx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
1857 					feat.max_queues.max_packet_tx_descs);
1858 	dev_info->tx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
1859 					feat.max_queues.max_packet_tx_descs);
1860 }
1861 
1862 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1863 				  uint16_t nb_pkts)
1864 {
1865 	struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1866 	unsigned int ring_size = rx_ring->ring_size;
1867 	unsigned int ring_mask = ring_size - 1;
1868 	uint16_t next_to_clean = rx_ring->next_to_clean;
1869 	uint16_t desc_in_use = 0;
1870 	uint16_t req_id;
1871 	unsigned int recv_idx = 0;
1872 	struct rte_mbuf *mbuf = NULL;
1873 	struct rte_mbuf *mbuf_head = NULL;
1874 	struct rte_mbuf *mbuf_prev = NULL;
1875 	struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1876 	unsigned int completed;
1877 
1878 	struct ena_com_rx_ctx ena_rx_ctx;
1879 	int rc = 0;
1880 
1881 	/* Check adapter state */
1882 	if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1883 		RTE_LOG(ALERT, PMD,
1884 			"Trying to receive pkts while device is NOT running\n");
1885 		return 0;
1886 	}
1887 
1888 	desc_in_use = rx_ring->next_to_use - next_to_clean;
1889 	if (unlikely(nb_pkts > desc_in_use))
1890 		nb_pkts = desc_in_use;
1891 
1892 	for (completed = 0; completed < nb_pkts; completed++) {
1893 		int segments = 0;
1894 
1895 		ena_rx_ctx.max_bufs = rx_ring->ring_size;
1896 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1897 		ena_rx_ctx.descs = 0;
1898 		/* receive packet context */
1899 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1900 				    rx_ring->ena_com_io_sq,
1901 				    &ena_rx_ctx);
1902 		if (unlikely(rc)) {
1903 			RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1904 			rx_ring->adapter->reset_reason =
1905 				ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1906 			rx_ring->adapter->trigger_reset = true;
1907 			return 0;
1908 		}
1909 
1910 		if (unlikely(ena_rx_ctx.descs == 0))
1911 			break;
1912 
1913 		while (segments < ena_rx_ctx.descs) {
1914 			req_id = ena_rx_ctx.ena_bufs[segments].req_id;
1915 			rc = validate_rx_req_id(rx_ring, req_id);
1916 			if (unlikely(rc)) {
1917 				if (segments != 0)
1918 					rte_mbuf_raw_free(mbuf_head);
1919 				break;
1920 			}
1921 
1922 			mbuf = rx_buff_info[req_id];
1923 			rx_buff_info[req_id] = NULL;
1924 			mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1925 			mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1926 			mbuf->refcnt = 1;
1927 			mbuf->next = NULL;
1928 			if (unlikely(segments == 0)) {
1929 				mbuf->nb_segs = ena_rx_ctx.descs;
1930 				mbuf->port = rx_ring->port_id;
1931 				mbuf->pkt_len = 0;
1932 				mbuf_head = mbuf;
1933 			} else {
1934 				/* for multi-segment pkts create mbuf chain */
1935 				mbuf_prev->next = mbuf;
1936 			}
1937 			mbuf_head->pkt_len += mbuf->data_len;
1938 
1939 			mbuf_prev = mbuf;
1940 			rx_ring->empty_rx_reqs[next_to_clean & ring_mask] =
1941 				req_id;
1942 			segments++;
1943 			next_to_clean++;
1944 		}
1945 		if (unlikely(rc))
1946 			break;
1947 
1948 		/* fill mbuf attributes if any */
1949 		ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1950 		mbuf_head->hash.rss = ena_rx_ctx.hash;
1951 
1952 		/* pass to DPDK application head mbuf */
1953 		rx_pkts[recv_idx] = mbuf_head;
1954 		recv_idx++;
1955 	}
1956 
1957 	rx_ring->next_to_clean = next_to_clean;
1958 
1959 	desc_in_use = desc_in_use - completed + 1;
1960 	/* Burst refill to save doorbells, memory barriers, const interval */
1961 	if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size)) {
1962 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1963 		ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1964 	}
1965 
1966 	return recv_idx;
1967 }
1968 
1969 static uint16_t
1970 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1971 		uint16_t nb_pkts)
1972 {
1973 	int32_t ret;
1974 	uint32_t i;
1975 	struct rte_mbuf *m;
1976 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1977 	struct ipv4_hdr *ip_hdr;
1978 	uint64_t ol_flags;
1979 	uint16_t frag_field;
1980 
1981 	for (i = 0; i != nb_pkts; i++) {
1982 		m = tx_pkts[i];
1983 		ol_flags = m->ol_flags;
1984 
1985 		if (!(ol_flags & PKT_TX_IPV4))
1986 			continue;
1987 
1988 		/* If there was not L2 header length specified, assume it is
1989 		 * length of the ethernet header.
1990 		 */
1991 		if (unlikely(m->l2_len == 0))
1992 			m->l2_len = sizeof(struct ether_hdr);
1993 
1994 		ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1995 						 m->l2_len);
1996 		frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1997 
1998 		if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1999 			m->packet_type |= RTE_PTYPE_L4_NONFRAG;
2000 
2001 			/* If IPv4 header has DF flag enabled and TSO support is
2002 			 * disabled, partial chcecksum should not be calculated.
2003 			 */
2004 			if (!tx_ring->adapter->tso4_supported)
2005 				continue;
2006 		}
2007 
2008 		if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
2009 				(ol_flags & PKT_TX_L4_MASK) ==
2010 				PKT_TX_SCTP_CKSUM) {
2011 			rte_errno = ENOTSUP;
2012 			return i;
2013 		}
2014 
2015 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2016 		ret = rte_validate_tx_offload(m);
2017 		if (ret != 0) {
2018 			rte_errno = -ret;
2019 			return i;
2020 		}
2021 #endif
2022 
2023 		/* In case we are supposed to TSO and have DF not set (DF=0)
2024 		 * hardware must be provided with partial checksum, otherwise
2025 		 * it will take care of necessary calculations.
2026 		 */
2027 
2028 		ret = rte_net_intel_cksum_flags_prepare(m,
2029 			ol_flags & ~PKT_TX_TCP_SEG);
2030 		if (ret != 0) {
2031 			rte_errno = -ret;
2032 			return i;
2033 		}
2034 	}
2035 
2036 	return i;
2037 }
2038 
2039 static void ena_update_hints(struct ena_adapter *adapter,
2040 			     struct ena_admin_ena_hw_hints *hints)
2041 {
2042 	if (hints->admin_completion_tx_timeout)
2043 		adapter->ena_dev.admin_queue.completion_timeout =
2044 			hints->admin_completion_tx_timeout * 1000;
2045 
2046 	if (hints->mmio_read_timeout)
2047 		/* convert to usec */
2048 		adapter->ena_dev.mmio_read.reg_read_to =
2049 			hints->mmio_read_timeout * 1000;
2050 
2051 	if (hints->driver_watchdog_timeout) {
2052 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2053 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2054 		else
2055 			// Convert msecs to ticks
2056 			adapter->keep_alive_timeout =
2057 				(hints->driver_watchdog_timeout *
2058 				rte_get_timer_hz()) / 1000;
2059 	}
2060 }
2061 
2062 static int ena_check_and_linearize_mbuf(struct ena_ring *tx_ring,
2063 					struct rte_mbuf *mbuf)
2064 {
2065 	int num_segments, rc;
2066 
2067 	num_segments = mbuf->nb_segs;
2068 
2069 	if (likely(num_segments < tx_ring->sgl_size))
2070 		return 0;
2071 
2072 	rc = rte_pktmbuf_linearize(mbuf);
2073 	if (unlikely(rc))
2074 		RTE_LOG(WARNING, PMD, "Mbuf linearize failed\n");
2075 
2076 	return rc;
2077 }
2078 
2079 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2080 				  uint16_t nb_pkts)
2081 {
2082 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2083 	uint16_t next_to_use = tx_ring->next_to_use;
2084 	uint16_t next_to_clean = tx_ring->next_to_clean;
2085 	struct rte_mbuf *mbuf;
2086 	unsigned int ring_size = tx_ring->ring_size;
2087 	unsigned int ring_mask = ring_size - 1;
2088 	struct ena_com_tx_ctx ena_tx_ctx;
2089 	struct ena_tx_buffer *tx_info;
2090 	struct ena_com_buf *ebuf;
2091 	uint16_t rc, req_id, total_tx_descs = 0;
2092 	uint16_t sent_idx = 0, empty_tx_reqs;
2093 	int nb_hw_desc;
2094 
2095 	/* Check adapter state */
2096 	if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2097 		RTE_LOG(ALERT, PMD,
2098 			"Trying to xmit pkts while device is NOT running\n");
2099 		return 0;
2100 	}
2101 
2102 	empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
2103 	if (nb_pkts > empty_tx_reqs)
2104 		nb_pkts = empty_tx_reqs;
2105 
2106 	for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
2107 		mbuf = tx_pkts[sent_idx];
2108 
2109 		rc = ena_check_and_linearize_mbuf(tx_ring, mbuf);
2110 		if (unlikely(rc))
2111 			break;
2112 
2113 		req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
2114 		tx_info = &tx_ring->tx_buffer_info[req_id];
2115 		tx_info->mbuf = mbuf;
2116 		tx_info->num_of_bufs = 0;
2117 		ebuf = tx_info->bufs;
2118 
2119 		/* Prepare TX context */
2120 		memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2121 		memset(&ena_tx_ctx.ena_meta, 0x0,
2122 		       sizeof(struct ena_com_tx_meta));
2123 		ena_tx_ctx.ena_bufs = ebuf;
2124 		ena_tx_ctx.req_id = req_id;
2125 		if (tx_ring->tx_mem_queue_type ==
2126 				ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2127 			/* prepare the push buffer with
2128 			 * virtual address of the data
2129 			 */
2130 			ena_tx_ctx.header_len =
2131 				RTE_MIN(mbuf->data_len,
2132 					tx_ring->tx_max_header_size);
2133 			ena_tx_ctx.push_header =
2134 				(void *)((char *)mbuf->buf_addr +
2135 					 mbuf->data_off);
2136 		} /* there's no else as we take advantage of memset zeroing */
2137 
2138 		/* Set TX offloads flags, if applicable */
2139 		ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads);
2140 
2141 		if (unlikely(mbuf->ol_flags &
2142 			     (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
2143 			rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
2144 
2145 		rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
2146 
2147 		/* Process first segment taking into
2148 		 * consideration pushed header
2149 		 */
2150 		if (mbuf->data_len > ena_tx_ctx.header_len) {
2151 			ebuf->paddr = mbuf->buf_iova +
2152 				      mbuf->data_off +
2153 				      ena_tx_ctx.header_len;
2154 			ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
2155 			ebuf++;
2156 			tx_info->num_of_bufs++;
2157 		}
2158 
2159 		while ((mbuf = mbuf->next) != NULL) {
2160 			ebuf->paddr = mbuf->buf_iova + mbuf->data_off;
2161 			ebuf->len = mbuf->data_len;
2162 			ebuf++;
2163 			tx_info->num_of_bufs++;
2164 		}
2165 
2166 		ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2167 
2168 		/* Write data to device */
2169 		rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
2170 					&ena_tx_ctx, &nb_hw_desc);
2171 		if (unlikely(rc))
2172 			break;
2173 
2174 		tx_info->tx_descs = nb_hw_desc;
2175 
2176 		next_to_use++;
2177 	}
2178 
2179 	/* If there are ready packets to be xmitted... */
2180 	if (sent_idx > 0) {
2181 		/* ...let HW do its best :-) */
2182 		rte_wmb();
2183 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2184 
2185 		tx_ring->next_to_use = next_to_use;
2186 	}
2187 
2188 	/* Clear complete packets  */
2189 	while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
2190 		rc = validate_tx_req_id(tx_ring, req_id);
2191 		if (rc)
2192 			break;
2193 
2194 		/* Get Tx info & store how many descs were processed  */
2195 		tx_info = &tx_ring->tx_buffer_info[req_id];
2196 		total_tx_descs += tx_info->tx_descs;
2197 
2198 		/* Free whole mbuf chain  */
2199 		mbuf = tx_info->mbuf;
2200 		rte_pktmbuf_free(mbuf);
2201 		tx_info->mbuf = NULL;
2202 
2203 		/* Put back descriptor to the ring for reuse */
2204 		tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
2205 		next_to_clean++;
2206 
2207 		/* If too many descs to clean, leave it for another run */
2208 		if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
2209 			break;
2210 	}
2211 
2212 	if (total_tx_descs > 0) {
2213 		/* acknowledge completion of sent packets */
2214 		tx_ring->next_to_clean = next_to_clean;
2215 		ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
2216 		ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
2217 	}
2218 
2219 	return sent_idx;
2220 }
2221 
2222 /*********************************************************************
2223  *  PMD configuration
2224  *********************************************************************/
2225 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2226 	struct rte_pci_device *pci_dev)
2227 {
2228 	return rte_eth_dev_pci_generic_probe(pci_dev,
2229 		sizeof(struct ena_adapter), eth_ena_dev_init);
2230 }
2231 
2232 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
2233 {
2234 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
2235 }
2236 
2237 static struct rte_pci_driver rte_ena_pmd = {
2238 	.id_table = pci_id_ena_map,
2239 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
2240 		     RTE_PCI_DRV_WC_ACTIVATE,
2241 	.probe = eth_ena_pci_probe,
2242 	.remove = eth_ena_pci_remove,
2243 };
2244 
2245 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
2246 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
2247 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
2248 
2249 RTE_INIT(ena_init_log)
2250 {
2251 	ena_logtype_init = rte_log_register("pmd.net.ena.init");
2252 	if (ena_logtype_init >= 0)
2253 		rte_log_set_level(ena_logtype_init, RTE_LOG_NOTICE);
2254 	ena_logtype_driver = rte_log_register("pmd.net.ena.driver");
2255 	if (ena_logtype_driver >= 0)
2256 		rte_log_set_level(ena_logtype_driver, RTE_LOG_NOTICE);
2257 }
2258 
2259 /******************************************************************************
2260  ******************************** AENQ Handlers *******************************
2261  *****************************************************************************/
2262 static void ena_update_on_link_change(void *adapter_data,
2263 				      struct ena_admin_aenq_entry *aenq_e)
2264 {
2265 	struct rte_eth_dev *eth_dev;
2266 	struct ena_adapter *adapter;
2267 	struct ena_admin_aenq_link_change_desc *aenq_link_desc;
2268 	uint32_t status;
2269 
2270 	adapter = (struct ena_adapter *)adapter_data;
2271 	aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
2272 	eth_dev = adapter->rte_dev;
2273 
2274 	status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
2275 	adapter->link_status = status;
2276 
2277 	ena_link_update(eth_dev, 0);
2278 	_rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2279 }
2280 
2281 static void ena_notification(void *data,
2282 			     struct ena_admin_aenq_entry *aenq_e)
2283 {
2284 	struct ena_adapter *adapter = (struct ena_adapter *)data;
2285 	struct ena_admin_ena_hw_hints *hints;
2286 
2287 	if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
2288 		RTE_LOG(WARNING, PMD, "Invalid group(%x) expected %x\n",
2289 			aenq_e->aenq_common_desc.group,
2290 			ENA_ADMIN_NOTIFICATION);
2291 
2292 	switch (aenq_e->aenq_common_desc.syndrom) {
2293 	case ENA_ADMIN_UPDATE_HINTS:
2294 		hints = (struct ena_admin_ena_hw_hints *)
2295 			(&aenq_e->inline_data_w4);
2296 		ena_update_hints(adapter, hints);
2297 		break;
2298 	default:
2299 		RTE_LOG(ERR, PMD, "Invalid aenq notification link state %d\n",
2300 			aenq_e->aenq_common_desc.syndrom);
2301 	}
2302 }
2303 
2304 static void ena_keep_alive(void *adapter_data,
2305 			   __rte_unused struct ena_admin_aenq_entry *aenq_e)
2306 {
2307 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
2308 
2309 	adapter->timestamp_wd = rte_get_timer_cycles();
2310 }
2311 
2312 /**
2313  * This handler will called for unknown event group or unimplemented handlers
2314  **/
2315 static void unimplemented_aenq_handler(__rte_unused void *data,
2316 				       __rte_unused struct ena_admin_aenq_entry *aenq_e)
2317 {
2318 	RTE_LOG(ERR, PMD, "Unknown event was received or event with "
2319 			  "unimplemented handler\n");
2320 }
2321 
2322 static struct ena_aenq_handlers aenq_handlers = {
2323 	.handlers = {
2324 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
2325 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
2326 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive
2327 	},
2328 	.unimplemented_handler = unimplemented_aenq_handler
2329 };
2330