xref: /dpdk/drivers/net/ena/ena_ethdev.c (revision 1be097dc96e05a10966db1096feadf956e7f147e)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright (c) 2015-2020 Amazon.com, Inc. or its affiliates.
3  * All rights reserved.
4  */
5 
6 #include <rte_string_fns.h>
7 #include <rte_ether.h>
8 #include <rte_ethdev_driver.h>
9 #include <rte_ethdev_pci.h>
10 #include <rte_tcp.h>
11 #include <rte_atomic.h>
12 #include <rte_dev.h>
13 #include <rte_errno.h>
14 #include <rte_version.h>
15 #include <rte_net.h>
16 #include <rte_kvargs.h>
17 
18 #include "ena_ethdev.h"
19 #include "ena_logs.h"
20 #include "ena_platform.h"
21 #include "ena_com.h"
22 #include "ena_eth_com.h"
23 
24 #include <ena_common_defs.h>
25 #include <ena_regs_defs.h>
26 #include <ena_admin_defs.h>
27 #include <ena_eth_io_defs.h>
28 
29 #define DRV_MODULE_VER_MAJOR	2
30 #define DRV_MODULE_VER_MINOR	0
31 #define DRV_MODULE_VER_SUBMINOR	3
32 
33 #define ENA_IO_TXQ_IDX(q)	(2 * (q))
34 #define ENA_IO_RXQ_IDX(q)	(2 * (q) + 1)
35 /*reverse version of ENA_IO_RXQ_IDX*/
36 #define ENA_IO_RXQ_IDX_REV(q)	((q - 1) / 2)
37 
38 /* While processing submitted and completed descriptors (rx and tx path
39  * respectively) in a loop it is desired to:
40  *  - perform batch submissions while populating sumbissmion queue
41  *  - avoid blocking transmission of other packets during cleanup phase
42  * Hence the utilization ratio of 1/8 of a queue size.
43  */
44 #define ENA_RING_DESCS_RATIO(ring_size)	(ring_size / 8)
45 
46 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
47 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
48 
49 #define GET_L4_HDR_LEN(mbuf)					\
50 	((rte_pktmbuf_mtod_offset(mbuf,	struct rte_tcp_hdr *,	\
51 		mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
52 
53 #define ENA_RX_RSS_TABLE_LOG_SIZE  7
54 #define ENA_RX_RSS_TABLE_SIZE	(1 << ENA_RX_RSS_TABLE_LOG_SIZE)
55 #define ENA_HASH_KEY_SIZE	40
56 #define ETH_GSTRING_LEN	32
57 
58 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
59 
60 #define ENA_MIN_RING_DESC	128
61 
62 enum ethtool_stringset {
63 	ETH_SS_TEST             = 0,
64 	ETH_SS_STATS,
65 };
66 
67 struct ena_stats {
68 	char name[ETH_GSTRING_LEN];
69 	int stat_offset;
70 };
71 
72 #define ENA_STAT_ENTRY(stat, stat_type) { \
73 	.name = #stat, \
74 	.stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
75 }
76 
77 #define ENA_STAT_RX_ENTRY(stat) \
78 	ENA_STAT_ENTRY(stat, rx)
79 
80 #define ENA_STAT_TX_ENTRY(stat) \
81 	ENA_STAT_ENTRY(stat, tx)
82 
83 #define ENA_STAT_GLOBAL_ENTRY(stat) \
84 	ENA_STAT_ENTRY(stat, dev)
85 
86 /* Device arguments */
87 #define ENA_DEVARG_LARGE_LLQ_HDR "large_llq_hdr"
88 
89 /*
90  * Each rte_memzone should have unique name.
91  * To satisfy it, count number of allocation and add it to name.
92  */
93 rte_atomic32_t ena_alloc_cnt;
94 
95 static const struct ena_stats ena_stats_global_strings[] = {
96 	ENA_STAT_GLOBAL_ENTRY(wd_expired),
97 	ENA_STAT_GLOBAL_ENTRY(dev_start),
98 	ENA_STAT_GLOBAL_ENTRY(dev_stop),
99 	ENA_STAT_GLOBAL_ENTRY(tx_drops),
100 };
101 
102 static const struct ena_stats ena_stats_tx_strings[] = {
103 	ENA_STAT_TX_ENTRY(cnt),
104 	ENA_STAT_TX_ENTRY(bytes),
105 	ENA_STAT_TX_ENTRY(prepare_ctx_err),
106 	ENA_STAT_TX_ENTRY(linearize),
107 	ENA_STAT_TX_ENTRY(linearize_failed),
108 	ENA_STAT_TX_ENTRY(tx_poll),
109 	ENA_STAT_TX_ENTRY(doorbells),
110 	ENA_STAT_TX_ENTRY(bad_req_id),
111 	ENA_STAT_TX_ENTRY(available_desc),
112 };
113 
114 static const struct ena_stats ena_stats_rx_strings[] = {
115 	ENA_STAT_RX_ENTRY(cnt),
116 	ENA_STAT_RX_ENTRY(bytes),
117 	ENA_STAT_RX_ENTRY(refill_partial),
118 	ENA_STAT_RX_ENTRY(bad_csum),
119 	ENA_STAT_RX_ENTRY(mbuf_alloc_fail),
120 	ENA_STAT_RX_ENTRY(bad_desc_num),
121 	ENA_STAT_RX_ENTRY(bad_req_id),
122 };
123 
124 #define ENA_STATS_ARRAY_GLOBAL	ARRAY_SIZE(ena_stats_global_strings)
125 #define ENA_STATS_ARRAY_TX	ARRAY_SIZE(ena_stats_tx_strings)
126 #define ENA_STATS_ARRAY_RX	ARRAY_SIZE(ena_stats_rx_strings)
127 
128 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
129 			DEV_TX_OFFLOAD_UDP_CKSUM |\
130 			DEV_TX_OFFLOAD_IPV4_CKSUM |\
131 			DEV_TX_OFFLOAD_TCP_TSO)
132 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
133 		       PKT_TX_IP_CKSUM |\
134 		       PKT_TX_TCP_SEG)
135 
136 /** Vendor ID used by Amazon devices */
137 #define PCI_VENDOR_ID_AMAZON 0x1D0F
138 /** Amazon devices */
139 #define PCI_DEVICE_ID_ENA_VF	0xEC20
140 #define PCI_DEVICE_ID_ENA_LLQ_VF	0xEC21
141 
142 #define	ENA_TX_OFFLOAD_MASK	(\
143 	PKT_TX_L4_MASK |         \
144 	PKT_TX_IPV6 |            \
145 	PKT_TX_IPV4 |            \
146 	PKT_TX_IP_CKSUM |        \
147 	PKT_TX_TCP_SEG)
148 
149 #define	ENA_TX_OFFLOAD_NOTSUP_MASK	\
150 	(PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
151 
152 int ena_logtype_init;
153 int ena_logtype_driver;
154 
155 #ifdef RTE_LIBRTE_ENA_DEBUG_RX
156 int ena_logtype_rx;
157 #endif
158 #ifdef RTE_LIBRTE_ENA_DEBUG_TX
159 int ena_logtype_tx;
160 #endif
161 #ifdef RTE_LIBRTE_ENA_DEBUG_TX_FREE
162 int ena_logtype_tx_free;
163 #endif
164 #ifdef RTE_LIBRTE_ENA_COM_DEBUG
165 int ena_logtype_com;
166 #endif
167 
168 static const struct rte_pci_id pci_id_ena_map[] = {
169 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
170 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
171 	{ .device_id = 0 },
172 };
173 
174 static struct ena_aenq_handlers aenq_handlers;
175 
176 static int ena_device_init(struct ena_com_dev *ena_dev,
177 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
178 			   bool *wd_state);
179 static int ena_dev_configure(struct rte_eth_dev *dev);
180 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
181 				  uint16_t nb_pkts);
182 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
183 		uint16_t nb_pkts);
184 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
185 			      uint16_t nb_desc, unsigned int socket_id,
186 			      const struct rte_eth_txconf *tx_conf);
187 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
188 			      uint16_t nb_desc, unsigned int socket_id,
189 			      const struct rte_eth_rxconf *rx_conf,
190 			      struct rte_mempool *mp);
191 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len);
192 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
193 				    struct ena_com_rx_buf_info *ena_bufs,
194 				    uint32_t descs,
195 				    uint16_t *next_to_clean,
196 				    uint8_t offset);
197 static uint16_t eth_ena_recv_pkts(void *rx_queue,
198 				  struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
199 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
200 static void ena_init_rings(struct ena_adapter *adapter,
201 			   bool disable_meta_caching);
202 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
203 static int ena_start(struct rte_eth_dev *dev);
204 static void ena_stop(struct rte_eth_dev *dev);
205 static void ena_close(struct rte_eth_dev *dev);
206 static int ena_dev_reset(struct rte_eth_dev *dev);
207 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
208 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
209 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
210 static void ena_rx_queue_release(void *queue);
211 static void ena_tx_queue_release(void *queue);
212 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
213 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
214 static int ena_link_update(struct rte_eth_dev *dev,
215 			   int wait_to_complete);
216 static int ena_create_io_queue(struct ena_ring *ring);
217 static void ena_queue_stop(struct ena_ring *ring);
218 static void ena_queue_stop_all(struct rte_eth_dev *dev,
219 			      enum ena_ring_type ring_type);
220 static int ena_queue_start(struct ena_ring *ring);
221 static int ena_queue_start_all(struct rte_eth_dev *dev,
222 			       enum ena_ring_type ring_type);
223 static void ena_stats_restart(struct rte_eth_dev *dev);
224 static int ena_infos_get(struct rte_eth_dev *dev,
225 			 struct rte_eth_dev_info *dev_info);
226 static int ena_rss_reta_update(struct rte_eth_dev *dev,
227 			       struct rte_eth_rss_reta_entry64 *reta_conf,
228 			       uint16_t reta_size);
229 static int ena_rss_reta_query(struct rte_eth_dev *dev,
230 			      struct rte_eth_rss_reta_entry64 *reta_conf,
231 			      uint16_t reta_size);
232 static void ena_interrupt_handler_rte(void *cb_arg);
233 static void ena_timer_wd_callback(struct rte_timer *timer, void *arg);
234 static void ena_destroy_device(struct rte_eth_dev *eth_dev);
235 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev);
236 static int ena_xstats_get_names(struct rte_eth_dev *dev,
237 				struct rte_eth_xstat_name *xstats_names,
238 				unsigned int n);
239 static int ena_xstats_get(struct rte_eth_dev *dev,
240 			  struct rte_eth_xstat *stats,
241 			  unsigned int n);
242 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
243 				const uint64_t *ids,
244 				uint64_t *values,
245 				unsigned int n);
246 static int ena_process_bool_devarg(const char *key,
247 				   const char *value,
248 				   void *opaque);
249 static int ena_parse_devargs(struct ena_adapter *adapter,
250 			     struct rte_devargs *devargs);
251 
252 static const struct eth_dev_ops ena_dev_ops = {
253 	.dev_configure        = ena_dev_configure,
254 	.dev_infos_get        = ena_infos_get,
255 	.rx_queue_setup       = ena_rx_queue_setup,
256 	.tx_queue_setup       = ena_tx_queue_setup,
257 	.dev_start            = ena_start,
258 	.dev_stop             = ena_stop,
259 	.link_update          = ena_link_update,
260 	.stats_get            = ena_stats_get,
261 	.xstats_get_names     = ena_xstats_get_names,
262 	.xstats_get	      = ena_xstats_get,
263 	.xstats_get_by_id     = ena_xstats_get_by_id,
264 	.mtu_set              = ena_mtu_set,
265 	.rx_queue_release     = ena_rx_queue_release,
266 	.tx_queue_release     = ena_tx_queue_release,
267 	.dev_close            = ena_close,
268 	.dev_reset            = ena_dev_reset,
269 	.reta_update          = ena_rss_reta_update,
270 	.reta_query           = ena_rss_reta_query,
271 };
272 
273 void ena_rss_key_fill(void *key, size_t size)
274 {
275 	static bool key_generated;
276 	static uint8_t default_key[ENA_HASH_KEY_SIZE];
277 	size_t i;
278 
279 	RTE_ASSERT(size <= ENA_HASH_KEY_SIZE);
280 
281 	if (!key_generated) {
282 		for (i = 0; i < ENA_HASH_KEY_SIZE; ++i)
283 			default_key[i] = rte_rand() & 0xff;
284 		key_generated = true;
285 	}
286 
287 	rte_memcpy(key, default_key, size);
288 }
289 
290 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
291 				       struct ena_com_rx_ctx *ena_rx_ctx)
292 {
293 	uint64_t ol_flags = 0;
294 	uint32_t packet_type = 0;
295 
296 	if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
297 		packet_type |= RTE_PTYPE_L4_TCP;
298 	else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
299 		packet_type |= RTE_PTYPE_L4_UDP;
300 
301 	if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
302 		packet_type |= RTE_PTYPE_L3_IPV4;
303 	else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
304 		packet_type |= RTE_PTYPE_L3_IPV6;
305 
306 	if (!ena_rx_ctx->l4_csum_checked)
307 		ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
308 	else
309 		if (unlikely(ena_rx_ctx->l4_csum_err) && !ena_rx_ctx->frag)
310 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
311 		else
312 			ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
313 
314 	if (unlikely(ena_rx_ctx->l3_csum_err))
315 		ol_flags |= PKT_RX_IP_CKSUM_BAD;
316 
317 	mbuf->ol_flags = ol_flags;
318 	mbuf->packet_type = packet_type;
319 }
320 
321 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
322 				       struct ena_com_tx_ctx *ena_tx_ctx,
323 				       uint64_t queue_offloads,
324 				       bool disable_meta_caching)
325 {
326 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
327 
328 	if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
329 	    (queue_offloads & QUEUE_OFFLOADS)) {
330 		/* check if TSO is required */
331 		if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
332 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
333 			ena_tx_ctx->tso_enable = true;
334 
335 			ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
336 		}
337 
338 		/* check if L3 checksum is needed */
339 		if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
340 		    (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
341 			ena_tx_ctx->l3_csum_enable = true;
342 
343 		if (mbuf->ol_flags & PKT_TX_IPV6) {
344 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
345 		} else {
346 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
347 
348 			/* set don't fragment (DF) flag */
349 			if (mbuf->packet_type &
350 				(RTE_PTYPE_L4_NONFRAG
351 				 | RTE_PTYPE_INNER_L4_NONFRAG))
352 				ena_tx_ctx->df = true;
353 		}
354 
355 		/* check if L4 checksum is needed */
356 		if (((mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) &&
357 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
358 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
359 			ena_tx_ctx->l4_csum_enable = true;
360 		} else if (((mbuf->ol_flags & PKT_TX_L4_MASK) ==
361 				PKT_TX_UDP_CKSUM) &&
362 				(queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
363 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
364 			ena_tx_ctx->l4_csum_enable = true;
365 		} else {
366 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
367 			ena_tx_ctx->l4_csum_enable = false;
368 		}
369 
370 		ena_meta->mss = mbuf->tso_segsz;
371 		ena_meta->l3_hdr_len = mbuf->l3_len;
372 		ena_meta->l3_hdr_offset = mbuf->l2_len;
373 
374 		ena_tx_ctx->meta_valid = true;
375 	} else if (disable_meta_caching) {
376 		memset(ena_meta, 0, sizeof(*ena_meta));
377 		ena_tx_ctx->meta_valid = true;
378 	} else {
379 		ena_tx_ctx->meta_valid = false;
380 	}
381 }
382 
383 static inline int validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
384 {
385 	if (likely(req_id < rx_ring->ring_size))
386 		return 0;
387 
388 	PMD_DRV_LOG(ERR, "Invalid rx req_id: %hu\n", req_id);
389 
390 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
391 	rx_ring->adapter->trigger_reset = true;
392 	++rx_ring->rx_stats.bad_req_id;
393 
394 	return -EFAULT;
395 }
396 
397 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
398 {
399 	struct ena_tx_buffer *tx_info = NULL;
400 
401 	if (likely(req_id < tx_ring->ring_size)) {
402 		tx_info = &tx_ring->tx_buffer_info[req_id];
403 		if (likely(tx_info->mbuf))
404 			return 0;
405 	}
406 
407 	if (tx_info)
408 		PMD_DRV_LOG(ERR, "tx_info doesn't have valid mbuf\n");
409 	else
410 		PMD_DRV_LOG(ERR, "Invalid req_id: %hu\n", req_id);
411 
412 	/* Trigger device reset */
413 	++tx_ring->tx_stats.bad_req_id;
414 	tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
415 	tx_ring->adapter->trigger_reset	= true;
416 	return -EFAULT;
417 }
418 
419 static void ena_config_host_info(struct ena_com_dev *ena_dev)
420 {
421 	struct ena_admin_host_info *host_info;
422 	int rc;
423 
424 	/* Allocate only the host info */
425 	rc = ena_com_allocate_host_info(ena_dev);
426 	if (rc) {
427 		PMD_DRV_LOG(ERR, "Cannot allocate host info\n");
428 		return;
429 	}
430 
431 	host_info = ena_dev->host_attr.host_info;
432 
433 	host_info->os_type = ENA_ADMIN_OS_DPDK;
434 	host_info->kernel_ver = RTE_VERSION;
435 	strlcpy((char *)host_info->kernel_ver_str, rte_version(),
436 		sizeof(host_info->kernel_ver_str));
437 	host_info->os_dist = RTE_VERSION;
438 	strlcpy((char *)host_info->os_dist_str, rte_version(),
439 		sizeof(host_info->os_dist_str));
440 	host_info->driver_version =
441 		(DRV_MODULE_VER_MAJOR) |
442 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
443 		(DRV_MODULE_VER_SUBMINOR <<
444 			ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
445 	host_info->num_cpus = rte_lcore_count();
446 
447 	host_info->driver_supported_features =
448 		ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK;
449 
450 	rc = ena_com_set_host_attributes(ena_dev);
451 	if (rc) {
452 		if (rc == -ENA_COM_UNSUPPORTED)
453 			PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
454 		else
455 			PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
456 
457 		goto err;
458 	}
459 
460 	return;
461 
462 err:
463 	ena_com_delete_host_info(ena_dev);
464 }
465 
466 /* This function calculates the number of xstats based on the current config */
467 static unsigned int ena_xstats_calc_num(struct rte_eth_dev *dev)
468 {
469 	return ENA_STATS_ARRAY_GLOBAL +
470 		(dev->data->nb_tx_queues * ENA_STATS_ARRAY_TX) +
471 		(dev->data->nb_rx_queues * ENA_STATS_ARRAY_RX);
472 }
473 
474 static void ena_config_debug_area(struct ena_adapter *adapter)
475 {
476 	u32 debug_area_size;
477 	int rc, ss_count;
478 
479 	ss_count = ena_xstats_calc_num(adapter->rte_dev);
480 
481 	/* allocate 32 bytes for each string and 64bit for the value */
482 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
483 
484 	rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
485 	if (rc) {
486 		PMD_DRV_LOG(ERR, "Cannot allocate debug area\n");
487 		return;
488 	}
489 
490 	rc = ena_com_set_host_attributes(&adapter->ena_dev);
491 	if (rc) {
492 		if (rc == -ENA_COM_UNSUPPORTED)
493 			PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
494 		else
495 			PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
496 
497 		goto err;
498 	}
499 
500 	return;
501 err:
502 	ena_com_delete_debug_area(&adapter->ena_dev);
503 }
504 
505 static void ena_close(struct rte_eth_dev *dev)
506 {
507 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
508 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
509 	struct ena_adapter *adapter = dev->data->dev_private;
510 
511 	if (adapter->state == ENA_ADAPTER_STATE_RUNNING)
512 		ena_stop(dev);
513 	adapter->state = ENA_ADAPTER_STATE_CLOSED;
514 
515 	ena_rx_queue_release_all(dev);
516 	ena_tx_queue_release_all(dev);
517 
518 	rte_free(adapter->drv_stats);
519 	adapter->drv_stats = NULL;
520 
521 	rte_intr_disable(intr_handle);
522 	rte_intr_callback_unregister(intr_handle,
523 				     ena_interrupt_handler_rte,
524 				     adapter);
525 
526 	/*
527 	 * MAC is not allocated dynamically. Setting NULL should prevent from
528 	 * release of the resource in the rte_eth_dev_release_port().
529 	 */
530 	dev->data->mac_addrs = NULL;
531 }
532 
533 static int
534 ena_dev_reset(struct rte_eth_dev *dev)
535 {
536 	int rc = 0;
537 
538 	ena_destroy_device(dev);
539 	rc = eth_ena_dev_init(dev);
540 	if (rc)
541 		PMD_INIT_LOG(CRIT, "Cannot initialize device");
542 
543 	return rc;
544 }
545 
546 static int ena_rss_reta_update(struct rte_eth_dev *dev,
547 			       struct rte_eth_rss_reta_entry64 *reta_conf,
548 			       uint16_t reta_size)
549 {
550 	struct ena_adapter *adapter = dev->data->dev_private;
551 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
552 	int rc, i;
553 	u16 entry_value;
554 	int conf_idx;
555 	int idx;
556 
557 	if ((reta_size == 0) || (reta_conf == NULL))
558 		return -EINVAL;
559 
560 	if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
561 		PMD_DRV_LOG(WARNING,
562 			"indirection table %d is bigger than supported (%d)\n",
563 			reta_size, ENA_RX_RSS_TABLE_SIZE);
564 		return -EINVAL;
565 	}
566 
567 	for (i = 0 ; i < reta_size ; i++) {
568 		/* each reta_conf is for 64 entries.
569 		 * to support 128 we use 2 conf of 64
570 		 */
571 		conf_idx = i / RTE_RETA_GROUP_SIZE;
572 		idx = i % RTE_RETA_GROUP_SIZE;
573 		if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
574 			entry_value =
575 				ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
576 
577 			rc = ena_com_indirect_table_fill_entry(ena_dev,
578 							       i,
579 							       entry_value);
580 			if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
581 				PMD_DRV_LOG(ERR,
582 					"Cannot fill indirect table\n");
583 				return rc;
584 			}
585 		}
586 	}
587 
588 	rc = ena_com_indirect_table_set(ena_dev);
589 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
590 		PMD_DRV_LOG(ERR, "Cannot flush the indirect table\n");
591 		return rc;
592 	}
593 
594 	PMD_DRV_LOG(DEBUG, "%s(): RSS configured %d entries  for port %d\n",
595 		__func__, reta_size, adapter->rte_dev->data->port_id);
596 
597 	return 0;
598 }
599 
600 /* Query redirection table. */
601 static int ena_rss_reta_query(struct rte_eth_dev *dev,
602 			      struct rte_eth_rss_reta_entry64 *reta_conf,
603 			      uint16_t reta_size)
604 {
605 	struct ena_adapter *adapter = dev->data->dev_private;
606 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
607 	int rc;
608 	int i;
609 	u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
610 	int reta_conf_idx;
611 	int reta_idx;
612 
613 	if (reta_size == 0 || reta_conf == NULL ||
614 	    (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
615 		return -EINVAL;
616 
617 	rc = ena_com_indirect_table_get(ena_dev, indirect_table);
618 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
619 		PMD_DRV_LOG(ERR, "cannot get indirect table\n");
620 		return -ENOTSUP;
621 	}
622 
623 	for (i = 0 ; i < reta_size ; i++) {
624 		reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
625 		reta_idx = i % RTE_RETA_GROUP_SIZE;
626 		if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
627 			reta_conf[reta_conf_idx].reta[reta_idx] =
628 				ENA_IO_RXQ_IDX_REV(indirect_table[i]);
629 	}
630 
631 	return 0;
632 }
633 
634 static int ena_rss_init_default(struct ena_adapter *adapter)
635 {
636 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
637 	uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
638 	int rc, i;
639 	u32 val;
640 
641 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
642 	if (unlikely(rc)) {
643 		PMD_DRV_LOG(ERR, "Cannot init indirect table\n");
644 		goto err_rss_init;
645 	}
646 
647 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
648 		val = i % nb_rx_queues;
649 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
650 						       ENA_IO_RXQ_IDX(val));
651 		if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
652 			PMD_DRV_LOG(ERR, "Cannot fill indirect table\n");
653 			goto err_fill_indir;
654 		}
655 	}
656 
657 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
658 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
659 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
660 		PMD_DRV_LOG(INFO, "Cannot fill hash function\n");
661 		goto err_fill_indir;
662 	}
663 
664 	rc = ena_com_set_default_hash_ctrl(ena_dev);
665 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
666 		PMD_DRV_LOG(INFO, "Cannot fill hash control\n");
667 		goto err_fill_indir;
668 	}
669 
670 	rc = ena_com_indirect_table_set(ena_dev);
671 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
672 		PMD_DRV_LOG(ERR, "Cannot flush the indirect table\n");
673 		goto err_fill_indir;
674 	}
675 	PMD_DRV_LOG(DEBUG, "RSS configured for port %d\n",
676 		adapter->rte_dev->data->port_id);
677 
678 	return 0;
679 
680 err_fill_indir:
681 	ena_com_rss_destroy(ena_dev);
682 err_rss_init:
683 
684 	return rc;
685 }
686 
687 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
688 {
689 	struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
690 	int nb_queues = dev->data->nb_rx_queues;
691 	int i;
692 
693 	for (i = 0; i < nb_queues; i++)
694 		ena_rx_queue_release(queues[i]);
695 }
696 
697 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
698 {
699 	struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
700 	int nb_queues = dev->data->nb_tx_queues;
701 	int i;
702 
703 	for (i = 0; i < nb_queues; i++)
704 		ena_tx_queue_release(queues[i]);
705 }
706 
707 static void ena_rx_queue_release(void *queue)
708 {
709 	struct ena_ring *ring = (struct ena_ring *)queue;
710 
711 	/* Free ring resources */
712 	if (ring->rx_buffer_info)
713 		rte_free(ring->rx_buffer_info);
714 	ring->rx_buffer_info = NULL;
715 
716 	if (ring->rx_refill_buffer)
717 		rte_free(ring->rx_refill_buffer);
718 	ring->rx_refill_buffer = NULL;
719 
720 	if (ring->empty_rx_reqs)
721 		rte_free(ring->empty_rx_reqs);
722 	ring->empty_rx_reqs = NULL;
723 
724 	ring->configured = 0;
725 
726 	PMD_DRV_LOG(NOTICE, "RX Queue %d:%d released\n",
727 		ring->port_id, ring->id);
728 }
729 
730 static void ena_tx_queue_release(void *queue)
731 {
732 	struct ena_ring *ring = (struct ena_ring *)queue;
733 
734 	/* Free ring resources */
735 	if (ring->push_buf_intermediate_buf)
736 		rte_free(ring->push_buf_intermediate_buf);
737 
738 	if (ring->tx_buffer_info)
739 		rte_free(ring->tx_buffer_info);
740 
741 	if (ring->empty_tx_reqs)
742 		rte_free(ring->empty_tx_reqs);
743 
744 	ring->empty_tx_reqs = NULL;
745 	ring->tx_buffer_info = NULL;
746 	ring->push_buf_intermediate_buf = NULL;
747 
748 	ring->configured = 0;
749 
750 	PMD_DRV_LOG(NOTICE, "TX Queue %d:%d released\n",
751 		ring->port_id, ring->id);
752 }
753 
754 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
755 {
756 	unsigned int i;
757 
758 	for (i = 0; i < ring->ring_size; ++i) {
759 		struct ena_rx_buffer *rx_info = &ring->rx_buffer_info[i];
760 		if (rx_info->mbuf) {
761 			rte_mbuf_raw_free(rx_info->mbuf);
762 			rx_info->mbuf = NULL;
763 		}
764 	}
765 }
766 
767 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
768 {
769 	unsigned int i;
770 
771 	for (i = 0; i < ring->ring_size; ++i) {
772 		struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
773 
774 		if (tx_buf->mbuf)
775 			rte_pktmbuf_free(tx_buf->mbuf);
776 	}
777 }
778 
779 static int ena_link_update(struct rte_eth_dev *dev,
780 			   __rte_unused int wait_to_complete)
781 {
782 	struct rte_eth_link *link = &dev->data->dev_link;
783 	struct ena_adapter *adapter = dev->data->dev_private;
784 
785 	link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
786 	link->link_speed = ETH_SPEED_NUM_NONE;
787 	link->link_duplex = ETH_LINK_FULL_DUPLEX;
788 
789 	return 0;
790 }
791 
792 static int ena_queue_start_all(struct rte_eth_dev *dev,
793 			       enum ena_ring_type ring_type)
794 {
795 	struct ena_adapter *adapter = dev->data->dev_private;
796 	struct ena_ring *queues = NULL;
797 	int nb_queues;
798 	int i = 0;
799 	int rc = 0;
800 
801 	if (ring_type == ENA_RING_TYPE_RX) {
802 		queues = adapter->rx_ring;
803 		nb_queues = dev->data->nb_rx_queues;
804 	} else {
805 		queues = adapter->tx_ring;
806 		nb_queues = dev->data->nb_tx_queues;
807 	}
808 	for (i = 0; i < nb_queues; i++) {
809 		if (queues[i].configured) {
810 			if (ring_type == ENA_RING_TYPE_RX) {
811 				ena_assert_msg(
812 					dev->data->rx_queues[i] == &queues[i],
813 					"Inconsistent state of rx queues\n");
814 			} else {
815 				ena_assert_msg(
816 					dev->data->tx_queues[i] == &queues[i],
817 					"Inconsistent state of tx queues\n");
818 			}
819 
820 			rc = ena_queue_start(&queues[i]);
821 
822 			if (rc) {
823 				PMD_INIT_LOG(ERR,
824 					     "failed to start queue %d type(%d)",
825 					     i, ring_type);
826 				goto err;
827 			}
828 		}
829 	}
830 
831 	return 0;
832 
833 err:
834 	while (i--)
835 		if (queues[i].configured)
836 			ena_queue_stop(&queues[i]);
837 
838 	return rc;
839 }
840 
841 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
842 {
843 	uint32_t max_frame_len = adapter->max_mtu;
844 
845 	if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
846 	    DEV_RX_OFFLOAD_JUMBO_FRAME)
847 		max_frame_len =
848 			adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
849 
850 	return max_frame_len;
851 }
852 
853 static int ena_check_valid_conf(struct ena_adapter *adapter)
854 {
855 	uint32_t max_frame_len = ena_get_mtu_conf(adapter);
856 
857 	if (max_frame_len > adapter->max_mtu || max_frame_len < ENA_MIN_MTU) {
858 		PMD_INIT_LOG(ERR, "Unsupported MTU of %d. "
859 				  "max mtu: %d, min mtu: %d",
860 			     max_frame_len, adapter->max_mtu, ENA_MIN_MTU);
861 		return ENA_COM_UNSUPPORTED;
862 	}
863 
864 	return 0;
865 }
866 
867 static int
868 ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx,
869 		       bool use_large_llq_hdr)
870 {
871 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
872 	struct ena_com_dev *ena_dev = ctx->ena_dev;
873 	uint32_t max_tx_queue_size;
874 	uint32_t max_rx_queue_size;
875 
876 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
877 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
878 			&ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
879 		max_rx_queue_size = RTE_MIN(max_queue_ext->max_rx_cq_depth,
880 			max_queue_ext->max_rx_sq_depth);
881 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
882 
883 		if (ena_dev->tx_mem_queue_type ==
884 		    ENA_ADMIN_PLACEMENT_POLICY_DEV) {
885 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
886 				llq->max_llq_depth);
887 		} else {
888 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
889 				max_queue_ext->max_tx_sq_depth);
890 		}
891 
892 		ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
893 			max_queue_ext->max_per_packet_rx_descs);
894 		ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
895 			max_queue_ext->max_per_packet_tx_descs);
896 	} else {
897 		struct ena_admin_queue_feature_desc *max_queues =
898 			&ctx->get_feat_ctx->max_queues;
899 		max_rx_queue_size = RTE_MIN(max_queues->max_cq_depth,
900 			max_queues->max_sq_depth);
901 		max_tx_queue_size = max_queues->max_cq_depth;
902 
903 		if (ena_dev->tx_mem_queue_type ==
904 		    ENA_ADMIN_PLACEMENT_POLICY_DEV) {
905 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
906 				llq->max_llq_depth);
907 		} else {
908 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
909 				max_queues->max_sq_depth);
910 		}
911 
912 		ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
913 			max_queues->max_packet_rx_descs);
914 		ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
915 			max_queues->max_packet_tx_descs);
916 	}
917 
918 	/* Round down to the nearest power of 2 */
919 	max_rx_queue_size = rte_align32prevpow2(max_rx_queue_size);
920 	max_tx_queue_size = rte_align32prevpow2(max_tx_queue_size);
921 
922 	if (use_large_llq_hdr) {
923 		if ((llq->entry_size_ctrl_supported &
924 		     ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
925 		    (ena_dev->tx_mem_queue_type ==
926 		     ENA_ADMIN_PLACEMENT_POLICY_DEV)) {
927 			max_tx_queue_size /= 2;
928 			PMD_INIT_LOG(INFO,
929 				"Forcing large headers and decreasing maximum TX queue size to %d\n",
930 				max_tx_queue_size);
931 		} else {
932 			PMD_INIT_LOG(ERR,
933 				"Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
934 		}
935 	}
936 
937 	if (unlikely(max_rx_queue_size == 0 || max_tx_queue_size == 0)) {
938 		PMD_INIT_LOG(ERR, "Invalid queue size");
939 		return -EFAULT;
940 	}
941 
942 	ctx->max_tx_queue_size = max_tx_queue_size;
943 	ctx->max_rx_queue_size = max_rx_queue_size;
944 
945 	return 0;
946 }
947 
948 static void ena_stats_restart(struct rte_eth_dev *dev)
949 {
950 	struct ena_adapter *adapter = dev->data->dev_private;
951 
952 	rte_atomic64_init(&adapter->drv_stats->ierrors);
953 	rte_atomic64_init(&adapter->drv_stats->oerrors);
954 	rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
955 	adapter->drv_stats->rx_drops = 0;
956 }
957 
958 static int ena_stats_get(struct rte_eth_dev *dev,
959 			  struct rte_eth_stats *stats)
960 {
961 	struct ena_admin_basic_stats ena_stats;
962 	struct ena_adapter *adapter = dev->data->dev_private;
963 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
964 	int rc;
965 	int i;
966 	int max_rings_stats;
967 
968 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
969 		return -ENOTSUP;
970 
971 	memset(&ena_stats, 0, sizeof(ena_stats));
972 	rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
973 	if (unlikely(rc)) {
974 		PMD_DRV_LOG(ERR, "Could not retrieve statistics from ENA\n");
975 		return rc;
976 	}
977 
978 	/* Set of basic statistics from ENA */
979 	stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
980 					  ena_stats.rx_pkts_low);
981 	stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
982 					  ena_stats.tx_pkts_low);
983 	stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
984 					ena_stats.rx_bytes_low);
985 	stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
986 					ena_stats.tx_bytes_low);
987 
988 	/* Driver related stats */
989 	stats->imissed = adapter->drv_stats->rx_drops;
990 	stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
991 	stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
992 	stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
993 
994 	max_rings_stats = RTE_MIN(dev->data->nb_rx_queues,
995 		RTE_ETHDEV_QUEUE_STAT_CNTRS);
996 	for (i = 0; i < max_rings_stats; ++i) {
997 		struct ena_stats_rx *rx_stats = &adapter->rx_ring[i].rx_stats;
998 
999 		stats->q_ibytes[i] = rx_stats->bytes;
1000 		stats->q_ipackets[i] = rx_stats->cnt;
1001 		stats->q_errors[i] = rx_stats->bad_desc_num +
1002 			rx_stats->bad_req_id;
1003 	}
1004 
1005 	max_rings_stats = RTE_MIN(dev->data->nb_tx_queues,
1006 		RTE_ETHDEV_QUEUE_STAT_CNTRS);
1007 	for (i = 0; i < max_rings_stats; ++i) {
1008 		struct ena_stats_tx *tx_stats = &adapter->tx_ring[i].tx_stats;
1009 
1010 		stats->q_obytes[i] = tx_stats->bytes;
1011 		stats->q_opackets[i] = tx_stats->cnt;
1012 	}
1013 
1014 	return 0;
1015 }
1016 
1017 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1018 {
1019 	struct ena_adapter *adapter;
1020 	struct ena_com_dev *ena_dev;
1021 	int rc = 0;
1022 
1023 	ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
1024 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
1025 	adapter = dev->data->dev_private;
1026 
1027 	ena_dev = &adapter->ena_dev;
1028 	ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
1029 
1030 	if (mtu > ena_get_mtu_conf(adapter) || mtu < ENA_MIN_MTU) {
1031 		PMD_DRV_LOG(ERR,
1032 			"Invalid MTU setting. new_mtu: %d "
1033 			"max mtu: %d min mtu: %d\n",
1034 			mtu, ena_get_mtu_conf(adapter), ENA_MIN_MTU);
1035 		return -EINVAL;
1036 	}
1037 
1038 	rc = ena_com_set_dev_mtu(ena_dev, mtu);
1039 	if (rc)
1040 		PMD_DRV_LOG(ERR, "Could not set MTU: %d\n", mtu);
1041 	else
1042 		PMD_DRV_LOG(NOTICE, "Set MTU: %d\n", mtu);
1043 
1044 	return rc;
1045 }
1046 
1047 static int ena_start(struct rte_eth_dev *dev)
1048 {
1049 	struct ena_adapter *adapter = dev->data->dev_private;
1050 	uint64_t ticks;
1051 	int rc = 0;
1052 
1053 	rc = ena_check_valid_conf(adapter);
1054 	if (rc)
1055 		return rc;
1056 
1057 	rc = ena_queue_start_all(dev, ENA_RING_TYPE_RX);
1058 	if (rc)
1059 		return rc;
1060 
1061 	rc = ena_queue_start_all(dev, ENA_RING_TYPE_TX);
1062 	if (rc)
1063 		goto err_start_tx;
1064 
1065 	if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
1066 	    ETH_MQ_RX_RSS_FLAG && adapter->rte_dev->data->nb_rx_queues > 0) {
1067 		rc = ena_rss_init_default(adapter);
1068 		if (rc)
1069 			goto err_rss_init;
1070 	}
1071 
1072 	ena_stats_restart(dev);
1073 
1074 	adapter->timestamp_wd = rte_get_timer_cycles();
1075 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
1076 
1077 	ticks = rte_get_timer_hz();
1078 	rte_timer_reset(&adapter->timer_wd, ticks, PERIODICAL, rte_lcore_id(),
1079 			ena_timer_wd_callback, adapter);
1080 
1081 	++adapter->dev_stats.dev_start;
1082 	adapter->state = ENA_ADAPTER_STATE_RUNNING;
1083 
1084 	return 0;
1085 
1086 err_rss_init:
1087 	ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1088 err_start_tx:
1089 	ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1090 	return rc;
1091 }
1092 
1093 static void ena_stop(struct rte_eth_dev *dev)
1094 {
1095 	struct ena_adapter *adapter = dev->data->dev_private;
1096 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1097 	int rc;
1098 
1099 	rte_timer_stop_sync(&adapter->timer_wd);
1100 	ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1101 	ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1102 
1103 	if (adapter->trigger_reset) {
1104 		rc = ena_com_dev_reset(ena_dev, adapter->reset_reason);
1105 		if (rc)
1106 			PMD_DRV_LOG(ERR, "Device reset failed rc=%d\n", rc);
1107 	}
1108 
1109 	++adapter->dev_stats.dev_stop;
1110 	adapter->state = ENA_ADAPTER_STATE_STOPPED;
1111 }
1112 
1113 static int ena_create_io_queue(struct ena_ring *ring)
1114 {
1115 	struct ena_adapter *adapter;
1116 	struct ena_com_dev *ena_dev;
1117 	struct ena_com_create_io_ctx ctx =
1118 		/* policy set to _HOST just to satisfy icc compiler */
1119 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
1120 		  0, 0, 0, 0, 0 };
1121 	uint16_t ena_qid;
1122 	unsigned int i;
1123 	int rc;
1124 
1125 	adapter = ring->adapter;
1126 	ena_dev = &adapter->ena_dev;
1127 
1128 	if (ring->type == ENA_RING_TYPE_TX) {
1129 		ena_qid = ENA_IO_TXQ_IDX(ring->id);
1130 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1131 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1132 		for (i = 0; i < ring->ring_size; i++)
1133 			ring->empty_tx_reqs[i] = i;
1134 	} else {
1135 		ena_qid = ENA_IO_RXQ_IDX(ring->id);
1136 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1137 		for (i = 0; i < ring->ring_size; i++)
1138 			ring->empty_rx_reqs[i] = i;
1139 	}
1140 	ctx.queue_size = ring->ring_size;
1141 	ctx.qid = ena_qid;
1142 	ctx.msix_vector = -1; /* interrupts not used */
1143 	ctx.numa_node = ring->numa_socket_id;
1144 
1145 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1146 	if (rc) {
1147 		PMD_DRV_LOG(ERR,
1148 			"failed to create io queue #%d (qid:%d) rc: %d\n",
1149 			ring->id, ena_qid, rc);
1150 		return rc;
1151 	}
1152 
1153 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1154 				     &ring->ena_com_io_sq,
1155 				     &ring->ena_com_io_cq);
1156 	if (rc) {
1157 		PMD_DRV_LOG(ERR,
1158 			"Failed to get io queue handlers. queue num %d rc: %d\n",
1159 			ring->id, rc);
1160 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1161 		return rc;
1162 	}
1163 
1164 	if (ring->type == ENA_RING_TYPE_TX)
1165 		ena_com_update_numa_node(ring->ena_com_io_cq, ctx.numa_node);
1166 
1167 	return 0;
1168 }
1169 
1170 static void ena_queue_stop(struct ena_ring *ring)
1171 {
1172 	struct ena_com_dev *ena_dev = &ring->adapter->ena_dev;
1173 
1174 	if (ring->type == ENA_RING_TYPE_RX) {
1175 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(ring->id));
1176 		ena_rx_queue_release_bufs(ring);
1177 	} else {
1178 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(ring->id));
1179 		ena_tx_queue_release_bufs(ring);
1180 	}
1181 }
1182 
1183 static void ena_queue_stop_all(struct rte_eth_dev *dev,
1184 			      enum ena_ring_type ring_type)
1185 {
1186 	struct ena_adapter *adapter = dev->data->dev_private;
1187 	struct ena_ring *queues = NULL;
1188 	uint16_t nb_queues, i;
1189 
1190 	if (ring_type == ENA_RING_TYPE_RX) {
1191 		queues = adapter->rx_ring;
1192 		nb_queues = dev->data->nb_rx_queues;
1193 	} else {
1194 		queues = adapter->tx_ring;
1195 		nb_queues = dev->data->nb_tx_queues;
1196 	}
1197 
1198 	for (i = 0; i < nb_queues; ++i)
1199 		if (queues[i].configured)
1200 			ena_queue_stop(&queues[i]);
1201 }
1202 
1203 static int ena_queue_start(struct ena_ring *ring)
1204 {
1205 	int rc, bufs_num;
1206 
1207 	ena_assert_msg(ring->configured == 1,
1208 		       "Trying to start unconfigured queue\n");
1209 
1210 	rc = ena_create_io_queue(ring);
1211 	if (rc) {
1212 		PMD_INIT_LOG(ERR, "Failed to create IO queue!");
1213 		return rc;
1214 	}
1215 
1216 	ring->next_to_clean = 0;
1217 	ring->next_to_use = 0;
1218 
1219 	if (ring->type == ENA_RING_TYPE_TX) {
1220 		ring->tx_stats.available_desc =
1221 			ena_com_free_q_entries(ring->ena_com_io_sq);
1222 		return 0;
1223 	}
1224 
1225 	bufs_num = ring->ring_size - 1;
1226 	rc = ena_populate_rx_queue(ring, bufs_num);
1227 	if (rc != bufs_num) {
1228 		ena_com_destroy_io_queue(&ring->adapter->ena_dev,
1229 					 ENA_IO_RXQ_IDX(ring->id));
1230 		PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
1231 		return ENA_COM_FAULT;
1232 	}
1233 
1234 	return 0;
1235 }
1236 
1237 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
1238 			      uint16_t queue_idx,
1239 			      uint16_t nb_desc,
1240 			      unsigned int socket_id,
1241 			      const struct rte_eth_txconf *tx_conf)
1242 {
1243 	struct ena_ring *txq = NULL;
1244 	struct ena_adapter *adapter = dev->data->dev_private;
1245 	unsigned int i;
1246 
1247 	txq = &adapter->tx_ring[queue_idx];
1248 
1249 	if (txq->configured) {
1250 		PMD_DRV_LOG(CRIT,
1251 			"API violation. Queue %d is already configured\n",
1252 			queue_idx);
1253 		return ENA_COM_FAULT;
1254 	}
1255 
1256 	if (!rte_is_power_of_2(nb_desc)) {
1257 		PMD_DRV_LOG(ERR,
1258 			"Unsupported size of TX queue: %d is not a power of 2.\n",
1259 			nb_desc);
1260 		return -EINVAL;
1261 	}
1262 
1263 	if (nb_desc > adapter->max_tx_ring_size) {
1264 		PMD_DRV_LOG(ERR,
1265 			"Unsupported size of TX queue (max size: %d)\n",
1266 			adapter->max_tx_ring_size);
1267 		return -EINVAL;
1268 	}
1269 
1270 	if (nb_desc == RTE_ETH_DEV_FALLBACK_TX_RINGSIZE)
1271 		nb_desc = adapter->max_tx_ring_size;
1272 
1273 	txq->port_id = dev->data->port_id;
1274 	txq->next_to_clean = 0;
1275 	txq->next_to_use = 0;
1276 	txq->ring_size = nb_desc;
1277 	txq->numa_socket_id = socket_id;
1278 
1279 	txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1280 					  sizeof(struct ena_tx_buffer) *
1281 					  txq->ring_size,
1282 					  RTE_CACHE_LINE_SIZE);
1283 	if (!txq->tx_buffer_info) {
1284 		PMD_DRV_LOG(ERR, "failed to alloc mem for tx buffer info\n");
1285 		return -ENOMEM;
1286 	}
1287 
1288 	txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1289 					 sizeof(u16) * txq->ring_size,
1290 					 RTE_CACHE_LINE_SIZE);
1291 	if (!txq->empty_tx_reqs) {
1292 		PMD_DRV_LOG(ERR, "failed to alloc mem for tx reqs\n");
1293 		rte_free(txq->tx_buffer_info);
1294 		return -ENOMEM;
1295 	}
1296 
1297 	txq->push_buf_intermediate_buf =
1298 		rte_zmalloc("txq->push_buf_intermediate_buf",
1299 			    txq->tx_max_header_size,
1300 			    RTE_CACHE_LINE_SIZE);
1301 	if (!txq->push_buf_intermediate_buf) {
1302 		PMD_DRV_LOG(ERR, "failed to alloc push buff for LLQ\n");
1303 		rte_free(txq->tx_buffer_info);
1304 		rte_free(txq->empty_tx_reqs);
1305 		return -ENOMEM;
1306 	}
1307 
1308 	for (i = 0; i < txq->ring_size; i++)
1309 		txq->empty_tx_reqs[i] = i;
1310 
1311 	if (tx_conf != NULL) {
1312 		txq->offloads =
1313 			tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1314 	}
1315 	/* Store pointer to this queue in upper layer */
1316 	txq->configured = 1;
1317 	dev->data->tx_queues[queue_idx] = txq;
1318 
1319 	return 0;
1320 }
1321 
1322 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1323 			      uint16_t queue_idx,
1324 			      uint16_t nb_desc,
1325 			      unsigned int socket_id,
1326 			      __rte_unused const struct rte_eth_rxconf *rx_conf,
1327 			      struct rte_mempool *mp)
1328 {
1329 	struct ena_adapter *adapter = dev->data->dev_private;
1330 	struct ena_ring *rxq = NULL;
1331 	size_t buffer_size;
1332 	int i;
1333 
1334 	rxq = &adapter->rx_ring[queue_idx];
1335 	if (rxq->configured) {
1336 		PMD_DRV_LOG(CRIT,
1337 			"API violation. Queue %d is already configured\n",
1338 			queue_idx);
1339 		return ENA_COM_FAULT;
1340 	}
1341 
1342 	if (nb_desc == RTE_ETH_DEV_FALLBACK_RX_RINGSIZE)
1343 		nb_desc = adapter->max_rx_ring_size;
1344 
1345 	if (!rte_is_power_of_2(nb_desc)) {
1346 		PMD_DRV_LOG(ERR,
1347 			"Unsupported size of RX queue: %d is not a power of 2.\n",
1348 			nb_desc);
1349 		return -EINVAL;
1350 	}
1351 
1352 	if (nb_desc > adapter->max_rx_ring_size) {
1353 		PMD_DRV_LOG(ERR,
1354 			"Unsupported size of RX queue (max size: %d)\n",
1355 			adapter->max_rx_ring_size);
1356 		return -EINVAL;
1357 	}
1358 
1359 	/* ENA isn't supporting buffers smaller than 1400 bytes */
1360 	buffer_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
1361 	if (buffer_size < ENA_RX_BUF_MIN_SIZE) {
1362 		PMD_DRV_LOG(ERR,
1363 			"Unsupported size of RX buffer: %zu (min size: %d)\n",
1364 			buffer_size, ENA_RX_BUF_MIN_SIZE);
1365 		return -EINVAL;
1366 	}
1367 
1368 	rxq->port_id = dev->data->port_id;
1369 	rxq->next_to_clean = 0;
1370 	rxq->next_to_use = 0;
1371 	rxq->ring_size = nb_desc;
1372 	rxq->numa_socket_id = socket_id;
1373 	rxq->mb_pool = mp;
1374 
1375 	rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1376 		sizeof(struct ena_rx_buffer) * nb_desc,
1377 		RTE_CACHE_LINE_SIZE);
1378 	if (!rxq->rx_buffer_info) {
1379 		PMD_DRV_LOG(ERR, "failed to alloc mem for rx buffer info\n");
1380 		return -ENOMEM;
1381 	}
1382 
1383 	rxq->rx_refill_buffer = rte_zmalloc("rxq->rx_refill_buffer",
1384 					    sizeof(struct rte_mbuf *) * nb_desc,
1385 					    RTE_CACHE_LINE_SIZE);
1386 
1387 	if (!rxq->rx_refill_buffer) {
1388 		PMD_DRV_LOG(ERR, "failed to alloc mem for rx refill buffer\n");
1389 		rte_free(rxq->rx_buffer_info);
1390 		rxq->rx_buffer_info = NULL;
1391 		return -ENOMEM;
1392 	}
1393 
1394 	rxq->empty_rx_reqs = rte_zmalloc("rxq->empty_rx_reqs",
1395 					 sizeof(uint16_t) * nb_desc,
1396 					 RTE_CACHE_LINE_SIZE);
1397 	if (!rxq->empty_rx_reqs) {
1398 		PMD_DRV_LOG(ERR, "failed to alloc mem for empty rx reqs\n");
1399 		rte_free(rxq->rx_buffer_info);
1400 		rxq->rx_buffer_info = NULL;
1401 		rte_free(rxq->rx_refill_buffer);
1402 		rxq->rx_refill_buffer = NULL;
1403 		return -ENOMEM;
1404 	}
1405 
1406 	for (i = 0; i < nb_desc; i++)
1407 		rxq->empty_rx_reqs[i] = i;
1408 
1409 	/* Store pointer to this queue in upper layer */
1410 	rxq->configured = 1;
1411 	dev->data->rx_queues[queue_idx] = rxq;
1412 
1413 	return 0;
1414 }
1415 
1416 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1417 {
1418 	unsigned int i;
1419 	int rc;
1420 	uint16_t ring_size = rxq->ring_size;
1421 	uint16_t ring_mask = ring_size - 1;
1422 	uint16_t next_to_use = rxq->next_to_use;
1423 	uint16_t in_use, req_id;
1424 	struct rte_mbuf **mbufs = rxq->rx_refill_buffer;
1425 
1426 	if (unlikely(!count))
1427 		return 0;
1428 
1429 	in_use = rxq->next_to_use - rxq->next_to_clean;
1430 	ena_assert_msg(((in_use + count) < ring_size), "bad ring state\n");
1431 
1432 	/* get resources for incoming packets */
1433 	rc = rte_mempool_get_bulk(rxq->mb_pool, (void **)mbufs, count);
1434 	if (unlikely(rc < 0)) {
1435 		rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1436 		++rxq->rx_stats.mbuf_alloc_fail;
1437 		PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1438 		return 0;
1439 	}
1440 
1441 	for (i = 0; i < count; i++) {
1442 		uint16_t next_to_use_masked = next_to_use & ring_mask;
1443 		struct rte_mbuf *mbuf = mbufs[i];
1444 		struct ena_com_buf ebuf;
1445 		struct ena_rx_buffer *rx_info;
1446 
1447 		if (likely((i + 4) < count))
1448 			rte_prefetch0(mbufs[i + 4]);
1449 
1450 		req_id = rxq->empty_rx_reqs[next_to_use_masked];
1451 		rc = validate_rx_req_id(rxq, req_id);
1452 		if (unlikely(rc))
1453 			break;
1454 
1455 		rx_info = &rxq->rx_buffer_info[req_id];
1456 
1457 		/* prepare physical address for DMA transaction */
1458 		ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1459 		ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1460 		/* pass resource to device */
1461 		rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1462 						&ebuf, req_id);
1463 		if (unlikely(rc)) {
1464 			PMD_DRV_LOG(WARNING, "failed adding rx desc\n");
1465 			break;
1466 		}
1467 		rx_info->mbuf = mbuf;
1468 		next_to_use++;
1469 	}
1470 
1471 	if (unlikely(i < count)) {
1472 		PMD_DRV_LOG(WARNING, "refilled rx qid %d with only %d "
1473 			"buffers (from %d)\n", rxq->id, i, count);
1474 		rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbufs[i]),
1475 				     count - i);
1476 		++rxq->rx_stats.refill_partial;
1477 	}
1478 
1479 	/* When we submitted free recources to device... */
1480 	if (likely(i > 0)) {
1481 		/* ...let HW know that it can fill buffers with data. */
1482 		ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1483 
1484 		rxq->next_to_use = next_to_use;
1485 	}
1486 
1487 	return i;
1488 }
1489 
1490 static int ena_device_init(struct ena_com_dev *ena_dev,
1491 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
1492 			   bool *wd_state)
1493 {
1494 	uint32_t aenq_groups;
1495 	int rc;
1496 	bool readless_supported;
1497 
1498 	/* Initialize mmio registers */
1499 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
1500 	if (rc) {
1501 		PMD_DRV_LOG(ERR, "failed to init mmio read less\n");
1502 		return rc;
1503 	}
1504 
1505 	/* The PCIe configuration space revision id indicate if mmio reg
1506 	 * read is disabled.
1507 	 */
1508 	readless_supported =
1509 		!(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1510 			       & ENA_MMIO_DISABLE_REG_READ);
1511 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1512 
1513 	/* reset device */
1514 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1515 	if (rc) {
1516 		PMD_DRV_LOG(ERR, "cannot reset device\n");
1517 		goto err_mmio_read_less;
1518 	}
1519 
1520 	/* check FW version */
1521 	rc = ena_com_validate_version(ena_dev);
1522 	if (rc) {
1523 		PMD_DRV_LOG(ERR, "device version is too low\n");
1524 		goto err_mmio_read_less;
1525 	}
1526 
1527 	ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1528 
1529 	/* ENA device administration layer init */
1530 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
1531 	if (rc) {
1532 		PMD_DRV_LOG(ERR,
1533 			"cannot initialize ena admin queue with device\n");
1534 		goto err_mmio_read_less;
1535 	}
1536 
1537 	/* To enable the msix interrupts the driver needs to know the number
1538 	 * of queues. So the driver uses polling mode to retrieve this
1539 	 * information.
1540 	 */
1541 	ena_com_set_admin_polling_mode(ena_dev, true);
1542 
1543 	ena_config_host_info(ena_dev);
1544 
1545 	/* Get Device Attributes and features */
1546 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1547 	if (rc) {
1548 		PMD_DRV_LOG(ERR,
1549 			"cannot get attribute for ena device rc= %d\n", rc);
1550 		goto err_admin_init;
1551 	}
1552 
1553 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1554 		      BIT(ENA_ADMIN_NOTIFICATION) |
1555 		      BIT(ENA_ADMIN_KEEP_ALIVE) |
1556 		      BIT(ENA_ADMIN_FATAL_ERROR) |
1557 		      BIT(ENA_ADMIN_WARNING);
1558 
1559 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
1560 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1561 	if (rc) {
1562 		PMD_DRV_LOG(ERR, "Cannot configure aenq groups rc: %d\n", rc);
1563 		goto err_admin_init;
1564 	}
1565 
1566 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
1567 
1568 	return 0;
1569 
1570 err_admin_init:
1571 	ena_com_admin_destroy(ena_dev);
1572 
1573 err_mmio_read_less:
1574 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1575 
1576 	return rc;
1577 }
1578 
1579 static void ena_interrupt_handler_rte(void *cb_arg)
1580 {
1581 	struct ena_adapter *adapter = cb_arg;
1582 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1583 
1584 	ena_com_admin_q_comp_intr_handler(ena_dev);
1585 	if (likely(adapter->state != ENA_ADAPTER_STATE_CLOSED))
1586 		ena_com_aenq_intr_handler(ena_dev, adapter);
1587 }
1588 
1589 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
1590 {
1591 	if (!adapter->wd_state)
1592 		return;
1593 
1594 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
1595 		return;
1596 
1597 	if (unlikely((rte_get_timer_cycles() - adapter->timestamp_wd) >=
1598 	    adapter->keep_alive_timeout)) {
1599 		PMD_DRV_LOG(ERR, "Keep alive timeout\n");
1600 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
1601 		adapter->trigger_reset = true;
1602 		++adapter->dev_stats.wd_expired;
1603 	}
1604 }
1605 
1606 /* Check if admin queue is enabled */
1607 static void check_for_admin_com_state(struct ena_adapter *adapter)
1608 {
1609 	if (unlikely(!ena_com_get_admin_running_state(&adapter->ena_dev))) {
1610 		PMD_DRV_LOG(ERR, "ENA admin queue is not in running state!\n");
1611 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
1612 		adapter->trigger_reset = true;
1613 	}
1614 }
1615 
1616 static void ena_timer_wd_callback(__rte_unused struct rte_timer *timer,
1617 				  void *arg)
1618 {
1619 	struct ena_adapter *adapter = arg;
1620 	struct rte_eth_dev *dev = adapter->rte_dev;
1621 
1622 	check_for_missing_keep_alive(adapter);
1623 	check_for_admin_com_state(adapter);
1624 
1625 	if (unlikely(adapter->trigger_reset)) {
1626 		PMD_DRV_LOG(ERR, "Trigger reset is on\n");
1627 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1628 			NULL);
1629 	}
1630 }
1631 
1632 static inline void
1633 set_default_llq_configurations(struct ena_llq_configurations *llq_config,
1634 			       struct ena_admin_feature_llq_desc *llq,
1635 			       bool use_large_llq_hdr)
1636 {
1637 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
1638 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
1639 	llq_config->llq_num_decs_before_header =
1640 		ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
1641 
1642 	if (use_large_llq_hdr &&
1643 	    (llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B)) {
1644 		llq_config->llq_ring_entry_size =
1645 			ENA_ADMIN_LIST_ENTRY_SIZE_256B;
1646 		llq_config->llq_ring_entry_size_value = 256;
1647 	} else {
1648 		llq_config->llq_ring_entry_size =
1649 			ENA_ADMIN_LIST_ENTRY_SIZE_128B;
1650 		llq_config->llq_ring_entry_size_value = 128;
1651 	}
1652 }
1653 
1654 static int
1655 ena_set_queues_placement_policy(struct ena_adapter *adapter,
1656 				struct ena_com_dev *ena_dev,
1657 				struct ena_admin_feature_llq_desc *llq,
1658 				struct ena_llq_configurations *llq_default_configurations)
1659 {
1660 	int rc;
1661 	u32 llq_feature_mask;
1662 
1663 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
1664 	if (!(ena_dev->supported_features & llq_feature_mask)) {
1665 		PMD_DRV_LOG(INFO,
1666 			"LLQ is not supported. Fallback to host mode policy.\n");
1667 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1668 		return 0;
1669 	}
1670 
1671 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
1672 	if (unlikely(rc)) {
1673 		PMD_INIT_LOG(WARNING, "Failed to config dev mode. "
1674 			"Fallback to host mode policy.");
1675 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1676 		return 0;
1677 	}
1678 
1679 	/* Nothing to config, exit */
1680 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1681 		return 0;
1682 
1683 	if (!adapter->dev_mem_base) {
1684 		PMD_DRV_LOG(ERR, "Unable to access LLQ bar resource. "
1685 			"Fallback to host mode policy.\n.");
1686 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1687 		return 0;
1688 	}
1689 
1690 	ena_dev->mem_bar = adapter->dev_mem_base;
1691 
1692 	return 0;
1693 }
1694 
1695 static uint32_t ena_calc_max_io_queue_num(struct ena_com_dev *ena_dev,
1696 	struct ena_com_dev_get_features_ctx *get_feat_ctx)
1697 {
1698 	uint32_t io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
1699 
1700 	/* Regular queues capabilities */
1701 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
1702 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
1703 			&get_feat_ctx->max_queue_ext.max_queue_ext;
1704 		io_rx_num = RTE_MIN(max_queue_ext->max_rx_sq_num,
1705 				    max_queue_ext->max_rx_cq_num);
1706 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
1707 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
1708 	} else {
1709 		struct ena_admin_queue_feature_desc *max_queues =
1710 			&get_feat_ctx->max_queues;
1711 		io_tx_sq_num = max_queues->max_sq_num;
1712 		io_tx_cq_num = max_queues->max_cq_num;
1713 		io_rx_num = RTE_MIN(io_tx_sq_num, io_tx_cq_num);
1714 	}
1715 
1716 	/* In case of LLQ use the llq number in the get feature cmd */
1717 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
1718 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
1719 
1720 	max_num_io_queues = RTE_MIN(ENA_MAX_NUM_IO_QUEUES, io_rx_num);
1721 	max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_sq_num);
1722 	max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_cq_num);
1723 
1724 	if (unlikely(max_num_io_queues == 0)) {
1725 		PMD_DRV_LOG(ERR, "Number of IO queues should not be 0\n");
1726 		return -EFAULT;
1727 	}
1728 
1729 	return max_num_io_queues;
1730 }
1731 
1732 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1733 {
1734 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
1735 	struct rte_pci_device *pci_dev;
1736 	struct rte_intr_handle *intr_handle;
1737 	struct ena_adapter *adapter = eth_dev->data->dev_private;
1738 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1739 	struct ena_com_dev_get_features_ctx get_feat_ctx;
1740 	struct ena_llq_configurations llq_config;
1741 	const char *queue_type_str;
1742 	uint32_t max_num_io_queues;
1743 	int rc;
1744 	static int adapters_found;
1745 	bool disable_meta_caching;
1746 	bool wd_state;
1747 
1748 	eth_dev->dev_ops = &ena_dev_ops;
1749 	eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1750 	eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1751 	eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1752 
1753 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1754 		return 0;
1755 
1756 	memset(adapter, 0, sizeof(struct ena_adapter));
1757 	ena_dev = &adapter->ena_dev;
1758 
1759 	adapter->rte_eth_dev_data = eth_dev->data;
1760 	adapter->rte_dev = eth_dev;
1761 
1762 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1763 	adapter->pdev = pci_dev;
1764 
1765 	PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1766 		     pci_dev->addr.domain,
1767 		     pci_dev->addr.bus,
1768 		     pci_dev->addr.devid,
1769 		     pci_dev->addr.function);
1770 
1771 	intr_handle = &pci_dev->intr_handle;
1772 
1773 	adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1774 	adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1775 
1776 	if (!adapter->regs) {
1777 		PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1778 			     ENA_REGS_BAR);
1779 		return -ENXIO;
1780 	}
1781 
1782 	ena_dev->reg_bar = adapter->regs;
1783 	ena_dev->dmadev = adapter->pdev;
1784 
1785 	adapter->id_number = adapters_found;
1786 
1787 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1788 		 adapter->id_number);
1789 
1790 	rc = ena_parse_devargs(adapter, pci_dev->device.devargs);
1791 	if (rc != 0) {
1792 		PMD_INIT_LOG(CRIT, "Failed to parse devargs\n");
1793 		goto err;
1794 	}
1795 
1796 	/* device specific initialization routine */
1797 	rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
1798 	if (rc) {
1799 		PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1800 		goto err;
1801 	}
1802 	adapter->wd_state = wd_state;
1803 
1804 	set_default_llq_configurations(&llq_config, &get_feat_ctx.llq,
1805 		adapter->use_large_llq_hdr);
1806 	rc = ena_set_queues_placement_policy(adapter, ena_dev,
1807 					     &get_feat_ctx.llq, &llq_config);
1808 	if (unlikely(rc)) {
1809 		PMD_INIT_LOG(CRIT, "Failed to set placement policy");
1810 		return rc;
1811 	}
1812 
1813 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1814 		queue_type_str = "Regular";
1815 	else
1816 		queue_type_str = "Low latency";
1817 	PMD_DRV_LOG(INFO, "Placement policy: %s\n", queue_type_str);
1818 
1819 	calc_queue_ctx.ena_dev = ena_dev;
1820 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
1821 
1822 	max_num_io_queues = ena_calc_max_io_queue_num(ena_dev, &get_feat_ctx);
1823 	rc = ena_calc_io_queue_size(&calc_queue_ctx,
1824 		adapter->use_large_llq_hdr);
1825 	if (unlikely((rc != 0) || (max_num_io_queues == 0))) {
1826 		rc = -EFAULT;
1827 		goto err_device_destroy;
1828 	}
1829 
1830 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
1831 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
1832 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
1833 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
1834 	adapter->max_num_io_queues = max_num_io_queues;
1835 
1836 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1837 		disable_meta_caching =
1838 			!!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
1839 			BIT(ENA_ADMIN_DISABLE_META_CACHING));
1840 	} else {
1841 		disable_meta_caching = false;
1842 	}
1843 
1844 	/* prepare ring structures */
1845 	ena_init_rings(adapter, disable_meta_caching);
1846 
1847 	ena_config_debug_area(adapter);
1848 
1849 	/* Set max MTU for this device */
1850 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1851 
1852 	/* set device support for offloads */
1853 	adapter->offloads.tso4_supported = (get_feat_ctx.offload.tx &
1854 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0;
1855 	adapter->offloads.tx_csum_supported = (get_feat_ctx.offload.tx &
1856 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK) != 0;
1857 	adapter->offloads.rx_csum_supported =
1858 		(get_feat_ctx.offload.rx_supported &
1859 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK) != 0;
1860 
1861 	/* Copy MAC address and point DPDK to it */
1862 	eth_dev->data->mac_addrs = (struct rte_ether_addr *)adapter->mac_addr;
1863 	rte_ether_addr_copy((struct rte_ether_addr *)
1864 			get_feat_ctx.dev_attr.mac_addr,
1865 			(struct rte_ether_addr *)adapter->mac_addr);
1866 
1867 	/*
1868 	 * Pass the information to the rte_eth_dev_close() that it should also
1869 	 * release the private port resources.
1870 	 */
1871 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1872 
1873 	adapter->drv_stats = rte_zmalloc("adapter stats",
1874 					 sizeof(*adapter->drv_stats),
1875 					 RTE_CACHE_LINE_SIZE);
1876 	if (!adapter->drv_stats) {
1877 		PMD_DRV_LOG(ERR, "failed to alloc mem for adapter stats\n");
1878 		rc = -ENOMEM;
1879 		goto err_delete_debug_area;
1880 	}
1881 
1882 	rte_intr_callback_register(intr_handle,
1883 				   ena_interrupt_handler_rte,
1884 				   adapter);
1885 	rte_intr_enable(intr_handle);
1886 	ena_com_set_admin_polling_mode(ena_dev, false);
1887 	ena_com_admin_aenq_enable(ena_dev);
1888 
1889 	if (adapters_found == 0)
1890 		rte_timer_subsystem_init();
1891 	rte_timer_init(&adapter->timer_wd);
1892 
1893 	adapters_found++;
1894 	adapter->state = ENA_ADAPTER_STATE_INIT;
1895 
1896 	return 0;
1897 
1898 err_delete_debug_area:
1899 	ena_com_delete_debug_area(ena_dev);
1900 
1901 err_device_destroy:
1902 	ena_com_delete_host_info(ena_dev);
1903 	ena_com_admin_destroy(ena_dev);
1904 
1905 err:
1906 	return rc;
1907 }
1908 
1909 static void ena_destroy_device(struct rte_eth_dev *eth_dev)
1910 {
1911 	struct ena_adapter *adapter = eth_dev->data->dev_private;
1912 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1913 
1914 	if (adapter->state == ENA_ADAPTER_STATE_FREE)
1915 		return;
1916 
1917 	ena_com_set_admin_running_state(ena_dev, false);
1918 
1919 	if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1920 		ena_close(eth_dev);
1921 
1922 	ena_com_delete_debug_area(ena_dev);
1923 	ena_com_delete_host_info(ena_dev);
1924 
1925 	ena_com_abort_admin_commands(ena_dev);
1926 	ena_com_wait_for_abort_completion(ena_dev);
1927 	ena_com_admin_destroy(ena_dev);
1928 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1929 
1930 	adapter->state = ENA_ADAPTER_STATE_FREE;
1931 }
1932 
1933 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1934 {
1935 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1936 		return 0;
1937 
1938 	ena_destroy_device(eth_dev);
1939 
1940 	eth_dev->dev_ops = NULL;
1941 	eth_dev->rx_pkt_burst = NULL;
1942 	eth_dev->tx_pkt_burst = NULL;
1943 	eth_dev->tx_pkt_prepare = NULL;
1944 
1945 	return 0;
1946 }
1947 
1948 static int ena_dev_configure(struct rte_eth_dev *dev)
1949 {
1950 	struct ena_adapter *adapter = dev->data->dev_private;
1951 
1952 	adapter->state = ENA_ADAPTER_STATE_CONFIG;
1953 
1954 	adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1955 	adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1956 	return 0;
1957 }
1958 
1959 static void ena_init_rings(struct ena_adapter *adapter,
1960 			   bool disable_meta_caching)
1961 {
1962 	size_t i;
1963 
1964 	for (i = 0; i < adapter->max_num_io_queues; i++) {
1965 		struct ena_ring *ring = &adapter->tx_ring[i];
1966 
1967 		ring->configured = 0;
1968 		ring->type = ENA_RING_TYPE_TX;
1969 		ring->adapter = adapter;
1970 		ring->id = i;
1971 		ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1972 		ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1973 		ring->sgl_size = adapter->max_tx_sgl_size;
1974 		ring->disable_meta_caching = disable_meta_caching;
1975 	}
1976 
1977 	for (i = 0; i < adapter->max_num_io_queues; i++) {
1978 		struct ena_ring *ring = &adapter->rx_ring[i];
1979 
1980 		ring->configured = 0;
1981 		ring->type = ENA_RING_TYPE_RX;
1982 		ring->adapter = adapter;
1983 		ring->id = i;
1984 		ring->sgl_size = adapter->max_rx_sgl_size;
1985 	}
1986 }
1987 
1988 static int ena_infos_get(struct rte_eth_dev *dev,
1989 			  struct rte_eth_dev_info *dev_info)
1990 {
1991 	struct ena_adapter *adapter;
1992 	struct ena_com_dev *ena_dev;
1993 	uint64_t rx_feat = 0, tx_feat = 0;
1994 
1995 	ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
1996 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
1997 	adapter = dev->data->dev_private;
1998 
1999 	ena_dev = &adapter->ena_dev;
2000 	ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
2001 
2002 	dev_info->speed_capa =
2003 			ETH_LINK_SPEED_1G   |
2004 			ETH_LINK_SPEED_2_5G |
2005 			ETH_LINK_SPEED_5G   |
2006 			ETH_LINK_SPEED_10G  |
2007 			ETH_LINK_SPEED_25G  |
2008 			ETH_LINK_SPEED_40G  |
2009 			ETH_LINK_SPEED_50G  |
2010 			ETH_LINK_SPEED_100G;
2011 
2012 	/* Set Tx & Rx features available for device */
2013 	if (adapter->offloads.tso4_supported)
2014 		tx_feat	|= DEV_TX_OFFLOAD_TCP_TSO;
2015 
2016 	if (adapter->offloads.tx_csum_supported)
2017 		tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
2018 			DEV_TX_OFFLOAD_UDP_CKSUM |
2019 			DEV_TX_OFFLOAD_TCP_CKSUM;
2020 
2021 	if (adapter->offloads.rx_csum_supported)
2022 		rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
2023 			DEV_RX_OFFLOAD_UDP_CKSUM  |
2024 			DEV_RX_OFFLOAD_TCP_CKSUM;
2025 
2026 	rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2027 
2028 	/* Inform framework about available features */
2029 	dev_info->rx_offload_capa = rx_feat;
2030 	dev_info->rx_queue_offload_capa = rx_feat;
2031 	dev_info->tx_offload_capa = tx_feat;
2032 	dev_info->tx_queue_offload_capa = tx_feat;
2033 
2034 	dev_info->flow_type_rss_offloads = ETH_RSS_IP | ETH_RSS_TCP |
2035 					   ETH_RSS_UDP;
2036 
2037 	dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
2038 	dev_info->max_rx_pktlen  = adapter->max_mtu;
2039 	dev_info->max_mac_addrs = 1;
2040 
2041 	dev_info->max_rx_queues = adapter->max_num_io_queues;
2042 	dev_info->max_tx_queues = adapter->max_num_io_queues;
2043 	dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
2044 
2045 	adapter->tx_supported_offloads = tx_feat;
2046 	adapter->rx_supported_offloads = rx_feat;
2047 
2048 	dev_info->rx_desc_lim.nb_max = adapter->max_rx_ring_size;
2049 	dev_info->rx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2050 	dev_info->rx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2051 					adapter->max_rx_sgl_size);
2052 	dev_info->rx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2053 					adapter->max_rx_sgl_size);
2054 
2055 	dev_info->tx_desc_lim.nb_max = adapter->max_tx_ring_size;
2056 	dev_info->tx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2057 	dev_info->tx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2058 					adapter->max_tx_sgl_size);
2059 	dev_info->tx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2060 					adapter->max_tx_sgl_size);
2061 
2062 	return 0;
2063 }
2064 
2065 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len)
2066 {
2067 	mbuf->data_len = len;
2068 	mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2069 	mbuf->refcnt = 1;
2070 	mbuf->next = NULL;
2071 }
2072 
2073 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
2074 				    struct ena_com_rx_buf_info *ena_bufs,
2075 				    uint32_t descs,
2076 				    uint16_t *next_to_clean,
2077 				    uint8_t offset)
2078 {
2079 	struct rte_mbuf *mbuf;
2080 	struct rte_mbuf *mbuf_head;
2081 	struct ena_rx_buffer *rx_info;
2082 	unsigned int ring_mask = rx_ring->ring_size - 1;
2083 	uint16_t ntc, len, req_id, buf = 0;
2084 
2085 	if (unlikely(descs == 0))
2086 		return NULL;
2087 
2088 	ntc = *next_to_clean;
2089 
2090 	len = ena_bufs[buf].len;
2091 	req_id = ena_bufs[buf].req_id;
2092 	if (unlikely(validate_rx_req_id(rx_ring, req_id)))
2093 		return NULL;
2094 
2095 	rx_info = &rx_ring->rx_buffer_info[req_id];
2096 
2097 	mbuf = rx_info->mbuf;
2098 	RTE_ASSERT(mbuf != NULL);
2099 
2100 	ena_init_rx_mbuf(mbuf, len);
2101 
2102 	/* Fill the mbuf head with the data specific for 1st segment. */
2103 	mbuf_head = mbuf;
2104 	mbuf_head->nb_segs = descs;
2105 	mbuf_head->port = rx_ring->port_id;
2106 	mbuf_head->pkt_len = len;
2107 	mbuf_head->data_off += offset;
2108 
2109 	rx_info->mbuf = NULL;
2110 	rx_ring->empty_rx_reqs[ntc & ring_mask] = req_id;
2111 	++ntc;
2112 
2113 	while (--descs) {
2114 		++buf;
2115 		len = ena_bufs[buf].len;
2116 		req_id = ena_bufs[buf].req_id;
2117 		if (unlikely(validate_rx_req_id(rx_ring, req_id))) {
2118 			rte_mbuf_raw_free(mbuf_head);
2119 			return NULL;
2120 		}
2121 
2122 		rx_info = &rx_ring->rx_buffer_info[req_id];
2123 		RTE_ASSERT(rx_info->mbuf != NULL);
2124 
2125 		/* Create an mbuf chain. */
2126 		mbuf->next = rx_info->mbuf;
2127 		mbuf = mbuf->next;
2128 
2129 		ena_init_rx_mbuf(mbuf, len);
2130 		mbuf_head->pkt_len += len;
2131 
2132 		rx_info->mbuf = NULL;
2133 		rx_ring->empty_rx_reqs[ntc & ring_mask] = req_id;
2134 		++ntc;
2135 	}
2136 
2137 	*next_to_clean = ntc;
2138 
2139 	return mbuf_head;
2140 }
2141 
2142 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
2143 				  uint16_t nb_pkts)
2144 {
2145 	struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
2146 	unsigned int ring_size = rx_ring->ring_size;
2147 	unsigned int ring_mask = ring_size - 1;
2148 	uint16_t next_to_clean = rx_ring->next_to_clean;
2149 	uint16_t desc_in_use = 0;
2150 	struct rte_mbuf *mbuf;
2151 	uint16_t completed;
2152 	struct ena_com_rx_ctx ena_rx_ctx;
2153 	int i, rc = 0;
2154 
2155 	/* Check adapter state */
2156 	if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2157 		PMD_DRV_LOG(ALERT,
2158 			"Trying to receive pkts while device is NOT running\n");
2159 		return 0;
2160 	}
2161 
2162 	desc_in_use = rx_ring->next_to_use - next_to_clean;
2163 	if (unlikely(nb_pkts > desc_in_use))
2164 		nb_pkts = desc_in_use;
2165 
2166 	for (completed = 0; completed < nb_pkts; completed++) {
2167 		ena_rx_ctx.max_bufs = rx_ring->sgl_size;
2168 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
2169 		ena_rx_ctx.descs = 0;
2170 		ena_rx_ctx.pkt_offset = 0;
2171 		/* receive packet context */
2172 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
2173 				    rx_ring->ena_com_io_sq,
2174 				    &ena_rx_ctx);
2175 		if (unlikely(rc)) {
2176 			PMD_DRV_LOG(ERR, "ena_com_rx_pkt error %d\n", rc);
2177 			rx_ring->adapter->reset_reason =
2178 				ENA_REGS_RESET_TOO_MANY_RX_DESCS;
2179 			rx_ring->adapter->trigger_reset = true;
2180 			++rx_ring->rx_stats.bad_desc_num;
2181 			return 0;
2182 		}
2183 
2184 		mbuf = ena_rx_mbuf(rx_ring,
2185 			ena_rx_ctx.ena_bufs,
2186 			ena_rx_ctx.descs,
2187 			&next_to_clean,
2188 			ena_rx_ctx.pkt_offset);
2189 		if (unlikely(mbuf == NULL)) {
2190 			for (i = 0; i < ena_rx_ctx.descs; ++i) {
2191 				rx_ring->empty_rx_reqs[next_to_clean & ring_mask] =
2192 					rx_ring->ena_bufs[i].req_id;
2193 				++next_to_clean;
2194 			}
2195 			break;
2196 		}
2197 
2198 		/* fill mbuf attributes if any */
2199 		ena_rx_mbuf_prepare(mbuf, &ena_rx_ctx);
2200 
2201 		if (unlikely(mbuf->ol_flags &
2202 				(PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD))) {
2203 			rte_atomic64_inc(&rx_ring->adapter->drv_stats->ierrors);
2204 			++rx_ring->rx_stats.bad_csum;
2205 		}
2206 
2207 		mbuf->hash.rss = ena_rx_ctx.hash;
2208 
2209 		rx_pkts[completed] = mbuf;
2210 		rx_ring->rx_stats.bytes += mbuf->pkt_len;
2211 	}
2212 
2213 	rx_ring->rx_stats.cnt += completed;
2214 	rx_ring->next_to_clean = next_to_clean;
2215 
2216 	desc_in_use = desc_in_use - completed + 1;
2217 	/* Burst refill to save doorbells, memory barriers, const interval */
2218 	if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size)) {
2219 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
2220 		ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
2221 	}
2222 
2223 	return completed;
2224 }
2225 
2226 static uint16_t
2227 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2228 		uint16_t nb_pkts)
2229 {
2230 	int32_t ret;
2231 	uint32_t i;
2232 	struct rte_mbuf *m;
2233 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2234 	struct rte_ipv4_hdr *ip_hdr;
2235 	uint64_t ol_flags;
2236 	uint16_t frag_field;
2237 
2238 	for (i = 0; i != nb_pkts; i++) {
2239 		m = tx_pkts[i];
2240 		ol_flags = m->ol_flags;
2241 
2242 		if (!(ol_flags & PKT_TX_IPV4))
2243 			continue;
2244 
2245 		/* If there was not L2 header length specified, assume it is
2246 		 * length of the ethernet header.
2247 		 */
2248 		if (unlikely(m->l2_len == 0))
2249 			m->l2_len = sizeof(struct rte_ether_hdr);
2250 
2251 		ip_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
2252 						 m->l2_len);
2253 		frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
2254 
2255 		if ((frag_field & RTE_IPV4_HDR_DF_FLAG) != 0) {
2256 			m->packet_type |= RTE_PTYPE_L4_NONFRAG;
2257 
2258 			/* If IPv4 header has DF flag enabled and TSO support is
2259 			 * disabled, partial chcecksum should not be calculated.
2260 			 */
2261 			if (!tx_ring->adapter->offloads.tso4_supported)
2262 				continue;
2263 		}
2264 
2265 		if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
2266 				(ol_flags & PKT_TX_L4_MASK) ==
2267 				PKT_TX_SCTP_CKSUM) {
2268 			rte_errno = ENOTSUP;
2269 			return i;
2270 		}
2271 
2272 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2273 		ret = rte_validate_tx_offload(m);
2274 		if (ret != 0) {
2275 			rte_errno = -ret;
2276 			return i;
2277 		}
2278 #endif
2279 
2280 		/* In case we are supposed to TSO and have DF not set (DF=0)
2281 		 * hardware must be provided with partial checksum, otherwise
2282 		 * it will take care of necessary calculations.
2283 		 */
2284 
2285 		ret = rte_net_intel_cksum_flags_prepare(m,
2286 			ol_flags & ~PKT_TX_TCP_SEG);
2287 		if (ret != 0) {
2288 			rte_errno = -ret;
2289 			return i;
2290 		}
2291 	}
2292 
2293 	return i;
2294 }
2295 
2296 static void ena_update_hints(struct ena_adapter *adapter,
2297 			     struct ena_admin_ena_hw_hints *hints)
2298 {
2299 	if (hints->admin_completion_tx_timeout)
2300 		adapter->ena_dev.admin_queue.completion_timeout =
2301 			hints->admin_completion_tx_timeout * 1000;
2302 
2303 	if (hints->mmio_read_timeout)
2304 		/* convert to usec */
2305 		adapter->ena_dev.mmio_read.reg_read_to =
2306 			hints->mmio_read_timeout * 1000;
2307 
2308 	if (hints->driver_watchdog_timeout) {
2309 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2310 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2311 		else
2312 			// Convert msecs to ticks
2313 			adapter->keep_alive_timeout =
2314 				(hints->driver_watchdog_timeout *
2315 				rte_get_timer_hz()) / 1000;
2316 	}
2317 }
2318 
2319 static int ena_check_and_linearize_mbuf(struct ena_ring *tx_ring,
2320 					struct rte_mbuf *mbuf)
2321 {
2322 	struct ena_com_dev *ena_dev;
2323 	int num_segments, header_len, rc;
2324 
2325 	ena_dev = &tx_ring->adapter->ena_dev;
2326 	num_segments = mbuf->nb_segs;
2327 	header_len = mbuf->data_len;
2328 
2329 	if (likely(num_segments < tx_ring->sgl_size))
2330 		return 0;
2331 
2332 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV &&
2333 	    (num_segments == tx_ring->sgl_size) &&
2334 	    (header_len < tx_ring->tx_max_header_size))
2335 		return 0;
2336 
2337 	++tx_ring->tx_stats.linearize;
2338 	rc = rte_pktmbuf_linearize(mbuf);
2339 	if (unlikely(rc)) {
2340 		PMD_DRV_LOG(WARNING, "Mbuf linearize failed\n");
2341 		rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
2342 		++tx_ring->tx_stats.linearize_failed;
2343 		return rc;
2344 	}
2345 
2346 	return rc;
2347 }
2348 
2349 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2350 				  uint16_t nb_pkts)
2351 {
2352 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2353 	uint16_t next_to_use = tx_ring->next_to_use;
2354 	uint16_t next_to_clean = tx_ring->next_to_clean;
2355 	struct rte_mbuf *mbuf;
2356 	uint16_t seg_len;
2357 	unsigned int ring_size = tx_ring->ring_size;
2358 	unsigned int ring_mask = ring_size - 1;
2359 	struct ena_com_tx_ctx ena_tx_ctx;
2360 	struct ena_tx_buffer *tx_info;
2361 	struct ena_com_buf *ebuf;
2362 	uint16_t rc, req_id, total_tx_descs = 0;
2363 	uint16_t sent_idx = 0, empty_tx_reqs;
2364 	uint16_t push_len = 0;
2365 	uint16_t delta = 0;
2366 	int nb_hw_desc;
2367 	uint32_t total_length;
2368 
2369 	/* Check adapter state */
2370 	if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2371 		PMD_DRV_LOG(ALERT,
2372 			"Trying to xmit pkts while device is NOT running\n");
2373 		return 0;
2374 	}
2375 
2376 	empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
2377 	if (nb_pkts > empty_tx_reqs)
2378 		nb_pkts = empty_tx_reqs;
2379 
2380 	for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
2381 		mbuf = tx_pkts[sent_idx];
2382 		total_length = 0;
2383 
2384 		rc = ena_check_and_linearize_mbuf(tx_ring, mbuf);
2385 		if (unlikely(rc))
2386 			break;
2387 
2388 		req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
2389 		tx_info = &tx_ring->tx_buffer_info[req_id];
2390 		tx_info->mbuf = mbuf;
2391 		tx_info->num_of_bufs = 0;
2392 		ebuf = tx_info->bufs;
2393 
2394 		/* Prepare TX context */
2395 		memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2396 		memset(&ena_tx_ctx.ena_meta, 0x0,
2397 		       sizeof(struct ena_com_tx_meta));
2398 		ena_tx_ctx.ena_bufs = ebuf;
2399 		ena_tx_ctx.req_id = req_id;
2400 
2401 		delta = 0;
2402 		seg_len = mbuf->data_len;
2403 
2404 		if (tx_ring->tx_mem_queue_type ==
2405 				ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2406 			push_len = RTE_MIN(mbuf->pkt_len,
2407 					   tx_ring->tx_max_header_size);
2408 			ena_tx_ctx.header_len = push_len;
2409 
2410 			if (likely(push_len <= seg_len)) {
2411 				/* If the push header is in the single segment,
2412 				 * then just point it to the 1st mbuf data.
2413 				 */
2414 				ena_tx_ctx.push_header =
2415 					rte_pktmbuf_mtod(mbuf, uint8_t *);
2416 			} else {
2417 				/* If the push header lays in the several
2418 				 * segments, copy it to the intermediate buffer.
2419 				 */
2420 				rte_pktmbuf_read(mbuf, 0, push_len,
2421 					tx_ring->push_buf_intermediate_buf);
2422 				ena_tx_ctx.push_header =
2423 					tx_ring->push_buf_intermediate_buf;
2424 				delta = push_len - seg_len;
2425 			}
2426 		} /* there's no else as we take advantage of memset zeroing */
2427 
2428 		/* Set TX offloads flags, if applicable */
2429 		ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads,
2430 			tx_ring->disable_meta_caching);
2431 
2432 		rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
2433 
2434 		/* Process first segment taking into
2435 		 * consideration pushed header
2436 		 */
2437 		if (seg_len > push_len) {
2438 			ebuf->paddr = mbuf->buf_iova +
2439 				      mbuf->data_off +
2440 				      push_len;
2441 			ebuf->len = seg_len - push_len;
2442 			ebuf++;
2443 			tx_info->num_of_bufs++;
2444 		}
2445 		total_length += mbuf->data_len;
2446 
2447 		while ((mbuf = mbuf->next) != NULL) {
2448 			seg_len = mbuf->data_len;
2449 
2450 			/* Skip mbufs if whole data is pushed as a header */
2451 			if (unlikely(delta > seg_len)) {
2452 				delta -= seg_len;
2453 				continue;
2454 			}
2455 
2456 			ebuf->paddr = mbuf->buf_iova + mbuf->data_off + delta;
2457 			ebuf->len = seg_len - delta;
2458 			total_length += ebuf->len;
2459 			ebuf++;
2460 			tx_info->num_of_bufs++;
2461 
2462 			delta = 0;
2463 		}
2464 
2465 		ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2466 
2467 		if (ena_com_is_doorbell_needed(tx_ring->ena_com_io_sq,
2468 					       &ena_tx_ctx)) {
2469 			PMD_DRV_LOG(DEBUG, "llq tx max burst size of queue %d"
2470 				" achieved, writing doorbell to send burst\n",
2471 				tx_ring->id);
2472 			ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2473 		}
2474 
2475 		/* prepare the packet's descriptors to dma engine */
2476 		rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
2477 					&ena_tx_ctx, &nb_hw_desc);
2478 		if (unlikely(rc)) {
2479 			++tx_ring->tx_stats.prepare_ctx_err;
2480 			break;
2481 		}
2482 		tx_info->tx_descs = nb_hw_desc;
2483 
2484 		next_to_use++;
2485 		tx_ring->tx_stats.cnt++;
2486 		tx_ring->tx_stats.bytes += total_length;
2487 	}
2488 	tx_ring->tx_stats.available_desc =
2489 		ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2490 
2491 	/* If there are ready packets to be xmitted... */
2492 	if (sent_idx > 0) {
2493 		/* ...let HW do its best :-) */
2494 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2495 		tx_ring->tx_stats.doorbells++;
2496 		tx_ring->next_to_use = next_to_use;
2497 	}
2498 
2499 	/* Clear complete packets  */
2500 	while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
2501 		rc = validate_tx_req_id(tx_ring, req_id);
2502 		if (rc)
2503 			break;
2504 
2505 		/* Get Tx info & store how many descs were processed  */
2506 		tx_info = &tx_ring->tx_buffer_info[req_id];
2507 		total_tx_descs += tx_info->tx_descs;
2508 
2509 		/* Free whole mbuf chain  */
2510 		mbuf = tx_info->mbuf;
2511 		rte_pktmbuf_free(mbuf);
2512 		tx_info->mbuf = NULL;
2513 
2514 		/* Put back descriptor to the ring for reuse */
2515 		tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
2516 		next_to_clean++;
2517 
2518 		/* If too many descs to clean, leave it for another run */
2519 		if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
2520 			break;
2521 	}
2522 	tx_ring->tx_stats.available_desc =
2523 		ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2524 
2525 	if (total_tx_descs > 0) {
2526 		/* acknowledge completion of sent packets */
2527 		tx_ring->next_to_clean = next_to_clean;
2528 		ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
2529 		ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
2530 	}
2531 
2532 	tx_ring->tx_stats.tx_poll++;
2533 
2534 	return sent_idx;
2535 }
2536 
2537 /**
2538  * DPDK callback to retrieve names of extended device statistics
2539  *
2540  * @param dev
2541  *   Pointer to Ethernet device structure.
2542  * @param[out] xstats_names
2543  *   Buffer to insert names into.
2544  * @param n
2545  *   Number of names.
2546  *
2547  * @return
2548  *   Number of xstats names.
2549  */
2550 static int ena_xstats_get_names(struct rte_eth_dev *dev,
2551 				struct rte_eth_xstat_name *xstats_names,
2552 				unsigned int n)
2553 {
2554 	unsigned int xstats_count = ena_xstats_calc_num(dev);
2555 	unsigned int stat, i, count = 0;
2556 
2557 	if (n < xstats_count || !xstats_names)
2558 		return xstats_count;
2559 
2560 	for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++)
2561 		strcpy(xstats_names[count].name,
2562 			ena_stats_global_strings[stat].name);
2563 
2564 	for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++)
2565 		for (i = 0; i < dev->data->nb_rx_queues; i++, count++)
2566 			snprintf(xstats_names[count].name,
2567 				sizeof(xstats_names[count].name),
2568 				"rx_q%d_%s", i,
2569 				ena_stats_rx_strings[stat].name);
2570 
2571 	for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++)
2572 		for (i = 0; i < dev->data->nb_tx_queues; i++, count++)
2573 			snprintf(xstats_names[count].name,
2574 				sizeof(xstats_names[count].name),
2575 				"tx_q%d_%s", i,
2576 				ena_stats_tx_strings[stat].name);
2577 
2578 	return xstats_count;
2579 }
2580 
2581 /**
2582  * DPDK callback to get extended device statistics.
2583  *
2584  * @param dev
2585  *   Pointer to Ethernet device structure.
2586  * @param[out] stats
2587  *   Stats table output buffer.
2588  * @param n
2589  *   The size of the stats table.
2590  *
2591  * @return
2592  *   Number of xstats on success, negative on failure.
2593  */
2594 static int ena_xstats_get(struct rte_eth_dev *dev,
2595 			  struct rte_eth_xstat *xstats,
2596 			  unsigned int n)
2597 {
2598 	struct ena_adapter *adapter = dev->data->dev_private;
2599 	unsigned int xstats_count = ena_xstats_calc_num(dev);
2600 	unsigned int stat, i, count = 0;
2601 	int stat_offset;
2602 	void *stats_begin;
2603 
2604 	if (n < xstats_count)
2605 		return xstats_count;
2606 
2607 	if (!xstats)
2608 		return 0;
2609 
2610 	for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++) {
2611 		stat_offset = ena_stats_rx_strings[stat].stat_offset;
2612 		stats_begin = &adapter->dev_stats;
2613 
2614 		xstats[count].id = count;
2615 		xstats[count].value = *((uint64_t *)
2616 			((char *)stats_begin + stat_offset));
2617 	}
2618 
2619 	for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++) {
2620 		for (i = 0; i < dev->data->nb_rx_queues; i++, count++) {
2621 			stat_offset = ena_stats_rx_strings[stat].stat_offset;
2622 			stats_begin = &adapter->rx_ring[i].rx_stats;
2623 
2624 			xstats[count].id = count;
2625 			xstats[count].value = *((uint64_t *)
2626 				((char *)stats_begin + stat_offset));
2627 		}
2628 	}
2629 
2630 	for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++) {
2631 		for (i = 0; i < dev->data->nb_tx_queues; i++, count++) {
2632 			stat_offset = ena_stats_tx_strings[stat].stat_offset;
2633 			stats_begin = &adapter->tx_ring[i].rx_stats;
2634 
2635 			xstats[count].id = count;
2636 			xstats[count].value = *((uint64_t *)
2637 				((char *)stats_begin + stat_offset));
2638 		}
2639 	}
2640 
2641 	return count;
2642 }
2643 
2644 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
2645 				const uint64_t *ids,
2646 				uint64_t *values,
2647 				unsigned int n)
2648 {
2649 	struct ena_adapter *adapter = dev->data->dev_private;
2650 	uint64_t id;
2651 	uint64_t rx_entries, tx_entries;
2652 	unsigned int i;
2653 	int qid;
2654 	int valid = 0;
2655 	for (i = 0; i < n; ++i) {
2656 		id = ids[i];
2657 		/* Check if id belongs to global statistics */
2658 		if (id < ENA_STATS_ARRAY_GLOBAL) {
2659 			values[i] = *((uint64_t *)&adapter->dev_stats + id);
2660 			++valid;
2661 			continue;
2662 		}
2663 
2664 		/* Check if id belongs to rx queue statistics */
2665 		id -= ENA_STATS_ARRAY_GLOBAL;
2666 		rx_entries = ENA_STATS_ARRAY_RX * dev->data->nb_rx_queues;
2667 		if (id < rx_entries) {
2668 			qid = id % dev->data->nb_rx_queues;
2669 			id /= dev->data->nb_rx_queues;
2670 			values[i] = *((uint64_t *)
2671 				&adapter->rx_ring[qid].rx_stats + id);
2672 			++valid;
2673 			continue;
2674 		}
2675 				/* Check if id belongs to rx queue statistics */
2676 		id -= rx_entries;
2677 		tx_entries = ENA_STATS_ARRAY_TX * dev->data->nb_tx_queues;
2678 		if (id < tx_entries) {
2679 			qid = id % dev->data->nb_tx_queues;
2680 			id /= dev->data->nb_tx_queues;
2681 			values[i] = *((uint64_t *)
2682 				&adapter->tx_ring[qid].tx_stats + id);
2683 			++valid;
2684 			continue;
2685 		}
2686 	}
2687 
2688 	return valid;
2689 }
2690 
2691 static int ena_process_bool_devarg(const char *key,
2692 				   const char *value,
2693 				   void *opaque)
2694 {
2695 	struct ena_adapter *adapter = opaque;
2696 	bool bool_value;
2697 
2698 	/* Parse the value. */
2699 	if (strcmp(value, "1") == 0) {
2700 		bool_value = true;
2701 	} else if (strcmp(value, "0") == 0) {
2702 		bool_value = false;
2703 	} else {
2704 		PMD_INIT_LOG(ERR,
2705 			"Invalid value: '%s' for key '%s'. Accepted: '0' or '1'\n",
2706 			value, key);
2707 		return -EINVAL;
2708 	}
2709 
2710 	/* Now, assign it to the proper adapter field. */
2711 	if (strcmp(key, ENA_DEVARG_LARGE_LLQ_HDR))
2712 		adapter->use_large_llq_hdr = bool_value;
2713 
2714 	return 0;
2715 }
2716 
2717 static int ena_parse_devargs(struct ena_adapter *adapter,
2718 			     struct rte_devargs *devargs)
2719 {
2720 	static const char * const allowed_args[] = {
2721 		ENA_DEVARG_LARGE_LLQ_HDR,
2722 	};
2723 	struct rte_kvargs *kvlist;
2724 	int rc;
2725 
2726 	if (devargs == NULL)
2727 		return 0;
2728 
2729 	kvlist = rte_kvargs_parse(devargs->args, allowed_args);
2730 	if (kvlist == NULL) {
2731 		PMD_INIT_LOG(ERR, "Invalid device arguments: %s\n",
2732 			devargs->args);
2733 		return -EINVAL;
2734 	}
2735 
2736 	rc = rte_kvargs_process(kvlist, ENA_DEVARG_LARGE_LLQ_HDR,
2737 		ena_process_bool_devarg, adapter);
2738 
2739 	rte_kvargs_free(kvlist);
2740 
2741 	return rc;
2742 }
2743 
2744 /*********************************************************************
2745  *  PMD configuration
2746  *********************************************************************/
2747 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2748 	struct rte_pci_device *pci_dev)
2749 {
2750 	return rte_eth_dev_pci_generic_probe(pci_dev,
2751 		sizeof(struct ena_adapter), eth_ena_dev_init);
2752 }
2753 
2754 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
2755 {
2756 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
2757 }
2758 
2759 static struct rte_pci_driver rte_ena_pmd = {
2760 	.id_table = pci_id_ena_map,
2761 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
2762 		     RTE_PCI_DRV_WC_ACTIVATE,
2763 	.probe = eth_ena_pci_probe,
2764 	.remove = eth_ena_pci_remove,
2765 };
2766 
2767 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
2768 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
2769 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
2770 RTE_PMD_REGISTER_PARAM_STRING(net_ena, ENA_DEVARG_LARGE_LLQ_HDR "=<0|1>");
2771 
2772 RTE_INIT(ena_init_log)
2773 {
2774 	ena_logtype_init = rte_log_register("pmd.net.ena.init");
2775 	if (ena_logtype_init >= 0)
2776 		rte_log_set_level(ena_logtype_init, RTE_LOG_NOTICE);
2777 	ena_logtype_driver = rte_log_register("pmd.net.ena.driver");
2778 	if (ena_logtype_driver >= 0)
2779 		rte_log_set_level(ena_logtype_driver, RTE_LOG_NOTICE);
2780 
2781 #ifdef RTE_LIBRTE_ENA_DEBUG_RX
2782 	ena_logtype_rx = rte_log_register("pmd.net.ena.rx");
2783 	if (ena_logtype_rx >= 0)
2784 		rte_log_set_level(ena_logtype_rx, RTE_LOG_NOTICE);
2785 #endif
2786 
2787 #ifdef RTE_LIBRTE_ENA_DEBUG_TX
2788 	ena_logtype_tx = rte_log_register("pmd.net.ena.tx");
2789 	if (ena_logtype_tx >= 0)
2790 		rte_log_set_level(ena_logtype_tx, RTE_LOG_NOTICE);
2791 #endif
2792 
2793 #ifdef RTE_LIBRTE_ENA_DEBUG_TX_FREE
2794 	ena_logtype_tx_free = rte_log_register("pmd.net.ena.tx_free");
2795 	if (ena_logtype_tx_free >= 0)
2796 		rte_log_set_level(ena_logtype_tx_free, RTE_LOG_NOTICE);
2797 #endif
2798 
2799 #ifdef RTE_LIBRTE_ENA_COM_DEBUG
2800 	ena_logtype_com = rte_log_register("pmd.net.ena.com");
2801 	if (ena_logtype_com >= 0)
2802 		rte_log_set_level(ena_logtype_com, RTE_LOG_NOTICE);
2803 #endif
2804 }
2805 
2806 /******************************************************************************
2807  ******************************** AENQ Handlers *******************************
2808  *****************************************************************************/
2809 static void ena_update_on_link_change(void *adapter_data,
2810 				      struct ena_admin_aenq_entry *aenq_e)
2811 {
2812 	struct rte_eth_dev *eth_dev;
2813 	struct ena_adapter *adapter;
2814 	struct ena_admin_aenq_link_change_desc *aenq_link_desc;
2815 	uint32_t status;
2816 
2817 	adapter = adapter_data;
2818 	aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
2819 	eth_dev = adapter->rte_dev;
2820 
2821 	status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
2822 	adapter->link_status = status;
2823 
2824 	ena_link_update(eth_dev, 0);
2825 	_rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2826 }
2827 
2828 static void ena_notification(void *data,
2829 			     struct ena_admin_aenq_entry *aenq_e)
2830 {
2831 	struct ena_adapter *adapter = data;
2832 	struct ena_admin_ena_hw_hints *hints;
2833 
2834 	if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
2835 		PMD_DRV_LOG(WARNING, "Invalid group(%x) expected %x\n",
2836 			aenq_e->aenq_common_desc.group,
2837 			ENA_ADMIN_NOTIFICATION);
2838 
2839 	switch (aenq_e->aenq_common_desc.syndrom) {
2840 	case ENA_ADMIN_UPDATE_HINTS:
2841 		hints = (struct ena_admin_ena_hw_hints *)
2842 			(&aenq_e->inline_data_w4);
2843 		ena_update_hints(adapter, hints);
2844 		break;
2845 	default:
2846 		PMD_DRV_LOG(ERR, "Invalid aenq notification link state %d\n",
2847 			aenq_e->aenq_common_desc.syndrom);
2848 	}
2849 }
2850 
2851 static void ena_keep_alive(void *adapter_data,
2852 			   __rte_unused struct ena_admin_aenq_entry *aenq_e)
2853 {
2854 	struct ena_adapter *adapter = adapter_data;
2855 	struct ena_admin_aenq_keep_alive_desc *desc;
2856 	uint64_t rx_drops;
2857 	uint64_t tx_drops;
2858 
2859 	adapter->timestamp_wd = rte_get_timer_cycles();
2860 
2861 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
2862 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
2863 	tx_drops = ((uint64_t)desc->tx_drops_high << 32) | desc->tx_drops_low;
2864 
2865 	adapter->drv_stats->rx_drops = rx_drops;
2866 	adapter->dev_stats.tx_drops = tx_drops;
2867 }
2868 
2869 /**
2870  * This handler will called for unknown event group or unimplemented handlers
2871  **/
2872 static void unimplemented_aenq_handler(__rte_unused void *data,
2873 				       __rte_unused struct ena_admin_aenq_entry *aenq_e)
2874 {
2875 	PMD_DRV_LOG(ERR, "Unknown event was received or event with "
2876 			  "unimplemented handler\n");
2877 }
2878 
2879 static struct ena_aenq_handlers aenq_handlers = {
2880 	.handlers = {
2881 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
2882 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
2883 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive
2884 	},
2885 	.unimplemented_handler = unimplemented_aenq_handler
2886 };
2887