xref: /dpdk/drivers/net/ena/ena_ethdev.c (revision e11bdd37745229bf26b557305c07d118c3dbaad7)
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	1
31 #define DRV_MODULE_VER_SUBMINOR	0
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 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
39 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
40 
41 #define GET_L4_HDR_LEN(mbuf)					\
42 	((rte_pktmbuf_mtod_offset(mbuf,	struct rte_tcp_hdr *,	\
43 		mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
44 
45 #define ENA_RX_RSS_TABLE_LOG_SIZE  7
46 #define ENA_RX_RSS_TABLE_SIZE	(1 << ENA_RX_RSS_TABLE_LOG_SIZE)
47 #define ENA_HASH_KEY_SIZE	40
48 #define ETH_GSTRING_LEN	32
49 
50 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
51 
52 #define ENA_MIN_RING_DESC	128
53 
54 enum ethtool_stringset {
55 	ETH_SS_TEST             = 0,
56 	ETH_SS_STATS,
57 };
58 
59 struct ena_stats {
60 	char name[ETH_GSTRING_LEN];
61 	int stat_offset;
62 };
63 
64 #define ENA_STAT_ENTRY(stat, stat_type) { \
65 	.name = #stat, \
66 	.stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
67 }
68 
69 #define ENA_STAT_RX_ENTRY(stat) \
70 	ENA_STAT_ENTRY(stat, rx)
71 
72 #define ENA_STAT_TX_ENTRY(stat) \
73 	ENA_STAT_ENTRY(stat, tx)
74 
75 #define ENA_STAT_GLOBAL_ENTRY(stat) \
76 	ENA_STAT_ENTRY(stat, dev)
77 
78 /* Device arguments */
79 #define ENA_DEVARG_LARGE_LLQ_HDR "large_llq_hdr"
80 
81 /*
82  * Each rte_memzone should have unique name.
83  * To satisfy it, count number of allocation and add it to name.
84  */
85 rte_atomic32_t ena_alloc_cnt;
86 
87 static const struct ena_stats ena_stats_global_strings[] = {
88 	ENA_STAT_GLOBAL_ENTRY(wd_expired),
89 	ENA_STAT_GLOBAL_ENTRY(dev_start),
90 	ENA_STAT_GLOBAL_ENTRY(dev_stop),
91 	ENA_STAT_GLOBAL_ENTRY(tx_drops),
92 };
93 
94 static const struct ena_stats ena_stats_tx_strings[] = {
95 	ENA_STAT_TX_ENTRY(cnt),
96 	ENA_STAT_TX_ENTRY(bytes),
97 	ENA_STAT_TX_ENTRY(prepare_ctx_err),
98 	ENA_STAT_TX_ENTRY(linearize),
99 	ENA_STAT_TX_ENTRY(linearize_failed),
100 	ENA_STAT_TX_ENTRY(tx_poll),
101 	ENA_STAT_TX_ENTRY(doorbells),
102 	ENA_STAT_TX_ENTRY(bad_req_id),
103 	ENA_STAT_TX_ENTRY(available_desc),
104 };
105 
106 static const struct ena_stats ena_stats_rx_strings[] = {
107 	ENA_STAT_RX_ENTRY(cnt),
108 	ENA_STAT_RX_ENTRY(bytes),
109 	ENA_STAT_RX_ENTRY(refill_partial),
110 	ENA_STAT_RX_ENTRY(bad_csum),
111 	ENA_STAT_RX_ENTRY(mbuf_alloc_fail),
112 	ENA_STAT_RX_ENTRY(bad_desc_num),
113 	ENA_STAT_RX_ENTRY(bad_req_id),
114 };
115 
116 #define ENA_STATS_ARRAY_GLOBAL	ARRAY_SIZE(ena_stats_global_strings)
117 #define ENA_STATS_ARRAY_TX	ARRAY_SIZE(ena_stats_tx_strings)
118 #define ENA_STATS_ARRAY_RX	ARRAY_SIZE(ena_stats_rx_strings)
119 
120 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
121 			DEV_TX_OFFLOAD_UDP_CKSUM |\
122 			DEV_TX_OFFLOAD_IPV4_CKSUM |\
123 			DEV_TX_OFFLOAD_TCP_TSO)
124 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
125 		       PKT_TX_IP_CKSUM |\
126 		       PKT_TX_TCP_SEG)
127 
128 /** Vendor ID used by Amazon devices */
129 #define PCI_VENDOR_ID_AMAZON 0x1D0F
130 /** Amazon devices */
131 #define PCI_DEVICE_ID_ENA_VF	0xEC20
132 #define PCI_DEVICE_ID_ENA_LLQ_VF	0xEC21
133 
134 #define	ENA_TX_OFFLOAD_MASK	(\
135 	PKT_TX_L4_MASK |         \
136 	PKT_TX_IPV6 |            \
137 	PKT_TX_IPV4 |            \
138 	PKT_TX_IP_CKSUM |        \
139 	PKT_TX_TCP_SEG)
140 
141 #define	ENA_TX_OFFLOAD_NOTSUP_MASK	\
142 	(PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
143 
144 int ena_logtype_init;
145 int ena_logtype_driver;
146 
147 #ifdef RTE_LIBRTE_ENA_DEBUG_RX
148 int ena_logtype_rx;
149 #endif
150 #ifdef RTE_LIBRTE_ENA_DEBUG_TX
151 int ena_logtype_tx;
152 #endif
153 #ifdef RTE_LIBRTE_ENA_DEBUG_TX_FREE
154 int ena_logtype_tx_free;
155 #endif
156 #ifdef RTE_LIBRTE_ENA_COM_DEBUG
157 int ena_logtype_com;
158 #endif
159 
160 static const struct rte_pci_id pci_id_ena_map[] = {
161 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
162 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
163 	{ .device_id = 0 },
164 };
165 
166 static struct ena_aenq_handlers aenq_handlers;
167 
168 static int ena_device_init(struct ena_com_dev *ena_dev,
169 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
170 			   bool *wd_state);
171 static int ena_dev_configure(struct rte_eth_dev *dev);
172 static void ena_tx_map_mbuf(struct ena_ring *tx_ring,
173 	struct ena_tx_buffer *tx_info,
174 	struct rte_mbuf *mbuf,
175 	void **push_header,
176 	uint16_t *header_len);
177 static int ena_xmit_mbuf(struct ena_ring *tx_ring, struct rte_mbuf *mbuf);
178 static void ena_tx_cleanup(struct ena_ring *tx_ring);
179 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
180 				  uint16_t nb_pkts);
181 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
182 		uint16_t nb_pkts);
183 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
184 			      uint16_t nb_desc, unsigned int socket_id,
185 			      const struct rte_eth_txconf *tx_conf);
186 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
187 			      uint16_t nb_desc, unsigned int socket_id,
188 			      const struct rte_eth_rxconf *rx_conf,
189 			      struct rte_mempool *mp);
190 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len);
191 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
192 				    struct ena_com_rx_buf_info *ena_bufs,
193 				    uint32_t descs,
194 				    uint16_t *next_to_clean,
195 				    uint8_t offset);
196 static uint16_t eth_ena_recv_pkts(void *rx_queue,
197 				  struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
198 static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
199 				  struct rte_mbuf *mbuf, uint16_t id);
200 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
201 static void ena_init_rings(struct ena_adapter *adapter,
202 			   bool disable_meta_caching);
203 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
204 static int ena_start(struct rte_eth_dev *dev);
205 static void ena_stop(struct rte_eth_dev *dev);
206 static void ena_close(struct rte_eth_dev *dev);
207 static int ena_dev_reset(struct rte_eth_dev *dev);
208 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
209 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
210 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
211 static void ena_rx_queue_release(void *queue);
212 static void ena_tx_queue_release(void *queue);
213 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
214 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
215 static int ena_link_update(struct rte_eth_dev *dev,
216 			   int wait_to_complete);
217 static int ena_create_io_queue(struct ena_ring *ring);
218 static void ena_queue_stop(struct ena_ring *ring);
219 static void ena_queue_stop_all(struct rte_eth_dev *dev,
220 			      enum ena_ring_type ring_type);
221 static int ena_queue_start(struct ena_ring *ring);
222 static int ena_queue_start_all(struct rte_eth_dev *dev,
223 			       enum ena_ring_type ring_type);
224 static void ena_stats_restart(struct rte_eth_dev *dev);
225 static int ena_infos_get(struct rte_eth_dev *dev,
226 			 struct rte_eth_dev_info *dev_info);
227 static int ena_rss_reta_update(struct rte_eth_dev *dev,
228 			       struct rte_eth_rss_reta_entry64 *reta_conf,
229 			       uint16_t reta_size);
230 static int ena_rss_reta_query(struct rte_eth_dev *dev,
231 			      struct rte_eth_rss_reta_entry64 *reta_conf,
232 			      uint16_t reta_size);
233 static void ena_interrupt_handler_rte(void *cb_arg);
234 static void ena_timer_wd_callback(struct rte_timer *timer, void *arg);
235 static void ena_destroy_device(struct rte_eth_dev *eth_dev);
236 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev);
237 static int ena_xstats_get_names(struct rte_eth_dev *dev,
238 				struct rte_eth_xstat_name *xstats_names,
239 				unsigned int n);
240 static int ena_xstats_get(struct rte_eth_dev *dev,
241 			  struct rte_eth_xstat *stats,
242 			  unsigned int n);
243 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
244 				const uint64_t *ids,
245 				uint64_t *values,
246 				unsigned int n);
247 static int ena_process_bool_devarg(const char *key,
248 				   const char *value,
249 				   void *opaque);
250 static int ena_parse_devargs(struct ena_adapter *adapter,
251 			     struct rte_devargs *devargs);
252 
253 static const struct eth_dev_ops ena_dev_ops = {
254 	.dev_configure        = ena_dev_configure,
255 	.dev_infos_get        = ena_infos_get,
256 	.rx_queue_setup       = ena_rx_queue_setup,
257 	.tx_queue_setup       = ena_tx_queue_setup,
258 	.dev_start            = ena_start,
259 	.dev_stop             = ena_stop,
260 	.link_update          = ena_link_update,
261 	.stats_get            = ena_stats_get,
262 	.xstats_get_names     = ena_xstats_get_names,
263 	.xstats_get	      = ena_xstats_get,
264 	.xstats_get_by_id     = ena_xstats_get_by_id,
265 	.mtu_set              = ena_mtu_set,
266 	.rx_queue_release     = ena_rx_queue_release,
267 	.tx_queue_release     = ena_tx_queue_release,
268 	.dev_close            = ena_close,
269 	.dev_reset            = ena_dev_reset,
270 	.reta_update          = ena_rss_reta_update,
271 	.reta_query           = ena_rss_reta_query,
272 };
273 
274 void ena_rss_key_fill(void *key, size_t size)
275 {
276 	static bool key_generated;
277 	static uint8_t default_key[ENA_HASH_KEY_SIZE];
278 	size_t i;
279 
280 	RTE_ASSERT(size <= ENA_HASH_KEY_SIZE);
281 
282 	if (!key_generated) {
283 		for (i = 0; i < ENA_HASH_KEY_SIZE; ++i)
284 			default_key[i] = rte_rand() & 0xff;
285 		key_generated = true;
286 	}
287 
288 	rte_memcpy(key, default_key, size);
289 }
290 
291 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
292 				       struct ena_com_rx_ctx *ena_rx_ctx)
293 {
294 	uint64_t ol_flags = 0;
295 	uint32_t packet_type = 0;
296 
297 	if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
298 		packet_type |= RTE_PTYPE_L4_TCP;
299 	else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
300 		packet_type |= RTE_PTYPE_L4_UDP;
301 
302 	if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
303 		packet_type |= RTE_PTYPE_L3_IPV4;
304 	else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
305 		packet_type |= RTE_PTYPE_L3_IPV6;
306 
307 	if (!ena_rx_ctx->l4_csum_checked)
308 		ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
309 	else
310 		if (unlikely(ena_rx_ctx->l4_csum_err) && !ena_rx_ctx->frag)
311 			ol_flags |= PKT_RX_L4_CKSUM_BAD;
312 		else
313 			ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
314 
315 	if (unlikely(ena_rx_ctx->l3_csum_err))
316 		ol_flags |= PKT_RX_IP_CKSUM_BAD;
317 
318 	mbuf->ol_flags = ol_flags;
319 	mbuf->packet_type = packet_type;
320 }
321 
322 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
323 				       struct ena_com_tx_ctx *ena_tx_ctx,
324 				       uint64_t queue_offloads,
325 				       bool disable_meta_caching)
326 {
327 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
328 
329 	if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
330 	    (queue_offloads & QUEUE_OFFLOADS)) {
331 		/* check if TSO is required */
332 		if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
333 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
334 			ena_tx_ctx->tso_enable = true;
335 
336 			ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
337 		}
338 
339 		/* check if L3 checksum is needed */
340 		if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
341 		    (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
342 			ena_tx_ctx->l3_csum_enable = true;
343 
344 		if (mbuf->ol_flags & PKT_TX_IPV6) {
345 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
346 		} else {
347 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
348 
349 			/* set don't fragment (DF) flag */
350 			if (mbuf->packet_type &
351 				(RTE_PTYPE_L4_NONFRAG
352 				 | RTE_PTYPE_INNER_L4_NONFRAG))
353 				ena_tx_ctx->df = true;
354 		}
355 
356 		/* check if L4 checksum is needed */
357 		if (((mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) &&
358 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
359 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
360 			ena_tx_ctx->l4_csum_enable = true;
361 		} else if (((mbuf->ol_flags & PKT_TX_L4_MASK) ==
362 				PKT_TX_UDP_CKSUM) &&
363 				(queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
364 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
365 			ena_tx_ctx->l4_csum_enable = true;
366 		} else {
367 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
368 			ena_tx_ctx->l4_csum_enable = false;
369 		}
370 
371 		ena_meta->mss = mbuf->tso_segsz;
372 		ena_meta->l3_hdr_len = mbuf->l3_len;
373 		ena_meta->l3_hdr_offset = mbuf->l2_len;
374 
375 		ena_tx_ctx->meta_valid = true;
376 	} else if (disable_meta_caching) {
377 		memset(ena_meta, 0, sizeof(*ena_meta));
378 		ena_tx_ctx->meta_valid = true;
379 	} else {
380 		ena_tx_ctx->meta_valid = false;
381 	}
382 }
383 
384 static inline int validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
385 {
386 	if (likely(req_id < rx_ring->ring_size))
387 		return 0;
388 
389 	PMD_DRV_LOG(ERR, "Invalid rx req_id: %hu\n", req_id);
390 
391 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
392 	rx_ring->adapter->trigger_reset = true;
393 	++rx_ring->rx_stats.bad_req_id;
394 
395 	return -EFAULT;
396 }
397 
398 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
399 {
400 	struct ena_tx_buffer *tx_info = NULL;
401 
402 	if (likely(req_id < tx_ring->ring_size)) {
403 		tx_info = &tx_ring->tx_buffer_info[req_id];
404 		if (likely(tx_info->mbuf))
405 			return 0;
406 	}
407 
408 	if (tx_info)
409 		PMD_DRV_LOG(ERR, "tx_info doesn't have valid mbuf\n");
410 	else
411 		PMD_DRV_LOG(ERR, "Invalid req_id: %hu\n", req_id);
412 
413 	/* Trigger device reset */
414 	++tx_ring->tx_stats.bad_req_id;
415 	tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
416 	tx_ring->adapter->trigger_reset	= true;
417 	return -EFAULT;
418 }
419 
420 static void ena_config_host_info(struct ena_com_dev *ena_dev)
421 {
422 	struct ena_admin_host_info *host_info;
423 	int rc;
424 
425 	/* Allocate only the host info */
426 	rc = ena_com_allocate_host_info(ena_dev);
427 	if (rc) {
428 		PMD_DRV_LOG(ERR, "Cannot allocate host info\n");
429 		return;
430 	}
431 
432 	host_info = ena_dev->host_attr.host_info;
433 
434 	host_info->os_type = ENA_ADMIN_OS_DPDK;
435 	host_info->kernel_ver = RTE_VERSION;
436 	strlcpy((char *)host_info->kernel_ver_str, rte_version(),
437 		sizeof(host_info->kernel_ver_str));
438 	host_info->os_dist = RTE_VERSION;
439 	strlcpy((char *)host_info->os_dist_str, rte_version(),
440 		sizeof(host_info->os_dist_str));
441 	host_info->driver_version =
442 		(DRV_MODULE_VER_MAJOR) |
443 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
444 		(DRV_MODULE_VER_SUBMINOR <<
445 			ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
446 	host_info->num_cpus = rte_lcore_count();
447 
448 	host_info->driver_supported_features =
449 		ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK;
450 
451 	rc = ena_com_set_host_attributes(ena_dev);
452 	if (rc) {
453 		if (rc == -ENA_COM_UNSUPPORTED)
454 			PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
455 		else
456 			PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
457 
458 		goto err;
459 	}
460 
461 	return;
462 
463 err:
464 	ena_com_delete_host_info(ena_dev);
465 }
466 
467 /* This function calculates the number of xstats based on the current config */
468 static unsigned int ena_xstats_calc_num(struct rte_eth_dev *dev)
469 {
470 	return ENA_STATS_ARRAY_GLOBAL +
471 		(dev->data->nb_tx_queues * ENA_STATS_ARRAY_TX) +
472 		(dev->data->nb_rx_queues * ENA_STATS_ARRAY_RX);
473 }
474 
475 static void ena_config_debug_area(struct ena_adapter *adapter)
476 {
477 	u32 debug_area_size;
478 	int rc, ss_count;
479 
480 	ss_count = ena_xstats_calc_num(adapter->rte_dev);
481 
482 	/* allocate 32 bytes for each string and 64bit for the value */
483 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
484 
485 	rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
486 	if (rc) {
487 		PMD_DRV_LOG(ERR, "Cannot allocate debug area\n");
488 		return;
489 	}
490 
491 	rc = ena_com_set_host_attributes(&adapter->ena_dev);
492 	if (rc) {
493 		if (rc == -ENA_COM_UNSUPPORTED)
494 			PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
495 		else
496 			PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
497 
498 		goto err;
499 	}
500 
501 	return;
502 err:
503 	ena_com_delete_debug_area(&adapter->ena_dev);
504 }
505 
506 static void ena_close(struct rte_eth_dev *dev)
507 {
508 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
509 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
510 	struct ena_adapter *adapter = dev->data->dev_private;
511 
512 	if (adapter->state == ENA_ADAPTER_STATE_RUNNING)
513 		ena_stop(dev);
514 	adapter->state = ENA_ADAPTER_STATE_CLOSED;
515 
516 	ena_rx_queue_release_all(dev);
517 	ena_tx_queue_release_all(dev);
518 
519 	rte_free(adapter->drv_stats);
520 	adapter->drv_stats = NULL;
521 
522 	rte_intr_disable(intr_handle);
523 	rte_intr_callback_unregister(intr_handle,
524 				     ena_interrupt_handler_rte,
525 				     adapter);
526 
527 	/*
528 	 * MAC is not allocated dynamically. Setting NULL should prevent from
529 	 * release of the resource in the rte_eth_dev_release_port().
530 	 */
531 	dev->data->mac_addrs = NULL;
532 }
533 
534 static int
535 ena_dev_reset(struct rte_eth_dev *dev)
536 {
537 	int rc = 0;
538 
539 	ena_destroy_device(dev);
540 	rc = eth_ena_dev_init(dev);
541 	if (rc)
542 		PMD_INIT_LOG(CRIT, "Cannot initialize device");
543 
544 	return rc;
545 }
546 
547 static int ena_rss_reta_update(struct rte_eth_dev *dev,
548 			       struct rte_eth_rss_reta_entry64 *reta_conf,
549 			       uint16_t reta_size)
550 {
551 	struct ena_adapter *adapter = dev->data->dev_private;
552 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
553 	int rc, i;
554 	u16 entry_value;
555 	int conf_idx;
556 	int idx;
557 
558 	if ((reta_size == 0) || (reta_conf == NULL))
559 		return -EINVAL;
560 
561 	if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
562 		PMD_DRV_LOG(WARNING,
563 			"indirection table %d is bigger than supported (%d)\n",
564 			reta_size, ENA_RX_RSS_TABLE_SIZE);
565 		return -EINVAL;
566 	}
567 
568 	for (i = 0 ; i < reta_size ; i++) {
569 		/* each reta_conf is for 64 entries.
570 		 * to support 128 we use 2 conf of 64
571 		 */
572 		conf_idx = i / RTE_RETA_GROUP_SIZE;
573 		idx = i % RTE_RETA_GROUP_SIZE;
574 		if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
575 			entry_value =
576 				ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
577 
578 			rc = ena_com_indirect_table_fill_entry(ena_dev,
579 							       i,
580 							       entry_value);
581 			if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
582 				PMD_DRV_LOG(ERR,
583 					"Cannot fill indirect table\n");
584 				return rc;
585 			}
586 		}
587 	}
588 
589 	rc = ena_com_indirect_table_set(ena_dev);
590 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
591 		PMD_DRV_LOG(ERR, "Cannot flush the indirect table\n");
592 		return rc;
593 	}
594 
595 	PMD_DRV_LOG(DEBUG, "%s(): RSS configured %d entries  for port %d\n",
596 		__func__, reta_size, adapter->rte_dev->data->port_id);
597 
598 	return 0;
599 }
600 
601 /* Query redirection table. */
602 static int ena_rss_reta_query(struct rte_eth_dev *dev,
603 			      struct rte_eth_rss_reta_entry64 *reta_conf,
604 			      uint16_t reta_size)
605 {
606 	struct ena_adapter *adapter = dev->data->dev_private;
607 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
608 	int rc;
609 	int i;
610 	u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
611 	int reta_conf_idx;
612 	int reta_idx;
613 
614 	if (reta_size == 0 || reta_conf == NULL ||
615 	    (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
616 		return -EINVAL;
617 
618 	rc = ena_com_indirect_table_get(ena_dev, indirect_table);
619 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
620 		PMD_DRV_LOG(ERR, "cannot get indirect table\n");
621 		return -ENOTSUP;
622 	}
623 
624 	for (i = 0 ; i < reta_size ; i++) {
625 		reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
626 		reta_idx = i % RTE_RETA_GROUP_SIZE;
627 		if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
628 			reta_conf[reta_conf_idx].reta[reta_idx] =
629 				ENA_IO_RXQ_IDX_REV(indirect_table[i]);
630 	}
631 
632 	return 0;
633 }
634 
635 static int ena_rss_init_default(struct ena_adapter *adapter)
636 {
637 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
638 	uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
639 	int rc, i;
640 	u32 val;
641 
642 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
643 	if (unlikely(rc)) {
644 		PMD_DRV_LOG(ERR, "Cannot init indirect table\n");
645 		goto err_rss_init;
646 	}
647 
648 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
649 		val = i % nb_rx_queues;
650 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
651 						       ENA_IO_RXQ_IDX(val));
652 		if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
653 			PMD_DRV_LOG(ERR, "Cannot fill indirect table\n");
654 			goto err_fill_indir;
655 		}
656 	}
657 
658 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
659 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
660 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
661 		PMD_DRV_LOG(INFO, "Cannot fill hash function\n");
662 		goto err_fill_indir;
663 	}
664 
665 	rc = ena_com_set_default_hash_ctrl(ena_dev);
666 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
667 		PMD_DRV_LOG(INFO, "Cannot fill hash control\n");
668 		goto err_fill_indir;
669 	}
670 
671 	rc = ena_com_indirect_table_set(ena_dev);
672 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
673 		PMD_DRV_LOG(ERR, "Cannot flush the indirect table\n");
674 		goto err_fill_indir;
675 	}
676 	PMD_DRV_LOG(DEBUG, "RSS configured for port %d\n",
677 		adapter->rte_dev->data->port_id);
678 
679 	return 0;
680 
681 err_fill_indir:
682 	ena_com_rss_destroy(ena_dev);
683 err_rss_init:
684 
685 	return rc;
686 }
687 
688 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
689 {
690 	struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
691 	int nb_queues = dev->data->nb_rx_queues;
692 	int i;
693 
694 	for (i = 0; i < nb_queues; i++)
695 		ena_rx_queue_release(queues[i]);
696 }
697 
698 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
699 {
700 	struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
701 	int nb_queues = dev->data->nb_tx_queues;
702 	int i;
703 
704 	for (i = 0; i < nb_queues; i++)
705 		ena_tx_queue_release(queues[i]);
706 }
707 
708 static void ena_rx_queue_release(void *queue)
709 {
710 	struct ena_ring *ring = (struct ena_ring *)queue;
711 
712 	/* Free ring resources */
713 	if (ring->rx_buffer_info)
714 		rte_free(ring->rx_buffer_info);
715 	ring->rx_buffer_info = NULL;
716 
717 	if (ring->rx_refill_buffer)
718 		rte_free(ring->rx_refill_buffer);
719 	ring->rx_refill_buffer = NULL;
720 
721 	if (ring->empty_rx_reqs)
722 		rte_free(ring->empty_rx_reqs);
723 	ring->empty_rx_reqs = NULL;
724 
725 	ring->configured = 0;
726 
727 	PMD_DRV_LOG(NOTICE, "RX Queue %d:%d released\n",
728 		ring->port_id, ring->id);
729 }
730 
731 static void ena_tx_queue_release(void *queue)
732 {
733 	struct ena_ring *ring = (struct ena_ring *)queue;
734 
735 	/* Free ring resources */
736 	if (ring->push_buf_intermediate_buf)
737 		rte_free(ring->push_buf_intermediate_buf);
738 
739 	if (ring->tx_buffer_info)
740 		rte_free(ring->tx_buffer_info);
741 
742 	if (ring->empty_tx_reqs)
743 		rte_free(ring->empty_tx_reqs);
744 
745 	ring->empty_tx_reqs = NULL;
746 	ring->tx_buffer_info = NULL;
747 	ring->push_buf_intermediate_buf = NULL;
748 
749 	ring->configured = 0;
750 
751 	PMD_DRV_LOG(NOTICE, "TX Queue %d:%d released\n",
752 		ring->port_id, ring->id);
753 }
754 
755 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
756 {
757 	unsigned int i;
758 
759 	for (i = 0; i < ring->ring_size; ++i) {
760 		struct ena_rx_buffer *rx_info = &ring->rx_buffer_info[i];
761 		if (rx_info->mbuf) {
762 			rte_mbuf_raw_free(rx_info->mbuf);
763 			rx_info->mbuf = NULL;
764 		}
765 	}
766 }
767 
768 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
769 {
770 	unsigned int i;
771 
772 	for (i = 0; i < ring->ring_size; ++i) {
773 		struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
774 
775 		if (tx_buf->mbuf)
776 			rte_pktmbuf_free(tx_buf->mbuf);
777 	}
778 }
779 
780 static int ena_link_update(struct rte_eth_dev *dev,
781 			   __rte_unused int wait_to_complete)
782 {
783 	struct rte_eth_link *link = &dev->data->dev_link;
784 	struct ena_adapter *adapter = dev->data->dev_private;
785 
786 	link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
787 	link->link_speed = ETH_SPEED_NUM_NONE;
788 	link->link_duplex = ETH_LINK_FULL_DUPLEX;
789 
790 	return 0;
791 }
792 
793 static int ena_queue_start_all(struct rte_eth_dev *dev,
794 			       enum ena_ring_type ring_type)
795 {
796 	struct ena_adapter *adapter = dev->data->dev_private;
797 	struct ena_ring *queues = NULL;
798 	int nb_queues;
799 	int i = 0;
800 	int rc = 0;
801 
802 	if (ring_type == ENA_RING_TYPE_RX) {
803 		queues = adapter->rx_ring;
804 		nb_queues = dev->data->nb_rx_queues;
805 	} else {
806 		queues = adapter->tx_ring;
807 		nb_queues = dev->data->nb_tx_queues;
808 	}
809 	for (i = 0; i < nb_queues; i++) {
810 		if (queues[i].configured) {
811 			if (ring_type == ENA_RING_TYPE_RX) {
812 				ena_assert_msg(
813 					dev->data->rx_queues[i] == &queues[i],
814 					"Inconsistent state of rx queues\n");
815 			} else {
816 				ena_assert_msg(
817 					dev->data->tx_queues[i] == &queues[i],
818 					"Inconsistent state of tx queues\n");
819 			}
820 
821 			rc = ena_queue_start(&queues[i]);
822 
823 			if (rc) {
824 				PMD_INIT_LOG(ERR,
825 					     "failed to start queue %d type(%d)",
826 					     i, ring_type);
827 				goto err;
828 			}
829 		}
830 	}
831 
832 	return 0;
833 
834 err:
835 	while (i--)
836 		if (queues[i].configured)
837 			ena_queue_stop(&queues[i]);
838 
839 	return rc;
840 }
841 
842 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
843 {
844 	uint32_t max_frame_len = adapter->max_mtu;
845 
846 	if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
847 	    DEV_RX_OFFLOAD_JUMBO_FRAME)
848 		max_frame_len =
849 			adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
850 
851 	return max_frame_len;
852 }
853 
854 static int ena_check_valid_conf(struct ena_adapter *adapter)
855 {
856 	uint32_t max_frame_len = ena_get_mtu_conf(adapter);
857 
858 	if (max_frame_len > adapter->max_mtu || max_frame_len < ENA_MIN_MTU) {
859 		PMD_INIT_LOG(ERR, "Unsupported MTU of %d. "
860 				  "max mtu: %d, min mtu: %d",
861 			     max_frame_len, adapter->max_mtu, ENA_MIN_MTU);
862 		return ENA_COM_UNSUPPORTED;
863 	}
864 
865 	return 0;
866 }
867 
868 static int
869 ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx,
870 		       bool use_large_llq_hdr)
871 {
872 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
873 	struct ena_com_dev *ena_dev = ctx->ena_dev;
874 	uint32_t max_tx_queue_size;
875 	uint32_t max_rx_queue_size;
876 
877 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
878 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
879 			&ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
880 		max_rx_queue_size = RTE_MIN(max_queue_ext->max_rx_cq_depth,
881 			max_queue_ext->max_rx_sq_depth);
882 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
883 
884 		if (ena_dev->tx_mem_queue_type ==
885 		    ENA_ADMIN_PLACEMENT_POLICY_DEV) {
886 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
887 				llq->max_llq_depth);
888 		} else {
889 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
890 				max_queue_ext->max_tx_sq_depth);
891 		}
892 
893 		ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
894 			max_queue_ext->max_per_packet_rx_descs);
895 		ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
896 			max_queue_ext->max_per_packet_tx_descs);
897 	} else {
898 		struct ena_admin_queue_feature_desc *max_queues =
899 			&ctx->get_feat_ctx->max_queues;
900 		max_rx_queue_size = RTE_MIN(max_queues->max_cq_depth,
901 			max_queues->max_sq_depth);
902 		max_tx_queue_size = max_queues->max_cq_depth;
903 
904 		if (ena_dev->tx_mem_queue_type ==
905 		    ENA_ADMIN_PLACEMENT_POLICY_DEV) {
906 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
907 				llq->max_llq_depth);
908 		} else {
909 			max_tx_queue_size = RTE_MIN(max_tx_queue_size,
910 				max_queues->max_sq_depth);
911 		}
912 
913 		ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
914 			max_queues->max_packet_rx_descs);
915 		ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
916 			max_queues->max_packet_tx_descs);
917 	}
918 
919 	/* Round down to the nearest power of 2 */
920 	max_rx_queue_size = rte_align32prevpow2(max_rx_queue_size);
921 	max_tx_queue_size = rte_align32prevpow2(max_tx_queue_size);
922 
923 	if (use_large_llq_hdr) {
924 		if ((llq->entry_size_ctrl_supported &
925 		     ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
926 		    (ena_dev->tx_mem_queue_type ==
927 		     ENA_ADMIN_PLACEMENT_POLICY_DEV)) {
928 			max_tx_queue_size /= 2;
929 			PMD_INIT_LOG(INFO,
930 				"Forcing large headers and decreasing maximum TX queue size to %d\n",
931 				max_tx_queue_size);
932 		} else {
933 			PMD_INIT_LOG(ERR,
934 				"Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
935 		}
936 	}
937 
938 	if (unlikely(max_rx_queue_size == 0 || max_tx_queue_size == 0)) {
939 		PMD_INIT_LOG(ERR, "Invalid queue size");
940 		return -EFAULT;
941 	}
942 
943 	ctx->max_tx_queue_size = max_tx_queue_size;
944 	ctx->max_rx_queue_size = max_rx_queue_size;
945 
946 	return 0;
947 }
948 
949 static void ena_stats_restart(struct rte_eth_dev *dev)
950 {
951 	struct ena_adapter *adapter = dev->data->dev_private;
952 
953 	rte_atomic64_init(&adapter->drv_stats->ierrors);
954 	rte_atomic64_init(&adapter->drv_stats->oerrors);
955 	rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
956 	adapter->drv_stats->rx_drops = 0;
957 }
958 
959 static int ena_stats_get(struct rte_eth_dev *dev,
960 			  struct rte_eth_stats *stats)
961 {
962 	struct ena_admin_basic_stats ena_stats;
963 	struct ena_adapter *adapter = dev->data->dev_private;
964 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
965 	int rc;
966 	int i;
967 	int max_rings_stats;
968 
969 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
970 		return -ENOTSUP;
971 
972 	memset(&ena_stats, 0, sizeof(ena_stats));
973 	rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
974 	if (unlikely(rc)) {
975 		PMD_DRV_LOG(ERR, "Could not retrieve statistics from ENA\n");
976 		return rc;
977 	}
978 
979 	/* Set of basic statistics from ENA */
980 	stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
981 					  ena_stats.rx_pkts_low);
982 	stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
983 					  ena_stats.tx_pkts_low);
984 	stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
985 					ena_stats.rx_bytes_low);
986 	stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
987 					ena_stats.tx_bytes_low);
988 
989 	/* Driver related stats */
990 	stats->imissed = adapter->drv_stats->rx_drops;
991 	stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
992 	stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
993 	stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
994 
995 	max_rings_stats = RTE_MIN(dev->data->nb_rx_queues,
996 		RTE_ETHDEV_QUEUE_STAT_CNTRS);
997 	for (i = 0; i < max_rings_stats; ++i) {
998 		struct ena_stats_rx *rx_stats = &adapter->rx_ring[i].rx_stats;
999 
1000 		stats->q_ibytes[i] = rx_stats->bytes;
1001 		stats->q_ipackets[i] = rx_stats->cnt;
1002 		stats->q_errors[i] = rx_stats->bad_desc_num +
1003 			rx_stats->bad_req_id;
1004 	}
1005 
1006 	max_rings_stats = RTE_MIN(dev->data->nb_tx_queues,
1007 		RTE_ETHDEV_QUEUE_STAT_CNTRS);
1008 	for (i = 0; i < max_rings_stats; ++i) {
1009 		struct ena_stats_tx *tx_stats = &adapter->tx_ring[i].tx_stats;
1010 
1011 		stats->q_obytes[i] = tx_stats->bytes;
1012 		stats->q_opackets[i] = tx_stats->cnt;
1013 	}
1014 
1015 	return 0;
1016 }
1017 
1018 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1019 {
1020 	struct ena_adapter *adapter;
1021 	struct ena_com_dev *ena_dev;
1022 	int rc = 0;
1023 
1024 	ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
1025 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
1026 	adapter = dev->data->dev_private;
1027 
1028 	ena_dev = &adapter->ena_dev;
1029 	ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
1030 
1031 	if (mtu > ena_get_mtu_conf(adapter) || mtu < ENA_MIN_MTU) {
1032 		PMD_DRV_LOG(ERR,
1033 			"Invalid MTU setting. new_mtu: %d "
1034 			"max mtu: %d min mtu: %d\n",
1035 			mtu, ena_get_mtu_conf(adapter), ENA_MIN_MTU);
1036 		return -EINVAL;
1037 	}
1038 
1039 	rc = ena_com_set_dev_mtu(ena_dev, mtu);
1040 	if (rc)
1041 		PMD_DRV_LOG(ERR, "Could not set MTU: %d\n", mtu);
1042 	else
1043 		PMD_DRV_LOG(NOTICE, "Set MTU: %d\n", mtu);
1044 
1045 	return rc;
1046 }
1047 
1048 static int ena_start(struct rte_eth_dev *dev)
1049 {
1050 	struct ena_adapter *adapter = dev->data->dev_private;
1051 	uint64_t ticks;
1052 	int rc = 0;
1053 
1054 	rc = ena_check_valid_conf(adapter);
1055 	if (rc)
1056 		return rc;
1057 
1058 	rc = ena_queue_start_all(dev, ENA_RING_TYPE_RX);
1059 	if (rc)
1060 		return rc;
1061 
1062 	rc = ena_queue_start_all(dev, ENA_RING_TYPE_TX);
1063 	if (rc)
1064 		goto err_start_tx;
1065 
1066 	if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
1067 	    ETH_MQ_RX_RSS_FLAG && adapter->rte_dev->data->nb_rx_queues > 0) {
1068 		rc = ena_rss_init_default(adapter);
1069 		if (rc)
1070 			goto err_rss_init;
1071 	}
1072 
1073 	ena_stats_restart(dev);
1074 
1075 	adapter->timestamp_wd = rte_get_timer_cycles();
1076 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
1077 
1078 	ticks = rte_get_timer_hz();
1079 	rte_timer_reset(&adapter->timer_wd, ticks, PERIODICAL, rte_lcore_id(),
1080 			ena_timer_wd_callback, adapter);
1081 
1082 	++adapter->dev_stats.dev_start;
1083 	adapter->state = ENA_ADAPTER_STATE_RUNNING;
1084 
1085 	return 0;
1086 
1087 err_rss_init:
1088 	ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1089 err_start_tx:
1090 	ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1091 	return rc;
1092 }
1093 
1094 static void ena_stop(struct rte_eth_dev *dev)
1095 {
1096 	struct ena_adapter *adapter = dev->data->dev_private;
1097 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1098 	int rc;
1099 
1100 	rte_timer_stop_sync(&adapter->timer_wd);
1101 	ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1102 	ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1103 
1104 	if (adapter->trigger_reset) {
1105 		rc = ena_com_dev_reset(ena_dev, adapter->reset_reason);
1106 		if (rc)
1107 			PMD_DRV_LOG(ERR, "Device reset failed rc=%d\n", rc);
1108 	}
1109 
1110 	++adapter->dev_stats.dev_stop;
1111 	adapter->state = ENA_ADAPTER_STATE_STOPPED;
1112 }
1113 
1114 static int ena_create_io_queue(struct ena_ring *ring)
1115 {
1116 	struct ena_adapter *adapter;
1117 	struct ena_com_dev *ena_dev;
1118 	struct ena_com_create_io_ctx ctx =
1119 		/* policy set to _HOST just to satisfy icc compiler */
1120 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
1121 		  0, 0, 0, 0, 0 };
1122 	uint16_t ena_qid;
1123 	unsigned int i;
1124 	int rc;
1125 
1126 	adapter = ring->adapter;
1127 	ena_dev = &adapter->ena_dev;
1128 
1129 	if (ring->type == ENA_RING_TYPE_TX) {
1130 		ena_qid = ENA_IO_TXQ_IDX(ring->id);
1131 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1132 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1133 		for (i = 0; i < ring->ring_size; i++)
1134 			ring->empty_tx_reqs[i] = i;
1135 	} else {
1136 		ena_qid = ENA_IO_RXQ_IDX(ring->id);
1137 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1138 		for (i = 0; i < ring->ring_size; i++)
1139 			ring->empty_rx_reqs[i] = i;
1140 	}
1141 	ctx.queue_size = ring->ring_size;
1142 	ctx.qid = ena_qid;
1143 	ctx.msix_vector = -1; /* interrupts not used */
1144 	ctx.numa_node = ring->numa_socket_id;
1145 
1146 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1147 	if (rc) {
1148 		PMD_DRV_LOG(ERR,
1149 			"failed to create io queue #%d (qid:%d) rc: %d\n",
1150 			ring->id, ena_qid, rc);
1151 		return rc;
1152 	}
1153 
1154 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1155 				     &ring->ena_com_io_sq,
1156 				     &ring->ena_com_io_cq);
1157 	if (rc) {
1158 		PMD_DRV_LOG(ERR,
1159 			"Failed to get io queue handlers. queue num %d rc: %d\n",
1160 			ring->id, rc);
1161 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1162 		return rc;
1163 	}
1164 
1165 	if (ring->type == ENA_RING_TYPE_TX)
1166 		ena_com_update_numa_node(ring->ena_com_io_cq, ctx.numa_node);
1167 
1168 	return 0;
1169 }
1170 
1171 static void ena_queue_stop(struct ena_ring *ring)
1172 {
1173 	struct ena_com_dev *ena_dev = &ring->adapter->ena_dev;
1174 
1175 	if (ring->type == ENA_RING_TYPE_RX) {
1176 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(ring->id));
1177 		ena_rx_queue_release_bufs(ring);
1178 	} else {
1179 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(ring->id));
1180 		ena_tx_queue_release_bufs(ring);
1181 	}
1182 }
1183 
1184 static void ena_queue_stop_all(struct rte_eth_dev *dev,
1185 			      enum ena_ring_type ring_type)
1186 {
1187 	struct ena_adapter *adapter = dev->data->dev_private;
1188 	struct ena_ring *queues = NULL;
1189 	uint16_t nb_queues, i;
1190 
1191 	if (ring_type == ENA_RING_TYPE_RX) {
1192 		queues = adapter->rx_ring;
1193 		nb_queues = dev->data->nb_rx_queues;
1194 	} else {
1195 		queues = adapter->tx_ring;
1196 		nb_queues = dev->data->nb_tx_queues;
1197 	}
1198 
1199 	for (i = 0; i < nb_queues; ++i)
1200 		if (queues[i].configured)
1201 			ena_queue_stop(&queues[i]);
1202 }
1203 
1204 static int ena_queue_start(struct ena_ring *ring)
1205 {
1206 	int rc, bufs_num;
1207 
1208 	ena_assert_msg(ring->configured == 1,
1209 		       "Trying to start unconfigured queue\n");
1210 
1211 	rc = ena_create_io_queue(ring);
1212 	if (rc) {
1213 		PMD_INIT_LOG(ERR, "Failed to create IO queue!");
1214 		return rc;
1215 	}
1216 
1217 	ring->next_to_clean = 0;
1218 	ring->next_to_use = 0;
1219 
1220 	if (ring->type == ENA_RING_TYPE_TX) {
1221 		ring->tx_stats.available_desc =
1222 			ena_com_free_q_entries(ring->ena_com_io_sq);
1223 		return 0;
1224 	}
1225 
1226 	bufs_num = ring->ring_size - 1;
1227 	rc = ena_populate_rx_queue(ring, bufs_num);
1228 	if (rc != bufs_num) {
1229 		ena_com_destroy_io_queue(&ring->adapter->ena_dev,
1230 					 ENA_IO_RXQ_IDX(ring->id));
1231 		PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
1232 		return ENA_COM_FAULT;
1233 	}
1234 
1235 	return 0;
1236 }
1237 
1238 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
1239 			      uint16_t queue_idx,
1240 			      uint16_t nb_desc,
1241 			      unsigned int socket_id,
1242 			      const struct rte_eth_txconf *tx_conf)
1243 {
1244 	struct ena_ring *txq = NULL;
1245 	struct ena_adapter *adapter = dev->data->dev_private;
1246 	unsigned int i;
1247 
1248 	txq = &adapter->tx_ring[queue_idx];
1249 
1250 	if (txq->configured) {
1251 		PMD_DRV_LOG(CRIT,
1252 			"API violation. Queue %d is already configured\n",
1253 			queue_idx);
1254 		return ENA_COM_FAULT;
1255 	}
1256 
1257 	if (!rte_is_power_of_2(nb_desc)) {
1258 		PMD_DRV_LOG(ERR,
1259 			"Unsupported size of TX queue: %d is not a power of 2.\n",
1260 			nb_desc);
1261 		return -EINVAL;
1262 	}
1263 
1264 	if (nb_desc > adapter->max_tx_ring_size) {
1265 		PMD_DRV_LOG(ERR,
1266 			"Unsupported size of TX queue (max size: %d)\n",
1267 			adapter->max_tx_ring_size);
1268 		return -EINVAL;
1269 	}
1270 
1271 	if (nb_desc == RTE_ETH_DEV_FALLBACK_TX_RINGSIZE)
1272 		nb_desc = adapter->max_tx_ring_size;
1273 
1274 	txq->port_id = dev->data->port_id;
1275 	txq->next_to_clean = 0;
1276 	txq->next_to_use = 0;
1277 	txq->ring_size = nb_desc;
1278 	txq->size_mask = nb_desc - 1;
1279 	txq->numa_socket_id = socket_id;
1280 
1281 	txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1282 					  sizeof(struct ena_tx_buffer) *
1283 					  txq->ring_size,
1284 					  RTE_CACHE_LINE_SIZE);
1285 	if (!txq->tx_buffer_info) {
1286 		PMD_DRV_LOG(ERR, "failed to alloc mem for tx buffer info\n");
1287 		return -ENOMEM;
1288 	}
1289 
1290 	txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1291 					 sizeof(u16) * txq->ring_size,
1292 					 RTE_CACHE_LINE_SIZE);
1293 	if (!txq->empty_tx_reqs) {
1294 		PMD_DRV_LOG(ERR, "failed to alloc mem for tx reqs\n");
1295 		rte_free(txq->tx_buffer_info);
1296 		return -ENOMEM;
1297 	}
1298 
1299 	txq->push_buf_intermediate_buf =
1300 		rte_zmalloc("txq->push_buf_intermediate_buf",
1301 			    txq->tx_max_header_size,
1302 			    RTE_CACHE_LINE_SIZE);
1303 	if (!txq->push_buf_intermediate_buf) {
1304 		PMD_DRV_LOG(ERR, "failed to alloc push buff for LLQ\n");
1305 		rte_free(txq->tx_buffer_info);
1306 		rte_free(txq->empty_tx_reqs);
1307 		return -ENOMEM;
1308 	}
1309 
1310 	for (i = 0; i < txq->ring_size; i++)
1311 		txq->empty_tx_reqs[i] = i;
1312 
1313 	if (tx_conf != NULL) {
1314 		txq->offloads =
1315 			tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1316 	}
1317 	/* Store pointer to this queue in upper layer */
1318 	txq->configured = 1;
1319 	dev->data->tx_queues[queue_idx] = txq;
1320 
1321 	return 0;
1322 }
1323 
1324 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1325 			      uint16_t queue_idx,
1326 			      uint16_t nb_desc,
1327 			      unsigned int socket_id,
1328 			      __rte_unused const struct rte_eth_rxconf *rx_conf,
1329 			      struct rte_mempool *mp)
1330 {
1331 	struct ena_adapter *adapter = dev->data->dev_private;
1332 	struct ena_ring *rxq = NULL;
1333 	size_t buffer_size;
1334 	int i;
1335 
1336 	rxq = &adapter->rx_ring[queue_idx];
1337 	if (rxq->configured) {
1338 		PMD_DRV_LOG(CRIT,
1339 			"API violation. Queue %d is already configured\n",
1340 			queue_idx);
1341 		return ENA_COM_FAULT;
1342 	}
1343 
1344 	if (nb_desc == RTE_ETH_DEV_FALLBACK_RX_RINGSIZE)
1345 		nb_desc = adapter->max_rx_ring_size;
1346 
1347 	if (!rte_is_power_of_2(nb_desc)) {
1348 		PMD_DRV_LOG(ERR,
1349 			"Unsupported size of RX queue: %d is not a power of 2.\n",
1350 			nb_desc);
1351 		return -EINVAL;
1352 	}
1353 
1354 	if (nb_desc > adapter->max_rx_ring_size) {
1355 		PMD_DRV_LOG(ERR,
1356 			"Unsupported size of RX queue (max size: %d)\n",
1357 			adapter->max_rx_ring_size);
1358 		return -EINVAL;
1359 	}
1360 
1361 	/* ENA isn't supporting buffers smaller than 1400 bytes */
1362 	buffer_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
1363 	if (buffer_size < ENA_RX_BUF_MIN_SIZE) {
1364 		PMD_DRV_LOG(ERR,
1365 			"Unsupported size of RX buffer: %zu (min size: %d)\n",
1366 			buffer_size, ENA_RX_BUF_MIN_SIZE);
1367 		return -EINVAL;
1368 	}
1369 
1370 	rxq->port_id = dev->data->port_id;
1371 	rxq->next_to_clean = 0;
1372 	rxq->next_to_use = 0;
1373 	rxq->ring_size = nb_desc;
1374 	rxq->size_mask = nb_desc - 1;
1375 	rxq->numa_socket_id = socket_id;
1376 	rxq->mb_pool = mp;
1377 
1378 	rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1379 		sizeof(struct ena_rx_buffer) * nb_desc,
1380 		RTE_CACHE_LINE_SIZE);
1381 	if (!rxq->rx_buffer_info) {
1382 		PMD_DRV_LOG(ERR, "failed to alloc mem for rx buffer info\n");
1383 		return -ENOMEM;
1384 	}
1385 
1386 	rxq->rx_refill_buffer = rte_zmalloc("rxq->rx_refill_buffer",
1387 					    sizeof(struct rte_mbuf *) * nb_desc,
1388 					    RTE_CACHE_LINE_SIZE);
1389 
1390 	if (!rxq->rx_refill_buffer) {
1391 		PMD_DRV_LOG(ERR, "failed to alloc mem for rx refill buffer\n");
1392 		rte_free(rxq->rx_buffer_info);
1393 		rxq->rx_buffer_info = NULL;
1394 		return -ENOMEM;
1395 	}
1396 
1397 	rxq->empty_rx_reqs = rte_zmalloc("rxq->empty_rx_reqs",
1398 					 sizeof(uint16_t) * nb_desc,
1399 					 RTE_CACHE_LINE_SIZE);
1400 	if (!rxq->empty_rx_reqs) {
1401 		PMD_DRV_LOG(ERR, "failed to alloc mem for empty rx reqs\n");
1402 		rte_free(rxq->rx_buffer_info);
1403 		rxq->rx_buffer_info = NULL;
1404 		rte_free(rxq->rx_refill_buffer);
1405 		rxq->rx_refill_buffer = NULL;
1406 		return -ENOMEM;
1407 	}
1408 
1409 	for (i = 0; i < nb_desc; i++)
1410 		rxq->empty_rx_reqs[i] = i;
1411 
1412 	/* Store pointer to this queue in upper layer */
1413 	rxq->configured = 1;
1414 	dev->data->rx_queues[queue_idx] = rxq;
1415 
1416 	return 0;
1417 }
1418 
1419 static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
1420 				  struct rte_mbuf *mbuf, uint16_t id)
1421 {
1422 	struct ena_com_buf ebuf;
1423 	int rc;
1424 
1425 	/* prepare physical address for DMA transaction */
1426 	ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1427 	ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1428 
1429 	/* pass resource to device */
1430 	rc = ena_com_add_single_rx_desc(io_sq, &ebuf, id);
1431 	if (unlikely(rc != 0))
1432 		PMD_DRV_LOG(WARNING, "failed adding rx desc\n");
1433 
1434 	return rc;
1435 }
1436 
1437 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1438 {
1439 	unsigned int i;
1440 	int rc;
1441 	uint16_t next_to_use = rxq->next_to_use;
1442 	uint16_t in_use, req_id;
1443 	struct rte_mbuf **mbufs = rxq->rx_refill_buffer;
1444 
1445 	if (unlikely(!count))
1446 		return 0;
1447 
1448 	in_use = rxq->ring_size - 1 -
1449 		ena_com_free_q_entries(rxq->ena_com_io_sq);
1450 	ena_assert_msg(((in_use + count) < rxq->ring_size),
1451 		"bad ring state\n");
1452 
1453 	/* get resources for incoming packets */
1454 	rc = rte_mempool_get_bulk(rxq->mb_pool, (void **)mbufs, count);
1455 	if (unlikely(rc < 0)) {
1456 		rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1457 		++rxq->rx_stats.mbuf_alloc_fail;
1458 		PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1459 		return 0;
1460 	}
1461 
1462 	for (i = 0; i < count; i++) {
1463 		struct rte_mbuf *mbuf = mbufs[i];
1464 		struct ena_rx_buffer *rx_info;
1465 
1466 		if (likely((i + 4) < count))
1467 			rte_prefetch0(mbufs[i + 4]);
1468 
1469 		req_id = rxq->empty_rx_reqs[next_to_use];
1470 		rc = validate_rx_req_id(rxq, req_id);
1471 		if (unlikely(rc))
1472 			break;
1473 
1474 		rx_info = &rxq->rx_buffer_info[req_id];
1475 
1476 		rc = ena_add_single_rx_desc(rxq->ena_com_io_sq, mbuf, req_id);
1477 		if (unlikely(rc != 0))
1478 			break;
1479 
1480 		rx_info->mbuf = mbuf;
1481 		next_to_use = ENA_IDX_NEXT_MASKED(next_to_use, rxq->size_mask);
1482 	}
1483 
1484 	if (unlikely(i < count)) {
1485 		PMD_DRV_LOG(WARNING, "refilled rx qid %d with only %d "
1486 			"buffers (from %d)\n", rxq->id, i, count);
1487 		rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbufs[i]),
1488 				     count - i);
1489 		++rxq->rx_stats.refill_partial;
1490 	}
1491 
1492 	/* When we submitted free recources to device... */
1493 	if (likely(i > 0)) {
1494 		/* ...let HW know that it can fill buffers with data. */
1495 		ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1496 
1497 		rxq->next_to_use = next_to_use;
1498 	}
1499 
1500 	return i;
1501 }
1502 
1503 static int ena_device_init(struct ena_com_dev *ena_dev,
1504 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
1505 			   bool *wd_state)
1506 {
1507 	uint32_t aenq_groups;
1508 	int rc;
1509 	bool readless_supported;
1510 
1511 	/* Initialize mmio registers */
1512 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
1513 	if (rc) {
1514 		PMD_DRV_LOG(ERR, "failed to init mmio read less\n");
1515 		return rc;
1516 	}
1517 
1518 	/* The PCIe configuration space revision id indicate if mmio reg
1519 	 * read is disabled.
1520 	 */
1521 	readless_supported =
1522 		!(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1523 			       & ENA_MMIO_DISABLE_REG_READ);
1524 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1525 
1526 	/* reset device */
1527 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1528 	if (rc) {
1529 		PMD_DRV_LOG(ERR, "cannot reset device\n");
1530 		goto err_mmio_read_less;
1531 	}
1532 
1533 	/* check FW version */
1534 	rc = ena_com_validate_version(ena_dev);
1535 	if (rc) {
1536 		PMD_DRV_LOG(ERR, "device version is too low\n");
1537 		goto err_mmio_read_less;
1538 	}
1539 
1540 	ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1541 
1542 	/* ENA device administration layer init */
1543 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
1544 	if (rc) {
1545 		PMD_DRV_LOG(ERR,
1546 			"cannot initialize ena admin queue with device\n");
1547 		goto err_mmio_read_less;
1548 	}
1549 
1550 	/* To enable the msix interrupts the driver needs to know the number
1551 	 * of queues. So the driver uses polling mode to retrieve this
1552 	 * information.
1553 	 */
1554 	ena_com_set_admin_polling_mode(ena_dev, true);
1555 
1556 	ena_config_host_info(ena_dev);
1557 
1558 	/* Get Device Attributes and features */
1559 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1560 	if (rc) {
1561 		PMD_DRV_LOG(ERR,
1562 			"cannot get attribute for ena device rc= %d\n", rc);
1563 		goto err_admin_init;
1564 	}
1565 
1566 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1567 		      BIT(ENA_ADMIN_NOTIFICATION) |
1568 		      BIT(ENA_ADMIN_KEEP_ALIVE) |
1569 		      BIT(ENA_ADMIN_FATAL_ERROR) |
1570 		      BIT(ENA_ADMIN_WARNING);
1571 
1572 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
1573 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1574 	if (rc) {
1575 		PMD_DRV_LOG(ERR, "Cannot configure aenq groups rc: %d\n", rc);
1576 		goto err_admin_init;
1577 	}
1578 
1579 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
1580 
1581 	return 0;
1582 
1583 err_admin_init:
1584 	ena_com_admin_destroy(ena_dev);
1585 
1586 err_mmio_read_less:
1587 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1588 
1589 	return rc;
1590 }
1591 
1592 static void ena_interrupt_handler_rte(void *cb_arg)
1593 {
1594 	struct ena_adapter *adapter = cb_arg;
1595 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1596 
1597 	ena_com_admin_q_comp_intr_handler(ena_dev);
1598 	if (likely(adapter->state != ENA_ADAPTER_STATE_CLOSED))
1599 		ena_com_aenq_intr_handler(ena_dev, adapter);
1600 }
1601 
1602 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
1603 {
1604 	if (!adapter->wd_state)
1605 		return;
1606 
1607 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
1608 		return;
1609 
1610 	if (unlikely((rte_get_timer_cycles() - adapter->timestamp_wd) >=
1611 	    adapter->keep_alive_timeout)) {
1612 		PMD_DRV_LOG(ERR, "Keep alive timeout\n");
1613 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
1614 		adapter->trigger_reset = true;
1615 		++adapter->dev_stats.wd_expired;
1616 	}
1617 }
1618 
1619 /* Check if admin queue is enabled */
1620 static void check_for_admin_com_state(struct ena_adapter *adapter)
1621 {
1622 	if (unlikely(!ena_com_get_admin_running_state(&adapter->ena_dev))) {
1623 		PMD_DRV_LOG(ERR, "ENA admin queue is not in running state!\n");
1624 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
1625 		adapter->trigger_reset = true;
1626 	}
1627 }
1628 
1629 static void ena_timer_wd_callback(__rte_unused struct rte_timer *timer,
1630 				  void *arg)
1631 {
1632 	struct ena_adapter *adapter = arg;
1633 	struct rte_eth_dev *dev = adapter->rte_dev;
1634 
1635 	check_for_missing_keep_alive(adapter);
1636 	check_for_admin_com_state(adapter);
1637 
1638 	if (unlikely(adapter->trigger_reset)) {
1639 		PMD_DRV_LOG(ERR, "Trigger reset is on\n");
1640 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1641 			NULL);
1642 	}
1643 }
1644 
1645 static inline void
1646 set_default_llq_configurations(struct ena_llq_configurations *llq_config,
1647 			       struct ena_admin_feature_llq_desc *llq,
1648 			       bool use_large_llq_hdr)
1649 {
1650 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
1651 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
1652 	llq_config->llq_num_decs_before_header =
1653 		ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
1654 
1655 	if (use_large_llq_hdr &&
1656 	    (llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B)) {
1657 		llq_config->llq_ring_entry_size =
1658 			ENA_ADMIN_LIST_ENTRY_SIZE_256B;
1659 		llq_config->llq_ring_entry_size_value = 256;
1660 	} else {
1661 		llq_config->llq_ring_entry_size =
1662 			ENA_ADMIN_LIST_ENTRY_SIZE_128B;
1663 		llq_config->llq_ring_entry_size_value = 128;
1664 	}
1665 }
1666 
1667 static int
1668 ena_set_queues_placement_policy(struct ena_adapter *adapter,
1669 				struct ena_com_dev *ena_dev,
1670 				struct ena_admin_feature_llq_desc *llq,
1671 				struct ena_llq_configurations *llq_default_configurations)
1672 {
1673 	int rc;
1674 	u32 llq_feature_mask;
1675 
1676 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
1677 	if (!(ena_dev->supported_features & llq_feature_mask)) {
1678 		PMD_DRV_LOG(INFO,
1679 			"LLQ is not supported. Fallback to host mode policy.\n");
1680 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1681 		return 0;
1682 	}
1683 
1684 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
1685 	if (unlikely(rc)) {
1686 		PMD_INIT_LOG(WARNING, "Failed to config dev mode. "
1687 			"Fallback to host mode policy.");
1688 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1689 		return 0;
1690 	}
1691 
1692 	/* Nothing to config, exit */
1693 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1694 		return 0;
1695 
1696 	if (!adapter->dev_mem_base) {
1697 		PMD_DRV_LOG(ERR, "Unable to access LLQ bar resource. "
1698 			"Fallback to host mode policy.\n.");
1699 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1700 		return 0;
1701 	}
1702 
1703 	ena_dev->mem_bar = adapter->dev_mem_base;
1704 
1705 	return 0;
1706 }
1707 
1708 static uint32_t ena_calc_max_io_queue_num(struct ena_com_dev *ena_dev,
1709 	struct ena_com_dev_get_features_ctx *get_feat_ctx)
1710 {
1711 	uint32_t io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
1712 
1713 	/* Regular queues capabilities */
1714 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
1715 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
1716 			&get_feat_ctx->max_queue_ext.max_queue_ext;
1717 		io_rx_num = RTE_MIN(max_queue_ext->max_rx_sq_num,
1718 				    max_queue_ext->max_rx_cq_num);
1719 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
1720 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
1721 	} else {
1722 		struct ena_admin_queue_feature_desc *max_queues =
1723 			&get_feat_ctx->max_queues;
1724 		io_tx_sq_num = max_queues->max_sq_num;
1725 		io_tx_cq_num = max_queues->max_cq_num;
1726 		io_rx_num = RTE_MIN(io_tx_sq_num, io_tx_cq_num);
1727 	}
1728 
1729 	/* In case of LLQ use the llq number in the get feature cmd */
1730 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
1731 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
1732 
1733 	max_num_io_queues = RTE_MIN(ENA_MAX_NUM_IO_QUEUES, io_rx_num);
1734 	max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_sq_num);
1735 	max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_cq_num);
1736 
1737 	if (unlikely(max_num_io_queues == 0)) {
1738 		PMD_DRV_LOG(ERR, "Number of IO queues should not be 0\n");
1739 		return -EFAULT;
1740 	}
1741 
1742 	return max_num_io_queues;
1743 }
1744 
1745 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1746 {
1747 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
1748 	struct rte_pci_device *pci_dev;
1749 	struct rte_intr_handle *intr_handle;
1750 	struct ena_adapter *adapter = eth_dev->data->dev_private;
1751 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1752 	struct ena_com_dev_get_features_ctx get_feat_ctx;
1753 	struct ena_llq_configurations llq_config;
1754 	const char *queue_type_str;
1755 	uint32_t max_num_io_queues;
1756 	int rc;
1757 	static int adapters_found;
1758 	bool disable_meta_caching;
1759 	bool wd_state = false;
1760 
1761 	eth_dev->dev_ops = &ena_dev_ops;
1762 	eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1763 	eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1764 	eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1765 
1766 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1767 		return 0;
1768 
1769 	memset(adapter, 0, sizeof(struct ena_adapter));
1770 	ena_dev = &adapter->ena_dev;
1771 
1772 	adapter->rte_eth_dev_data = eth_dev->data;
1773 	adapter->rte_dev = eth_dev;
1774 
1775 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1776 	adapter->pdev = pci_dev;
1777 
1778 	PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1779 		     pci_dev->addr.domain,
1780 		     pci_dev->addr.bus,
1781 		     pci_dev->addr.devid,
1782 		     pci_dev->addr.function);
1783 
1784 	intr_handle = &pci_dev->intr_handle;
1785 
1786 	adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1787 	adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1788 
1789 	if (!adapter->regs) {
1790 		PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1791 			     ENA_REGS_BAR);
1792 		return -ENXIO;
1793 	}
1794 
1795 	ena_dev->reg_bar = adapter->regs;
1796 	ena_dev->dmadev = adapter->pdev;
1797 
1798 	adapter->id_number = adapters_found;
1799 
1800 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1801 		 adapter->id_number);
1802 
1803 	rc = ena_parse_devargs(adapter, pci_dev->device.devargs);
1804 	if (rc != 0) {
1805 		PMD_INIT_LOG(CRIT, "Failed to parse devargs\n");
1806 		goto err;
1807 	}
1808 
1809 	/* device specific initialization routine */
1810 	rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
1811 	if (rc) {
1812 		PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1813 		goto err;
1814 	}
1815 	adapter->wd_state = wd_state;
1816 
1817 	set_default_llq_configurations(&llq_config, &get_feat_ctx.llq,
1818 		adapter->use_large_llq_hdr);
1819 	rc = ena_set_queues_placement_policy(adapter, ena_dev,
1820 					     &get_feat_ctx.llq, &llq_config);
1821 	if (unlikely(rc)) {
1822 		PMD_INIT_LOG(CRIT, "Failed to set placement policy");
1823 		return rc;
1824 	}
1825 
1826 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1827 		queue_type_str = "Regular";
1828 	else
1829 		queue_type_str = "Low latency";
1830 	PMD_DRV_LOG(INFO, "Placement policy: %s\n", queue_type_str);
1831 
1832 	calc_queue_ctx.ena_dev = ena_dev;
1833 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
1834 
1835 	max_num_io_queues = ena_calc_max_io_queue_num(ena_dev, &get_feat_ctx);
1836 	rc = ena_calc_io_queue_size(&calc_queue_ctx,
1837 		adapter->use_large_llq_hdr);
1838 	if (unlikely((rc != 0) || (max_num_io_queues == 0))) {
1839 		rc = -EFAULT;
1840 		goto err_device_destroy;
1841 	}
1842 
1843 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
1844 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
1845 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
1846 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
1847 	adapter->max_num_io_queues = max_num_io_queues;
1848 
1849 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1850 		disable_meta_caching =
1851 			!!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
1852 			BIT(ENA_ADMIN_DISABLE_META_CACHING));
1853 	} else {
1854 		disable_meta_caching = false;
1855 	}
1856 
1857 	/* prepare ring structures */
1858 	ena_init_rings(adapter, disable_meta_caching);
1859 
1860 	ena_config_debug_area(adapter);
1861 
1862 	/* Set max MTU for this device */
1863 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1864 
1865 	/* set device support for offloads */
1866 	adapter->offloads.tso4_supported = (get_feat_ctx.offload.tx &
1867 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0;
1868 	adapter->offloads.tx_csum_supported = (get_feat_ctx.offload.tx &
1869 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK) != 0;
1870 	adapter->offloads.rx_csum_supported =
1871 		(get_feat_ctx.offload.rx_supported &
1872 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK) != 0;
1873 
1874 	/* Copy MAC address and point DPDK to it */
1875 	eth_dev->data->mac_addrs = (struct rte_ether_addr *)adapter->mac_addr;
1876 	rte_ether_addr_copy((struct rte_ether_addr *)
1877 			get_feat_ctx.dev_attr.mac_addr,
1878 			(struct rte_ether_addr *)adapter->mac_addr);
1879 
1880 	/*
1881 	 * Pass the information to the rte_eth_dev_close() that it should also
1882 	 * release the private port resources.
1883 	 */
1884 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1885 
1886 	adapter->drv_stats = rte_zmalloc("adapter stats",
1887 					 sizeof(*adapter->drv_stats),
1888 					 RTE_CACHE_LINE_SIZE);
1889 	if (!adapter->drv_stats) {
1890 		PMD_DRV_LOG(ERR, "failed to alloc mem for adapter stats\n");
1891 		rc = -ENOMEM;
1892 		goto err_delete_debug_area;
1893 	}
1894 
1895 	rte_intr_callback_register(intr_handle,
1896 				   ena_interrupt_handler_rte,
1897 				   adapter);
1898 	rte_intr_enable(intr_handle);
1899 	ena_com_set_admin_polling_mode(ena_dev, false);
1900 	ena_com_admin_aenq_enable(ena_dev);
1901 
1902 	if (adapters_found == 0)
1903 		rte_timer_subsystem_init();
1904 	rte_timer_init(&adapter->timer_wd);
1905 
1906 	adapters_found++;
1907 	adapter->state = ENA_ADAPTER_STATE_INIT;
1908 
1909 	return 0;
1910 
1911 err_delete_debug_area:
1912 	ena_com_delete_debug_area(ena_dev);
1913 
1914 err_device_destroy:
1915 	ena_com_delete_host_info(ena_dev);
1916 	ena_com_admin_destroy(ena_dev);
1917 
1918 err:
1919 	return rc;
1920 }
1921 
1922 static void ena_destroy_device(struct rte_eth_dev *eth_dev)
1923 {
1924 	struct ena_adapter *adapter = eth_dev->data->dev_private;
1925 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1926 
1927 	if (adapter->state == ENA_ADAPTER_STATE_FREE)
1928 		return;
1929 
1930 	ena_com_set_admin_running_state(ena_dev, false);
1931 
1932 	if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1933 		ena_close(eth_dev);
1934 
1935 	ena_com_delete_debug_area(ena_dev);
1936 	ena_com_delete_host_info(ena_dev);
1937 
1938 	ena_com_abort_admin_commands(ena_dev);
1939 	ena_com_wait_for_abort_completion(ena_dev);
1940 	ena_com_admin_destroy(ena_dev);
1941 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1942 
1943 	adapter->state = ENA_ADAPTER_STATE_FREE;
1944 }
1945 
1946 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1947 {
1948 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1949 		return 0;
1950 
1951 	ena_destroy_device(eth_dev);
1952 
1953 	eth_dev->dev_ops = NULL;
1954 	eth_dev->rx_pkt_burst = NULL;
1955 	eth_dev->tx_pkt_burst = NULL;
1956 	eth_dev->tx_pkt_prepare = NULL;
1957 
1958 	return 0;
1959 }
1960 
1961 static int ena_dev_configure(struct rte_eth_dev *dev)
1962 {
1963 	struct ena_adapter *adapter = dev->data->dev_private;
1964 
1965 	adapter->state = ENA_ADAPTER_STATE_CONFIG;
1966 
1967 	adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1968 	adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1969 	return 0;
1970 }
1971 
1972 static void ena_init_rings(struct ena_adapter *adapter,
1973 			   bool disable_meta_caching)
1974 {
1975 	size_t i;
1976 
1977 	for (i = 0; i < adapter->max_num_io_queues; i++) {
1978 		struct ena_ring *ring = &adapter->tx_ring[i];
1979 
1980 		ring->configured = 0;
1981 		ring->type = ENA_RING_TYPE_TX;
1982 		ring->adapter = adapter;
1983 		ring->id = i;
1984 		ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1985 		ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1986 		ring->sgl_size = adapter->max_tx_sgl_size;
1987 		ring->disable_meta_caching = disable_meta_caching;
1988 	}
1989 
1990 	for (i = 0; i < adapter->max_num_io_queues; i++) {
1991 		struct ena_ring *ring = &adapter->rx_ring[i];
1992 
1993 		ring->configured = 0;
1994 		ring->type = ENA_RING_TYPE_RX;
1995 		ring->adapter = adapter;
1996 		ring->id = i;
1997 		ring->sgl_size = adapter->max_rx_sgl_size;
1998 	}
1999 }
2000 
2001 static int ena_infos_get(struct rte_eth_dev *dev,
2002 			  struct rte_eth_dev_info *dev_info)
2003 {
2004 	struct ena_adapter *adapter;
2005 	struct ena_com_dev *ena_dev;
2006 	uint64_t rx_feat = 0, tx_feat = 0;
2007 
2008 	ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
2009 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
2010 	adapter = dev->data->dev_private;
2011 
2012 	ena_dev = &adapter->ena_dev;
2013 	ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
2014 
2015 	dev_info->speed_capa =
2016 			ETH_LINK_SPEED_1G   |
2017 			ETH_LINK_SPEED_2_5G |
2018 			ETH_LINK_SPEED_5G   |
2019 			ETH_LINK_SPEED_10G  |
2020 			ETH_LINK_SPEED_25G  |
2021 			ETH_LINK_SPEED_40G  |
2022 			ETH_LINK_SPEED_50G  |
2023 			ETH_LINK_SPEED_100G;
2024 
2025 	/* Set Tx & Rx features available for device */
2026 	if (adapter->offloads.tso4_supported)
2027 		tx_feat	|= DEV_TX_OFFLOAD_TCP_TSO;
2028 
2029 	if (adapter->offloads.tx_csum_supported)
2030 		tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
2031 			DEV_TX_OFFLOAD_UDP_CKSUM |
2032 			DEV_TX_OFFLOAD_TCP_CKSUM;
2033 
2034 	if (adapter->offloads.rx_csum_supported)
2035 		rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
2036 			DEV_RX_OFFLOAD_UDP_CKSUM  |
2037 			DEV_RX_OFFLOAD_TCP_CKSUM;
2038 
2039 	rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2040 
2041 	/* Inform framework about available features */
2042 	dev_info->rx_offload_capa = rx_feat;
2043 	dev_info->rx_queue_offload_capa = rx_feat;
2044 	dev_info->tx_offload_capa = tx_feat;
2045 	dev_info->tx_queue_offload_capa = tx_feat;
2046 
2047 	dev_info->flow_type_rss_offloads = ETH_RSS_IP | ETH_RSS_TCP |
2048 					   ETH_RSS_UDP;
2049 
2050 	dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
2051 	dev_info->max_rx_pktlen  = adapter->max_mtu;
2052 	dev_info->max_mac_addrs = 1;
2053 
2054 	dev_info->max_rx_queues = adapter->max_num_io_queues;
2055 	dev_info->max_tx_queues = adapter->max_num_io_queues;
2056 	dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
2057 
2058 	adapter->tx_supported_offloads = tx_feat;
2059 	adapter->rx_supported_offloads = rx_feat;
2060 
2061 	dev_info->rx_desc_lim.nb_max = adapter->max_rx_ring_size;
2062 	dev_info->rx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2063 	dev_info->rx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2064 					adapter->max_rx_sgl_size);
2065 	dev_info->rx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2066 					adapter->max_rx_sgl_size);
2067 
2068 	dev_info->tx_desc_lim.nb_max = adapter->max_tx_ring_size;
2069 	dev_info->tx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2070 	dev_info->tx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2071 					adapter->max_tx_sgl_size);
2072 	dev_info->tx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2073 					adapter->max_tx_sgl_size);
2074 
2075 	return 0;
2076 }
2077 
2078 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len)
2079 {
2080 	mbuf->data_len = len;
2081 	mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2082 	mbuf->refcnt = 1;
2083 	mbuf->next = NULL;
2084 }
2085 
2086 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
2087 				    struct ena_com_rx_buf_info *ena_bufs,
2088 				    uint32_t descs,
2089 				    uint16_t *next_to_clean,
2090 				    uint8_t offset)
2091 {
2092 	struct rte_mbuf *mbuf;
2093 	struct rte_mbuf *mbuf_head;
2094 	struct ena_rx_buffer *rx_info;
2095 	int rc;
2096 	uint16_t ntc, len, req_id, buf = 0;
2097 
2098 	if (unlikely(descs == 0))
2099 		return NULL;
2100 
2101 	ntc = *next_to_clean;
2102 
2103 	len = ena_bufs[buf].len;
2104 	req_id = ena_bufs[buf].req_id;
2105 	if (unlikely(validate_rx_req_id(rx_ring, req_id)))
2106 		return NULL;
2107 
2108 	rx_info = &rx_ring->rx_buffer_info[req_id];
2109 
2110 	mbuf = rx_info->mbuf;
2111 	RTE_ASSERT(mbuf != NULL);
2112 
2113 	ena_init_rx_mbuf(mbuf, len);
2114 
2115 	/* Fill the mbuf head with the data specific for 1st segment. */
2116 	mbuf_head = mbuf;
2117 	mbuf_head->nb_segs = descs;
2118 	mbuf_head->port = rx_ring->port_id;
2119 	mbuf_head->pkt_len = len;
2120 	mbuf_head->data_off += offset;
2121 
2122 	rx_info->mbuf = NULL;
2123 	rx_ring->empty_rx_reqs[ntc] = req_id;
2124 	ntc = ENA_IDX_NEXT_MASKED(ntc, rx_ring->size_mask);
2125 
2126 	while (--descs) {
2127 		++buf;
2128 		len = ena_bufs[buf].len;
2129 		req_id = ena_bufs[buf].req_id;
2130 		if (unlikely(validate_rx_req_id(rx_ring, req_id))) {
2131 			rte_mbuf_raw_free(mbuf_head);
2132 			return NULL;
2133 		}
2134 
2135 		rx_info = &rx_ring->rx_buffer_info[req_id];
2136 		RTE_ASSERT(rx_info->mbuf != NULL);
2137 
2138 		if (unlikely(len == 0)) {
2139 			/*
2140 			 * Some devices can pass descriptor with the length 0.
2141 			 * To avoid confusion, the PMD is simply putting the
2142 			 * descriptor back, as it was never used. We'll avoid
2143 			 * mbuf allocation that way.
2144 			 */
2145 			rc = ena_add_single_rx_desc(rx_ring->ena_com_io_sq,
2146 				rx_info->mbuf, req_id);
2147 			if (unlikely(rc != 0)) {
2148 				/* Free the mbuf in case of an error. */
2149 				rte_mbuf_raw_free(rx_info->mbuf);
2150 			} else {
2151 				/*
2152 				 * If there was no error, just exit the loop as
2153 				 * 0 length descriptor is always the last one.
2154 				 */
2155 				break;
2156 			}
2157 		} else {
2158 			/* Create an mbuf chain. */
2159 			mbuf->next = rx_info->mbuf;
2160 			mbuf = mbuf->next;
2161 
2162 			ena_init_rx_mbuf(mbuf, len);
2163 			mbuf_head->pkt_len += len;
2164 		}
2165 
2166 		/*
2167 		 * Mark the descriptor as depleted and perform necessary
2168 		 * cleanup.
2169 		 * This code will execute in two cases:
2170 		 *  1. Descriptor len was greater than 0 - normal situation.
2171 		 *  2. Descriptor len was 0 and we failed to add the descriptor
2172 		 *     to the device. In that situation, we should try to add
2173 		 *     the mbuf again in the populate routine and mark the
2174 		 *     descriptor as used up by the device.
2175 		 */
2176 		rx_info->mbuf = NULL;
2177 		rx_ring->empty_rx_reqs[ntc] = req_id;
2178 		ntc = ENA_IDX_NEXT_MASKED(ntc, rx_ring->size_mask);
2179 	}
2180 
2181 	*next_to_clean = ntc;
2182 
2183 	return mbuf_head;
2184 }
2185 
2186 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
2187 				  uint16_t nb_pkts)
2188 {
2189 	struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
2190 	unsigned int free_queue_entries;
2191 	unsigned int refill_threshold;
2192 	uint16_t next_to_clean = rx_ring->next_to_clean;
2193 	uint16_t descs_in_use;
2194 	struct rte_mbuf *mbuf;
2195 	uint16_t completed;
2196 	struct ena_com_rx_ctx ena_rx_ctx;
2197 	int i, rc = 0;
2198 
2199 	/* Check adapter state */
2200 	if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2201 		PMD_DRV_LOG(ALERT,
2202 			"Trying to receive pkts while device is NOT running\n");
2203 		return 0;
2204 	}
2205 
2206 	descs_in_use = rx_ring->ring_size -
2207 		ena_com_free_q_entries(rx_ring->ena_com_io_sq) - 1;
2208 	nb_pkts = RTE_MIN(descs_in_use, nb_pkts);
2209 
2210 	for (completed = 0; completed < nb_pkts; completed++) {
2211 		ena_rx_ctx.max_bufs = rx_ring->sgl_size;
2212 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
2213 		ena_rx_ctx.descs = 0;
2214 		ena_rx_ctx.pkt_offset = 0;
2215 		/* receive packet context */
2216 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
2217 				    rx_ring->ena_com_io_sq,
2218 				    &ena_rx_ctx);
2219 		if (unlikely(rc)) {
2220 			PMD_DRV_LOG(ERR, "ena_com_rx_pkt error %d\n", rc);
2221 			rx_ring->adapter->reset_reason =
2222 				ENA_REGS_RESET_TOO_MANY_RX_DESCS;
2223 			rx_ring->adapter->trigger_reset = true;
2224 			++rx_ring->rx_stats.bad_desc_num;
2225 			return 0;
2226 		}
2227 
2228 		mbuf = ena_rx_mbuf(rx_ring,
2229 			ena_rx_ctx.ena_bufs,
2230 			ena_rx_ctx.descs,
2231 			&next_to_clean,
2232 			ena_rx_ctx.pkt_offset);
2233 		if (unlikely(mbuf == NULL)) {
2234 			for (i = 0; i < ena_rx_ctx.descs; ++i) {
2235 				rx_ring->empty_rx_reqs[next_to_clean] =
2236 					rx_ring->ena_bufs[i].req_id;
2237 				next_to_clean = ENA_IDX_NEXT_MASKED(
2238 					next_to_clean, rx_ring->size_mask);
2239 			}
2240 			break;
2241 		}
2242 
2243 		/* fill mbuf attributes if any */
2244 		ena_rx_mbuf_prepare(mbuf, &ena_rx_ctx);
2245 
2246 		if (unlikely(mbuf->ol_flags &
2247 				(PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD))) {
2248 			rte_atomic64_inc(&rx_ring->adapter->drv_stats->ierrors);
2249 			++rx_ring->rx_stats.bad_csum;
2250 		}
2251 
2252 		mbuf->hash.rss = ena_rx_ctx.hash;
2253 
2254 		rx_pkts[completed] = mbuf;
2255 		rx_ring->rx_stats.bytes += mbuf->pkt_len;
2256 	}
2257 
2258 	rx_ring->rx_stats.cnt += completed;
2259 	rx_ring->next_to_clean = next_to_clean;
2260 
2261 	free_queue_entries = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
2262 	refill_threshold =
2263 		RTE_MIN(rx_ring->ring_size / ENA_REFILL_THRESH_DIVIDER,
2264 		(unsigned int)ENA_REFILL_THRESH_PACKET);
2265 
2266 	/* Burst refill to save doorbells, memory barriers, const interval */
2267 	if (free_queue_entries > refill_threshold) {
2268 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
2269 		ena_populate_rx_queue(rx_ring, free_queue_entries);
2270 	}
2271 
2272 	return completed;
2273 }
2274 
2275 static uint16_t
2276 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2277 		uint16_t nb_pkts)
2278 {
2279 	int32_t ret;
2280 	uint32_t i;
2281 	struct rte_mbuf *m;
2282 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2283 	struct rte_ipv4_hdr *ip_hdr;
2284 	uint64_t ol_flags;
2285 	uint16_t frag_field;
2286 
2287 	for (i = 0; i != nb_pkts; i++) {
2288 		m = tx_pkts[i];
2289 		ol_flags = m->ol_flags;
2290 
2291 		if (!(ol_flags & PKT_TX_IPV4))
2292 			continue;
2293 
2294 		/* If there was not L2 header length specified, assume it is
2295 		 * length of the ethernet header.
2296 		 */
2297 		if (unlikely(m->l2_len == 0))
2298 			m->l2_len = sizeof(struct rte_ether_hdr);
2299 
2300 		ip_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
2301 						 m->l2_len);
2302 		frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
2303 
2304 		if ((frag_field & RTE_IPV4_HDR_DF_FLAG) != 0) {
2305 			m->packet_type |= RTE_PTYPE_L4_NONFRAG;
2306 
2307 			/* If IPv4 header has DF flag enabled and TSO support is
2308 			 * disabled, partial chcecksum should not be calculated.
2309 			 */
2310 			if (!tx_ring->adapter->offloads.tso4_supported)
2311 				continue;
2312 		}
2313 
2314 		if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
2315 				(ol_flags & PKT_TX_L4_MASK) ==
2316 				PKT_TX_SCTP_CKSUM) {
2317 			rte_errno = ENOTSUP;
2318 			return i;
2319 		}
2320 
2321 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2322 		ret = rte_validate_tx_offload(m);
2323 		if (ret != 0) {
2324 			rte_errno = -ret;
2325 			return i;
2326 		}
2327 #endif
2328 
2329 		/* In case we are supposed to TSO and have DF not set (DF=0)
2330 		 * hardware must be provided with partial checksum, otherwise
2331 		 * it will take care of necessary calculations.
2332 		 */
2333 
2334 		ret = rte_net_intel_cksum_flags_prepare(m,
2335 			ol_flags & ~PKT_TX_TCP_SEG);
2336 		if (ret != 0) {
2337 			rte_errno = -ret;
2338 			return i;
2339 		}
2340 	}
2341 
2342 	return i;
2343 }
2344 
2345 static void ena_update_hints(struct ena_adapter *adapter,
2346 			     struct ena_admin_ena_hw_hints *hints)
2347 {
2348 	if (hints->admin_completion_tx_timeout)
2349 		adapter->ena_dev.admin_queue.completion_timeout =
2350 			hints->admin_completion_tx_timeout * 1000;
2351 
2352 	if (hints->mmio_read_timeout)
2353 		/* convert to usec */
2354 		adapter->ena_dev.mmio_read.reg_read_to =
2355 			hints->mmio_read_timeout * 1000;
2356 
2357 	if (hints->driver_watchdog_timeout) {
2358 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2359 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2360 		else
2361 			// Convert msecs to ticks
2362 			adapter->keep_alive_timeout =
2363 				(hints->driver_watchdog_timeout *
2364 				rte_get_timer_hz()) / 1000;
2365 	}
2366 }
2367 
2368 static int ena_check_and_linearize_mbuf(struct ena_ring *tx_ring,
2369 					struct rte_mbuf *mbuf)
2370 {
2371 	struct ena_com_dev *ena_dev;
2372 	int num_segments, header_len, rc;
2373 
2374 	ena_dev = &tx_ring->adapter->ena_dev;
2375 	num_segments = mbuf->nb_segs;
2376 	header_len = mbuf->data_len;
2377 
2378 	if (likely(num_segments < tx_ring->sgl_size))
2379 		return 0;
2380 
2381 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV &&
2382 	    (num_segments == tx_ring->sgl_size) &&
2383 	    (header_len < tx_ring->tx_max_header_size))
2384 		return 0;
2385 
2386 	++tx_ring->tx_stats.linearize;
2387 	rc = rte_pktmbuf_linearize(mbuf);
2388 	if (unlikely(rc)) {
2389 		PMD_DRV_LOG(WARNING, "Mbuf linearize failed\n");
2390 		rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
2391 		++tx_ring->tx_stats.linearize_failed;
2392 		return rc;
2393 	}
2394 
2395 	return rc;
2396 }
2397 
2398 static void ena_tx_map_mbuf(struct ena_ring *tx_ring,
2399 	struct ena_tx_buffer *tx_info,
2400 	struct rte_mbuf *mbuf,
2401 	void **push_header,
2402 	uint16_t *header_len)
2403 {
2404 	struct ena_com_buf *ena_buf;
2405 	uint16_t delta, seg_len, push_len;
2406 
2407 	delta = 0;
2408 	seg_len = mbuf->data_len;
2409 
2410 	tx_info->mbuf = mbuf;
2411 	ena_buf = tx_info->bufs;
2412 
2413 	if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2414 		/*
2415 		 * Tx header might be (and will be in most cases) smaller than
2416 		 * tx_max_header_size. But it's not an issue to send more data
2417 		 * to the device, than actually needed if the mbuf size is
2418 		 * greater than tx_max_header_size.
2419 		 */
2420 		push_len = RTE_MIN(mbuf->pkt_len, tx_ring->tx_max_header_size);
2421 		*header_len = push_len;
2422 
2423 		if (likely(push_len <= seg_len)) {
2424 			/* If the push header is in the single segment, then
2425 			 * just point it to the 1st mbuf data.
2426 			 */
2427 			*push_header = rte_pktmbuf_mtod(mbuf, uint8_t *);
2428 		} else {
2429 			/* If the push header lays in the several segments, copy
2430 			 * it to the intermediate buffer.
2431 			 */
2432 			rte_pktmbuf_read(mbuf, 0, push_len,
2433 				tx_ring->push_buf_intermediate_buf);
2434 			*push_header = tx_ring->push_buf_intermediate_buf;
2435 			delta = push_len - seg_len;
2436 		}
2437 	} else {
2438 		*push_header = NULL;
2439 		*header_len = 0;
2440 		push_len = 0;
2441 	}
2442 
2443 	/* Process first segment taking into consideration pushed header */
2444 	if (seg_len > push_len) {
2445 		ena_buf->paddr = mbuf->buf_iova +
2446 				mbuf->data_off +
2447 				push_len;
2448 		ena_buf->len = seg_len - push_len;
2449 		ena_buf++;
2450 		tx_info->num_of_bufs++;
2451 	}
2452 
2453 	while ((mbuf = mbuf->next) != NULL) {
2454 		seg_len = mbuf->data_len;
2455 
2456 		/* Skip mbufs if whole data is pushed as a header */
2457 		if (unlikely(delta > seg_len)) {
2458 			delta -= seg_len;
2459 			continue;
2460 		}
2461 
2462 		ena_buf->paddr = mbuf->buf_iova + mbuf->data_off + delta;
2463 		ena_buf->len = seg_len - delta;
2464 		ena_buf++;
2465 		tx_info->num_of_bufs++;
2466 
2467 		delta = 0;
2468 	}
2469 }
2470 
2471 static int ena_xmit_mbuf(struct ena_ring *tx_ring, struct rte_mbuf *mbuf)
2472 {
2473 	struct ena_tx_buffer *tx_info;
2474 	struct ena_com_tx_ctx ena_tx_ctx = { { 0 } };
2475 	uint16_t next_to_use;
2476 	uint16_t header_len;
2477 	uint16_t req_id;
2478 	void *push_header;
2479 	int nb_hw_desc;
2480 	int rc;
2481 
2482 	rc = ena_check_and_linearize_mbuf(tx_ring, mbuf);
2483 	if (unlikely(rc))
2484 		return rc;
2485 
2486 	next_to_use = tx_ring->next_to_use;
2487 
2488 	req_id = tx_ring->empty_tx_reqs[next_to_use];
2489 	tx_info = &tx_ring->tx_buffer_info[req_id];
2490 	tx_info->num_of_bufs = 0;
2491 
2492 	ena_tx_map_mbuf(tx_ring, tx_info, mbuf, &push_header, &header_len);
2493 
2494 	ena_tx_ctx.ena_bufs = tx_info->bufs;
2495 	ena_tx_ctx.push_header = push_header;
2496 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2497 	ena_tx_ctx.req_id = req_id;
2498 	ena_tx_ctx.header_len = header_len;
2499 
2500 	/* Set Tx offloads flags, if applicable */
2501 	ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads,
2502 		tx_ring->disable_meta_caching);
2503 
2504 	if (unlikely(ena_com_is_doorbell_needed(tx_ring->ena_com_io_sq,
2505 			&ena_tx_ctx))) {
2506 		PMD_DRV_LOG(DEBUG,
2507 			"llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
2508 			tx_ring->id);
2509 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2510 	}
2511 
2512 	/* prepare the packet's descriptors to dma engine */
2513 	rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,	&ena_tx_ctx,
2514 		&nb_hw_desc);
2515 	if (unlikely(rc)) {
2516 		++tx_ring->tx_stats.prepare_ctx_err;
2517 		return rc;
2518 	}
2519 
2520 	tx_info->tx_descs = nb_hw_desc;
2521 
2522 	tx_ring->tx_stats.cnt++;
2523 	tx_ring->tx_stats.bytes += mbuf->pkt_len;
2524 
2525 	tx_ring->next_to_use = ENA_IDX_NEXT_MASKED(next_to_use,
2526 		tx_ring->size_mask);
2527 
2528 	return 0;
2529 }
2530 
2531 static void ena_tx_cleanup(struct ena_ring *tx_ring)
2532 {
2533 	unsigned int cleanup_budget;
2534 	unsigned int total_tx_descs = 0;
2535 	uint16_t next_to_clean = tx_ring->next_to_clean;
2536 
2537 	cleanup_budget = RTE_MIN(tx_ring->ring_size / ENA_REFILL_THRESH_DIVIDER,
2538 		(unsigned int)ENA_REFILL_THRESH_PACKET);
2539 
2540 	while (likely(total_tx_descs < cleanup_budget)) {
2541 		struct rte_mbuf *mbuf;
2542 		struct ena_tx_buffer *tx_info;
2543 		uint16_t req_id;
2544 
2545 		if (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) != 0)
2546 			break;
2547 
2548 		if (unlikely(validate_tx_req_id(tx_ring, req_id) != 0))
2549 			break;
2550 
2551 		/* Get Tx info & store how many descs were processed  */
2552 		tx_info = &tx_ring->tx_buffer_info[req_id];
2553 
2554 		mbuf = tx_info->mbuf;
2555 		rte_pktmbuf_free(mbuf);
2556 
2557 		tx_info->mbuf = NULL;
2558 		tx_ring->empty_tx_reqs[next_to_clean] = req_id;
2559 
2560 		total_tx_descs += tx_info->tx_descs;
2561 
2562 		/* Put back descriptor to the ring for reuse */
2563 		next_to_clean = ENA_IDX_NEXT_MASKED(next_to_clean,
2564 			tx_ring->size_mask);
2565 	}
2566 
2567 	if (likely(total_tx_descs > 0)) {
2568 		/* acknowledge completion of sent packets */
2569 		tx_ring->next_to_clean = next_to_clean;
2570 		ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
2571 		ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
2572 	}
2573 }
2574 
2575 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2576 				  uint16_t nb_pkts)
2577 {
2578 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2579 	uint16_t sent_idx = 0;
2580 
2581 	/* Check adapter state */
2582 	if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2583 		PMD_DRV_LOG(ALERT,
2584 			"Trying to xmit pkts while device is NOT running\n");
2585 		return 0;
2586 	}
2587 
2588 	nb_pkts = RTE_MIN(ena_com_free_q_entries(tx_ring->ena_com_io_sq),
2589 		nb_pkts);
2590 
2591 	for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
2592 		if (ena_xmit_mbuf(tx_ring, tx_pkts[sent_idx]))
2593 			break;
2594 
2595 		rte_prefetch0(tx_pkts[ENA_IDX_ADD_MASKED(sent_idx, 4,
2596 			tx_ring->size_mask)]);
2597 	}
2598 
2599 	tx_ring->tx_stats.available_desc =
2600 		ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2601 
2602 	/* If there are ready packets to be xmitted... */
2603 	if (sent_idx > 0) {
2604 		/* ...let HW do its best :-) */
2605 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2606 		tx_ring->tx_stats.doorbells++;
2607 	}
2608 
2609 	ena_tx_cleanup(tx_ring);
2610 
2611 	tx_ring->tx_stats.available_desc =
2612 		ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2613 	tx_ring->tx_stats.tx_poll++;
2614 
2615 	return sent_idx;
2616 }
2617 
2618 /**
2619  * DPDK callback to retrieve names of extended device statistics
2620  *
2621  * @param dev
2622  *   Pointer to Ethernet device structure.
2623  * @param[out] xstats_names
2624  *   Buffer to insert names into.
2625  * @param n
2626  *   Number of names.
2627  *
2628  * @return
2629  *   Number of xstats names.
2630  */
2631 static int ena_xstats_get_names(struct rte_eth_dev *dev,
2632 				struct rte_eth_xstat_name *xstats_names,
2633 				unsigned int n)
2634 {
2635 	unsigned int xstats_count = ena_xstats_calc_num(dev);
2636 	unsigned int stat, i, count = 0;
2637 
2638 	if (n < xstats_count || !xstats_names)
2639 		return xstats_count;
2640 
2641 	for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++)
2642 		strcpy(xstats_names[count].name,
2643 			ena_stats_global_strings[stat].name);
2644 
2645 	for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++)
2646 		for (i = 0; i < dev->data->nb_rx_queues; i++, count++)
2647 			snprintf(xstats_names[count].name,
2648 				sizeof(xstats_names[count].name),
2649 				"rx_q%d_%s", i,
2650 				ena_stats_rx_strings[stat].name);
2651 
2652 	for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++)
2653 		for (i = 0; i < dev->data->nb_tx_queues; i++, count++)
2654 			snprintf(xstats_names[count].name,
2655 				sizeof(xstats_names[count].name),
2656 				"tx_q%d_%s", i,
2657 				ena_stats_tx_strings[stat].name);
2658 
2659 	return xstats_count;
2660 }
2661 
2662 /**
2663  * DPDK callback to get extended device statistics.
2664  *
2665  * @param dev
2666  *   Pointer to Ethernet device structure.
2667  * @param[out] stats
2668  *   Stats table output buffer.
2669  * @param n
2670  *   The size of the stats table.
2671  *
2672  * @return
2673  *   Number of xstats on success, negative on failure.
2674  */
2675 static int ena_xstats_get(struct rte_eth_dev *dev,
2676 			  struct rte_eth_xstat *xstats,
2677 			  unsigned int n)
2678 {
2679 	struct ena_adapter *adapter = dev->data->dev_private;
2680 	unsigned int xstats_count = ena_xstats_calc_num(dev);
2681 	unsigned int stat, i, count = 0;
2682 	int stat_offset;
2683 	void *stats_begin;
2684 
2685 	if (n < xstats_count)
2686 		return xstats_count;
2687 
2688 	if (!xstats)
2689 		return 0;
2690 
2691 	for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++) {
2692 		stat_offset = ena_stats_rx_strings[stat].stat_offset;
2693 		stats_begin = &adapter->dev_stats;
2694 
2695 		xstats[count].id = count;
2696 		xstats[count].value = *((uint64_t *)
2697 			((char *)stats_begin + stat_offset));
2698 	}
2699 
2700 	for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++) {
2701 		for (i = 0; i < dev->data->nb_rx_queues; i++, count++) {
2702 			stat_offset = ena_stats_rx_strings[stat].stat_offset;
2703 			stats_begin = &adapter->rx_ring[i].rx_stats;
2704 
2705 			xstats[count].id = count;
2706 			xstats[count].value = *((uint64_t *)
2707 				((char *)stats_begin + stat_offset));
2708 		}
2709 	}
2710 
2711 	for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++) {
2712 		for (i = 0; i < dev->data->nb_tx_queues; i++, count++) {
2713 			stat_offset = ena_stats_tx_strings[stat].stat_offset;
2714 			stats_begin = &adapter->tx_ring[i].rx_stats;
2715 
2716 			xstats[count].id = count;
2717 			xstats[count].value = *((uint64_t *)
2718 				((char *)stats_begin + stat_offset));
2719 		}
2720 	}
2721 
2722 	return count;
2723 }
2724 
2725 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
2726 				const uint64_t *ids,
2727 				uint64_t *values,
2728 				unsigned int n)
2729 {
2730 	struct ena_adapter *adapter = dev->data->dev_private;
2731 	uint64_t id;
2732 	uint64_t rx_entries, tx_entries;
2733 	unsigned int i;
2734 	int qid;
2735 	int valid = 0;
2736 	for (i = 0; i < n; ++i) {
2737 		id = ids[i];
2738 		/* Check if id belongs to global statistics */
2739 		if (id < ENA_STATS_ARRAY_GLOBAL) {
2740 			values[i] = *((uint64_t *)&adapter->dev_stats + id);
2741 			++valid;
2742 			continue;
2743 		}
2744 
2745 		/* Check if id belongs to rx queue statistics */
2746 		id -= ENA_STATS_ARRAY_GLOBAL;
2747 		rx_entries = ENA_STATS_ARRAY_RX * dev->data->nb_rx_queues;
2748 		if (id < rx_entries) {
2749 			qid = id % dev->data->nb_rx_queues;
2750 			id /= dev->data->nb_rx_queues;
2751 			values[i] = *((uint64_t *)
2752 				&adapter->rx_ring[qid].rx_stats + id);
2753 			++valid;
2754 			continue;
2755 		}
2756 				/* Check if id belongs to rx queue statistics */
2757 		id -= rx_entries;
2758 		tx_entries = ENA_STATS_ARRAY_TX * dev->data->nb_tx_queues;
2759 		if (id < tx_entries) {
2760 			qid = id % dev->data->nb_tx_queues;
2761 			id /= dev->data->nb_tx_queues;
2762 			values[i] = *((uint64_t *)
2763 				&adapter->tx_ring[qid].tx_stats + id);
2764 			++valid;
2765 			continue;
2766 		}
2767 	}
2768 
2769 	return valid;
2770 }
2771 
2772 static int ena_process_bool_devarg(const char *key,
2773 				   const char *value,
2774 				   void *opaque)
2775 {
2776 	struct ena_adapter *adapter = opaque;
2777 	bool bool_value;
2778 
2779 	/* Parse the value. */
2780 	if (strcmp(value, "1") == 0) {
2781 		bool_value = true;
2782 	} else if (strcmp(value, "0") == 0) {
2783 		bool_value = false;
2784 	} else {
2785 		PMD_INIT_LOG(ERR,
2786 			"Invalid value: '%s' for key '%s'. Accepted: '0' or '1'\n",
2787 			value, key);
2788 		return -EINVAL;
2789 	}
2790 
2791 	/* Now, assign it to the proper adapter field. */
2792 	if (strcmp(key, ENA_DEVARG_LARGE_LLQ_HDR))
2793 		adapter->use_large_llq_hdr = bool_value;
2794 
2795 	return 0;
2796 }
2797 
2798 static int ena_parse_devargs(struct ena_adapter *adapter,
2799 			     struct rte_devargs *devargs)
2800 {
2801 	static const char * const allowed_args[] = {
2802 		ENA_DEVARG_LARGE_LLQ_HDR,
2803 	};
2804 	struct rte_kvargs *kvlist;
2805 	int rc;
2806 
2807 	if (devargs == NULL)
2808 		return 0;
2809 
2810 	kvlist = rte_kvargs_parse(devargs->args, allowed_args);
2811 	if (kvlist == NULL) {
2812 		PMD_INIT_LOG(ERR, "Invalid device arguments: %s\n",
2813 			devargs->args);
2814 		return -EINVAL;
2815 	}
2816 
2817 	rc = rte_kvargs_process(kvlist, ENA_DEVARG_LARGE_LLQ_HDR,
2818 		ena_process_bool_devarg, adapter);
2819 
2820 	rte_kvargs_free(kvlist);
2821 
2822 	return rc;
2823 }
2824 
2825 /*********************************************************************
2826  *  PMD configuration
2827  *********************************************************************/
2828 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2829 	struct rte_pci_device *pci_dev)
2830 {
2831 	return rte_eth_dev_pci_generic_probe(pci_dev,
2832 		sizeof(struct ena_adapter), eth_ena_dev_init);
2833 }
2834 
2835 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
2836 {
2837 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
2838 }
2839 
2840 static struct rte_pci_driver rte_ena_pmd = {
2841 	.id_table = pci_id_ena_map,
2842 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
2843 		     RTE_PCI_DRV_WC_ACTIVATE,
2844 	.probe = eth_ena_pci_probe,
2845 	.remove = eth_ena_pci_remove,
2846 };
2847 
2848 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
2849 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
2850 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
2851 RTE_PMD_REGISTER_PARAM_STRING(net_ena, ENA_DEVARG_LARGE_LLQ_HDR "=<0|1>");
2852 
2853 RTE_INIT(ena_init_log)
2854 {
2855 	ena_logtype_init = rte_log_register("pmd.net.ena.init");
2856 	if (ena_logtype_init >= 0)
2857 		rte_log_set_level(ena_logtype_init, RTE_LOG_NOTICE);
2858 	ena_logtype_driver = rte_log_register("pmd.net.ena.driver");
2859 	if (ena_logtype_driver >= 0)
2860 		rte_log_set_level(ena_logtype_driver, RTE_LOG_NOTICE);
2861 
2862 #ifdef RTE_LIBRTE_ENA_DEBUG_RX
2863 	ena_logtype_rx = rte_log_register("pmd.net.ena.rx");
2864 	if (ena_logtype_rx >= 0)
2865 		rte_log_set_level(ena_logtype_rx, RTE_LOG_NOTICE);
2866 #endif
2867 
2868 #ifdef RTE_LIBRTE_ENA_DEBUG_TX
2869 	ena_logtype_tx = rte_log_register("pmd.net.ena.tx");
2870 	if (ena_logtype_tx >= 0)
2871 		rte_log_set_level(ena_logtype_tx, RTE_LOG_NOTICE);
2872 #endif
2873 
2874 #ifdef RTE_LIBRTE_ENA_DEBUG_TX_FREE
2875 	ena_logtype_tx_free = rte_log_register("pmd.net.ena.tx_free");
2876 	if (ena_logtype_tx_free >= 0)
2877 		rte_log_set_level(ena_logtype_tx_free, RTE_LOG_NOTICE);
2878 #endif
2879 
2880 #ifdef RTE_LIBRTE_ENA_COM_DEBUG
2881 	ena_logtype_com = rte_log_register("pmd.net.ena.com");
2882 	if (ena_logtype_com >= 0)
2883 		rte_log_set_level(ena_logtype_com, RTE_LOG_NOTICE);
2884 #endif
2885 }
2886 
2887 /******************************************************************************
2888  ******************************** AENQ Handlers *******************************
2889  *****************************************************************************/
2890 static void ena_update_on_link_change(void *adapter_data,
2891 				      struct ena_admin_aenq_entry *aenq_e)
2892 {
2893 	struct rte_eth_dev *eth_dev;
2894 	struct ena_adapter *adapter;
2895 	struct ena_admin_aenq_link_change_desc *aenq_link_desc;
2896 	uint32_t status;
2897 
2898 	adapter = adapter_data;
2899 	aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
2900 	eth_dev = adapter->rte_dev;
2901 
2902 	status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
2903 	adapter->link_status = status;
2904 
2905 	ena_link_update(eth_dev, 0);
2906 	_rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2907 }
2908 
2909 static void ena_notification(void *data,
2910 			     struct ena_admin_aenq_entry *aenq_e)
2911 {
2912 	struct ena_adapter *adapter = data;
2913 	struct ena_admin_ena_hw_hints *hints;
2914 
2915 	if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
2916 		PMD_DRV_LOG(WARNING, "Invalid group(%x) expected %x\n",
2917 			aenq_e->aenq_common_desc.group,
2918 			ENA_ADMIN_NOTIFICATION);
2919 
2920 	switch (aenq_e->aenq_common_desc.syndrom) {
2921 	case ENA_ADMIN_UPDATE_HINTS:
2922 		hints = (struct ena_admin_ena_hw_hints *)
2923 			(&aenq_e->inline_data_w4);
2924 		ena_update_hints(adapter, hints);
2925 		break;
2926 	default:
2927 		PMD_DRV_LOG(ERR, "Invalid aenq notification link state %d\n",
2928 			aenq_e->aenq_common_desc.syndrom);
2929 	}
2930 }
2931 
2932 static void ena_keep_alive(void *adapter_data,
2933 			   __rte_unused struct ena_admin_aenq_entry *aenq_e)
2934 {
2935 	struct ena_adapter *adapter = adapter_data;
2936 	struct ena_admin_aenq_keep_alive_desc *desc;
2937 	uint64_t rx_drops;
2938 	uint64_t tx_drops;
2939 
2940 	adapter->timestamp_wd = rte_get_timer_cycles();
2941 
2942 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
2943 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
2944 	tx_drops = ((uint64_t)desc->tx_drops_high << 32) | desc->tx_drops_low;
2945 
2946 	adapter->drv_stats->rx_drops = rx_drops;
2947 	adapter->dev_stats.tx_drops = tx_drops;
2948 }
2949 
2950 /**
2951  * This handler will called for unknown event group or unimplemented handlers
2952  **/
2953 static void unimplemented_aenq_handler(__rte_unused void *data,
2954 				       __rte_unused struct ena_admin_aenq_entry *aenq_e)
2955 {
2956 	PMD_DRV_LOG(ERR, "Unknown event was received or event with "
2957 			  "unimplemented handler\n");
2958 }
2959 
2960 static struct ena_aenq_handlers aenq_handlers = {
2961 	.handlers = {
2962 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
2963 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
2964 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive
2965 	},
2966 	.unimplemented_handler = unimplemented_aenq_handler
2967 };
2968