xref: /dpdk/drivers/net/ena/ena_ethdev.c (revision ceccf8dc7c3d7797e380f12b45cd3ea1d7396b58)
1 /*-
2 * BSD LICENSE
3 *
4 * Copyright (c) 2015-2016 Amazon.com, Inc. or its affiliates.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * * Neither the name of copyright holder nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33 
34 #include <rte_ether.h>
35 #include <rte_ethdev_driver.h>
36 #include <rte_ethdev_pci.h>
37 #include <rte_tcp.h>
38 #include <rte_atomic.h>
39 #include <rte_dev.h>
40 #include <rte_errno.h>
41 #include <rte_version.h>
42 #include <rte_eal_memconfig.h>
43 #include <rte_net.h>
44 
45 #include "ena_ethdev.h"
46 #include "ena_logs.h"
47 #include "ena_platform.h"
48 #include "ena_com.h"
49 #include "ena_eth_com.h"
50 
51 #include <ena_common_defs.h>
52 #include <ena_regs_defs.h>
53 #include <ena_admin_defs.h>
54 #include <ena_eth_io_defs.h>
55 
56 #define DRV_MODULE_VER_MAJOR	1
57 #define DRV_MODULE_VER_MINOR	1
58 #define DRV_MODULE_VER_SUBMINOR	1
59 
60 #define ENA_IO_TXQ_IDX(q)	(2 * (q))
61 #define ENA_IO_RXQ_IDX(q)	(2 * (q) + 1)
62 /*reverse version of ENA_IO_RXQ_IDX*/
63 #define ENA_IO_RXQ_IDX_REV(q)	((q - 1) / 2)
64 
65 /* While processing submitted and completed descriptors (rx and tx path
66  * respectively) in a loop it is desired to:
67  *  - perform batch submissions while populating sumbissmion queue
68  *  - avoid blocking transmission of other packets during cleanup phase
69  * Hence the utilization ratio of 1/8 of a queue size.
70  */
71 #define ENA_RING_DESCS_RATIO(ring_size)	(ring_size / 8)
72 
73 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
74 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
75 
76 #define GET_L4_HDR_LEN(mbuf)					\
77 	((rte_pktmbuf_mtod_offset(mbuf,	struct tcp_hdr *,	\
78 		mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
79 
80 #define ENA_RX_RSS_TABLE_LOG_SIZE  7
81 #define ENA_RX_RSS_TABLE_SIZE	(1 << ENA_RX_RSS_TABLE_LOG_SIZE)
82 #define ENA_HASH_KEY_SIZE	40
83 #define ENA_ETH_SS_STATS	0xFF
84 #define ETH_GSTRING_LEN	32
85 
86 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
87 
88 #define ENA_MAX_RING_DESC	ENA_DEFAULT_RING_SIZE
89 #define ENA_MIN_RING_DESC	128
90 
91 enum ethtool_stringset {
92 	ETH_SS_TEST             = 0,
93 	ETH_SS_STATS,
94 };
95 
96 struct ena_stats {
97 	char name[ETH_GSTRING_LEN];
98 	int stat_offset;
99 };
100 
101 #define ENA_STAT_ENA_COM_ENTRY(stat) { \
102 	.name = #stat, \
103 	.stat_offset = offsetof(struct ena_com_stats_admin, stat) \
104 }
105 
106 #define ENA_STAT_ENTRY(stat, stat_type) { \
107 	.name = #stat, \
108 	.stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
109 }
110 
111 #define ENA_STAT_RX_ENTRY(stat) \
112 	ENA_STAT_ENTRY(stat, rx)
113 
114 #define ENA_STAT_TX_ENTRY(stat) \
115 	ENA_STAT_ENTRY(stat, tx)
116 
117 #define ENA_STAT_GLOBAL_ENTRY(stat) \
118 	ENA_STAT_ENTRY(stat, dev)
119 
120 /*
121  * Each rte_memzone should have unique name.
122  * To satisfy it, count number of allocation and add it to name.
123  */
124 uint32_t ena_alloc_cnt;
125 
126 static const struct ena_stats ena_stats_global_strings[] = {
127 	ENA_STAT_GLOBAL_ENTRY(tx_timeout),
128 	ENA_STAT_GLOBAL_ENTRY(io_suspend),
129 	ENA_STAT_GLOBAL_ENTRY(io_resume),
130 	ENA_STAT_GLOBAL_ENTRY(wd_expired),
131 	ENA_STAT_GLOBAL_ENTRY(interface_up),
132 	ENA_STAT_GLOBAL_ENTRY(interface_down),
133 	ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
134 };
135 
136 static const struct ena_stats ena_stats_tx_strings[] = {
137 	ENA_STAT_TX_ENTRY(cnt),
138 	ENA_STAT_TX_ENTRY(bytes),
139 	ENA_STAT_TX_ENTRY(queue_stop),
140 	ENA_STAT_TX_ENTRY(queue_wakeup),
141 	ENA_STAT_TX_ENTRY(dma_mapping_err),
142 	ENA_STAT_TX_ENTRY(linearize),
143 	ENA_STAT_TX_ENTRY(linearize_failed),
144 	ENA_STAT_TX_ENTRY(tx_poll),
145 	ENA_STAT_TX_ENTRY(doorbells),
146 	ENA_STAT_TX_ENTRY(prepare_ctx_err),
147 	ENA_STAT_TX_ENTRY(missing_tx_comp),
148 	ENA_STAT_TX_ENTRY(bad_req_id),
149 };
150 
151 static const struct ena_stats ena_stats_rx_strings[] = {
152 	ENA_STAT_RX_ENTRY(cnt),
153 	ENA_STAT_RX_ENTRY(bytes),
154 	ENA_STAT_RX_ENTRY(refil_partial),
155 	ENA_STAT_RX_ENTRY(bad_csum),
156 	ENA_STAT_RX_ENTRY(page_alloc_fail),
157 	ENA_STAT_RX_ENTRY(skb_alloc_fail),
158 	ENA_STAT_RX_ENTRY(dma_mapping_err),
159 	ENA_STAT_RX_ENTRY(bad_desc_num),
160 	ENA_STAT_RX_ENTRY(small_copy_len_pkt),
161 };
162 
163 static const struct ena_stats ena_stats_ena_com_strings[] = {
164 	ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
165 	ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
166 	ENA_STAT_ENA_COM_ENTRY(completed_cmd),
167 	ENA_STAT_ENA_COM_ENTRY(out_of_space),
168 	ENA_STAT_ENA_COM_ENTRY(no_completion),
169 };
170 
171 #define ENA_STATS_ARRAY_GLOBAL	ARRAY_SIZE(ena_stats_global_strings)
172 #define ENA_STATS_ARRAY_TX	ARRAY_SIZE(ena_stats_tx_strings)
173 #define ENA_STATS_ARRAY_RX	ARRAY_SIZE(ena_stats_rx_strings)
174 #define ENA_STATS_ARRAY_ENA_COM	ARRAY_SIZE(ena_stats_ena_com_strings)
175 
176 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
177 			DEV_TX_OFFLOAD_UDP_CKSUM |\
178 			DEV_TX_OFFLOAD_IPV4_CKSUM |\
179 			DEV_TX_OFFLOAD_TCP_TSO)
180 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
181 		       PKT_TX_IP_CKSUM |\
182 		       PKT_TX_TCP_SEG)
183 
184 /** Vendor ID used by Amazon devices */
185 #define PCI_VENDOR_ID_AMAZON 0x1D0F
186 /** Amazon devices */
187 #define PCI_DEVICE_ID_ENA_VF	0xEC20
188 #define PCI_DEVICE_ID_ENA_LLQ_VF	0xEC21
189 
190 #define	ENA_TX_OFFLOAD_MASK	(\
191 	PKT_TX_L4_MASK |         \
192 	PKT_TX_IP_CKSUM |        \
193 	PKT_TX_TCP_SEG)
194 
195 #define	ENA_TX_OFFLOAD_NOTSUP_MASK	\
196 	(PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
197 
198 int ena_logtype_init;
199 int ena_logtype_driver;
200 
201 static const struct rte_pci_id pci_id_ena_map[] = {
202 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
203 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
204 	{ .device_id = 0 },
205 };
206 
207 static struct ena_aenq_handlers aenq_handlers;
208 
209 static int ena_device_init(struct ena_com_dev *ena_dev,
210 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
211 			   bool *wd_state);
212 static int ena_dev_configure(struct rte_eth_dev *dev);
213 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
214 				  uint16_t nb_pkts);
215 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
216 		uint16_t nb_pkts);
217 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
218 			      uint16_t nb_desc, unsigned int socket_id,
219 			      const struct rte_eth_txconf *tx_conf);
220 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
221 			      uint16_t nb_desc, unsigned int socket_id,
222 			      const struct rte_eth_rxconf *rx_conf,
223 			      struct rte_mempool *mp);
224 static uint16_t eth_ena_recv_pkts(void *rx_queue,
225 				  struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
226 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
227 static void ena_init_rings(struct ena_adapter *adapter);
228 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
229 static int ena_start(struct rte_eth_dev *dev);
230 static void ena_stop(struct rte_eth_dev *dev);
231 static void ena_close(struct rte_eth_dev *dev);
232 static int ena_dev_reset(struct rte_eth_dev *dev);
233 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
234 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
235 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
236 static void ena_rx_queue_release(void *queue);
237 static void ena_tx_queue_release(void *queue);
238 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
239 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
240 static int ena_link_update(struct rte_eth_dev *dev,
241 			   int wait_to_complete);
242 static int ena_create_io_queue(struct ena_ring *ring);
243 static void ena_free_io_queues_all(struct ena_adapter *adapter);
244 static int ena_queue_restart(struct ena_ring *ring);
245 static int ena_queue_restart_all(struct rte_eth_dev *dev,
246 				 enum ena_ring_type ring_type);
247 static void ena_stats_restart(struct rte_eth_dev *dev);
248 static void ena_infos_get(struct rte_eth_dev *dev,
249 			  struct rte_eth_dev_info *dev_info);
250 static int ena_rss_reta_update(struct rte_eth_dev *dev,
251 			       struct rte_eth_rss_reta_entry64 *reta_conf,
252 			       uint16_t reta_size);
253 static int ena_rss_reta_query(struct rte_eth_dev *dev,
254 			      struct rte_eth_rss_reta_entry64 *reta_conf,
255 			      uint16_t reta_size);
256 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
257 static void ena_interrupt_handler_rte(void *cb_arg);
258 static void ena_timer_wd_callback(struct rte_timer *timer, void *arg);
259 
260 static const struct eth_dev_ops ena_dev_ops = {
261 	.dev_configure        = ena_dev_configure,
262 	.dev_infos_get        = ena_infos_get,
263 	.rx_queue_setup       = ena_rx_queue_setup,
264 	.tx_queue_setup       = ena_tx_queue_setup,
265 	.dev_start            = ena_start,
266 	.dev_stop             = ena_stop,
267 	.link_update          = ena_link_update,
268 	.stats_get            = ena_stats_get,
269 	.mtu_set              = ena_mtu_set,
270 	.rx_queue_release     = ena_rx_queue_release,
271 	.tx_queue_release     = ena_tx_queue_release,
272 	.dev_close            = ena_close,
273 	.dev_reset            = ena_dev_reset,
274 	.reta_update          = ena_rss_reta_update,
275 	.reta_query           = ena_rss_reta_query,
276 };
277 
278 #define NUMA_NO_NODE	SOCKET_ID_ANY
279 
280 static inline int ena_cpu_to_node(int cpu)
281 {
282 	struct rte_config *config = rte_eal_get_configuration();
283 	struct rte_fbarray *arr = &config->mem_config->memzones;
284 	const struct rte_memzone *mz;
285 
286 	if (unlikely(cpu >= RTE_MAX_MEMZONE))
287 		return NUMA_NO_NODE;
288 
289 	mz = rte_fbarray_get(arr, cpu);
290 
291 	return mz->socket_id;
292 }
293 
294 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
295 				       struct ena_com_rx_ctx *ena_rx_ctx)
296 {
297 	uint64_t ol_flags = 0;
298 	uint32_t packet_type = 0;
299 
300 	if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
301 		packet_type |= RTE_PTYPE_L4_TCP;
302 	else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
303 		packet_type |= RTE_PTYPE_L4_UDP;
304 
305 	if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
306 		packet_type |= RTE_PTYPE_L3_IPV4;
307 	else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
308 		packet_type |= RTE_PTYPE_L3_IPV6;
309 
310 	if (unlikely(ena_rx_ctx->l4_csum_err))
311 		ol_flags |= PKT_RX_L4_CKSUM_BAD;
312 	if (unlikely(ena_rx_ctx->l3_csum_err))
313 		ol_flags |= PKT_RX_IP_CKSUM_BAD;
314 
315 	mbuf->ol_flags = ol_flags;
316 	mbuf->packet_type = packet_type;
317 }
318 
319 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
320 				       struct ena_com_tx_ctx *ena_tx_ctx,
321 				       uint64_t queue_offloads)
322 {
323 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
324 
325 	if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
326 	    (queue_offloads & QUEUE_OFFLOADS)) {
327 		/* check if TSO is required */
328 		if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
329 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
330 			ena_tx_ctx->tso_enable = true;
331 
332 			ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
333 		}
334 
335 		/* check if L3 checksum is needed */
336 		if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
337 		    (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
338 			ena_tx_ctx->l3_csum_enable = true;
339 
340 		if (mbuf->ol_flags & PKT_TX_IPV6) {
341 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
342 		} else {
343 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
344 
345 			/* set don't fragment (DF) flag */
346 			if (mbuf->packet_type &
347 				(RTE_PTYPE_L4_NONFRAG
348 				 | RTE_PTYPE_INNER_L4_NONFRAG))
349 				ena_tx_ctx->df = true;
350 		}
351 
352 		/* check if L4 checksum is needed */
353 		if ((mbuf->ol_flags & PKT_TX_TCP_CKSUM) &&
354 		    (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
355 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
356 			ena_tx_ctx->l4_csum_enable = true;
357 		} else if ((mbuf->ol_flags & PKT_TX_UDP_CKSUM) &&
358 			   (queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
359 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
360 			ena_tx_ctx->l4_csum_enable = true;
361 		} else {
362 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
363 			ena_tx_ctx->l4_csum_enable = false;
364 		}
365 
366 		ena_meta->mss = mbuf->tso_segsz;
367 		ena_meta->l3_hdr_len = mbuf->l3_len;
368 		ena_meta->l3_hdr_offset = mbuf->l2_len;
369 
370 		ena_tx_ctx->meta_valid = true;
371 	} else {
372 		ena_tx_ctx->meta_valid = false;
373 	}
374 }
375 
376 static inline int validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
377 {
378 	if (likely(req_id < rx_ring->ring_size))
379 		return 0;
380 
381 	RTE_LOG(ERR, PMD, "Invalid rx req_id: %hu\n", req_id);
382 
383 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
384 	rx_ring->adapter->trigger_reset = true;
385 
386 	return -EFAULT;
387 }
388 
389 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
390 {
391 	struct ena_tx_buffer *tx_info = NULL;
392 
393 	if (likely(req_id < tx_ring->ring_size)) {
394 		tx_info = &tx_ring->tx_buffer_info[req_id];
395 		if (likely(tx_info->mbuf))
396 			return 0;
397 	}
398 
399 	if (tx_info)
400 		RTE_LOG(ERR, PMD, "tx_info doesn't have valid mbuf\n");
401 	else
402 		RTE_LOG(ERR, PMD, "Invalid req_id: %hu\n", req_id);
403 
404 	/* Trigger device reset */
405 	tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
406 	tx_ring->adapter->trigger_reset	= true;
407 	return -EFAULT;
408 }
409 
410 static void ena_config_host_info(struct ena_com_dev *ena_dev)
411 {
412 	struct ena_admin_host_info *host_info;
413 	int rc;
414 
415 	/* Allocate only the host info */
416 	rc = ena_com_allocate_host_info(ena_dev);
417 	if (rc) {
418 		RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
419 		return;
420 	}
421 
422 	host_info = ena_dev->host_attr.host_info;
423 
424 	host_info->os_type = ENA_ADMIN_OS_DPDK;
425 	host_info->kernel_ver = RTE_VERSION;
426 	snprintf((char *)host_info->kernel_ver_str,
427 		 sizeof(host_info->kernel_ver_str),
428 		 "%s", rte_version());
429 	host_info->os_dist = RTE_VERSION;
430 	snprintf((char *)host_info->os_dist_str,
431 		 sizeof(host_info->os_dist_str),
432 		 "%s", rte_version());
433 	host_info->driver_version =
434 		(DRV_MODULE_VER_MAJOR) |
435 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
436 		(DRV_MODULE_VER_SUBMINOR <<
437 			ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
438 
439 	rc = ena_com_set_host_attributes(ena_dev);
440 	if (rc) {
441 		if (rc == -ENA_COM_UNSUPPORTED)
442 			RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
443 		else
444 			RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
445 
446 		goto err;
447 	}
448 
449 	return;
450 
451 err:
452 	ena_com_delete_host_info(ena_dev);
453 }
454 
455 static int
456 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
457 {
458 	if (sset != ETH_SS_STATS)
459 		return -EOPNOTSUPP;
460 
461 	 /* Workaround for clang:
462 	 * touch internal structures to prevent
463 	 * compiler error
464 	 */
465 	ENA_TOUCH(ena_stats_global_strings);
466 	ENA_TOUCH(ena_stats_tx_strings);
467 	ENA_TOUCH(ena_stats_rx_strings);
468 	ENA_TOUCH(ena_stats_ena_com_strings);
469 
470 	return  dev->data->nb_tx_queues *
471 		(ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
472 		ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
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_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
481 	if (ss_count <= 0) {
482 		RTE_LOG(ERR, PMD, "SS count is negative\n");
483 		return;
484 	}
485 
486 	/* allocate 32 bytes for each string and 64bit for the value */
487 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
488 
489 	rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
490 	if (rc) {
491 		RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
492 		return;
493 	}
494 
495 	rc = ena_com_set_host_attributes(&adapter->ena_dev);
496 	if (rc) {
497 		if (rc == -ENA_COM_UNSUPPORTED)
498 			RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
499 		else
500 			RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
501 
502 		goto err;
503 	}
504 
505 	return;
506 err:
507 	ena_com_delete_debug_area(&adapter->ena_dev);
508 }
509 
510 static void ena_close(struct rte_eth_dev *dev)
511 {
512 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
513 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
514 	struct ena_adapter *adapter =
515 		(struct ena_adapter *)(dev->data->dev_private);
516 
517 	if (adapter->state == ENA_ADAPTER_STATE_RUNNING)
518 		ena_stop(dev);
519 	adapter->state = ENA_ADAPTER_STATE_CLOSED;
520 
521 	ena_rx_queue_release_all(dev);
522 	ena_tx_queue_release_all(dev);
523 
524 	rte_free(adapter->drv_stats);
525 	adapter->drv_stats = NULL;
526 
527 	rte_intr_disable(intr_handle);
528 	rte_intr_callback_unregister(intr_handle,
529 				     ena_interrupt_handler_rte,
530 				     adapter);
531 
532 	/*
533 	 * MAC is not allocated dynamically. Setting NULL should prevent from
534 	 * release of the resource in the rte_eth_dev_release_port().
535 	 */
536 	dev->data->mac_addrs = NULL;
537 }
538 
539 static int
540 ena_dev_reset(struct rte_eth_dev *dev)
541 {
542 	struct rte_mempool *mb_pool_rx[ENA_MAX_NUM_QUEUES];
543 	struct rte_eth_dev *eth_dev;
544 	struct rte_pci_device *pci_dev;
545 	struct rte_intr_handle *intr_handle;
546 	struct ena_com_dev *ena_dev;
547 	struct ena_com_dev_get_features_ctx get_feat_ctx;
548 	struct ena_adapter *adapter;
549 	int nb_queues;
550 	int rc, i;
551 	bool wd_state;
552 
553 	adapter = (struct ena_adapter *)(dev->data->dev_private);
554 	ena_dev = &adapter->ena_dev;
555 	eth_dev = adapter->rte_dev;
556 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
557 	intr_handle = &pci_dev->intr_handle;
558 	nb_queues = eth_dev->data->nb_rx_queues;
559 
560 	ena_com_set_admin_running_state(ena_dev, false);
561 
562 	rc = ena_com_dev_reset(ena_dev, adapter->reset_reason);
563 	if (rc)
564 		RTE_LOG(ERR, PMD, "Device reset failed\n");
565 
566 	for (i = 0; i < nb_queues; i++)
567 		mb_pool_rx[i] = adapter->rx_ring[i].mb_pool;
568 
569 	ena_rx_queue_release_all(eth_dev);
570 	ena_tx_queue_release_all(eth_dev);
571 
572 	rte_intr_disable(intr_handle);
573 
574 	ena_com_abort_admin_commands(ena_dev);
575 	ena_com_wait_for_abort_completion(ena_dev);
576 	ena_com_admin_destroy(ena_dev);
577 	ena_com_mmio_reg_read_request_destroy(ena_dev);
578 
579 	rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
580 	if (rc) {
581 		PMD_INIT_LOG(CRIT, "Cannot initialize device\n");
582 		return rc;
583 	}
584 	adapter->wd_state = wd_state;
585 
586 	rte_intr_enable(intr_handle);
587 	ena_com_set_admin_polling_mode(ena_dev, false);
588 	ena_com_admin_aenq_enable(ena_dev);
589 
590 	for (i = 0; i < nb_queues; ++i)
591 		ena_rx_queue_setup(eth_dev, i, adapter->rx_ring_size, 0, NULL,
592 			mb_pool_rx[i]);
593 
594 	for (i = 0; i < nb_queues; ++i)
595 		ena_tx_queue_setup(eth_dev, i, adapter->tx_ring_size, 0, NULL);
596 
597 	adapter->trigger_reset = false;
598 
599 	return 0;
600 }
601 
602 static int ena_rss_reta_update(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 =
607 		(struct ena_adapter *)(dev->data->dev_private);
608 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
609 	int rc, i;
610 	u16 entry_value;
611 	int conf_idx;
612 	int idx;
613 
614 	if ((reta_size == 0) || (reta_conf == NULL))
615 		return -EINVAL;
616 
617 	if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
618 		RTE_LOG(WARNING, PMD,
619 			"indirection table %d is bigger than supported (%d)\n",
620 			reta_size, ENA_RX_RSS_TABLE_SIZE);
621 		return -EINVAL;
622 	}
623 
624 	for (i = 0 ; i < reta_size ; i++) {
625 		/* each reta_conf is for 64 entries.
626 		 * to support 128 we use 2 conf of 64
627 		 */
628 		conf_idx = i / RTE_RETA_GROUP_SIZE;
629 		idx = i % RTE_RETA_GROUP_SIZE;
630 		if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
631 			entry_value =
632 				ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
633 
634 			rc = ena_com_indirect_table_fill_entry(ena_dev,
635 							       i,
636 							       entry_value);
637 			if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
638 				RTE_LOG(ERR, PMD,
639 					"Cannot fill indirect table\n");
640 				return rc;
641 			}
642 		}
643 	}
644 
645 	rc = ena_com_indirect_table_set(ena_dev);
646 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
647 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
648 		return rc;
649 	}
650 
651 	RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries  for port %d\n",
652 		__func__, reta_size, adapter->rte_dev->data->port_id);
653 
654 	return 0;
655 }
656 
657 /* Query redirection table. */
658 static int ena_rss_reta_query(struct rte_eth_dev *dev,
659 			      struct rte_eth_rss_reta_entry64 *reta_conf,
660 			      uint16_t reta_size)
661 {
662 	struct ena_adapter *adapter =
663 		(struct ena_adapter *)(dev->data->dev_private);
664 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
665 	int rc;
666 	int i;
667 	u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
668 	int reta_conf_idx;
669 	int reta_idx;
670 
671 	if (reta_size == 0 || reta_conf == NULL ||
672 	    (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
673 		return -EINVAL;
674 
675 	rc = ena_com_indirect_table_get(ena_dev, indirect_table);
676 	if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
677 		RTE_LOG(ERR, PMD, "cannot get indirect table\n");
678 		return -ENOTSUP;
679 	}
680 
681 	for (i = 0 ; i < reta_size ; i++) {
682 		reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
683 		reta_idx = i % RTE_RETA_GROUP_SIZE;
684 		if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
685 			reta_conf[reta_conf_idx].reta[reta_idx] =
686 				ENA_IO_RXQ_IDX_REV(indirect_table[i]);
687 	}
688 
689 	return 0;
690 }
691 
692 static int ena_rss_init_default(struct ena_adapter *adapter)
693 {
694 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
695 	uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
696 	int rc, i;
697 	u32 val;
698 
699 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
700 	if (unlikely(rc)) {
701 		RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
702 		goto err_rss_init;
703 	}
704 
705 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
706 		val = i % nb_rx_queues;
707 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
708 						       ENA_IO_RXQ_IDX(val));
709 		if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
710 			RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
711 			goto err_fill_indir;
712 		}
713 	}
714 
715 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
716 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
717 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
718 		RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
719 		goto err_fill_indir;
720 	}
721 
722 	rc = ena_com_set_default_hash_ctrl(ena_dev);
723 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
724 		RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
725 		goto err_fill_indir;
726 	}
727 
728 	rc = ena_com_indirect_table_set(ena_dev);
729 	if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
730 		RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
731 		goto err_fill_indir;
732 	}
733 	RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
734 		adapter->rte_dev->data->port_id);
735 
736 	return 0;
737 
738 err_fill_indir:
739 	ena_com_rss_destroy(ena_dev);
740 err_rss_init:
741 
742 	return rc;
743 }
744 
745 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
746 {
747 	struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
748 	int nb_queues = dev->data->nb_rx_queues;
749 	int i;
750 
751 	for (i = 0; i < nb_queues; i++)
752 		ena_rx_queue_release(queues[i]);
753 }
754 
755 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
756 {
757 	struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
758 	int nb_queues = dev->data->nb_tx_queues;
759 	int i;
760 
761 	for (i = 0; i < nb_queues; i++)
762 		ena_tx_queue_release(queues[i]);
763 }
764 
765 static void ena_rx_queue_release(void *queue)
766 {
767 	struct ena_ring *ring = (struct ena_ring *)queue;
768 
769 	ena_assert_msg(ring->configured,
770 		       "API violation - releasing not configured queue");
771 	ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
772 		       "API violation");
773 
774 	/* Free ring resources */
775 	if (ring->rx_buffer_info)
776 		rte_free(ring->rx_buffer_info);
777 	ring->rx_buffer_info = NULL;
778 
779 	if (ring->empty_rx_reqs)
780 		rte_free(ring->empty_rx_reqs);
781 	ring->empty_rx_reqs = NULL;
782 
783 	ring->configured = 0;
784 
785 	RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
786 		ring->port_id, ring->id);
787 }
788 
789 static void ena_tx_queue_release(void *queue)
790 {
791 	struct ena_ring *ring = (struct ena_ring *)queue;
792 
793 	ena_assert_msg(ring->configured,
794 		       "API violation. Releasing not configured queue");
795 	ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
796 		       "API violation");
797 
798 	/* Free all bufs */
799 	ena_tx_queue_release_bufs(ring);
800 
801 	/* Free ring resources */
802 	if (ring->tx_buffer_info)
803 		rte_free(ring->tx_buffer_info);
804 
805 	if (ring->empty_tx_reqs)
806 		rte_free(ring->empty_tx_reqs);
807 
808 	ring->empty_tx_reqs = NULL;
809 	ring->tx_buffer_info = NULL;
810 
811 	ring->configured = 0;
812 
813 	RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
814 		ring->port_id, ring->id);
815 }
816 
817 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
818 {
819 	unsigned int ring_mask = ring->ring_size - 1;
820 
821 	while (ring->next_to_clean != ring->next_to_use) {
822 		struct rte_mbuf *m =
823 			ring->rx_buffer_info[ring->next_to_clean & ring_mask];
824 
825 		if (m)
826 			rte_mbuf_raw_free(m);
827 
828 		ring->next_to_clean++;
829 	}
830 }
831 
832 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
833 {
834 	unsigned int i;
835 
836 	for (i = 0; i < ring->ring_size; ++i) {
837 		struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
838 
839 		if (tx_buf->mbuf)
840 			rte_pktmbuf_free(tx_buf->mbuf);
841 
842 		ring->next_to_clean++;
843 	}
844 }
845 
846 static int ena_link_update(struct rte_eth_dev *dev,
847 			   __rte_unused int wait_to_complete)
848 {
849 	struct rte_eth_link *link = &dev->data->dev_link;
850 	struct ena_adapter *adapter;
851 
852 	adapter = (struct ena_adapter *)(dev->data->dev_private);
853 
854 	link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
855 	link->link_speed = ETH_SPEED_NUM_NONE;
856 	link->link_duplex = ETH_LINK_FULL_DUPLEX;
857 
858 	return 0;
859 }
860 
861 static int ena_queue_restart_all(struct rte_eth_dev *dev,
862 				 enum ena_ring_type ring_type)
863 {
864 	struct ena_adapter *adapter =
865 		(struct ena_adapter *)(dev->data->dev_private);
866 	struct ena_ring *queues = NULL;
867 	int nb_queues;
868 	int i = 0;
869 	int rc = 0;
870 
871 	if (ring_type == ENA_RING_TYPE_RX) {
872 		queues = adapter->rx_ring;
873 		nb_queues = dev->data->nb_rx_queues;
874 	} else {
875 		queues = adapter->tx_ring;
876 		nb_queues = dev->data->nb_tx_queues;
877 	}
878 	for (i = 0; i < nb_queues; i++) {
879 		if (queues[i].configured) {
880 			if (ring_type == ENA_RING_TYPE_RX) {
881 				ena_assert_msg(
882 					dev->data->rx_queues[i] == &queues[i],
883 					"Inconsistent state of rx queues\n");
884 			} else {
885 				ena_assert_msg(
886 					dev->data->tx_queues[i] == &queues[i],
887 					"Inconsistent state of tx queues\n");
888 			}
889 
890 			rc = ena_queue_restart(&queues[i]);
891 
892 			if (rc) {
893 				PMD_INIT_LOG(ERR,
894 					     "failed to restart queue %d type(%d)",
895 					     i, ring_type);
896 				return rc;
897 			}
898 		}
899 	}
900 
901 	return 0;
902 }
903 
904 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
905 {
906 	uint32_t max_frame_len = adapter->max_mtu;
907 
908 	if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
909 	    DEV_RX_OFFLOAD_JUMBO_FRAME)
910 		max_frame_len =
911 			adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
912 
913 	return max_frame_len;
914 }
915 
916 static int ena_check_valid_conf(struct ena_adapter *adapter)
917 {
918 	uint32_t max_frame_len = ena_get_mtu_conf(adapter);
919 
920 	if (max_frame_len > adapter->max_mtu || max_frame_len < ENA_MIN_MTU) {
921 		PMD_INIT_LOG(ERR, "Unsupported MTU of %d. "
922 				  "max mtu: %d, min mtu: %d\n",
923 			     max_frame_len, adapter->max_mtu, ENA_MIN_MTU);
924 		return ENA_COM_UNSUPPORTED;
925 	}
926 
927 	return 0;
928 }
929 
930 static int
931 ena_calc_queue_size(struct ena_com_dev *ena_dev,
932 		    u16 *max_tx_sgl_size,
933 		    struct ena_com_dev_get_features_ctx *get_feat_ctx)
934 {
935 	uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
936 
937 	queue_size = RTE_MIN(queue_size,
938 			     get_feat_ctx->max_queues.max_cq_depth);
939 	queue_size = RTE_MIN(queue_size,
940 			     get_feat_ctx->max_queues.max_sq_depth);
941 
942 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
943 		queue_size = RTE_MIN(queue_size,
944 				     get_feat_ctx->max_queues.max_llq_depth);
945 
946 	/* Round down to power of 2 */
947 	if (!rte_is_power_of_2(queue_size))
948 		queue_size = rte_align32pow2(queue_size >> 1);
949 
950 	if (unlikely(queue_size == 0)) {
951 		PMD_INIT_LOG(ERR, "Invalid queue size");
952 		return -EFAULT;
953 	}
954 
955 	*max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
956 		get_feat_ctx->max_queues.max_packet_tx_descs);
957 
958 	return queue_size;
959 }
960 
961 static void ena_stats_restart(struct rte_eth_dev *dev)
962 {
963 	struct ena_adapter *adapter =
964 		(struct ena_adapter *)(dev->data->dev_private);
965 
966 	rte_atomic64_init(&adapter->drv_stats->ierrors);
967 	rte_atomic64_init(&adapter->drv_stats->oerrors);
968 	rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
969 }
970 
971 static int ena_stats_get(struct rte_eth_dev *dev,
972 			  struct rte_eth_stats *stats)
973 {
974 	struct ena_admin_basic_stats ena_stats;
975 	struct ena_adapter *adapter =
976 		(struct ena_adapter *)(dev->data->dev_private);
977 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
978 	int rc;
979 
980 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
981 		return -ENOTSUP;
982 
983 	memset(&ena_stats, 0, sizeof(ena_stats));
984 	rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
985 	if (unlikely(rc)) {
986 		RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
987 		return rc;
988 	}
989 
990 	/* Set of basic statistics from ENA */
991 	stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
992 					  ena_stats.rx_pkts_low);
993 	stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
994 					  ena_stats.tx_pkts_low);
995 	stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
996 					ena_stats.rx_bytes_low);
997 	stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
998 					ena_stats.tx_bytes_low);
999 	stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
1000 					 ena_stats.rx_drops_low);
1001 
1002 	/* Driver related stats */
1003 	stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
1004 	stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
1005 	stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
1006 	return 0;
1007 }
1008 
1009 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1010 {
1011 	struct ena_adapter *adapter;
1012 	struct ena_com_dev *ena_dev;
1013 	int rc = 0;
1014 
1015 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
1016 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1017 	adapter = (struct ena_adapter *)(dev->data->dev_private);
1018 
1019 	ena_dev = &adapter->ena_dev;
1020 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1021 
1022 	if (mtu > ena_get_mtu_conf(adapter) || mtu < ENA_MIN_MTU) {
1023 		RTE_LOG(ERR, PMD,
1024 			"Invalid MTU setting. new_mtu: %d "
1025 			"max mtu: %d min mtu: %d\n",
1026 			mtu, ena_get_mtu_conf(adapter), ENA_MIN_MTU);
1027 		return -EINVAL;
1028 	}
1029 
1030 	rc = ena_com_set_dev_mtu(ena_dev, mtu);
1031 	if (rc)
1032 		RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
1033 	else
1034 		RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
1035 
1036 	return rc;
1037 }
1038 
1039 static int ena_start(struct rte_eth_dev *dev)
1040 {
1041 	struct ena_adapter *adapter =
1042 		(struct ena_adapter *)(dev->data->dev_private);
1043 	uint64_t ticks;
1044 	int rc = 0;
1045 
1046 	rc = ena_check_valid_conf(adapter);
1047 	if (rc)
1048 		return rc;
1049 
1050 	rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
1051 	if (rc)
1052 		return rc;
1053 
1054 	rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
1055 	if (rc)
1056 		return rc;
1057 
1058 	if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
1059 	    ETH_MQ_RX_RSS_FLAG && adapter->rte_dev->data->nb_rx_queues > 0) {
1060 		rc = ena_rss_init_default(adapter);
1061 		if (rc)
1062 			return rc;
1063 	}
1064 
1065 	ena_stats_restart(dev);
1066 
1067 	adapter->timestamp_wd = rte_get_timer_cycles();
1068 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
1069 
1070 	ticks = rte_get_timer_hz();
1071 	rte_timer_reset(&adapter->timer_wd, ticks, PERIODICAL, rte_lcore_id(),
1072 			ena_timer_wd_callback, adapter);
1073 
1074 	adapter->state = ENA_ADAPTER_STATE_RUNNING;
1075 
1076 	return 0;
1077 }
1078 
1079 static void ena_stop(struct rte_eth_dev *dev)
1080 {
1081 	struct ena_adapter *adapter =
1082 		(struct ena_adapter *)(dev->data->dev_private);
1083 
1084 	rte_timer_stop_sync(&adapter->timer_wd);
1085 	ena_free_io_queues_all(adapter);
1086 
1087 	adapter->state = ENA_ADAPTER_STATE_STOPPED;
1088 }
1089 
1090 static int ena_create_io_queue(struct ena_ring *ring)
1091 {
1092 	struct ena_adapter *adapter;
1093 	struct ena_com_dev *ena_dev;
1094 	struct ena_com_create_io_ctx ctx =
1095 		/* policy set to _HOST just to satisfy icc compiler */
1096 		{ ENA_ADMIN_PLACEMENT_POLICY_HOST,
1097 		  0, 0, 0, 0, 0 };
1098 	uint16_t ena_qid;
1099 	unsigned int i;
1100 	int rc;
1101 
1102 	adapter = ring->adapter;
1103 	ena_dev = &adapter->ena_dev;
1104 
1105 	if (ring->type == ENA_RING_TYPE_TX) {
1106 		ena_qid = ENA_IO_TXQ_IDX(ring->id);
1107 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1108 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1109 		ctx.queue_size = adapter->tx_ring_size;
1110 		for (i = 0; i < ring->ring_size; i++)
1111 			ring->empty_tx_reqs[i] = i;
1112 	} else {
1113 		ena_qid = ENA_IO_RXQ_IDX(ring->id);
1114 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1115 		ctx.queue_size = adapter->rx_ring_size;
1116 		for (i = 0; i < ring->ring_size; i++)
1117 			ring->empty_rx_reqs[i] = i;
1118 	}
1119 	ctx.qid = ena_qid;
1120 	ctx.msix_vector = -1; /* interrupts not used */
1121 	ctx.numa_node = ena_cpu_to_node(ring->id);
1122 
1123 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1124 	if (rc) {
1125 		RTE_LOG(ERR, PMD,
1126 			"failed to create io queue #%d (qid:%d) rc: %d\n",
1127 			ring->id, ena_qid, rc);
1128 		return rc;
1129 	}
1130 
1131 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1132 				     &ring->ena_com_io_sq,
1133 				     &ring->ena_com_io_cq);
1134 	if (rc) {
1135 		RTE_LOG(ERR, PMD,
1136 			"Failed to get io queue handlers. queue num %d rc: %d\n",
1137 			ring->id, rc);
1138 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1139 		return rc;
1140 	}
1141 
1142 	if (ring->type == ENA_RING_TYPE_TX)
1143 		ena_com_update_numa_node(ring->ena_com_io_cq, ctx.numa_node);
1144 
1145 	return 0;
1146 }
1147 
1148 static void ena_free_io_queues_all(struct ena_adapter *adapter)
1149 {
1150 	struct rte_eth_dev *eth_dev = adapter->rte_dev;
1151 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1152 	int i;
1153 	uint16_t ena_qid;
1154 	uint16_t nb_rxq = eth_dev->data->nb_rx_queues;
1155 	uint16_t nb_txq = eth_dev->data->nb_tx_queues;
1156 
1157 	for (i = 0; i < nb_txq; ++i) {
1158 		ena_qid = ENA_IO_TXQ_IDX(i);
1159 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1160 
1161 		ena_tx_queue_release_bufs(&adapter->tx_ring[i]);
1162 	}
1163 
1164 	for (i = 0; i < nb_rxq; ++i) {
1165 		ena_qid = ENA_IO_RXQ_IDX(i);
1166 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1167 
1168 		ena_rx_queue_release_bufs(&adapter->rx_ring[i]);
1169 	}
1170 }
1171 
1172 static int ena_queue_restart(struct ena_ring *ring)
1173 {
1174 	int rc, bufs_num;
1175 
1176 	ena_assert_msg(ring->configured == 1,
1177 		       "Trying to restart unconfigured queue\n");
1178 
1179 	rc = ena_create_io_queue(ring);
1180 	if (rc) {
1181 		PMD_INIT_LOG(ERR, "Failed to create IO queue!\n");
1182 		return rc;
1183 	}
1184 
1185 	ring->next_to_clean = 0;
1186 	ring->next_to_use = 0;
1187 
1188 	if (ring->type == ENA_RING_TYPE_TX)
1189 		return 0;
1190 
1191 	bufs_num = ring->ring_size - 1;
1192 	rc = ena_populate_rx_queue(ring, bufs_num);
1193 	if (rc != bufs_num) {
1194 		PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
1195 		return ENA_COM_FAULT;
1196 	}
1197 
1198 	return 0;
1199 }
1200 
1201 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
1202 			      uint16_t queue_idx,
1203 			      uint16_t nb_desc,
1204 			      __rte_unused unsigned int socket_id,
1205 			      const struct rte_eth_txconf *tx_conf)
1206 {
1207 	struct ena_ring *txq = NULL;
1208 	struct ena_adapter *adapter =
1209 		(struct ena_adapter *)(dev->data->dev_private);
1210 	unsigned int i;
1211 
1212 	txq = &adapter->tx_ring[queue_idx];
1213 
1214 	if (txq->configured) {
1215 		RTE_LOG(CRIT, PMD,
1216 			"API violation. Queue %d is already configured\n",
1217 			queue_idx);
1218 		return ENA_COM_FAULT;
1219 	}
1220 
1221 	if (!rte_is_power_of_2(nb_desc)) {
1222 		RTE_LOG(ERR, PMD,
1223 			"Unsupported size of TX queue: %d is not a power of 2.",
1224 			nb_desc);
1225 		return -EINVAL;
1226 	}
1227 
1228 	if (nb_desc > adapter->tx_ring_size) {
1229 		RTE_LOG(ERR, PMD,
1230 			"Unsupported size of TX queue (max size: %d)\n",
1231 			adapter->tx_ring_size);
1232 		return -EINVAL;
1233 	}
1234 
1235 	txq->port_id = dev->data->port_id;
1236 	txq->next_to_clean = 0;
1237 	txq->next_to_use = 0;
1238 	txq->ring_size = nb_desc;
1239 
1240 	txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1241 					  sizeof(struct ena_tx_buffer) *
1242 					  txq->ring_size,
1243 					  RTE_CACHE_LINE_SIZE);
1244 	if (!txq->tx_buffer_info) {
1245 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1246 		return -ENOMEM;
1247 	}
1248 
1249 	txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1250 					 sizeof(u16) * txq->ring_size,
1251 					 RTE_CACHE_LINE_SIZE);
1252 	if (!txq->empty_tx_reqs) {
1253 		RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1254 		rte_free(txq->tx_buffer_info);
1255 		return -ENOMEM;
1256 	}
1257 
1258 	for (i = 0; i < txq->ring_size; i++)
1259 		txq->empty_tx_reqs[i] = i;
1260 
1261 	if (tx_conf != NULL) {
1262 		txq->offloads =
1263 			tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1264 	}
1265 
1266 	/* Store pointer to this queue in upper layer */
1267 	txq->configured = 1;
1268 	dev->data->tx_queues[queue_idx] = txq;
1269 
1270 	return 0;
1271 }
1272 
1273 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1274 			      uint16_t queue_idx,
1275 			      uint16_t nb_desc,
1276 			      __rte_unused unsigned int socket_id,
1277 			      __rte_unused const struct rte_eth_rxconf *rx_conf,
1278 			      struct rte_mempool *mp)
1279 {
1280 	struct ena_adapter *adapter =
1281 		(struct ena_adapter *)(dev->data->dev_private);
1282 	struct ena_ring *rxq = NULL;
1283 	int i;
1284 
1285 	rxq = &adapter->rx_ring[queue_idx];
1286 	if (rxq->configured) {
1287 		RTE_LOG(CRIT, PMD,
1288 			"API violation. Queue %d is already configured\n",
1289 			queue_idx);
1290 		return ENA_COM_FAULT;
1291 	}
1292 
1293 	if (!rte_is_power_of_2(nb_desc)) {
1294 		RTE_LOG(ERR, PMD,
1295 			"Unsupported size of RX queue: %d is not a power of 2.",
1296 			nb_desc);
1297 		return -EINVAL;
1298 	}
1299 
1300 	if (nb_desc > adapter->rx_ring_size) {
1301 		RTE_LOG(ERR, PMD,
1302 			"Unsupported size of RX queue (max size: %d)\n",
1303 			adapter->rx_ring_size);
1304 		return -EINVAL;
1305 	}
1306 
1307 	rxq->port_id = dev->data->port_id;
1308 	rxq->next_to_clean = 0;
1309 	rxq->next_to_use = 0;
1310 	rxq->ring_size = nb_desc;
1311 	rxq->mb_pool = mp;
1312 
1313 	rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1314 					  sizeof(struct rte_mbuf *) * nb_desc,
1315 					  RTE_CACHE_LINE_SIZE);
1316 	if (!rxq->rx_buffer_info) {
1317 		RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1318 		return -ENOMEM;
1319 	}
1320 
1321 	rxq->empty_rx_reqs = rte_zmalloc("rxq->empty_rx_reqs",
1322 					 sizeof(uint16_t) * nb_desc,
1323 					 RTE_CACHE_LINE_SIZE);
1324 	if (!rxq->empty_rx_reqs) {
1325 		RTE_LOG(ERR, PMD, "failed to alloc mem for empty rx reqs\n");
1326 		rte_free(rxq->rx_buffer_info);
1327 		rxq->rx_buffer_info = NULL;
1328 		return -ENOMEM;
1329 	}
1330 
1331 	for (i = 0; i < nb_desc; i++)
1332 		rxq->empty_tx_reqs[i] = i;
1333 
1334 	/* Store pointer to this queue in upper layer */
1335 	rxq->configured = 1;
1336 	dev->data->rx_queues[queue_idx] = rxq;
1337 
1338 	return 0;
1339 }
1340 
1341 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1342 {
1343 	unsigned int i;
1344 	int rc;
1345 	uint16_t ring_size = rxq->ring_size;
1346 	uint16_t ring_mask = ring_size - 1;
1347 	uint16_t next_to_use = rxq->next_to_use;
1348 	uint16_t in_use, req_id;
1349 	struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1350 
1351 	if (unlikely(!count))
1352 		return 0;
1353 
1354 	in_use = rxq->next_to_use - rxq->next_to_clean;
1355 	ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1356 
1357 	count = RTE_MIN(count,
1358 			(uint16_t)(ring_size - (next_to_use & ring_mask)));
1359 
1360 	/* get resources for incoming packets */
1361 	rc = rte_mempool_get_bulk(rxq->mb_pool,
1362 				  (void **)(&mbufs[next_to_use & ring_mask]),
1363 				  count);
1364 	if (unlikely(rc < 0)) {
1365 		rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1366 		PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1367 		return 0;
1368 	}
1369 
1370 	for (i = 0; i < count; i++) {
1371 		uint16_t next_to_use_masked = next_to_use & ring_mask;
1372 		struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
1373 		struct ena_com_buf ebuf;
1374 
1375 		rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1376 
1377 		req_id = rxq->empty_rx_reqs[next_to_use_masked];
1378 		rc = validate_rx_req_id(rxq, req_id);
1379 		if (unlikely(rc < 0))
1380 			break;
1381 
1382 		/* prepare physical address for DMA transaction */
1383 		ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1384 		ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1385 		/* pass resource to device */
1386 		rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1387 						&ebuf, req_id);
1388 		if (unlikely(rc)) {
1389 			rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbuf),
1390 					     count - i);
1391 			RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1392 			break;
1393 		}
1394 		next_to_use++;
1395 	}
1396 
1397 	if (unlikely(i < count))
1398 		RTE_LOG(WARNING, PMD, "refilled rx qid %d with only %d "
1399 			"buffers (from %d)\n", rxq->id, i, count);
1400 
1401 	/* When we submitted free recources to device... */
1402 	if (likely(i > 0)) {
1403 		/* ...let HW know that it can fill buffers with data
1404 		 *
1405 		 * Add memory barrier to make sure the desc were written before
1406 		 * issue a doorbell
1407 		 */
1408 		rte_wmb();
1409 		ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1410 
1411 		rxq->next_to_use = next_to_use;
1412 	}
1413 
1414 	return i;
1415 }
1416 
1417 static int ena_device_init(struct ena_com_dev *ena_dev,
1418 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
1419 			   bool *wd_state)
1420 {
1421 	uint32_t aenq_groups;
1422 	int rc;
1423 	bool readless_supported;
1424 
1425 	/* Initialize mmio registers */
1426 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
1427 	if (rc) {
1428 		RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1429 		return rc;
1430 	}
1431 
1432 	/* The PCIe configuration space revision id indicate if mmio reg
1433 	 * read is disabled.
1434 	 */
1435 	readless_supported =
1436 		!(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1437 			       & ENA_MMIO_DISABLE_REG_READ);
1438 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1439 
1440 	/* reset device */
1441 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1442 	if (rc) {
1443 		RTE_LOG(ERR, PMD, "cannot reset device\n");
1444 		goto err_mmio_read_less;
1445 	}
1446 
1447 	/* check FW version */
1448 	rc = ena_com_validate_version(ena_dev);
1449 	if (rc) {
1450 		RTE_LOG(ERR, PMD, "device version is too low\n");
1451 		goto err_mmio_read_less;
1452 	}
1453 
1454 	ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1455 
1456 	/* ENA device administration layer init */
1457 	rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
1458 	if (rc) {
1459 		RTE_LOG(ERR, PMD,
1460 			"cannot initialize ena admin queue with device\n");
1461 		goto err_mmio_read_less;
1462 	}
1463 
1464 	/* To enable the msix interrupts the driver needs to know the number
1465 	 * of queues. So the driver uses polling mode to retrieve this
1466 	 * information.
1467 	 */
1468 	ena_com_set_admin_polling_mode(ena_dev, true);
1469 
1470 	ena_config_host_info(ena_dev);
1471 
1472 	/* Get Device Attributes and features */
1473 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1474 	if (rc) {
1475 		RTE_LOG(ERR, PMD,
1476 			"cannot get attribute for ena device rc= %d\n", rc);
1477 		goto err_admin_init;
1478 	}
1479 
1480 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1481 		      BIT(ENA_ADMIN_NOTIFICATION) |
1482 		      BIT(ENA_ADMIN_KEEP_ALIVE) |
1483 		      BIT(ENA_ADMIN_FATAL_ERROR) |
1484 		      BIT(ENA_ADMIN_WARNING);
1485 
1486 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
1487 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1488 	if (rc) {
1489 		RTE_LOG(ERR, PMD, "Cannot configure aenq groups rc: %d\n", rc);
1490 		goto err_admin_init;
1491 	}
1492 
1493 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
1494 
1495 	return 0;
1496 
1497 err_admin_init:
1498 	ena_com_admin_destroy(ena_dev);
1499 
1500 err_mmio_read_less:
1501 	ena_com_mmio_reg_read_request_destroy(ena_dev);
1502 
1503 	return rc;
1504 }
1505 
1506 static void ena_interrupt_handler_rte(void *cb_arg)
1507 {
1508 	struct ena_adapter *adapter = (struct ena_adapter *)cb_arg;
1509 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1510 
1511 	ena_com_admin_q_comp_intr_handler(ena_dev);
1512 	if (likely(adapter->state != ENA_ADAPTER_STATE_CLOSED))
1513 		ena_com_aenq_intr_handler(ena_dev, adapter);
1514 }
1515 
1516 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
1517 {
1518 	if (!adapter->wd_state)
1519 		return;
1520 
1521 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
1522 		return;
1523 
1524 	if (unlikely((rte_get_timer_cycles() - adapter->timestamp_wd) >=
1525 	    adapter->keep_alive_timeout)) {
1526 		RTE_LOG(ERR, PMD, "Keep alive timeout\n");
1527 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
1528 		adapter->trigger_reset = true;
1529 	}
1530 }
1531 
1532 /* Check if admin queue is enabled */
1533 static void check_for_admin_com_state(struct ena_adapter *adapter)
1534 {
1535 	if (unlikely(!ena_com_get_admin_running_state(&adapter->ena_dev))) {
1536 		RTE_LOG(ERR, PMD, "ENA admin queue is not in running state!\n");
1537 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
1538 		adapter->trigger_reset = true;
1539 	}
1540 }
1541 
1542 static void ena_timer_wd_callback(__rte_unused struct rte_timer *timer,
1543 				  void *arg)
1544 {
1545 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1546 	struct rte_eth_dev *dev = adapter->rte_dev;
1547 
1548 	check_for_missing_keep_alive(adapter);
1549 	check_for_admin_com_state(adapter);
1550 
1551 	if (unlikely(adapter->trigger_reset)) {
1552 		RTE_LOG(ERR, PMD, "Trigger reset is on\n");
1553 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1554 			NULL);
1555 	}
1556 }
1557 
1558 static int ena_calc_io_queue_num(__rte_unused struct ena_com_dev *ena_dev,
1559 				 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1560 {
1561 	int io_sq_num, io_cq_num, io_queue_num;
1562 
1563 	io_sq_num = get_feat_ctx->max_queues.max_sq_num;
1564 	io_cq_num = get_feat_ctx->max_queues.max_cq_num;
1565 
1566 	io_queue_num = RTE_MIN(io_sq_num, io_cq_num);
1567 
1568 	if (unlikely(io_queue_num == 0)) {
1569 		RTE_LOG(ERR, PMD, "Number of IO queues should not be 0\n");
1570 		return -EFAULT;
1571 	}
1572 
1573 	return io_queue_num;
1574 }
1575 
1576 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1577 {
1578 	struct rte_pci_device *pci_dev;
1579 	struct rte_intr_handle *intr_handle;
1580 	struct ena_adapter *adapter =
1581 		(struct ena_adapter *)(eth_dev->data->dev_private);
1582 	struct ena_com_dev *ena_dev = &adapter->ena_dev;
1583 	struct ena_com_dev_get_features_ctx get_feat_ctx;
1584 	int queue_size, rc;
1585 	u16 tx_sgl_size = 0;
1586 
1587 	static int adapters_found;
1588 	bool wd_state;
1589 
1590 	memset(adapter, 0, sizeof(struct ena_adapter));
1591 	ena_dev = &adapter->ena_dev;
1592 
1593 	eth_dev->dev_ops = &ena_dev_ops;
1594 	eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1595 	eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1596 	eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1597 	adapter->rte_eth_dev_data = eth_dev->data;
1598 	adapter->rte_dev = eth_dev;
1599 
1600 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1601 		return 0;
1602 
1603 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1604 	adapter->pdev = pci_dev;
1605 
1606 	PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1607 		     pci_dev->addr.domain,
1608 		     pci_dev->addr.bus,
1609 		     pci_dev->addr.devid,
1610 		     pci_dev->addr.function);
1611 
1612 	intr_handle = &pci_dev->intr_handle;
1613 
1614 	adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1615 	adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1616 
1617 	if (!adapter->regs) {
1618 		PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1619 			     ENA_REGS_BAR);
1620 		return -ENXIO;
1621 	}
1622 
1623 	ena_dev->reg_bar = adapter->regs;
1624 	ena_dev->dmadev = adapter->pdev;
1625 
1626 	adapter->id_number = adapters_found;
1627 
1628 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1629 		 adapter->id_number);
1630 
1631 	/* device specific initialization routine */
1632 	rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
1633 	if (rc) {
1634 		PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1635 		goto err;
1636 	}
1637 	adapter->wd_state = wd_state;
1638 
1639 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1640 	adapter->num_queues = ena_calc_io_queue_num(ena_dev,
1641 						    &get_feat_ctx);
1642 
1643 	queue_size = ena_calc_queue_size(ena_dev, &tx_sgl_size, &get_feat_ctx);
1644 	if (queue_size <= 0 || adapter->num_queues <= 0) {
1645 		rc = -EFAULT;
1646 		goto err_device_destroy;
1647 	}
1648 
1649 	adapter->tx_ring_size = queue_size;
1650 	adapter->rx_ring_size = queue_size;
1651 
1652 	adapter->max_tx_sgl_size = tx_sgl_size;
1653 
1654 	/* prepare ring structures */
1655 	ena_init_rings(adapter);
1656 
1657 	ena_config_debug_area(adapter);
1658 
1659 	/* Set max MTU for this device */
1660 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1661 
1662 	/* set device support for TSO */
1663 	adapter->tso4_supported = get_feat_ctx.offload.tx &
1664 				  ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1665 
1666 	/* Copy MAC address and point DPDK to it */
1667 	eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1668 	ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1669 			(struct ether_addr *)adapter->mac_addr);
1670 
1671 	/*
1672 	 * Pass the information to the rte_eth_dev_close() that it should also
1673 	 * release the private port resources.
1674 	 */
1675 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1676 
1677 	adapter->drv_stats = rte_zmalloc("adapter stats",
1678 					 sizeof(*adapter->drv_stats),
1679 					 RTE_CACHE_LINE_SIZE);
1680 	if (!adapter->drv_stats) {
1681 		RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1682 		rc = -ENOMEM;
1683 		goto err_delete_debug_area;
1684 	}
1685 
1686 	rte_intr_callback_register(intr_handle,
1687 				   ena_interrupt_handler_rte,
1688 				   adapter);
1689 	rte_intr_enable(intr_handle);
1690 	ena_com_set_admin_polling_mode(ena_dev, false);
1691 	ena_com_admin_aenq_enable(ena_dev);
1692 
1693 	if (adapters_found == 0)
1694 		rte_timer_subsystem_init();
1695 	rte_timer_init(&adapter->timer_wd);
1696 
1697 	adapters_found++;
1698 	adapter->state = ENA_ADAPTER_STATE_INIT;
1699 
1700 	return 0;
1701 
1702 err_delete_debug_area:
1703 	ena_com_delete_debug_area(ena_dev);
1704 
1705 err_device_destroy:
1706 	ena_com_delete_host_info(ena_dev);
1707 	ena_com_admin_destroy(ena_dev);
1708 
1709 err:
1710 	return rc;
1711 }
1712 
1713 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1714 {
1715 	struct ena_adapter *adapter =
1716 		(struct ena_adapter *)(eth_dev->data->dev_private);
1717 
1718 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1719 		return 0;
1720 
1721 	if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1722 		ena_close(eth_dev);
1723 
1724 	eth_dev->dev_ops = NULL;
1725 	eth_dev->rx_pkt_burst = NULL;
1726 	eth_dev->tx_pkt_burst = NULL;
1727 	eth_dev->tx_pkt_prepare = NULL;
1728 
1729 	adapter->state = ENA_ADAPTER_STATE_FREE;
1730 
1731 	return 0;
1732 }
1733 
1734 static int ena_dev_configure(struct rte_eth_dev *dev)
1735 {
1736 	struct ena_adapter *adapter =
1737 		(struct ena_adapter *)(dev->data->dev_private);
1738 
1739 	adapter->state = ENA_ADAPTER_STATE_CONFIG;
1740 
1741 	adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1742 	adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1743 	return 0;
1744 }
1745 
1746 static void ena_init_rings(struct ena_adapter *adapter)
1747 {
1748 	int i;
1749 
1750 	for (i = 0; i < adapter->num_queues; i++) {
1751 		struct ena_ring *ring = &adapter->tx_ring[i];
1752 
1753 		ring->configured = 0;
1754 		ring->type = ENA_RING_TYPE_TX;
1755 		ring->adapter = adapter;
1756 		ring->id = i;
1757 		ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1758 		ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1759 		ring->sgl_size = adapter->max_tx_sgl_size;
1760 	}
1761 
1762 	for (i = 0; i < adapter->num_queues; i++) {
1763 		struct ena_ring *ring = &adapter->rx_ring[i];
1764 
1765 		ring->configured = 0;
1766 		ring->type = ENA_RING_TYPE_RX;
1767 		ring->adapter = adapter;
1768 		ring->id = i;
1769 	}
1770 }
1771 
1772 static void ena_infos_get(struct rte_eth_dev *dev,
1773 			  struct rte_eth_dev_info *dev_info)
1774 {
1775 	struct ena_adapter *adapter;
1776 	struct ena_com_dev *ena_dev;
1777 	struct ena_com_dev_get_features_ctx feat;
1778 	uint64_t rx_feat = 0, tx_feat = 0;
1779 	int rc = 0;
1780 
1781 	ena_assert_msg(dev->data != NULL, "Uninitialized device");
1782 	ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1783 	adapter = (struct ena_adapter *)(dev->data->dev_private);
1784 
1785 	ena_dev = &adapter->ena_dev;
1786 	ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1787 
1788 	dev_info->speed_capa =
1789 			ETH_LINK_SPEED_1G   |
1790 			ETH_LINK_SPEED_2_5G |
1791 			ETH_LINK_SPEED_5G   |
1792 			ETH_LINK_SPEED_10G  |
1793 			ETH_LINK_SPEED_25G  |
1794 			ETH_LINK_SPEED_40G  |
1795 			ETH_LINK_SPEED_50G  |
1796 			ETH_LINK_SPEED_100G;
1797 
1798 	/* Get supported features from HW */
1799 	rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1800 	if (unlikely(rc)) {
1801 		RTE_LOG(ERR, PMD,
1802 			"Cannot get attribute for ena device rc= %d\n", rc);
1803 		return;
1804 	}
1805 
1806 	/* Set Tx & Rx features available for device */
1807 	if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1808 		tx_feat	|= DEV_TX_OFFLOAD_TCP_TSO;
1809 
1810 	if (feat.offload.tx &
1811 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1812 		tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1813 			DEV_TX_OFFLOAD_UDP_CKSUM |
1814 			DEV_TX_OFFLOAD_TCP_CKSUM;
1815 
1816 	if (feat.offload.rx_supported &
1817 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1818 		rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1819 			DEV_RX_OFFLOAD_UDP_CKSUM  |
1820 			DEV_RX_OFFLOAD_TCP_CKSUM;
1821 
1822 	rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1823 
1824 	/* Inform framework about available features */
1825 	dev_info->rx_offload_capa = rx_feat;
1826 	dev_info->rx_queue_offload_capa = rx_feat;
1827 	dev_info->tx_offload_capa = tx_feat;
1828 	dev_info->tx_queue_offload_capa = tx_feat;
1829 
1830 	dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1831 	dev_info->max_rx_pktlen  = adapter->max_mtu;
1832 	dev_info->max_mac_addrs = 1;
1833 
1834 	dev_info->max_rx_queues = adapter->num_queues;
1835 	dev_info->max_tx_queues = adapter->num_queues;
1836 	dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1837 
1838 	adapter->tx_supported_offloads = tx_feat;
1839 	adapter->rx_supported_offloads = rx_feat;
1840 
1841 	dev_info->rx_desc_lim.nb_max = ENA_MAX_RING_DESC;
1842 	dev_info->rx_desc_lim.nb_min = ENA_MIN_RING_DESC;
1843 
1844 	dev_info->tx_desc_lim.nb_max = ENA_MAX_RING_DESC;
1845 	dev_info->tx_desc_lim.nb_min = ENA_MIN_RING_DESC;
1846 	dev_info->tx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
1847 					feat.max_queues.max_packet_tx_descs);
1848 	dev_info->tx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
1849 					feat.max_queues.max_packet_tx_descs);
1850 }
1851 
1852 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1853 				  uint16_t nb_pkts)
1854 {
1855 	struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1856 	unsigned int ring_size = rx_ring->ring_size;
1857 	unsigned int ring_mask = ring_size - 1;
1858 	uint16_t next_to_clean = rx_ring->next_to_clean;
1859 	uint16_t desc_in_use = 0;
1860 	uint16_t req_id;
1861 	unsigned int recv_idx = 0;
1862 	struct rte_mbuf *mbuf = NULL;
1863 	struct rte_mbuf *mbuf_head = NULL;
1864 	struct rte_mbuf *mbuf_prev = NULL;
1865 	struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1866 	unsigned int completed;
1867 
1868 	struct ena_com_rx_ctx ena_rx_ctx;
1869 	int rc = 0;
1870 
1871 	/* Check adapter state */
1872 	if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1873 		RTE_LOG(ALERT, PMD,
1874 			"Trying to receive pkts while device is NOT running\n");
1875 		return 0;
1876 	}
1877 
1878 	desc_in_use = rx_ring->next_to_use - next_to_clean;
1879 	if (unlikely(nb_pkts > desc_in_use))
1880 		nb_pkts = desc_in_use;
1881 
1882 	for (completed = 0; completed < nb_pkts; completed++) {
1883 		int segments = 0;
1884 
1885 		ena_rx_ctx.max_bufs = rx_ring->ring_size;
1886 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1887 		ena_rx_ctx.descs = 0;
1888 		/* receive packet context */
1889 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1890 				    rx_ring->ena_com_io_sq,
1891 				    &ena_rx_ctx);
1892 		if (unlikely(rc)) {
1893 			RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1894 			rx_ring->adapter->trigger_reset = true;
1895 			return 0;
1896 		}
1897 
1898 		if (unlikely(ena_rx_ctx.descs == 0))
1899 			break;
1900 
1901 		while (segments < ena_rx_ctx.descs) {
1902 			req_id = ena_rx_ctx.ena_bufs[segments].req_id;
1903 			rc = validate_rx_req_id(rx_ring, req_id);
1904 			if (unlikely(rc))
1905 				break;
1906 
1907 			mbuf = rx_buff_info[req_id];
1908 			mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1909 			mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1910 			mbuf->refcnt = 1;
1911 			mbuf->next = NULL;
1912 			if (unlikely(segments == 0)) {
1913 				mbuf->nb_segs = ena_rx_ctx.descs;
1914 				mbuf->port = rx_ring->port_id;
1915 				mbuf->pkt_len = 0;
1916 				mbuf_head = mbuf;
1917 			} else {
1918 				/* for multi-segment pkts create mbuf chain */
1919 				mbuf_prev->next = mbuf;
1920 			}
1921 			mbuf_head->pkt_len += mbuf->data_len;
1922 
1923 			mbuf_prev = mbuf;
1924 			rx_ring->empty_rx_reqs[next_to_clean & ring_mask] =
1925 				req_id;
1926 			segments++;
1927 			next_to_clean++;
1928 		}
1929 
1930 		/* fill mbuf attributes if any */
1931 		ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1932 		mbuf_head->hash.rss = ena_rx_ctx.hash;
1933 
1934 		/* pass to DPDK application head mbuf */
1935 		rx_pkts[recv_idx] = mbuf_head;
1936 		recv_idx++;
1937 	}
1938 
1939 	rx_ring->next_to_clean = next_to_clean;
1940 
1941 	desc_in_use = desc_in_use - completed + 1;
1942 	/* Burst refill to save doorbells, memory barriers, const interval */
1943 	if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1944 		ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1945 
1946 	return recv_idx;
1947 }
1948 
1949 static uint16_t
1950 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1951 		uint16_t nb_pkts)
1952 {
1953 	int32_t ret;
1954 	uint32_t i;
1955 	struct rte_mbuf *m;
1956 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1957 	struct ipv4_hdr *ip_hdr;
1958 	uint64_t ol_flags;
1959 	uint16_t frag_field;
1960 
1961 	for (i = 0; i != nb_pkts; i++) {
1962 		m = tx_pkts[i];
1963 		ol_flags = m->ol_flags;
1964 
1965 		if (!(ol_flags & PKT_TX_IPV4))
1966 			continue;
1967 
1968 		/* If there was not L2 header length specified, assume it is
1969 		 * length of the ethernet header.
1970 		 */
1971 		if (unlikely(m->l2_len == 0))
1972 			m->l2_len = sizeof(struct ether_hdr);
1973 
1974 		ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1975 						 m->l2_len);
1976 		frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1977 
1978 		if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1979 			m->packet_type |= RTE_PTYPE_L4_NONFRAG;
1980 
1981 			/* If IPv4 header has DF flag enabled and TSO support is
1982 			 * disabled, partial chcecksum should not be calculated.
1983 			 */
1984 			if (!tx_ring->adapter->tso4_supported)
1985 				continue;
1986 		}
1987 
1988 		if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
1989 				(ol_flags & PKT_TX_L4_MASK) ==
1990 				PKT_TX_SCTP_CKSUM) {
1991 			rte_errno = -ENOTSUP;
1992 			return i;
1993 		}
1994 
1995 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1996 		ret = rte_validate_tx_offload(m);
1997 		if (ret != 0) {
1998 			rte_errno = ret;
1999 			return i;
2000 		}
2001 #endif
2002 
2003 		/* In case we are supposed to TSO and have DF not set (DF=0)
2004 		 * hardware must be provided with partial checksum, otherwise
2005 		 * it will take care of necessary calculations.
2006 		 */
2007 
2008 		ret = rte_net_intel_cksum_flags_prepare(m,
2009 			ol_flags & ~PKT_TX_TCP_SEG);
2010 		if (ret != 0) {
2011 			rte_errno = ret;
2012 			return i;
2013 		}
2014 	}
2015 
2016 	return i;
2017 }
2018 
2019 static void ena_update_hints(struct ena_adapter *adapter,
2020 			     struct ena_admin_ena_hw_hints *hints)
2021 {
2022 	if (hints->admin_completion_tx_timeout)
2023 		adapter->ena_dev.admin_queue.completion_timeout =
2024 			hints->admin_completion_tx_timeout * 1000;
2025 
2026 	if (hints->mmio_read_timeout)
2027 		/* convert to usec */
2028 		adapter->ena_dev.mmio_read.reg_read_to =
2029 			hints->mmio_read_timeout * 1000;
2030 
2031 	if (hints->driver_watchdog_timeout) {
2032 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2033 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2034 		else
2035 			// Convert msecs to ticks
2036 			adapter->keep_alive_timeout =
2037 				(hints->driver_watchdog_timeout *
2038 				rte_get_timer_hz()) / 1000;
2039 	}
2040 }
2041 
2042 static int ena_check_and_linearize_mbuf(struct ena_ring *tx_ring,
2043 					struct rte_mbuf *mbuf)
2044 {
2045 	int num_segments, rc;
2046 
2047 	num_segments = mbuf->nb_segs;
2048 
2049 	if (likely(num_segments < tx_ring->sgl_size))
2050 		return 0;
2051 
2052 	rc = rte_pktmbuf_linearize(mbuf);
2053 	if (unlikely(rc))
2054 		RTE_LOG(WARNING, PMD, "Mbuf linearize failed\n");
2055 
2056 	return rc;
2057 }
2058 
2059 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2060 				  uint16_t nb_pkts)
2061 {
2062 	struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2063 	uint16_t next_to_use = tx_ring->next_to_use;
2064 	uint16_t next_to_clean = tx_ring->next_to_clean;
2065 	struct rte_mbuf *mbuf;
2066 	unsigned int ring_size = tx_ring->ring_size;
2067 	unsigned int ring_mask = ring_size - 1;
2068 	struct ena_com_tx_ctx ena_tx_ctx;
2069 	struct ena_tx_buffer *tx_info;
2070 	struct ena_com_buf *ebuf;
2071 	uint16_t rc, req_id, total_tx_descs = 0;
2072 	uint16_t sent_idx = 0, empty_tx_reqs;
2073 	int nb_hw_desc;
2074 
2075 	/* Check adapter state */
2076 	if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2077 		RTE_LOG(ALERT, PMD,
2078 			"Trying to xmit pkts while device is NOT running\n");
2079 		return 0;
2080 	}
2081 
2082 	empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
2083 	if (nb_pkts > empty_tx_reqs)
2084 		nb_pkts = empty_tx_reqs;
2085 
2086 	for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
2087 		mbuf = tx_pkts[sent_idx];
2088 
2089 		rc = ena_check_and_linearize_mbuf(tx_ring, mbuf);
2090 		if (unlikely(rc))
2091 			break;
2092 
2093 		req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
2094 		tx_info = &tx_ring->tx_buffer_info[req_id];
2095 		tx_info->mbuf = mbuf;
2096 		tx_info->num_of_bufs = 0;
2097 		ebuf = tx_info->bufs;
2098 
2099 		/* Prepare TX context */
2100 		memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2101 		memset(&ena_tx_ctx.ena_meta, 0x0,
2102 		       sizeof(struct ena_com_tx_meta));
2103 		ena_tx_ctx.ena_bufs = ebuf;
2104 		ena_tx_ctx.req_id = req_id;
2105 		if (tx_ring->tx_mem_queue_type ==
2106 				ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2107 			/* prepare the push buffer with
2108 			 * virtual address of the data
2109 			 */
2110 			ena_tx_ctx.header_len =
2111 				RTE_MIN(mbuf->data_len,
2112 					tx_ring->tx_max_header_size);
2113 			ena_tx_ctx.push_header =
2114 				(void *)((char *)mbuf->buf_addr +
2115 					 mbuf->data_off);
2116 		} /* there's no else as we take advantage of memset zeroing */
2117 
2118 		/* Set TX offloads flags, if applicable */
2119 		ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads);
2120 
2121 		if (unlikely(mbuf->ol_flags &
2122 			     (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
2123 			rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
2124 
2125 		rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
2126 
2127 		/* Process first segment taking into
2128 		 * consideration pushed header
2129 		 */
2130 		if (mbuf->data_len > ena_tx_ctx.header_len) {
2131 			ebuf->paddr = mbuf->buf_iova +
2132 				      mbuf->data_off +
2133 				      ena_tx_ctx.header_len;
2134 			ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
2135 			ebuf++;
2136 			tx_info->num_of_bufs++;
2137 		}
2138 
2139 		while ((mbuf = mbuf->next) != NULL) {
2140 			ebuf->paddr = mbuf->buf_iova + mbuf->data_off;
2141 			ebuf->len = mbuf->data_len;
2142 			ebuf++;
2143 			tx_info->num_of_bufs++;
2144 		}
2145 
2146 		ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2147 
2148 		/* Write data to device */
2149 		rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
2150 					&ena_tx_ctx, &nb_hw_desc);
2151 		if (unlikely(rc))
2152 			break;
2153 
2154 		tx_info->tx_descs = nb_hw_desc;
2155 
2156 		next_to_use++;
2157 	}
2158 
2159 	/* If there are ready packets to be xmitted... */
2160 	if (sent_idx > 0) {
2161 		/* ...let HW do its best :-) */
2162 		rte_wmb();
2163 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2164 
2165 		tx_ring->next_to_use = next_to_use;
2166 	}
2167 
2168 	/* Clear complete packets  */
2169 	while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
2170 		rc = validate_tx_req_id(tx_ring, req_id);
2171 		if (rc)
2172 			break;
2173 
2174 		/* Get Tx info & store how many descs were processed  */
2175 		tx_info = &tx_ring->tx_buffer_info[req_id];
2176 		total_tx_descs += tx_info->tx_descs;
2177 
2178 		/* Free whole mbuf chain  */
2179 		mbuf = tx_info->mbuf;
2180 		rte_pktmbuf_free(mbuf);
2181 		tx_info->mbuf = NULL;
2182 
2183 		/* Put back descriptor to the ring for reuse */
2184 		tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
2185 		next_to_clean++;
2186 
2187 		/* If too many descs to clean, leave it for another run */
2188 		if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
2189 			break;
2190 	}
2191 
2192 	if (total_tx_descs > 0) {
2193 		/* acknowledge completion of sent packets */
2194 		ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
2195 		tx_ring->next_to_clean = next_to_clean;
2196 	}
2197 
2198 	return sent_idx;
2199 }
2200 
2201 /*********************************************************************
2202  *  PMD configuration
2203  *********************************************************************/
2204 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2205 	struct rte_pci_device *pci_dev)
2206 {
2207 	return rte_eth_dev_pci_generic_probe(pci_dev,
2208 		sizeof(struct ena_adapter), eth_ena_dev_init);
2209 }
2210 
2211 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
2212 {
2213 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
2214 }
2215 
2216 static struct rte_pci_driver rte_ena_pmd = {
2217 	.id_table = pci_id_ena_map,
2218 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
2219 		     RTE_PCI_DRV_WC_ACTIVATE,
2220 	.probe = eth_ena_pci_probe,
2221 	.remove = eth_ena_pci_remove,
2222 };
2223 
2224 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
2225 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
2226 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
2227 
2228 RTE_INIT(ena_init_log)
2229 {
2230 	ena_logtype_init = rte_log_register("pmd.net.ena.init");
2231 	if (ena_logtype_init >= 0)
2232 		rte_log_set_level(ena_logtype_init, RTE_LOG_NOTICE);
2233 	ena_logtype_driver = rte_log_register("pmd.net.ena.driver");
2234 	if (ena_logtype_driver >= 0)
2235 		rte_log_set_level(ena_logtype_driver, RTE_LOG_NOTICE);
2236 }
2237 
2238 /******************************************************************************
2239  ******************************** AENQ Handlers *******************************
2240  *****************************************************************************/
2241 static void ena_update_on_link_change(void *adapter_data,
2242 				      struct ena_admin_aenq_entry *aenq_e)
2243 {
2244 	struct rte_eth_dev *eth_dev;
2245 	struct ena_adapter *adapter;
2246 	struct ena_admin_aenq_link_change_desc *aenq_link_desc;
2247 	uint32_t status;
2248 
2249 	adapter = (struct ena_adapter *)adapter_data;
2250 	aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
2251 	eth_dev = adapter->rte_dev;
2252 
2253 	status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
2254 	adapter->link_status = status;
2255 
2256 	ena_link_update(eth_dev, 0);
2257 	_rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2258 }
2259 
2260 static void ena_notification(void *data,
2261 			     struct ena_admin_aenq_entry *aenq_e)
2262 {
2263 	struct ena_adapter *adapter = (struct ena_adapter *)data;
2264 	struct ena_admin_ena_hw_hints *hints;
2265 
2266 	if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
2267 		RTE_LOG(WARNING, PMD, "Invalid group(%x) expected %x\n",
2268 			aenq_e->aenq_common_desc.group,
2269 			ENA_ADMIN_NOTIFICATION);
2270 
2271 	switch (aenq_e->aenq_common_desc.syndrom) {
2272 	case ENA_ADMIN_UPDATE_HINTS:
2273 		hints = (struct ena_admin_ena_hw_hints *)
2274 			(&aenq_e->inline_data_w4);
2275 		ena_update_hints(adapter, hints);
2276 		break;
2277 	default:
2278 		RTE_LOG(ERR, PMD, "Invalid aenq notification link state %d\n",
2279 			aenq_e->aenq_common_desc.syndrom);
2280 	}
2281 }
2282 
2283 static void ena_keep_alive(void *adapter_data,
2284 			   __rte_unused struct ena_admin_aenq_entry *aenq_e)
2285 {
2286 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
2287 
2288 	adapter->timestamp_wd = rte_get_timer_cycles();
2289 }
2290 
2291 /**
2292  * This handler will called for unknown event group or unimplemented handlers
2293  **/
2294 static void unimplemented_aenq_handler(__rte_unused void *data,
2295 				       __rte_unused struct ena_admin_aenq_entry *aenq_e)
2296 {
2297 	RTE_LOG(ERR, PMD, "Unknown event was received or event with "
2298 			  "unimplemented handler\n");
2299 }
2300 
2301 static struct ena_aenq_handlers aenq_handlers = {
2302 	.handlers = {
2303 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
2304 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
2305 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive
2306 	},
2307 	.unimplemented_handler = unimplemented_aenq_handler
2308 };
2309