xref: /linux/drivers/net/ethernet/amazon/ena/ena_netdev.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
1 /*
2  * Copyright 2015 Amazon.com, Inc. or its affiliates.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 
35 #ifdef CONFIG_RFS_ACCEL
36 #include <linux/cpu_rmap.h>
37 #endif /* CONFIG_RFS_ACCEL */
38 #include <linux/ethtool.h>
39 #include <linux/if_vlan.h>
40 #include <linux/kernel.h>
41 #include <linux/module.h>
42 #include <linux/moduleparam.h>
43 #include <linux/numa.h>
44 #include <linux/pci.h>
45 #include <linux/utsname.h>
46 #include <linux/version.h>
47 #include <linux/vmalloc.h>
48 #include <net/ip.h>
49 
50 #include "ena_netdev.h"
51 #include "ena_pci_id_tbl.h"
52 
53 static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
54 
55 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
56 MODULE_DESCRIPTION(DEVICE_NAME);
57 MODULE_LICENSE("GPL");
58 MODULE_VERSION(DRV_MODULE_VERSION);
59 
60 /* Time in jiffies before concluding the transmitter is hung. */
61 #define TX_TIMEOUT  (5 * HZ)
62 
63 #define ENA_NAPI_BUDGET 64
64 
65 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
66 		NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
67 static int debug = -1;
68 module_param(debug, int, 0);
69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
70 
71 static struct ena_aenq_handlers aenq_handlers;
72 
73 static struct workqueue_struct *ena_wq;
74 
75 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
76 
77 static int ena_rss_init_default(struct ena_adapter *adapter);
78 
79 static void ena_tx_timeout(struct net_device *dev)
80 {
81 	struct ena_adapter *adapter = netdev_priv(dev);
82 
83 	/* Change the state of the device to trigger reset
84 	 * Check that we are not in the middle or a trigger already
85 	 */
86 
87 	if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
88 		return;
89 
90 	u64_stats_update_begin(&adapter->syncp);
91 	adapter->dev_stats.tx_timeout++;
92 	u64_stats_update_end(&adapter->syncp);
93 
94 	netif_err(adapter, tx_err, dev, "Transmit time out\n");
95 }
96 
97 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
98 {
99 	int i;
100 
101 	for (i = 0; i < adapter->num_queues; i++)
102 		adapter->rx_ring[i].mtu = mtu;
103 }
104 
105 static int ena_change_mtu(struct net_device *dev, int new_mtu)
106 {
107 	struct ena_adapter *adapter = netdev_priv(dev);
108 	int ret;
109 
110 	ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
111 	if (!ret) {
112 		netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
113 		update_rx_ring_mtu(adapter, new_mtu);
114 		dev->mtu = new_mtu;
115 	} else {
116 		netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
117 			  new_mtu);
118 	}
119 
120 	return ret;
121 }
122 
123 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
124 {
125 #ifdef CONFIG_RFS_ACCEL
126 	u32 i;
127 	int rc;
128 
129 	adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_queues);
130 	if (!adapter->netdev->rx_cpu_rmap)
131 		return -ENOMEM;
132 	for (i = 0; i < adapter->num_queues; i++) {
133 		int irq_idx = ENA_IO_IRQ_IDX(i);
134 
135 		rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
136 				      pci_irq_vector(adapter->pdev, irq_idx));
137 		if (rc) {
138 			free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
139 			adapter->netdev->rx_cpu_rmap = NULL;
140 			return rc;
141 		}
142 	}
143 #endif /* CONFIG_RFS_ACCEL */
144 	return 0;
145 }
146 
147 static void ena_init_io_rings_common(struct ena_adapter *adapter,
148 				     struct ena_ring *ring, u16 qid)
149 {
150 	ring->qid = qid;
151 	ring->pdev = adapter->pdev;
152 	ring->dev = &adapter->pdev->dev;
153 	ring->netdev = adapter->netdev;
154 	ring->napi = &adapter->ena_napi[qid].napi;
155 	ring->adapter = adapter;
156 	ring->ena_dev = adapter->ena_dev;
157 	ring->per_napi_packets = 0;
158 	ring->per_napi_bytes = 0;
159 	ring->cpu = 0;
160 	u64_stats_init(&ring->syncp);
161 }
162 
163 static void ena_init_io_rings(struct ena_adapter *adapter)
164 {
165 	struct ena_com_dev *ena_dev;
166 	struct ena_ring *txr, *rxr;
167 	int i;
168 
169 	ena_dev = adapter->ena_dev;
170 
171 	for (i = 0; i < adapter->num_queues; i++) {
172 		txr = &adapter->tx_ring[i];
173 		rxr = &adapter->rx_ring[i];
174 
175 		/* TX/RX common ring state */
176 		ena_init_io_rings_common(adapter, txr, i);
177 		ena_init_io_rings_common(adapter, rxr, i);
178 
179 		/* TX specific ring state */
180 		txr->ring_size = adapter->tx_ring_size;
181 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
182 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
183 		txr->sgl_size = adapter->max_tx_sgl_size;
184 		txr->smoothed_interval =
185 			ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
186 
187 		/* RX specific ring state */
188 		rxr->ring_size = adapter->rx_ring_size;
189 		rxr->rx_copybreak = adapter->rx_copybreak;
190 		rxr->sgl_size = adapter->max_rx_sgl_size;
191 		rxr->smoothed_interval =
192 			ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
193 	}
194 }
195 
196 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
197  * @adapter: network interface device structure
198  * @qid: queue index
199  *
200  * Return 0 on success, negative on failure
201  */
202 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
203 {
204 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
205 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
206 	int size, i, node;
207 
208 	if (tx_ring->tx_buffer_info) {
209 		netif_err(adapter, ifup,
210 			  adapter->netdev, "tx_buffer_info info is not NULL");
211 		return -EEXIST;
212 	}
213 
214 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
215 	node = cpu_to_node(ena_irq->cpu);
216 
217 	tx_ring->tx_buffer_info = vzalloc_node(size, node);
218 	if (!tx_ring->tx_buffer_info) {
219 		tx_ring->tx_buffer_info = vzalloc(size);
220 		if (!tx_ring->tx_buffer_info)
221 			return -ENOMEM;
222 	}
223 
224 	size = sizeof(u16) * tx_ring->ring_size;
225 	tx_ring->free_tx_ids = vzalloc_node(size, node);
226 	if (!tx_ring->free_tx_ids) {
227 		tx_ring->free_tx_ids = vzalloc(size);
228 		if (!tx_ring->free_tx_ids) {
229 			vfree(tx_ring->tx_buffer_info);
230 			return -ENOMEM;
231 		}
232 	}
233 
234 	/* Req id ring for TX out of order completions */
235 	for (i = 0; i < tx_ring->ring_size; i++)
236 		tx_ring->free_tx_ids[i] = i;
237 
238 	/* Reset tx statistics */
239 	memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
240 
241 	tx_ring->next_to_use = 0;
242 	tx_ring->next_to_clean = 0;
243 	tx_ring->cpu = ena_irq->cpu;
244 	return 0;
245 }
246 
247 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
248  * @adapter: network interface device structure
249  * @qid: queue index
250  *
251  * Free all transmit software resources
252  */
253 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
254 {
255 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
256 
257 	vfree(tx_ring->tx_buffer_info);
258 	tx_ring->tx_buffer_info = NULL;
259 
260 	vfree(tx_ring->free_tx_ids);
261 	tx_ring->free_tx_ids = NULL;
262 }
263 
264 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
265  * @adapter: private structure
266  *
267  * Return 0 on success, negative on failure
268  */
269 static int ena_setup_all_tx_resources(struct ena_adapter *adapter)
270 {
271 	int i, rc = 0;
272 
273 	for (i = 0; i < adapter->num_queues; i++) {
274 		rc = ena_setup_tx_resources(adapter, i);
275 		if (rc)
276 			goto err_setup_tx;
277 	}
278 
279 	return 0;
280 
281 err_setup_tx:
282 
283 	netif_err(adapter, ifup, adapter->netdev,
284 		  "Tx queue %d: allocation failed\n", i);
285 
286 	/* rewind the index freeing the rings as we go */
287 	while (i--)
288 		ena_free_tx_resources(adapter, i);
289 	return rc;
290 }
291 
292 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
293  * @adapter: board private structure
294  *
295  * Free all transmit software resources
296  */
297 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
298 {
299 	int i;
300 
301 	for (i = 0; i < adapter->num_queues; i++)
302 		ena_free_tx_resources(adapter, i);
303 }
304 
305 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
306  * @adapter: network interface device structure
307  * @qid: queue index
308  *
309  * Returns 0 on success, negative on failure
310  */
311 static int ena_setup_rx_resources(struct ena_adapter *adapter,
312 				  u32 qid)
313 {
314 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
315 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
316 	int size, node;
317 
318 	if (rx_ring->rx_buffer_info) {
319 		netif_err(adapter, ifup, adapter->netdev,
320 			  "rx_buffer_info is not NULL");
321 		return -EEXIST;
322 	}
323 
324 	/* alloc extra element so in rx path
325 	 * we can always prefetch rx_info + 1
326 	 */
327 	size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
328 	node = cpu_to_node(ena_irq->cpu);
329 
330 	rx_ring->rx_buffer_info = vzalloc_node(size, node);
331 	if (!rx_ring->rx_buffer_info) {
332 		rx_ring->rx_buffer_info = vzalloc(size);
333 		if (!rx_ring->rx_buffer_info)
334 			return -ENOMEM;
335 	}
336 
337 	/* Reset rx statistics */
338 	memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
339 
340 	rx_ring->next_to_clean = 0;
341 	rx_ring->next_to_use = 0;
342 	rx_ring->cpu = ena_irq->cpu;
343 
344 	return 0;
345 }
346 
347 /* ena_free_rx_resources - Free I/O Rx Resources
348  * @adapter: network interface device structure
349  * @qid: queue index
350  *
351  * Free all receive software resources
352  */
353 static void ena_free_rx_resources(struct ena_adapter *adapter,
354 				  u32 qid)
355 {
356 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
357 
358 	vfree(rx_ring->rx_buffer_info);
359 	rx_ring->rx_buffer_info = NULL;
360 }
361 
362 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
363  * @adapter: board private structure
364  *
365  * Return 0 on success, negative on failure
366  */
367 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
368 {
369 	int i, rc = 0;
370 
371 	for (i = 0; i < adapter->num_queues; i++) {
372 		rc = ena_setup_rx_resources(adapter, i);
373 		if (rc)
374 			goto err_setup_rx;
375 	}
376 
377 	return 0;
378 
379 err_setup_rx:
380 
381 	netif_err(adapter, ifup, adapter->netdev,
382 		  "Rx queue %d: allocation failed\n", i);
383 
384 	/* rewind the index freeing the rings as we go */
385 	while (i--)
386 		ena_free_rx_resources(adapter, i);
387 	return rc;
388 }
389 
390 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
391  * @adapter: board private structure
392  *
393  * Free all receive software resources
394  */
395 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
396 {
397 	int i;
398 
399 	for (i = 0; i < adapter->num_queues; i++)
400 		ena_free_rx_resources(adapter, i);
401 }
402 
403 static inline int ena_alloc_rx_page(struct ena_ring *rx_ring,
404 				    struct ena_rx_buffer *rx_info, gfp_t gfp)
405 {
406 	struct ena_com_buf *ena_buf;
407 	struct page *page;
408 	dma_addr_t dma;
409 
410 	/* if previous allocated page is not used */
411 	if (unlikely(rx_info->page))
412 		return 0;
413 
414 	page = alloc_page(gfp);
415 	if (unlikely(!page)) {
416 		u64_stats_update_begin(&rx_ring->syncp);
417 		rx_ring->rx_stats.page_alloc_fail++;
418 		u64_stats_update_end(&rx_ring->syncp);
419 		return -ENOMEM;
420 	}
421 
422 	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE,
423 			   DMA_FROM_DEVICE);
424 	if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
425 		u64_stats_update_begin(&rx_ring->syncp);
426 		rx_ring->rx_stats.dma_mapping_err++;
427 		u64_stats_update_end(&rx_ring->syncp);
428 
429 		__free_page(page);
430 		return -EIO;
431 	}
432 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
433 		  "alloc page %p, rx_info %p\n", page, rx_info);
434 
435 	rx_info->page = page;
436 	rx_info->page_offset = 0;
437 	ena_buf = &rx_info->ena_buf;
438 	ena_buf->paddr = dma;
439 	ena_buf->len = PAGE_SIZE;
440 
441 	return 0;
442 }
443 
444 static void ena_free_rx_page(struct ena_ring *rx_ring,
445 			     struct ena_rx_buffer *rx_info)
446 {
447 	struct page *page = rx_info->page;
448 	struct ena_com_buf *ena_buf = &rx_info->ena_buf;
449 
450 	if (unlikely(!page)) {
451 		netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
452 			   "Trying to free unallocated buffer\n");
453 		return;
454 	}
455 
456 	dma_unmap_page(rx_ring->dev, ena_buf->paddr, PAGE_SIZE,
457 		       DMA_FROM_DEVICE);
458 
459 	__free_page(page);
460 	rx_info->page = NULL;
461 }
462 
463 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
464 {
465 	u16 next_to_use;
466 	u32 i;
467 	int rc;
468 
469 	next_to_use = rx_ring->next_to_use;
470 
471 	for (i = 0; i < num; i++) {
472 		struct ena_rx_buffer *rx_info =
473 			&rx_ring->rx_buffer_info[next_to_use];
474 
475 		rc = ena_alloc_rx_page(rx_ring, rx_info,
476 				       __GFP_COLD | GFP_ATOMIC | __GFP_COMP);
477 		if (unlikely(rc < 0)) {
478 			netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
479 				   "failed to alloc buffer for rx queue %d\n",
480 				   rx_ring->qid);
481 			break;
482 		}
483 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
484 						&rx_info->ena_buf,
485 						next_to_use);
486 		if (unlikely(rc)) {
487 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
488 				   "failed to add buffer for rx queue %d\n",
489 				   rx_ring->qid);
490 			break;
491 		}
492 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
493 						   rx_ring->ring_size);
494 	}
495 
496 	if (unlikely(i < num)) {
497 		u64_stats_update_begin(&rx_ring->syncp);
498 		rx_ring->rx_stats.refil_partial++;
499 		u64_stats_update_end(&rx_ring->syncp);
500 		netdev_warn(rx_ring->netdev,
501 			    "refilled rx qid %d with only %d buffers (from %d)\n",
502 			    rx_ring->qid, i, num);
503 	}
504 
505 	if (likely(i)) {
506 		/* Add memory barrier to make sure the desc were written before
507 		 * issue a doorbell
508 		 */
509 		wmb();
510 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
511 	}
512 
513 	rx_ring->next_to_use = next_to_use;
514 
515 	return i;
516 }
517 
518 static void ena_free_rx_bufs(struct ena_adapter *adapter,
519 			     u32 qid)
520 {
521 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
522 	u32 i;
523 
524 	for (i = 0; i < rx_ring->ring_size; i++) {
525 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
526 
527 		if (rx_info->page)
528 			ena_free_rx_page(rx_ring, rx_info);
529 	}
530 }
531 
532 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
533  * @adapter: board private structure
534  *
535  */
536 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
537 {
538 	struct ena_ring *rx_ring;
539 	int i, rc, bufs_num;
540 
541 	for (i = 0; i < adapter->num_queues; i++) {
542 		rx_ring = &adapter->rx_ring[i];
543 		bufs_num = rx_ring->ring_size - 1;
544 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
545 
546 		if (unlikely(rc != bufs_num))
547 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
548 				   "refilling Queue %d failed. allocated %d buffers from: %d\n",
549 				   i, rc, bufs_num);
550 	}
551 }
552 
553 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
554 {
555 	int i;
556 
557 	for (i = 0; i < adapter->num_queues; i++)
558 		ena_free_rx_bufs(adapter, i);
559 }
560 
561 /* ena_free_tx_bufs - Free Tx Buffers per Queue
562  * @tx_ring: TX ring for which buffers be freed
563  */
564 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
565 {
566 	bool print_once = true;
567 	u32 i;
568 
569 	for (i = 0; i < tx_ring->ring_size; i++) {
570 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
571 		struct ena_com_buf *ena_buf;
572 		int nr_frags;
573 		int j;
574 
575 		if (!tx_info->skb)
576 			continue;
577 
578 		if (print_once) {
579 			netdev_notice(tx_ring->netdev,
580 				      "free uncompleted tx skb qid %d idx 0x%x\n",
581 				      tx_ring->qid, i);
582 			print_once = false;
583 		} else {
584 			netdev_dbg(tx_ring->netdev,
585 				   "free uncompleted tx skb qid %d idx 0x%x\n",
586 				   tx_ring->qid, i);
587 		}
588 
589 		ena_buf = tx_info->bufs;
590 		dma_unmap_single(tx_ring->dev,
591 				 ena_buf->paddr,
592 				 ena_buf->len,
593 				 DMA_TO_DEVICE);
594 
595 		/* unmap remaining mapped pages */
596 		nr_frags = tx_info->num_of_bufs - 1;
597 		for (j = 0; j < nr_frags; j++) {
598 			ena_buf++;
599 			dma_unmap_page(tx_ring->dev,
600 				       ena_buf->paddr,
601 				       ena_buf->len,
602 				       DMA_TO_DEVICE);
603 		}
604 
605 		dev_kfree_skb_any(tx_info->skb);
606 	}
607 	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
608 						  tx_ring->qid));
609 }
610 
611 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
612 {
613 	struct ena_ring *tx_ring;
614 	int i;
615 
616 	for (i = 0; i < adapter->num_queues; i++) {
617 		tx_ring = &adapter->tx_ring[i];
618 		ena_free_tx_bufs(tx_ring);
619 	}
620 }
621 
622 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
623 {
624 	u16 ena_qid;
625 	int i;
626 
627 	for (i = 0; i < adapter->num_queues; i++) {
628 		ena_qid = ENA_IO_TXQ_IDX(i);
629 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
630 	}
631 }
632 
633 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
634 {
635 	u16 ena_qid;
636 	int i;
637 
638 	for (i = 0; i < adapter->num_queues; i++) {
639 		ena_qid = ENA_IO_RXQ_IDX(i);
640 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
641 	}
642 }
643 
644 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
645 {
646 	ena_destroy_all_tx_queues(adapter);
647 	ena_destroy_all_rx_queues(adapter);
648 }
649 
650 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
651 {
652 	struct ena_tx_buffer *tx_info = NULL;
653 
654 	if (likely(req_id < tx_ring->ring_size)) {
655 		tx_info = &tx_ring->tx_buffer_info[req_id];
656 		if (likely(tx_info->skb))
657 			return 0;
658 	}
659 
660 	if (tx_info)
661 		netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
662 			  "tx_info doesn't have valid skb\n");
663 	else
664 		netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
665 			  "Invalid req_id: %hu\n", req_id);
666 
667 	u64_stats_update_begin(&tx_ring->syncp);
668 	tx_ring->tx_stats.bad_req_id++;
669 	u64_stats_update_end(&tx_ring->syncp);
670 
671 	/* Trigger device reset */
672 	set_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags);
673 	return -EFAULT;
674 }
675 
676 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
677 {
678 	struct netdev_queue *txq;
679 	bool above_thresh;
680 	u32 tx_bytes = 0;
681 	u32 total_done = 0;
682 	u16 next_to_clean;
683 	u16 req_id;
684 	int tx_pkts = 0;
685 	int rc;
686 
687 	next_to_clean = tx_ring->next_to_clean;
688 	txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
689 
690 	while (tx_pkts < budget) {
691 		struct ena_tx_buffer *tx_info;
692 		struct sk_buff *skb;
693 		struct ena_com_buf *ena_buf;
694 		int i, nr_frags;
695 
696 		rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
697 						&req_id);
698 		if (rc)
699 			break;
700 
701 		rc = validate_tx_req_id(tx_ring, req_id);
702 		if (rc)
703 			break;
704 
705 		tx_info = &tx_ring->tx_buffer_info[req_id];
706 		skb = tx_info->skb;
707 
708 		/* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
709 		prefetch(&skb->end);
710 
711 		tx_info->skb = NULL;
712 		tx_info->last_jiffies = 0;
713 
714 		if (likely(tx_info->num_of_bufs != 0)) {
715 			ena_buf = tx_info->bufs;
716 
717 			dma_unmap_single(tx_ring->dev,
718 					 dma_unmap_addr(ena_buf, paddr),
719 					 dma_unmap_len(ena_buf, len),
720 					 DMA_TO_DEVICE);
721 
722 			/* unmap remaining mapped pages */
723 			nr_frags = tx_info->num_of_bufs - 1;
724 			for (i = 0; i < nr_frags; i++) {
725 				ena_buf++;
726 				dma_unmap_page(tx_ring->dev,
727 					       dma_unmap_addr(ena_buf, paddr),
728 					       dma_unmap_len(ena_buf, len),
729 					       DMA_TO_DEVICE);
730 			}
731 		}
732 
733 		netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
734 			  "tx_poll: q %d skb %p completed\n", tx_ring->qid,
735 			  skb);
736 
737 		tx_bytes += skb->len;
738 		dev_kfree_skb(skb);
739 		tx_pkts++;
740 		total_done += tx_info->tx_descs;
741 
742 		tx_ring->free_tx_ids[next_to_clean] = req_id;
743 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
744 						     tx_ring->ring_size);
745 	}
746 
747 	tx_ring->next_to_clean = next_to_clean;
748 	ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
749 	ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
750 
751 	netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
752 
753 	netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
754 		  "tx_poll: q %d done. total pkts: %d\n",
755 		  tx_ring->qid, tx_pkts);
756 
757 	/* need to make the rings circular update visible to
758 	 * ena_start_xmit() before checking for netif_queue_stopped().
759 	 */
760 	smp_mb();
761 
762 	above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
763 		ENA_TX_WAKEUP_THRESH;
764 	if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
765 		__netif_tx_lock(txq, smp_processor_id());
766 		above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
767 			ENA_TX_WAKEUP_THRESH;
768 		if (netif_tx_queue_stopped(txq) && above_thresh) {
769 			netif_tx_wake_queue(txq);
770 			u64_stats_update_begin(&tx_ring->syncp);
771 			tx_ring->tx_stats.queue_wakeup++;
772 			u64_stats_update_end(&tx_ring->syncp);
773 		}
774 		__netif_tx_unlock(txq);
775 	}
776 
777 	tx_ring->per_napi_bytes += tx_bytes;
778 	tx_ring->per_napi_packets += tx_pkts;
779 
780 	return tx_pkts;
781 }
782 
783 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
784 				  struct ena_com_rx_buf_info *ena_bufs,
785 				  u32 descs,
786 				  u16 *next_to_clean)
787 {
788 	struct sk_buff *skb;
789 	struct ena_rx_buffer *rx_info =
790 		&rx_ring->rx_buffer_info[*next_to_clean];
791 	u32 len;
792 	u32 buf = 0;
793 	void *va;
794 
795 	len = ena_bufs[0].len;
796 	if (unlikely(!rx_info->page)) {
797 		netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
798 			  "Page is NULL\n");
799 		return NULL;
800 	}
801 
802 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
803 		  "rx_info %p page %p\n",
804 		  rx_info, rx_info->page);
805 
806 	/* save virt address of first buffer */
807 	va = page_address(rx_info->page) + rx_info->page_offset;
808 	prefetch(va + NET_IP_ALIGN);
809 
810 	if (len <= rx_ring->rx_copybreak) {
811 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
812 						rx_ring->rx_copybreak);
813 		if (unlikely(!skb)) {
814 			u64_stats_update_begin(&rx_ring->syncp);
815 			rx_ring->rx_stats.skb_alloc_fail++;
816 			u64_stats_update_end(&rx_ring->syncp);
817 			netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
818 				  "Failed to allocate skb\n");
819 			return NULL;
820 		}
821 
822 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
823 			  "rx allocated small packet. len %d. data_len %d\n",
824 			  skb->len, skb->data_len);
825 
826 		/* sync this buffer for CPU use */
827 		dma_sync_single_for_cpu(rx_ring->dev,
828 					dma_unmap_addr(&rx_info->ena_buf, paddr),
829 					len,
830 					DMA_FROM_DEVICE);
831 		skb_copy_to_linear_data(skb, va, len);
832 		dma_sync_single_for_device(rx_ring->dev,
833 					   dma_unmap_addr(&rx_info->ena_buf, paddr),
834 					   len,
835 					   DMA_FROM_DEVICE);
836 
837 		skb_put(skb, len);
838 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
839 		*next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
840 						     rx_ring->ring_size);
841 		return skb;
842 	}
843 
844 	skb = napi_get_frags(rx_ring->napi);
845 	if (unlikely(!skb)) {
846 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
847 			  "Failed allocating skb\n");
848 		u64_stats_update_begin(&rx_ring->syncp);
849 		rx_ring->rx_stats.skb_alloc_fail++;
850 		u64_stats_update_end(&rx_ring->syncp);
851 		return NULL;
852 	}
853 
854 	do {
855 		dma_unmap_page(rx_ring->dev,
856 			       dma_unmap_addr(&rx_info->ena_buf, paddr),
857 			       PAGE_SIZE, DMA_FROM_DEVICE);
858 
859 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
860 				rx_info->page_offset, len, PAGE_SIZE);
861 
862 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
863 			  "rx skb updated. len %d. data_len %d\n",
864 			  skb->len, skb->data_len);
865 
866 		rx_info->page = NULL;
867 		*next_to_clean =
868 			ENA_RX_RING_IDX_NEXT(*next_to_clean,
869 					     rx_ring->ring_size);
870 		if (likely(--descs == 0))
871 			break;
872 		rx_info = &rx_ring->rx_buffer_info[*next_to_clean];
873 		len = ena_bufs[++buf].len;
874 	} while (1);
875 
876 	return skb;
877 }
878 
879 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
880  * @adapter: structure containing adapter specific data
881  * @ena_rx_ctx: received packet context/metadata
882  * @skb: skb currently being received and modified
883  */
884 static inline void ena_rx_checksum(struct ena_ring *rx_ring,
885 				   struct ena_com_rx_ctx *ena_rx_ctx,
886 				   struct sk_buff *skb)
887 {
888 	/* Rx csum disabled */
889 	if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
890 		skb->ip_summed = CHECKSUM_NONE;
891 		return;
892 	}
893 
894 	/* For fragmented packets the checksum isn't valid */
895 	if (ena_rx_ctx->frag) {
896 		skb->ip_summed = CHECKSUM_NONE;
897 		return;
898 	}
899 
900 	/* if IP and error */
901 	if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
902 		     (ena_rx_ctx->l3_csum_err))) {
903 		/* ipv4 checksum error */
904 		skb->ip_summed = CHECKSUM_NONE;
905 		u64_stats_update_begin(&rx_ring->syncp);
906 		rx_ring->rx_stats.bad_csum++;
907 		u64_stats_update_end(&rx_ring->syncp);
908 		netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
909 			  "RX IPv4 header checksum error\n");
910 		return;
911 	}
912 
913 	/* if TCP/UDP */
914 	if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
915 		   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
916 		if (unlikely(ena_rx_ctx->l4_csum_err)) {
917 			/* TCP/UDP checksum error */
918 			u64_stats_update_begin(&rx_ring->syncp);
919 			rx_ring->rx_stats.bad_csum++;
920 			u64_stats_update_end(&rx_ring->syncp);
921 			netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
922 				  "RX L4 checksum error\n");
923 			skb->ip_summed = CHECKSUM_NONE;
924 			return;
925 		}
926 
927 		skb->ip_summed = CHECKSUM_UNNECESSARY;
928 	}
929 }
930 
931 static void ena_set_rx_hash(struct ena_ring *rx_ring,
932 			    struct ena_com_rx_ctx *ena_rx_ctx,
933 			    struct sk_buff *skb)
934 {
935 	enum pkt_hash_types hash_type;
936 
937 	if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
938 		if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
939 			   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
940 
941 			hash_type = PKT_HASH_TYPE_L4;
942 		else
943 			hash_type = PKT_HASH_TYPE_NONE;
944 
945 		/* Override hash type if the packet is fragmented */
946 		if (ena_rx_ctx->frag)
947 			hash_type = PKT_HASH_TYPE_NONE;
948 
949 		skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
950 	}
951 }
952 
953 /* ena_clean_rx_irq - Cleanup RX irq
954  * @rx_ring: RX ring to clean
955  * @napi: napi handler
956  * @budget: how many packets driver is allowed to clean
957  *
958  * Returns the number of cleaned buffers.
959  */
960 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
961 			    u32 budget)
962 {
963 	u16 next_to_clean = rx_ring->next_to_clean;
964 	u32 res_budget, work_done;
965 
966 	struct ena_com_rx_ctx ena_rx_ctx;
967 	struct ena_adapter *adapter;
968 	struct sk_buff *skb;
969 	int refill_required;
970 	int refill_threshold;
971 	int rc = 0;
972 	int total_len = 0;
973 	int rx_copybreak_pkt = 0;
974 
975 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
976 		  "%s qid %d\n", __func__, rx_ring->qid);
977 	res_budget = budget;
978 
979 	do {
980 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
981 		ena_rx_ctx.max_bufs = rx_ring->sgl_size;
982 		ena_rx_ctx.descs = 0;
983 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
984 				    rx_ring->ena_com_io_sq,
985 				    &ena_rx_ctx);
986 		if (unlikely(rc))
987 			goto error;
988 
989 		if (unlikely(ena_rx_ctx.descs == 0))
990 			break;
991 
992 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
993 			  "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
994 			  rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
995 			  ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
996 
997 		/* allocate skb and fill it */
998 		skb = ena_rx_skb(rx_ring, rx_ring->ena_bufs, ena_rx_ctx.descs,
999 				 &next_to_clean);
1000 
1001 		/* exit if we failed to retrieve a buffer */
1002 		if (unlikely(!skb)) {
1003 			next_to_clean = ENA_RX_RING_IDX_ADD(next_to_clean,
1004 							    ena_rx_ctx.descs,
1005 							    rx_ring->ring_size);
1006 			break;
1007 		}
1008 
1009 		ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1010 
1011 		ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1012 
1013 		skb_record_rx_queue(skb, rx_ring->qid);
1014 
1015 		if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1016 			total_len += rx_ring->ena_bufs[0].len;
1017 			rx_copybreak_pkt++;
1018 			napi_gro_receive(napi, skb);
1019 		} else {
1020 			total_len += skb->len;
1021 			napi_gro_frags(napi);
1022 		}
1023 
1024 		res_budget--;
1025 	} while (likely(res_budget));
1026 
1027 	work_done = budget - res_budget;
1028 	rx_ring->per_napi_bytes += total_len;
1029 	rx_ring->per_napi_packets += work_done;
1030 	u64_stats_update_begin(&rx_ring->syncp);
1031 	rx_ring->rx_stats.bytes += total_len;
1032 	rx_ring->rx_stats.cnt += work_done;
1033 	rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1034 	u64_stats_update_end(&rx_ring->syncp);
1035 
1036 	rx_ring->next_to_clean = next_to_clean;
1037 
1038 	refill_required = ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
1039 	refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
1040 
1041 	/* Optimization, try to batch new rx buffers */
1042 	if (refill_required > refill_threshold) {
1043 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1044 		ena_refill_rx_bufs(rx_ring, refill_required);
1045 	}
1046 
1047 	return work_done;
1048 
1049 error:
1050 	adapter = netdev_priv(rx_ring->netdev);
1051 
1052 	u64_stats_update_begin(&rx_ring->syncp);
1053 	rx_ring->rx_stats.bad_desc_num++;
1054 	u64_stats_update_end(&rx_ring->syncp);
1055 
1056 	/* Too many desc from the device. Trigger reset */
1057 	set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1058 
1059 	return 0;
1060 }
1061 
1062 inline void ena_adjust_intr_moderation(struct ena_ring *rx_ring,
1063 				       struct ena_ring *tx_ring)
1064 {
1065 	/* We apply adaptive moderation on Rx path only.
1066 	 * Tx uses static interrupt moderation.
1067 	 */
1068 	ena_com_calculate_interrupt_delay(rx_ring->ena_dev,
1069 					  rx_ring->per_napi_packets,
1070 					  rx_ring->per_napi_bytes,
1071 					  &rx_ring->smoothed_interval,
1072 					  &rx_ring->moder_tbl_idx);
1073 
1074 	/* Reset per napi packets/bytes */
1075 	tx_ring->per_napi_packets = 0;
1076 	tx_ring->per_napi_bytes = 0;
1077 	rx_ring->per_napi_packets = 0;
1078 	rx_ring->per_napi_bytes = 0;
1079 }
1080 
1081 static inline void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1082 					     struct ena_ring *rx_ring)
1083 {
1084 	int cpu = get_cpu();
1085 	int numa_node;
1086 
1087 	/* Check only one ring since the 2 rings are running on the same cpu */
1088 	if (likely(tx_ring->cpu == cpu))
1089 		goto out;
1090 
1091 	numa_node = cpu_to_node(cpu);
1092 	put_cpu();
1093 
1094 	if (numa_node != NUMA_NO_NODE) {
1095 		ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1096 		ena_com_update_numa_node(rx_ring->ena_com_io_cq, numa_node);
1097 	}
1098 
1099 	tx_ring->cpu = cpu;
1100 	rx_ring->cpu = cpu;
1101 
1102 	return;
1103 out:
1104 	put_cpu();
1105 }
1106 
1107 static int ena_io_poll(struct napi_struct *napi, int budget)
1108 {
1109 	struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1110 	struct ena_ring *tx_ring, *rx_ring;
1111 	struct ena_eth_io_intr_reg intr_reg;
1112 
1113 	u32 tx_work_done;
1114 	u32 rx_work_done;
1115 	int tx_budget;
1116 	int napi_comp_call = 0;
1117 	int ret;
1118 
1119 	tx_ring = ena_napi->tx_ring;
1120 	rx_ring = ena_napi->rx_ring;
1121 
1122 	tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1123 
1124 	if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1125 	    test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1126 		napi_complete_done(napi, 0);
1127 		return 0;
1128 	}
1129 
1130 	tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1131 	rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1132 
1133 	/* If the device is about to reset or down, avoid unmask
1134 	 * the interrupt and return 0 so NAPI won't reschedule
1135 	 */
1136 	if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1137 		     test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1138 		napi_complete_done(napi, 0);
1139 		ret = 0;
1140 
1141 	} else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1142 		napi_comp_call = 1;
1143 
1144 		/* Update numa and unmask the interrupt only when schedule
1145 		 * from the interrupt context (vs from sk_busy_loop)
1146 		 */
1147 		if (napi_complete_done(napi, rx_work_done)) {
1148 			/* Tx and Rx share the same interrupt vector */
1149 			if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1150 				ena_adjust_intr_moderation(rx_ring, tx_ring);
1151 
1152 			/* Update intr register: rx intr delay,
1153 			 * tx intr delay and interrupt unmask
1154 			 */
1155 			ena_com_update_intr_reg(&intr_reg,
1156 						rx_ring->smoothed_interval,
1157 						tx_ring->smoothed_interval,
1158 						true);
1159 
1160 			/* It is a shared MSI-X.
1161 			 * Tx and Rx CQ have pointer to it.
1162 			 * So we use one of them to reach the intr reg
1163 			 */
1164 			ena_com_unmask_intr(rx_ring->ena_com_io_cq, &intr_reg);
1165 		}
1166 
1167 
1168 		ena_update_ring_numa_node(tx_ring, rx_ring);
1169 
1170 		ret = rx_work_done;
1171 	} else {
1172 		ret = budget;
1173 	}
1174 
1175 	u64_stats_update_begin(&tx_ring->syncp);
1176 	tx_ring->tx_stats.napi_comp += napi_comp_call;
1177 	tx_ring->tx_stats.tx_poll++;
1178 	u64_stats_update_end(&tx_ring->syncp);
1179 
1180 	return ret;
1181 }
1182 
1183 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1184 {
1185 	struct ena_adapter *adapter = (struct ena_adapter *)data;
1186 
1187 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1188 
1189 	/* Don't call the aenq handler before probe is done */
1190 	if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1191 		ena_com_aenq_intr_handler(adapter->ena_dev, data);
1192 
1193 	return IRQ_HANDLED;
1194 }
1195 
1196 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1197  * @irq: interrupt number
1198  * @data: pointer to a network interface private napi device structure
1199  */
1200 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1201 {
1202 	struct ena_napi *ena_napi = data;
1203 
1204 	napi_schedule(&ena_napi->napi);
1205 
1206 	return IRQ_HANDLED;
1207 }
1208 
1209 static int ena_enable_msix(struct ena_adapter *adapter, int num_queues)
1210 {
1211 	int msix_vecs, rc;
1212 
1213 	/* Reserved the max msix vectors we might need */
1214 	msix_vecs = ENA_MAX_MSIX_VEC(num_queues);
1215 
1216 	netif_dbg(adapter, probe, adapter->netdev,
1217 		  "trying to enable MSI-X, vectors %d\n", msix_vecs);
1218 
1219 	rc = pci_alloc_irq_vectors(adapter->pdev, msix_vecs, msix_vecs,
1220 			PCI_IRQ_MSIX);
1221 	if (rc < 0) {
1222 		netif_err(adapter, probe, adapter->netdev,
1223 			  "Failed to enable MSI-X, vectors %d rc %d\n",
1224 			  msix_vecs, rc);
1225 		return -ENOSPC;
1226 	}
1227 
1228 	netif_dbg(adapter, probe, adapter->netdev, "enable MSI-X, vectors %d\n",
1229 		  msix_vecs);
1230 
1231 	if (msix_vecs >= 1) {
1232 		if (ena_init_rx_cpu_rmap(adapter))
1233 			netif_warn(adapter, probe, adapter->netdev,
1234 				   "Failed to map IRQs to CPUs\n");
1235 	}
1236 
1237 	adapter->msix_vecs = msix_vecs;
1238 
1239 	return 0;
1240 }
1241 
1242 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1243 {
1244 	u32 cpu;
1245 
1246 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1247 		 ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1248 		 pci_name(adapter->pdev));
1249 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
1250 		ena_intr_msix_mgmnt;
1251 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1252 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1253 		pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
1254 	cpu = cpumask_first(cpu_online_mask);
1255 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
1256 	cpumask_set_cpu(cpu,
1257 			&adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
1258 }
1259 
1260 static void ena_setup_io_intr(struct ena_adapter *adapter)
1261 {
1262 	struct net_device *netdev;
1263 	int irq_idx, i, cpu;
1264 
1265 	netdev = adapter->netdev;
1266 
1267 	for (i = 0; i < adapter->num_queues; i++) {
1268 		irq_idx = ENA_IO_IRQ_IDX(i);
1269 		cpu = i % num_online_cpus();
1270 
1271 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1272 			 "%s-Tx-Rx-%d", netdev->name, i);
1273 		adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
1274 		adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
1275 		adapter->irq_tbl[irq_idx].vector =
1276 			pci_irq_vector(adapter->pdev, irq_idx);
1277 		adapter->irq_tbl[irq_idx].cpu = cpu;
1278 
1279 		cpumask_set_cpu(cpu,
1280 				&adapter->irq_tbl[irq_idx].affinity_hint_mask);
1281 	}
1282 }
1283 
1284 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
1285 {
1286 	unsigned long flags = 0;
1287 	struct ena_irq *irq;
1288 	int rc;
1289 
1290 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1291 	rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1292 			 irq->data);
1293 	if (rc) {
1294 		netif_err(adapter, probe, adapter->netdev,
1295 			  "failed to request admin irq\n");
1296 		return rc;
1297 	}
1298 
1299 	netif_dbg(adapter, probe, adapter->netdev,
1300 		  "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1301 		  irq->affinity_hint_mask.bits[0], irq->vector);
1302 
1303 	irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1304 
1305 	return rc;
1306 }
1307 
1308 static int ena_request_io_irq(struct ena_adapter *adapter)
1309 {
1310 	unsigned long flags = 0;
1311 	struct ena_irq *irq;
1312 	int rc = 0, i, k;
1313 
1314 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1315 		irq = &adapter->irq_tbl[i];
1316 		rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1317 				 irq->data);
1318 		if (rc) {
1319 			netif_err(adapter, ifup, adapter->netdev,
1320 				  "Failed to request I/O IRQ. index %d rc %d\n",
1321 				   i, rc);
1322 			goto err;
1323 		}
1324 
1325 		netif_dbg(adapter, ifup, adapter->netdev,
1326 			  "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1327 			  i, irq->affinity_hint_mask.bits[0], irq->vector);
1328 
1329 		irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1330 	}
1331 
1332 	return rc;
1333 
1334 err:
1335 	for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
1336 		irq = &adapter->irq_tbl[k];
1337 		free_irq(irq->vector, irq->data);
1338 	}
1339 
1340 	return rc;
1341 }
1342 
1343 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
1344 {
1345 	struct ena_irq *irq;
1346 
1347 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1348 	synchronize_irq(irq->vector);
1349 	irq_set_affinity_hint(irq->vector, NULL);
1350 	free_irq(irq->vector, irq->data);
1351 }
1352 
1353 static void ena_free_io_irq(struct ena_adapter *adapter)
1354 {
1355 	struct ena_irq *irq;
1356 	int i;
1357 
1358 #ifdef CONFIG_RFS_ACCEL
1359 	if (adapter->msix_vecs >= 1) {
1360 		free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
1361 		adapter->netdev->rx_cpu_rmap = NULL;
1362 	}
1363 #endif /* CONFIG_RFS_ACCEL */
1364 
1365 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1366 		irq = &adapter->irq_tbl[i];
1367 		irq_set_affinity_hint(irq->vector, NULL);
1368 		free_irq(irq->vector, irq->data);
1369 	}
1370 }
1371 
1372 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
1373 {
1374 	int i;
1375 
1376 	if (!netif_running(adapter->netdev))
1377 		return;
1378 
1379 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++)
1380 		synchronize_irq(adapter->irq_tbl[i].vector);
1381 }
1382 
1383 static void ena_del_napi(struct ena_adapter *adapter)
1384 {
1385 	int i;
1386 
1387 	for (i = 0; i < adapter->num_queues; i++)
1388 		netif_napi_del(&adapter->ena_napi[i].napi);
1389 }
1390 
1391 static void ena_init_napi(struct ena_adapter *adapter)
1392 {
1393 	struct ena_napi *napi;
1394 	int i;
1395 
1396 	for (i = 0; i < adapter->num_queues; i++) {
1397 		napi = &adapter->ena_napi[i];
1398 
1399 		netif_napi_add(adapter->netdev,
1400 			       &adapter->ena_napi[i].napi,
1401 			       ena_io_poll,
1402 			       ENA_NAPI_BUDGET);
1403 		napi->rx_ring = &adapter->rx_ring[i];
1404 		napi->tx_ring = &adapter->tx_ring[i];
1405 		napi->qid = i;
1406 	}
1407 }
1408 
1409 static void ena_napi_disable_all(struct ena_adapter *adapter)
1410 {
1411 	int i;
1412 
1413 	for (i = 0; i < adapter->num_queues; i++)
1414 		napi_disable(&adapter->ena_napi[i].napi);
1415 }
1416 
1417 static void ena_napi_enable_all(struct ena_adapter *adapter)
1418 {
1419 	int i;
1420 
1421 	for (i = 0; i < adapter->num_queues; i++)
1422 		napi_enable(&adapter->ena_napi[i].napi);
1423 }
1424 
1425 static void ena_restore_ethtool_params(struct ena_adapter *adapter)
1426 {
1427 	adapter->tx_usecs = 0;
1428 	adapter->rx_usecs = 0;
1429 	adapter->tx_frames = 1;
1430 	adapter->rx_frames = 1;
1431 }
1432 
1433 /* Configure the Rx forwarding */
1434 static int ena_rss_configure(struct ena_adapter *adapter)
1435 {
1436 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1437 	int rc;
1438 
1439 	/* In case the RSS table wasn't initialized by probe */
1440 	if (!ena_dev->rss.tbl_log_size) {
1441 		rc = ena_rss_init_default(adapter);
1442 		if (rc && (rc != -EPERM)) {
1443 			netif_err(adapter, ifup, adapter->netdev,
1444 				  "Failed to init RSS rc: %d\n", rc);
1445 			return rc;
1446 		}
1447 	}
1448 
1449 	/* Set indirect table */
1450 	rc = ena_com_indirect_table_set(ena_dev);
1451 	if (unlikely(rc && rc != -EPERM))
1452 		return rc;
1453 
1454 	/* Configure hash function (if supported) */
1455 	rc = ena_com_set_hash_function(ena_dev);
1456 	if (unlikely(rc && (rc != -EPERM)))
1457 		return rc;
1458 
1459 	/* Configure hash inputs (if supported) */
1460 	rc = ena_com_set_hash_ctrl(ena_dev);
1461 	if (unlikely(rc && (rc != -EPERM)))
1462 		return rc;
1463 
1464 	return 0;
1465 }
1466 
1467 static int ena_up_complete(struct ena_adapter *adapter)
1468 {
1469 	int rc, i;
1470 
1471 	rc = ena_rss_configure(adapter);
1472 	if (rc)
1473 		return rc;
1474 
1475 	ena_init_napi(adapter);
1476 
1477 	ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
1478 
1479 	ena_refill_all_rx_bufs(adapter);
1480 
1481 	/* enable transmits */
1482 	netif_tx_start_all_queues(adapter->netdev);
1483 
1484 	ena_restore_ethtool_params(adapter);
1485 
1486 	ena_napi_enable_all(adapter);
1487 
1488 	/* schedule napi in case we had pending packets
1489 	 * from the last time we disable napi
1490 	 */
1491 	for (i = 0; i < adapter->num_queues; i++)
1492 		napi_schedule(&adapter->ena_napi[i].napi);
1493 
1494 	return 0;
1495 }
1496 
1497 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
1498 {
1499 	struct ena_com_create_io_ctx ctx = { 0 };
1500 	struct ena_com_dev *ena_dev;
1501 	struct ena_ring *tx_ring;
1502 	u32 msix_vector;
1503 	u16 ena_qid;
1504 	int rc;
1505 
1506 	ena_dev = adapter->ena_dev;
1507 
1508 	tx_ring = &adapter->tx_ring[qid];
1509 	msix_vector = ENA_IO_IRQ_IDX(qid);
1510 	ena_qid = ENA_IO_TXQ_IDX(qid);
1511 
1512 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1513 	ctx.qid = ena_qid;
1514 	ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1515 	ctx.msix_vector = msix_vector;
1516 	ctx.queue_size = adapter->tx_ring_size;
1517 	ctx.numa_node = cpu_to_node(tx_ring->cpu);
1518 
1519 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1520 	if (rc) {
1521 		netif_err(adapter, ifup, adapter->netdev,
1522 			  "Failed to create I/O TX queue num %d rc: %d\n",
1523 			  qid, rc);
1524 		return rc;
1525 	}
1526 
1527 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1528 				     &tx_ring->ena_com_io_sq,
1529 				     &tx_ring->ena_com_io_cq);
1530 	if (rc) {
1531 		netif_err(adapter, ifup, adapter->netdev,
1532 			  "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1533 			  qid, rc);
1534 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1535 	}
1536 
1537 	ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
1538 	return rc;
1539 }
1540 
1541 static int ena_create_all_io_tx_queues(struct ena_adapter *adapter)
1542 {
1543 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1544 	int rc, i;
1545 
1546 	for (i = 0; i < adapter->num_queues; i++) {
1547 		rc = ena_create_io_tx_queue(adapter, i);
1548 		if (rc)
1549 			goto create_err;
1550 	}
1551 
1552 	return 0;
1553 
1554 create_err:
1555 	while (i--)
1556 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1557 
1558 	return rc;
1559 }
1560 
1561 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
1562 {
1563 	struct ena_com_dev *ena_dev;
1564 	struct ena_com_create_io_ctx ctx = { 0 };
1565 	struct ena_ring *rx_ring;
1566 	u32 msix_vector;
1567 	u16 ena_qid;
1568 	int rc;
1569 
1570 	ena_dev = adapter->ena_dev;
1571 
1572 	rx_ring = &adapter->rx_ring[qid];
1573 	msix_vector = ENA_IO_IRQ_IDX(qid);
1574 	ena_qid = ENA_IO_RXQ_IDX(qid);
1575 
1576 	ctx.qid = ena_qid;
1577 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1578 	ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1579 	ctx.msix_vector = msix_vector;
1580 	ctx.queue_size = adapter->rx_ring_size;
1581 	ctx.numa_node = cpu_to_node(rx_ring->cpu);
1582 
1583 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1584 	if (rc) {
1585 		netif_err(adapter, ifup, adapter->netdev,
1586 			  "Failed to create I/O RX queue num %d rc: %d\n",
1587 			  qid, rc);
1588 		return rc;
1589 	}
1590 
1591 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1592 				     &rx_ring->ena_com_io_sq,
1593 				     &rx_ring->ena_com_io_cq);
1594 	if (rc) {
1595 		netif_err(adapter, ifup, adapter->netdev,
1596 			  "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1597 			  qid, rc);
1598 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1599 	}
1600 
1601 	ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
1602 
1603 	return rc;
1604 }
1605 
1606 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
1607 {
1608 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1609 	int rc, i;
1610 
1611 	for (i = 0; i < adapter->num_queues; i++) {
1612 		rc = ena_create_io_rx_queue(adapter, i);
1613 		if (rc)
1614 			goto create_err;
1615 	}
1616 
1617 	return 0;
1618 
1619 create_err:
1620 	while (i--)
1621 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1622 
1623 	return rc;
1624 }
1625 
1626 static int ena_up(struct ena_adapter *adapter)
1627 {
1628 	int rc;
1629 
1630 	netdev_dbg(adapter->netdev, "%s\n", __func__);
1631 
1632 	ena_setup_io_intr(adapter);
1633 
1634 	rc = ena_request_io_irq(adapter);
1635 	if (rc)
1636 		goto err_req_irq;
1637 
1638 	/* allocate transmit descriptors */
1639 	rc = ena_setup_all_tx_resources(adapter);
1640 	if (rc)
1641 		goto err_setup_tx;
1642 
1643 	/* allocate receive descriptors */
1644 	rc = ena_setup_all_rx_resources(adapter);
1645 	if (rc)
1646 		goto err_setup_rx;
1647 
1648 	/* Create TX queues */
1649 	rc = ena_create_all_io_tx_queues(adapter);
1650 	if (rc)
1651 		goto err_create_tx_queues;
1652 
1653 	/* Create RX queues */
1654 	rc = ena_create_all_io_rx_queues(adapter);
1655 	if (rc)
1656 		goto err_create_rx_queues;
1657 
1658 	rc = ena_up_complete(adapter);
1659 	if (rc)
1660 		goto err_up;
1661 
1662 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
1663 		netif_carrier_on(adapter->netdev);
1664 
1665 	u64_stats_update_begin(&adapter->syncp);
1666 	adapter->dev_stats.interface_up++;
1667 	u64_stats_update_end(&adapter->syncp);
1668 
1669 	set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1670 
1671 	return rc;
1672 
1673 err_up:
1674 	ena_destroy_all_rx_queues(adapter);
1675 err_create_rx_queues:
1676 	ena_destroy_all_tx_queues(adapter);
1677 err_create_tx_queues:
1678 	ena_free_all_io_rx_resources(adapter);
1679 err_setup_rx:
1680 	ena_free_all_io_tx_resources(adapter);
1681 err_setup_tx:
1682 	ena_free_io_irq(adapter);
1683 err_req_irq:
1684 
1685 	return rc;
1686 }
1687 
1688 static void ena_down(struct ena_adapter *adapter)
1689 {
1690 	netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
1691 
1692 	clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1693 
1694 	u64_stats_update_begin(&adapter->syncp);
1695 	adapter->dev_stats.interface_down++;
1696 	u64_stats_update_end(&adapter->syncp);
1697 
1698 	netif_carrier_off(adapter->netdev);
1699 	netif_tx_disable(adapter->netdev);
1700 
1701 	/* After this point the napi handler won't enable the tx queue */
1702 	ena_napi_disable_all(adapter);
1703 
1704 	/* After destroy the queue there won't be any new interrupts */
1705 
1706 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
1707 		int rc;
1708 
1709 		rc = ena_com_dev_reset(adapter->ena_dev);
1710 		if (rc)
1711 			dev_err(&adapter->pdev->dev, "Device reset failed\n");
1712 	}
1713 
1714 	ena_destroy_all_io_queues(adapter);
1715 
1716 	ena_disable_io_intr_sync(adapter);
1717 	ena_free_io_irq(adapter);
1718 	ena_del_napi(adapter);
1719 
1720 	ena_free_all_tx_bufs(adapter);
1721 	ena_free_all_rx_bufs(adapter);
1722 	ena_free_all_io_tx_resources(adapter);
1723 	ena_free_all_io_rx_resources(adapter);
1724 }
1725 
1726 /* ena_open - Called when a network interface is made active
1727  * @netdev: network interface device structure
1728  *
1729  * Returns 0 on success, negative value on failure
1730  *
1731  * The open entry point is called when a network interface is made
1732  * active by the system (IFF_UP).  At this point all resources needed
1733  * for transmit and receive operations are allocated, the interrupt
1734  * handler is registered with the OS, the watchdog timer is started,
1735  * and the stack is notified that the interface is ready.
1736  */
1737 static int ena_open(struct net_device *netdev)
1738 {
1739 	struct ena_adapter *adapter = netdev_priv(netdev);
1740 	int rc;
1741 
1742 	/* Notify the stack of the actual queue counts. */
1743 	rc = netif_set_real_num_tx_queues(netdev, adapter->num_queues);
1744 	if (rc) {
1745 		netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
1746 		return rc;
1747 	}
1748 
1749 	rc = netif_set_real_num_rx_queues(netdev, adapter->num_queues);
1750 	if (rc) {
1751 		netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
1752 		return rc;
1753 	}
1754 
1755 	rc = ena_up(adapter);
1756 	if (rc)
1757 		return rc;
1758 
1759 	return rc;
1760 }
1761 
1762 /* ena_close - Disables a network interface
1763  * @netdev: network interface device structure
1764  *
1765  * Returns 0, this is not allowed to fail
1766  *
1767  * The close entry point is called when an interface is de-activated
1768  * by the OS.  The hardware is still under the drivers control, but
1769  * needs to be disabled.  A global MAC reset is issued to stop the
1770  * hardware, and all transmit and receive resources are freed.
1771  */
1772 static int ena_close(struct net_device *netdev)
1773 {
1774 	struct ena_adapter *adapter = netdev_priv(netdev);
1775 
1776 	netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
1777 
1778 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
1779 		ena_down(adapter);
1780 
1781 	return 0;
1782 }
1783 
1784 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
1785 {
1786 	u32 mss = skb_shinfo(skb)->gso_size;
1787 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
1788 	u8 l4_protocol = 0;
1789 
1790 	if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
1791 		ena_tx_ctx->l4_csum_enable = 1;
1792 		if (mss) {
1793 			ena_tx_ctx->tso_enable = 1;
1794 			ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
1795 			ena_tx_ctx->l4_csum_partial = 0;
1796 		} else {
1797 			ena_tx_ctx->tso_enable = 0;
1798 			ena_meta->l4_hdr_len = 0;
1799 			ena_tx_ctx->l4_csum_partial = 1;
1800 		}
1801 
1802 		switch (ip_hdr(skb)->version) {
1803 		case IPVERSION:
1804 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
1805 			if (ip_hdr(skb)->frag_off & htons(IP_DF))
1806 				ena_tx_ctx->df = 1;
1807 			if (mss)
1808 				ena_tx_ctx->l3_csum_enable = 1;
1809 			l4_protocol = ip_hdr(skb)->protocol;
1810 			break;
1811 		case 6:
1812 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
1813 			l4_protocol = ipv6_hdr(skb)->nexthdr;
1814 			break;
1815 		default:
1816 			break;
1817 		}
1818 
1819 		if (l4_protocol == IPPROTO_TCP)
1820 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
1821 		else
1822 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
1823 
1824 		ena_meta->mss = mss;
1825 		ena_meta->l3_hdr_len = skb_network_header_len(skb);
1826 		ena_meta->l3_hdr_offset = skb_network_offset(skb);
1827 		ena_tx_ctx->meta_valid = 1;
1828 
1829 	} else {
1830 		ena_tx_ctx->meta_valid = 0;
1831 	}
1832 }
1833 
1834 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
1835 				       struct sk_buff *skb)
1836 {
1837 	int num_frags, header_len, rc;
1838 
1839 	num_frags = skb_shinfo(skb)->nr_frags;
1840 	header_len = skb_headlen(skb);
1841 
1842 	if (num_frags < tx_ring->sgl_size)
1843 		return 0;
1844 
1845 	if ((num_frags == tx_ring->sgl_size) &&
1846 	    (header_len < tx_ring->tx_max_header_size))
1847 		return 0;
1848 
1849 	u64_stats_update_begin(&tx_ring->syncp);
1850 	tx_ring->tx_stats.linearize++;
1851 	u64_stats_update_end(&tx_ring->syncp);
1852 
1853 	rc = skb_linearize(skb);
1854 	if (unlikely(rc)) {
1855 		u64_stats_update_begin(&tx_ring->syncp);
1856 		tx_ring->tx_stats.linearize_failed++;
1857 		u64_stats_update_end(&tx_ring->syncp);
1858 	}
1859 
1860 	return rc;
1861 }
1862 
1863 /* Called with netif_tx_lock. */
1864 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
1865 {
1866 	struct ena_adapter *adapter = netdev_priv(dev);
1867 	struct ena_tx_buffer *tx_info;
1868 	struct ena_com_tx_ctx ena_tx_ctx;
1869 	struct ena_ring *tx_ring;
1870 	struct netdev_queue *txq;
1871 	struct ena_com_buf *ena_buf;
1872 	void *push_hdr;
1873 	u32 len, last_frag;
1874 	u16 next_to_use;
1875 	u16 req_id;
1876 	u16 push_len;
1877 	u16 header_len;
1878 	dma_addr_t dma;
1879 	int qid, rc, nb_hw_desc;
1880 	int i = -1;
1881 
1882 	netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
1883 	/*  Determine which tx ring we will be placed on */
1884 	qid = skb_get_queue_mapping(skb);
1885 	tx_ring = &adapter->tx_ring[qid];
1886 	txq = netdev_get_tx_queue(dev, qid);
1887 
1888 	rc = ena_check_and_linearize_skb(tx_ring, skb);
1889 	if (unlikely(rc))
1890 		goto error_drop_packet;
1891 
1892 	skb_tx_timestamp(skb);
1893 	len = skb_headlen(skb);
1894 
1895 	next_to_use = tx_ring->next_to_use;
1896 	req_id = tx_ring->free_tx_ids[next_to_use];
1897 	tx_info = &tx_ring->tx_buffer_info[req_id];
1898 	tx_info->num_of_bufs = 0;
1899 
1900 	WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
1901 	ena_buf = tx_info->bufs;
1902 	tx_info->skb = skb;
1903 
1904 	if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1905 		/* prepared the push buffer */
1906 		push_len = min_t(u32, len, tx_ring->tx_max_header_size);
1907 		header_len = push_len;
1908 		push_hdr = skb->data;
1909 	} else {
1910 		push_len = 0;
1911 		header_len = min_t(u32, len, tx_ring->tx_max_header_size);
1912 		push_hdr = NULL;
1913 	}
1914 
1915 	netif_dbg(adapter, tx_queued, dev,
1916 		  "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
1917 		  push_hdr, push_len);
1918 
1919 	if (len > push_len) {
1920 		dma = dma_map_single(tx_ring->dev, skb->data + push_len,
1921 				     len - push_len, DMA_TO_DEVICE);
1922 		if (dma_mapping_error(tx_ring->dev, dma))
1923 			goto error_report_dma_error;
1924 
1925 		ena_buf->paddr = dma;
1926 		ena_buf->len = len - push_len;
1927 
1928 		ena_buf++;
1929 		tx_info->num_of_bufs++;
1930 	}
1931 
1932 	last_frag = skb_shinfo(skb)->nr_frags;
1933 
1934 	for (i = 0; i < last_frag; i++) {
1935 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1936 
1937 		len = skb_frag_size(frag);
1938 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, len,
1939 				       DMA_TO_DEVICE);
1940 		if (dma_mapping_error(tx_ring->dev, dma))
1941 			goto error_report_dma_error;
1942 
1943 		ena_buf->paddr = dma;
1944 		ena_buf->len = len;
1945 		ena_buf++;
1946 	}
1947 
1948 	tx_info->num_of_bufs += last_frag;
1949 
1950 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1951 	ena_tx_ctx.ena_bufs = tx_info->bufs;
1952 	ena_tx_ctx.push_header = push_hdr;
1953 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1954 	ena_tx_ctx.req_id = req_id;
1955 	ena_tx_ctx.header_len = header_len;
1956 
1957 	/* set flags and meta data */
1958 	ena_tx_csum(&ena_tx_ctx, skb);
1959 
1960 	/* prepare the packet's descriptors to dma engine */
1961 	rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
1962 				&nb_hw_desc);
1963 
1964 	if (unlikely(rc)) {
1965 		netif_err(adapter, tx_queued, dev,
1966 			  "failed to prepare tx bufs\n");
1967 		u64_stats_update_begin(&tx_ring->syncp);
1968 		tx_ring->tx_stats.queue_stop++;
1969 		tx_ring->tx_stats.prepare_ctx_err++;
1970 		u64_stats_update_end(&tx_ring->syncp);
1971 		netif_tx_stop_queue(txq);
1972 		goto error_unmap_dma;
1973 	}
1974 
1975 	netdev_tx_sent_queue(txq, skb->len);
1976 
1977 	u64_stats_update_begin(&tx_ring->syncp);
1978 	tx_ring->tx_stats.cnt++;
1979 	tx_ring->tx_stats.bytes += skb->len;
1980 	u64_stats_update_end(&tx_ring->syncp);
1981 
1982 	tx_info->tx_descs = nb_hw_desc;
1983 	tx_info->last_jiffies = jiffies;
1984 
1985 	tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
1986 		tx_ring->ring_size);
1987 
1988 	/* This WMB is aimed to:
1989 	 * 1 - perform smp barrier before reading next_to_completion
1990 	 * 2 - make sure the desc were written before trigger DB
1991 	 */
1992 	wmb();
1993 
1994 	/* stop the queue when no more space available, the packet can have up
1995 	 * to sgl_size + 2. one for the meta descriptor and one for header
1996 	 * (if the header is larger than tx_max_header_size).
1997 	 */
1998 	if (unlikely(ena_com_sq_empty_space(tx_ring->ena_com_io_sq) <
1999 		     (tx_ring->sgl_size + 2))) {
2000 		netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
2001 			  __func__, qid);
2002 
2003 		netif_tx_stop_queue(txq);
2004 		u64_stats_update_begin(&tx_ring->syncp);
2005 		tx_ring->tx_stats.queue_stop++;
2006 		u64_stats_update_end(&tx_ring->syncp);
2007 
2008 		/* There is a rare condition where this function decide to
2009 		 * stop the queue but meanwhile clean_tx_irq updates
2010 		 * next_to_completion and terminates.
2011 		 * The queue will remain stopped forever.
2012 		 * To solve this issue this function perform rmb, check
2013 		 * the wakeup condition and wake up the queue if needed.
2014 		 */
2015 		smp_rmb();
2016 
2017 		if (ena_com_sq_empty_space(tx_ring->ena_com_io_sq)
2018 				> ENA_TX_WAKEUP_THRESH) {
2019 			netif_tx_wake_queue(txq);
2020 			u64_stats_update_begin(&tx_ring->syncp);
2021 			tx_ring->tx_stats.queue_wakeup++;
2022 			u64_stats_update_end(&tx_ring->syncp);
2023 		}
2024 	}
2025 
2026 	if (netif_xmit_stopped(txq) || !skb->xmit_more) {
2027 		/* trigger the dma engine */
2028 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2029 		u64_stats_update_begin(&tx_ring->syncp);
2030 		tx_ring->tx_stats.doorbells++;
2031 		u64_stats_update_end(&tx_ring->syncp);
2032 	}
2033 
2034 	return NETDEV_TX_OK;
2035 
2036 error_report_dma_error:
2037 	u64_stats_update_begin(&tx_ring->syncp);
2038 	tx_ring->tx_stats.dma_mapping_err++;
2039 	u64_stats_update_end(&tx_ring->syncp);
2040 	netdev_warn(adapter->netdev, "failed to map skb\n");
2041 
2042 	tx_info->skb = NULL;
2043 
2044 error_unmap_dma:
2045 	if (i >= 0) {
2046 		/* save value of frag that failed */
2047 		last_frag = i;
2048 
2049 		/* start back at beginning and unmap skb */
2050 		tx_info->skb = NULL;
2051 		ena_buf = tx_info->bufs;
2052 		dma_unmap_single(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2053 				 dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2054 
2055 		/* unmap remaining mapped pages */
2056 		for (i = 0; i < last_frag; i++) {
2057 			ena_buf++;
2058 			dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2059 				       dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2060 		}
2061 	}
2062 
2063 error_drop_packet:
2064 
2065 	dev_kfree_skb(skb);
2066 	return NETDEV_TX_OK;
2067 }
2068 
2069 #ifdef CONFIG_NET_POLL_CONTROLLER
2070 static void ena_netpoll(struct net_device *netdev)
2071 {
2072 	struct ena_adapter *adapter = netdev_priv(netdev);
2073 	int i;
2074 
2075 	/* Dont schedule NAPI if the driver is in the middle of reset
2076 	 * or netdev is down.
2077 	 */
2078 
2079 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags) ||
2080 	    test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2081 		return;
2082 
2083 	for (i = 0; i < adapter->num_queues; i++)
2084 		napi_schedule(&adapter->ena_napi[i].napi);
2085 }
2086 #endif /* CONFIG_NET_POLL_CONTROLLER */
2087 
2088 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
2089 			    void *accel_priv, select_queue_fallback_t fallback)
2090 {
2091 	u16 qid;
2092 	/* we suspect that this is good for in--kernel network services that
2093 	 * want to loop incoming skb rx to tx in normal user generated traffic,
2094 	 * most probably we will not get to this
2095 	 */
2096 	if (skb_rx_queue_recorded(skb))
2097 		qid = skb_get_rx_queue(skb);
2098 	else
2099 		qid = fallback(dev, skb);
2100 
2101 	return qid;
2102 }
2103 
2104 static void ena_config_host_info(struct ena_com_dev *ena_dev)
2105 {
2106 	struct ena_admin_host_info *host_info;
2107 	int rc;
2108 
2109 	/* Allocate only the host info */
2110 	rc = ena_com_allocate_host_info(ena_dev);
2111 	if (rc) {
2112 		pr_err("Cannot allocate host info\n");
2113 		return;
2114 	}
2115 
2116 	host_info = ena_dev->host_attr.host_info;
2117 
2118 	host_info->os_type = ENA_ADMIN_OS_LINUX;
2119 	host_info->kernel_ver = LINUX_VERSION_CODE;
2120 	strncpy(host_info->kernel_ver_str, utsname()->version,
2121 		sizeof(host_info->kernel_ver_str) - 1);
2122 	host_info->os_dist = 0;
2123 	strncpy(host_info->os_dist_str, utsname()->release,
2124 		sizeof(host_info->os_dist_str) - 1);
2125 	host_info->driver_version =
2126 		(DRV_MODULE_VER_MAJOR) |
2127 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2128 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2129 
2130 	rc = ena_com_set_host_attributes(ena_dev);
2131 	if (rc) {
2132 		if (rc == -EPERM)
2133 			pr_warn("Cannot set host attributes\n");
2134 		else
2135 			pr_err("Cannot set host attributes\n");
2136 
2137 		goto err;
2138 	}
2139 
2140 	return;
2141 
2142 err:
2143 	ena_com_delete_host_info(ena_dev);
2144 }
2145 
2146 static void ena_config_debug_area(struct ena_adapter *adapter)
2147 {
2148 	u32 debug_area_size;
2149 	int rc, ss_count;
2150 
2151 	ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
2152 	if (ss_count <= 0) {
2153 		netif_err(adapter, drv, adapter->netdev,
2154 			  "SS count is negative\n");
2155 		return;
2156 	}
2157 
2158 	/* allocate 32 bytes for each string and 64bit for the value */
2159 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
2160 
2161 	rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
2162 	if (rc) {
2163 		pr_err("Cannot allocate debug area\n");
2164 		return;
2165 	}
2166 
2167 	rc = ena_com_set_host_attributes(adapter->ena_dev);
2168 	if (rc) {
2169 		if (rc == -EPERM)
2170 			netif_warn(adapter, drv, adapter->netdev,
2171 				   "Cannot set host attributes\n");
2172 		else
2173 			netif_err(adapter, drv, adapter->netdev,
2174 				  "Cannot set host attributes\n");
2175 		goto err;
2176 	}
2177 
2178 	return;
2179 err:
2180 	ena_com_delete_debug_area(adapter->ena_dev);
2181 }
2182 
2183 static void ena_get_stats64(struct net_device *netdev,
2184 			    struct rtnl_link_stats64 *stats)
2185 {
2186 	struct ena_adapter *adapter = netdev_priv(netdev);
2187 	struct ena_ring *rx_ring, *tx_ring;
2188 	unsigned int start;
2189 	u64 rx_drops;
2190 	int i;
2191 
2192 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2193 		return;
2194 
2195 	for (i = 0; i < adapter->num_queues; i++) {
2196 		u64 bytes, packets;
2197 
2198 		tx_ring = &adapter->tx_ring[i];
2199 
2200 		do {
2201 			start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
2202 			packets = tx_ring->tx_stats.cnt;
2203 			bytes = tx_ring->tx_stats.bytes;
2204 		} while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
2205 
2206 		stats->tx_packets += packets;
2207 		stats->tx_bytes += bytes;
2208 
2209 		rx_ring = &adapter->rx_ring[i];
2210 
2211 		do {
2212 			start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
2213 			packets = rx_ring->rx_stats.cnt;
2214 			bytes = rx_ring->rx_stats.bytes;
2215 		} while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
2216 
2217 		stats->rx_packets += packets;
2218 		stats->rx_bytes += bytes;
2219 	}
2220 
2221 	do {
2222 		start = u64_stats_fetch_begin_irq(&adapter->syncp);
2223 		rx_drops = adapter->dev_stats.rx_drops;
2224 	} while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
2225 
2226 	stats->rx_dropped = rx_drops;
2227 
2228 	stats->multicast = 0;
2229 	stats->collisions = 0;
2230 
2231 	stats->rx_length_errors = 0;
2232 	stats->rx_crc_errors = 0;
2233 	stats->rx_frame_errors = 0;
2234 	stats->rx_fifo_errors = 0;
2235 	stats->rx_missed_errors = 0;
2236 	stats->tx_window_errors = 0;
2237 
2238 	stats->rx_errors = 0;
2239 	stats->tx_errors = 0;
2240 }
2241 
2242 static const struct net_device_ops ena_netdev_ops = {
2243 	.ndo_open		= ena_open,
2244 	.ndo_stop		= ena_close,
2245 	.ndo_start_xmit		= ena_start_xmit,
2246 	.ndo_select_queue	= ena_select_queue,
2247 	.ndo_get_stats64	= ena_get_stats64,
2248 	.ndo_tx_timeout		= ena_tx_timeout,
2249 	.ndo_change_mtu		= ena_change_mtu,
2250 	.ndo_set_mac_address	= NULL,
2251 	.ndo_validate_addr	= eth_validate_addr,
2252 #ifdef CONFIG_NET_POLL_CONTROLLER
2253 	.ndo_poll_controller	= ena_netpoll,
2254 #endif /* CONFIG_NET_POLL_CONTROLLER */
2255 };
2256 
2257 static void ena_device_io_suspend(struct work_struct *work)
2258 {
2259 	struct ena_adapter *adapter =
2260 		container_of(work, struct ena_adapter, suspend_io_task);
2261 	struct net_device *netdev = adapter->netdev;
2262 
2263 	/* ena_napi_disable_all disables only the IO handling.
2264 	 * We are still subject to AENQ keep alive watchdog.
2265 	 */
2266 	u64_stats_update_begin(&adapter->syncp);
2267 	adapter->dev_stats.io_suspend++;
2268 	u64_stats_update_begin(&adapter->syncp);
2269 	ena_napi_disable_all(adapter);
2270 	netif_tx_lock(netdev);
2271 	netif_device_detach(netdev);
2272 	netif_tx_unlock(netdev);
2273 }
2274 
2275 static void ena_device_io_resume(struct work_struct *work)
2276 {
2277 	struct ena_adapter *adapter =
2278 		container_of(work, struct ena_adapter, resume_io_task);
2279 	struct net_device *netdev = adapter->netdev;
2280 
2281 	u64_stats_update_begin(&adapter->syncp);
2282 	adapter->dev_stats.io_resume++;
2283 	u64_stats_update_end(&adapter->syncp);
2284 
2285 	netif_device_attach(netdev);
2286 	ena_napi_enable_all(adapter);
2287 }
2288 
2289 static int ena_device_validate_params(struct ena_adapter *adapter,
2290 				      struct ena_com_dev_get_features_ctx *get_feat_ctx)
2291 {
2292 	struct net_device *netdev = adapter->netdev;
2293 	int rc;
2294 
2295 	rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
2296 			      adapter->mac_addr);
2297 	if (!rc) {
2298 		netif_err(adapter, drv, netdev,
2299 			  "Error, mac address are different\n");
2300 		return -EINVAL;
2301 	}
2302 
2303 	if ((get_feat_ctx->max_queues.max_cq_num < adapter->num_queues) ||
2304 	    (get_feat_ctx->max_queues.max_sq_num < adapter->num_queues)) {
2305 		netif_err(adapter, drv, netdev,
2306 			  "Error, device doesn't support enough queues\n");
2307 		return -EINVAL;
2308 	}
2309 
2310 	if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
2311 		netif_err(adapter, drv, netdev,
2312 			  "Error, device max mtu is smaller than netdev MTU\n");
2313 		return -EINVAL;
2314 	}
2315 
2316 	return 0;
2317 }
2318 
2319 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
2320 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
2321 			   bool *wd_state)
2322 {
2323 	struct device *dev = &pdev->dev;
2324 	bool readless_supported;
2325 	u32 aenq_groups;
2326 	int dma_width;
2327 	int rc;
2328 
2329 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
2330 	if (rc) {
2331 		dev_err(dev, "failed to init mmio read less\n");
2332 		return rc;
2333 	}
2334 
2335 	/* The PCIe configuration space revision id indicate if mmio reg
2336 	 * read is disabled
2337 	 */
2338 	readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
2339 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2340 
2341 	rc = ena_com_dev_reset(ena_dev);
2342 	if (rc) {
2343 		dev_err(dev, "Can not reset device\n");
2344 		goto err_mmio_read_less;
2345 	}
2346 
2347 	rc = ena_com_validate_version(ena_dev);
2348 	if (rc) {
2349 		dev_err(dev, "device version is too low\n");
2350 		goto err_mmio_read_less;
2351 	}
2352 
2353 	dma_width = ena_com_get_dma_width(ena_dev);
2354 	if (dma_width < 0) {
2355 		dev_err(dev, "Invalid dma width value %d", dma_width);
2356 		rc = dma_width;
2357 		goto err_mmio_read_less;
2358 	}
2359 
2360 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2361 	if (rc) {
2362 		dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
2363 		goto err_mmio_read_less;
2364 	}
2365 
2366 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2367 	if (rc) {
2368 		dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2369 			rc);
2370 		goto err_mmio_read_less;
2371 	}
2372 
2373 	/* ENA admin level init */
2374 	rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
2375 	if (rc) {
2376 		dev_err(dev,
2377 			"Can not initialize ena admin queue with device\n");
2378 		goto err_mmio_read_less;
2379 	}
2380 
2381 	/* To enable the msix interrupts the driver needs to know the number
2382 	 * of queues. So the driver uses polling mode to retrieve this
2383 	 * information
2384 	 */
2385 	ena_com_set_admin_polling_mode(ena_dev, true);
2386 
2387 	ena_config_host_info(ena_dev);
2388 
2389 	/* Get Device Attributes*/
2390 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2391 	if (rc) {
2392 		dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
2393 		goto err_admin_init;
2394 	}
2395 
2396 	/* Try to turn all the available aenq groups */
2397 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2398 		BIT(ENA_ADMIN_FATAL_ERROR) |
2399 		BIT(ENA_ADMIN_WARNING) |
2400 		BIT(ENA_ADMIN_NOTIFICATION) |
2401 		BIT(ENA_ADMIN_KEEP_ALIVE);
2402 
2403 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
2404 
2405 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2406 	if (rc) {
2407 		dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
2408 		goto err_admin_init;
2409 	}
2410 
2411 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2412 
2413 	return 0;
2414 
2415 err_admin_init:
2416 	ena_com_delete_host_info(ena_dev);
2417 	ena_com_admin_destroy(ena_dev);
2418 err_mmio_read_less:
2419 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2420 
2421 	return rc;
2422 }
2423 
2424 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
2425 						    int io_vectors)
2426 {
2427 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2428 	struct device *dev = &adapter->pdev->dev;
2429 	int rc;
2430 
2431 	rc = ena_enable_msix(adapter, io_vectors);
2432 	if (rc) {
2433 		dev_err(dev, "Can not reserve msix vectors\n");
2434 		return rc;
2435 	}
2436 
2437 	ena_setup_mgmnt_intr(adapter);
2438 
2439 	rc = ena_request_mgmnt_irq(adapter);
2440 	if (rc) {
2441 		dev_err(dev, "Can not setup management interrupts\n");
2442 		goto err_disable_msix;
2443 	}
2444 
2445 	ena_com_set_admin_polling_mode(ena_dev, false);
2446 
2447 	ena_com_admin_aenq_enable(ena_dev);
2448 
2449 	return 0;
2450 
2451 err_disable_msix:
2452 	pci_free_irq_vectors(adapter->pdev);
2453 	return rc;
2454 }
2455 
2456 static void ena_fw_reset_device(struct work_struct *work)
2457 {
2458 	struct ena_com_dev_get_features_ctx get_feat_ctx;
2459 	struct ena_adapter *adapter =
2460 		container_of(work, struct ena_adapter, reset_task);
2461 	struct net_device *netdev = adapter->netdev;
2462 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2463 	struct pci_dev *pdev = adapter->pdev;
2464 	bool dev_up, wd_state;
2465 	int rc;
2466 
2467 	if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2468 		dev_err(&pdev->dev,
2469 			"device reset schedule while reset bit is off\n");
2470 		return;
2471 	}
2472 
2473 	netif_carrier_off(netdev);
2474 
2475 	del_timer_sync(&adapter->timer_service);
2476 
2477 	rtnl_lock();
2478 
2479 	dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2480 	ena_com_set_admin_running_state(ena_dev, false);
2481 
2482 	/* After calling ena_close the tx queues and the napi
2483 	 * are disabled so no one can interfere or touch the
2484 	 * data structures
2485 	 */
2486 	ena_close(netdev);
2487 
2488 	ena_free_mgmnt_irq(adapter);
2489 
2490 	pci_free_irq_vectors(adapter->pdev);
2491 
2492 	ena_com_abort_admin_commands(ena_dev);
2493 
2494 	ena_com_wait_for_abort_completion(ena_dev);
2495 
2496 	ena_com_admin_destroy(ena_dev);
2497 
2498 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2499 
2500 	clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2501 
2502 	/* Finish with the destroy part. Start the init part */
2503 
2504 	rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
2505 	if (rc) {
2506 		dev_err(&pdev->dev, "Can not initialize device\n");
2507 		goto err;
2508 	}
2509 	adapter->wd_state = wd_state;
2510 
2511 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
2512 	if (rc) {
2513 		dev_err(&pdev->dev, "Validation of device parameters failed\n");
2514 		goto err_device_destroy;
2515 	}
2516 
2517 	rc = ena_enable_msix_and_set_admin_interrupts(adapter,
2518 						      adapter->num_queues);
2519 	if (rc) {
2520 		dev_err(&pdev->dev, "Enable MSI-X failed\n");
2521 		goto err_device_destroy;
2522 	}
2523 	/* If the interface was up before the reset bring it up */
2524 	if (dev_up) {
2525 		rc = ena_up(adapter);
2526 		if (rc) {
2527 			dev_err(&pdev->dev, "Failed to create I/O queues\n");
2528 			goto err_disable_msix;
2529 		}
2530 	}
2531 
2532 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
2533 
2534 	rtnl_unlock();
2535 
2536 	dev_err(&pdev->dev, "Device reset completed successfully\n");
2537 
2538 	return;
2539 err_disable_msix:
2540 	ena_free_mgmnt_irq(adapter);
2541 	pci_free_irq_vectors(adapter->pdev);
2542 err_device_destroy:
2543 	ena_com_admin_destroy(ena_dev);
2544 err:
2545 	rtnl_unlock();
2546 
2547 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2548 
2549 	dev_err(&pdev->dev,
2550 		"Reset attempt failed. Can not reset the device\n");
2551 }
2552 
2553 static void check_for_missing_tx_completions(struct ena_adapter *adapter)
2554 {
2555 	struct ena_tx_buffer *tx_buf;
2556 	unsigned long last_jiffies;
2557 	struct ena_ring *tx_ring;
2558 	int i, j, budget;
2559 	u32 missed_tx;
2560 
2561 	/* Make sure the driver doesn't turn the device in other process */
2562 	smp_rmb();
2563 
2564 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2565 		return;
2566 
2567 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2568 		return;
2569 
2570 	budget = ENA_MONITORED_TX_QUEUES;
2571 
2572 	for (i = adapter->last_monitored_tx_qid; i < adapter->num_queues; i++) {
2573 		tx_ring = &adapter->tx_ring[i];
2574 
2575 		for (j = 0; j < tx_ring->ring_size; j++) {
2576 			tx_buf = &tx_ring->tx_buffer_info[j];
2577 			last_jiffies = tx_buf->last_jiffies;
2578 			if (unlikely(last_jiffies && time_is_before_jiffies(last_jiffies + TX_TIMEOUT))) {
2579 				netif_notice(adapter, tx_err, adapter->netdev,
2580 					     "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2581 					     tx_ring->qid, j);
2582 
2583 				u64_stats_update_begin(&tx_ring->syncp);
2584 				missed_tx = tx_ring->tx_stats.missing_tx_comp++;
2585 				u64_stats_update_end(&tx_ring->syncp);
2586 
2587 				/* Clear last jiffies so the lost buffer won't
2588 				 * be counted twice.
2589 				 */
2590 				tx_buf->last_jiffies = 0;
2591 
2592 				if (unlikely(missed_tx > MAX_NUM_OF_TIMEOUTED_PACKETS)) {
2593 					netif_err(adapter, tx_err, adapter->netdev,
2594 						  "The number of lost tx completion is above the threshold (%d > %d). Reset the device\n",
2595 						  missed_tx, MAX_NUM_OF_TIMEOUTED_PACKETS);
2596 					set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2597 				}
2598 			}
2599 		}
2600 
2601 		budget--;
2602 		if (!budget)
2603 			break;
2604 	}
2605 
2606 	adapter->last_monitored_tx_qid = i % adapter->num_queues;
2607 }
2608 
2609 /* Check for keep alive expiration */
2610 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
2611 {
2612 	unsigned long keep_alive_expired;
2613 
2614 	if (!adapter->wd_state)
2615 		return;
2616 
2617 	keep_alive_expired = round_jiffies(adapter->last_keep_alive_jiffies
2618 					   + ENA_DEVICE_KALIVE_TIMEOUT);
2619 	if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
2620 		netif_err(adapter, drv, adapter->netdev,
2621 			  "Keep alive watchdog timeout.\n");
2622 		u64_stats_update_begin(&adapter->syncp);
2623 		adapter->dev_stats.wd_expired++;
2624 		u64_stats_update_end(&adapter->syncp);
2625 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2626 	}
2627 }
2628 
2629 static void check_for_admin_com_state(struct ena_adapter *adapter)
2630 {
2631 	if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
2632 		netif_err(adapter, drv, adapter->netdev,
2633 			  "ENA admin queue is not in running state!\n");
2634 		u64_stats_update_begin(&adapter->syncp);
2635 		adapter->dev_stats.admin_q_pause++;
2636 		u64_stats_update_end(&adapter->syncp);
2637 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2638 	}
2639 }
2640 
2641 static void ena_update_host_info(struct ena_admin_host_info *host_info,
2642 				 struct net_device *netdev)
2643 {
2644 	host_info->supported_network_features[0] =
2645 		netdev->features & GENMASK_ULL(31, 0);
2646 	host_info->supported_network_features[1] =
2647 		(netdev->features & GENMASK_ULL(63, 32)) >> 32;
2648 }
2649 
2650 static void ena_timer_service(unsigned long data)
2651 {
2652 	struct ena_adapter *adapter = (struct ena_adapter *)data;
2653 	u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
2654 	struct ena_admin_host_info *host_info =
2655 		adapter->ena_dev->host_attr.host_info;
2656 
2657 	check_for_missing_keep_alive(adapter);
2658 
2659 	check_for_admin_com_state(adapter);
2660 
2661 	check_for_missing_tx_completions(adapter);
2662 
2663 	if (debug_area)
2664 		ena_dump_stats_to_buf(adapter, debug_area);
2665 
2666 	if (host_info)
2667 		ena_update_host_info(host_info, adapter->netdev);
2668 
2669 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2670 		netif_err(adapter, drv, adapter->netdev,
2671 			  "Trigger reset is on\n");
2672 		ena_dump_stats_to_dmesg(adapter);
2673 		queue_work(ena_wq, &adapter->reset_task);
2674 		return;
2675 	}
2676 
2677 	/* Reset the timer */
2678 	mod_timer(&adapter->timer_service, jiffies + HZ);
2679 }
2680 
2681 static int ena_calc_io_queue_num(struct pci_dev *pdev,
2682 				 struct ena_com_dev *ena_dev,
2683 				 struct ena_com_dev_get_features_ctx *get_feat_ctx)
2684 {
2685 	int io_sq_num, io_queue_num;
2686 
2687 	/* In case of LLQ use the llq number in the get feature cmd */
2688 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2689 		io_sq_num = get_feat_ctx->max_queues.max_llq_num;
2690 
2691 		if (io_sq_num == 0) {
2692 			dev_err(&pdev->dev,
2693 				"Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2694 
2695 			ena_dev->tx_mem_queue_type =
2696 				ENA_ADMIN_PLACEMENT_POLICY_HOST;
2697 			io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2698 		}
2699 	} else {
2700 		io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2701 	}
2702 
2703 	io_queue_num = min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
2704 	io_queue_num = min_t(int, io_queue_num, io_sq_num);
2705 	io_queue_num = min_t(int, io_queue_num,
2706 			     get_feat_ctx->max_queues.max_cq_num);
2707 	/* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2708 	io_queue_num = min_t(int, io_queue_num, pci_msix_vec_count(pdev) - 1);
2709 	if (unlikely(!io_queue_num)) {
2710 		dev_err(&pdev->dev, "The device doesn't have io queues\n");
2711 		return -EFAULT;
2712 	}
2713 
2714 	return io_queue_num;
2715 }
2716 
2717 static void ena_set_push_mode(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
2718 			      struct ena_com_dev_get_features_ctx *get_feat_ctx)
2719 {
2720 	bool has_mem_bar;
2721 
2722 	has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
2723 
2724 	/* Enable push mode if device supports LLQ */
2725 	if (has_mem_bar && (get_feat_ctx->max_queues.max_llq_num > 0))
2726 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
2727 	else
2728 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2729 }
2730 
2731 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
2732 				 struct net_device *netdev)
2733 {
2734 	netdev_features_t dev_features = 0;
2735 
2736 	/* Set offload features */
2737 	if (feat->offload.tx &
2738 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
2739 		dev_features |= NETIF_F_IP_CSUM;
2740 
2741 	if (feat->offload.tx &
2742 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
2743 		dev_features |= NETIF_F_IPV6_CSUM;
2744 
2745 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
2746 		dev_features |= NETIF_F_TSO;
2747 
2748 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
2749 		dev_features |= NETIF_F_TSO6;
2750 
2751 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
2752 		dev_features |= NETIF_F_TSO_ECN;
2753 
2754 	if (feat->offload.rx_supported &
2755 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
2756 		dev_features |= NETIF_F_RXCSUM;
2757 
2758 	if (feat->offload.rx_supported &
2759 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
2760 		dev_features |= NETIF_F_RXCSUM;
2761 
2762 	netdev->features =
2763 		dev_features |
2764 		NETIF_F_SG |
2765 		NETIF_F_RXHASH |
2766 		NETIF_F_HIGHDMA;
2767 
2768 	netdev->hw_features |= netdev->features;
2769 	netdev->vlan_features |= netdev->features;
2770 }
2771 
2772 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
2773 				     struct ena_com_dev_get_features_ctx *feat)
2774 {
2775 	struct net_device *netdev = adapter->netdev;
2776 
2777 	/* Copy mac address */
2778 	if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
2779 		eth_hw_addr_random(netdev);
2780 		ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2781 	} else {
2782 		ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
2783 		ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
2784 	}
2785 
2786 	/* Set offload features */
2787 	ena_set_dev_offloads(feat, netdev);
2788 
2789 	adapter->max_mtu = feat->dev_attr.max_mtu;
2790 	netdev->max_mtu = adapter->max_mtu;
2791 	netdev->min_mtu = ENA_MIN_MTU;
2792 }
2793 
2794 static int ena_rss_init_default(struct ena_adapter *adapter)
2795 {
2796 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2797 	struct device *dev = &adapter->pdev->dev;
2798 	int rc, i;
2799 	u32 val;
2800 
2801 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
2802 	if (unlikely(rc)) {
2803 		dev_err(dev, "Cannot init indirect table\n");
2804 		goto err_rss_init;
2805 	}
2806 
2807 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
2808 		val = ethtool_rxfh_indir_default(i, adapter->num_queues);
2809 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
2810 						       ENA_IO_RXQ_IDX(val));
2811 		if (unlikely(rc && (rc != -EPERM))) {
2812 			dev_err(dev, "Cannot fill indirect table\n");
2813 			goto err_fill_indir;
2814 		}
2815 	}
2816 
2817 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
2818 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
2819 	if (unlikely(rc && (rc != -EPERM))) {
2820 		dev_err(dev, "Cannot fill hash function\n");
2821 		goto err_fill_indir;
2822 	}
2823 
2824 	rc = ena_com_set_default_hash_ctrl(ena_dev);
2825 	if (unlikely(rc && (rc != -EPERM))) {
2826 		dev_err(dev, "Cannot fill hash control\n");
2827 		goto err_fill_indir;
2828 	}
2829 
2830 	return 0;
2831 
2832 err_fill_indir:
2833 	ena_com_rss_destroy(ena_dev);
2834 err_rss_init:
2835 
2836 	return rc;
2837 }
2838 
2839 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
2840 {
2841 	int release_bars;
2842 
2843 	release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
2844 	pci_release_selected_regions(pdev, release_bars);
2845 }
2846 
2847 static int ena_calc_queue_size(struct pci_dev *pdev,
2848 			       struct ena_com_dev *ena_dev,
2849 			       u16 *max_tx_sgl_size,
2850 			       u16 *max_rx_sgl_size,
2851 			       struct ena_com_dev_get_features_ctx *get_feat_ctx)
2852 {
2853 	u32 queue_size = ENA_DEFAULT_RING_SIZE;
2854 
2855 	queue_size = min_t(u32, queue_size,
2856 			   get_feat_ctx->max_queues.max_cq_depth);
2857 	queue_size = min_t(u32, queue_size,
2858 			   get_feat_ctx->max_queues.max_sq_depth);
2859 
2860 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
2861 		queue_size = min_t(u32, queue_size,
2862 				   get_feat_ctx->max_queues.max_llq_depth);
2863 
2864 	queue_size = rounddown_pow_of_two(queue_size);
2865 
2866 	if (unlikely(!queue_size)) {
2867 		dev_err(&pdev->dev, "Invalid queue size\n");
2868 		return -EFAULT;
2869 	}
2870 
2871 	*max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
2872 				 get_feat_ctx->max_queues.max_packet_tx_descs);
2873 	*max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
2874 				 get_feat_ctx->max_queues.max_packet_rx_descs);
2875 
2876 	return queue_size;
2877 }
2878 
2879 /* ena_probe - Device Initialization Routine
2880  * @pdev: PCI device information struct
2881  * @ent: entry in ena_pci_tbl
2882  *
2883  * Returns 0 on success, negative on failure
2884  *
2885  * ena_probe initializes an adapter identified by a pci_dev structure.
2886  * The OS initialization, configuring of the adapter private structure,
2887  * and a hardware reset occur.
2888  */
2889 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2890 {
2891 	struct ena_com_dev_get_features_ctx get_feat_ctx;
2892 	static int version_printed;
2893 	struct net_device *netdev;
2894 	struct ena_adapter *adapter;
2895 	struct ena_com_dev *ena_dev = NULL;
2896 	static int adapters_found;
2897 	int io_queue_num, bars, rc;
2898 	int queue_size;
2899 	u16 tx_sgl_size = 0;
2900 	u16 rx_sgl_size = 0;
2901 	bool wd_state;
2902 
2903 	dev_dbg(&pdev->dev, "%s\n", __func__);
2904 
2905 	if (version_printed++ == 0)
2906 		dev_info(&pdev->dev, "%s", version);
2907 
2908 	rc = pci_enable_device_mem(pdev);
2909 	if (rc) {
2910 		dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
2911 		return rc;
2912 	}
2913 
2914 	pci_set_master(pdev);
2915 
2916 	ena_dev = vzalloc(sizeof(*ena_dev));
2917 	if (!ena_dev) {
2918 		rc = -ENOMEM;
2919 		goto err_disable_device;
2920 	}
2921 
2922 	bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
2923 	rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
2924 	if (rc) {
2925 		dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
2926 			rc);
2927 		goto err_free_ena_dev;
2928 	}
2929 
2930 	ena_dev->reg_bar = ioremap(pci_resource_start(pdev, ENA_REG_BAR),
2931 				   pci_resource_len(pdev, ENA_REG_BAR));
2932 	if (!ena_dev->reg_bar) {
2933 		dev_err(&pdev->dev, "failed to remap regs bar\n");
2934 		rc = -EFAULT;
2935 		goto err_free_region;
2936 	}
2937 
2938 	ena_dev->dmadev = &pdev->dev;
2939 
2940 	rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
2941 	if (rc) {
2942 		dev_err(&pdev->dev, "ena device init failed\n");
2943 		if (rc == -ETIME)
2944 			rc = -EPROBE_DEFER;
2945 		goto err_free_region;
2946 	}
2947 
2948 	ena_set_push_mode(pdev, ena_dev, &get_feat_ctx);
2949 
2950 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2951 		ena_dev->mem_bar = ioremap_wc(pci_resource_start(pdev, ENA_MEM_BAR),
2952 					      pci_resource_len(pdev, ENA_MEM_BAR));
2953 		if (!ena_dev->mem_bar) {
2954 			rc = -EFAULT;
2955 			goto err_device_destroy;
2956 		}
2957 	}
2958 
2959 	/* initial Tx interrupt delay, Assumes 1 usec granularity.
2960 	* Updated during device initialization with the real granularity
2961 	*/
2962 	ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
2963 	io_queue_num = ena_calc_io_queue_num(pdev, ena_dev, &get_feat_ctx);
2964 	queue_size = ena_calc_queue_size(pdev, ena_dev, &tx_sgl_size,
2965 					 &rx_sgl_size, &get_feat_ctx);
2966 	if ((queue_size <= 0) || (io_queue_num <= 0)) {
2967 		rc = -EFAULT;
2968 		goto err_device_destroy;
2969 	}
2970 
2971 	dev_info(&pdev->dev, "creating %d io queues. queue size: %d\n",
2972 		 io_queue_num, queue_size);
2973 
2974 	/* dev zeroed in init_etherdev */
2975 	netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), io_queue_num);
2976 	if (!netdev) {
2977 		dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
2978 		rc = -ENOMEM;
2979 		goto err_device_destroy;
2980 	}
2981 
2982 	SET_NETDEV_DEV(netdev, &pdev->dev);
2983 
2984 	adapter = netdev_priv(netdev);
2985 	pci_set_drvdata(pdev, adapter);
2986 
2987 	adapter->ena_dev = ena_dev;
2988 	adapter->netdev = netdev;
2989 	adapter->pdev = pdev;
2990 
2991 	ena_set_conf_feat_params(adapter, &get_feat_ctx);
2992 
2993 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2994 
2995 	adapter->tx_ring_size = queue_size;
2996 	adapter->rx_ring_size = queue_size;
2997 
2998 	adapter->max_tx_sgl_size = tx_sgl_size;
2999 	adapter->max_rx_sgl_size = rx_sgl_size;
3000 
3001 	adapter->num_queues = io_queue_num;
3002 	adapter->last_monitored_tx_qid = 0;
3003 
3004 	adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
3005 	adapter->wd_state = wd_state;
3006 
3007 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
3008 
3009 	rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
3010 	if (rc) {
3011 		dev_err(&pdev->dev,
3012 			"Failed to query interrupt moderation feature\n");
3013 		goto err_netdev_destroy;
3014 	}
3015 	ena_init_io_rings(adapter);
3016 
3017 	netdev->netdev_ops = &ena_netdev_ops;
3018 	netdev->watchdog_timeo = TX_TIMEOUT;
3019 	ena_set_ethtool_ops(netdev);
3020 
3021 	netdev->priv_flags |= IFF_UNICAST_FLT;
3022 
3023 	u64_stats_init(&adapter->syncp);
3024 
3025 	rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3026 	if (rc) {
3027 		dev_err(&pdev->dev,
3028 			"Failed to enable and set the admin interrupts\n");
3029 		goto err_worker_destroy;
3030 	}
3031 	rc = ena_rss_init_default(adapter);
3032 	if (rc && (rc != -EPERM)) {
3033 		dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
3034 		goto err_free_msix;
3035 	}
3036 
3037 	ena_config_debug_area(adapter);
3038 
3039 	memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
3040 
3041 	netif_carrier_off(netdev);
3042 
3043 	rc = register_netdev(netdev);
3044 	if (rc) {
3045 		dev_err(&pdev->dev, "Cannot register net device\n");
3046 		goto err_rss;
3047 	}
3048 
3049 	INIT_WORK(&adapter->suspend_io_task, ena_device_io_suspend);
3050 	INIT_WORK(&adapter->resume_io_task, ena_device_io_resume);
3051 	INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
3052 
3053 	adapter->last_keep_alive_jiffies = jiffies;
3054 
3055 	setup_timer(&adapter->timer_service, ena_timer_service,
3056 		    (unsigned long)adapter);
3057 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3058 
3059 	dev_info(&pdev->dev, "%s found at mem %lx, mac addr %pM Queues %d\n",
3060 		 DEVICE_NAME, (long)pci_resource_start(pdev, 0),
3061 		 netdev->dev_addr, io_queue_num);
3062 
3063 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3064 
3065 	adapters_found++;
3066 
3067 	return 0;
3068 
3069 err_rss:
3070 	ena_com_delete_debug_area(ena_dev);
3071 	ena_com_rss_destroy(ena_dev);
3072 err_free_msix:
3073 	ena_com_dev_reset(ena_dev);
3074 	ena_free_mgmnt_irq(adapter);
3075 	pci_free_irq_vectors(adapter->pdev);
3076 err_worker_destroy:
3077 	ena_com_destroy_interrupt_moderation(ena_dev);
3078 	del_timer(&adapter->timer_service);
3079 	cancel_work_sync(&adapter->suspend_io_task);
3080 	cancel_work_sync(&adapter->resume_io_task);
3081 err_netdev_destroy:
3082 	free_netdev(netdev);
3083 err_device_destroy:
3084 	ena_com_delete_host_info(ena_dev);
3085 	ena_com_admin_destroy(ena_dev);
3086 err_free_region:
3087 	ena_release_bars(ena_dev, pdev);
3088 err_free_ena_dev:
3089 	vfree(ena_dev);
3090 err_disable_device:
3091 	pci_disable_device(pdev);
3092 	return rc;
3093 }
3094 
3095 /*****************************************************************************/
3096 static int ena_sriov_configure(struct pci_dev *dev, int numvfs)
3097 {
3098 	int rc;
3099 
3100 	if (numvfs > 0) {
3101 		rc = pci_enable_sriov(dev, numvfs);
3102 		if (rc != 0) {
3103 			dev_err(&dev->dev,
3104 				"pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
3105 				numvfs, rc);
3106 			return rc;
3107 		}
3108 
3109 		return numvfs;
3110 	}
3111 
3112 	if (numvfs == 0) {
3113 		pci_disable_sriov(dev);
3114 		return 0;
3115 	}
3116 
3117 	return -EINVAL;
3118 }
3119 
3120 /*****************************************************************************/
3121 /*****************************************************************************/
3122 
3123 /* ena_remove - Device Removal Routine
3124  * @pdev: PCI device information struct
3125  *
3126  * ena_remove is called by the PCI subsystem to alert the driver
3127  * that it should release a PCI device.
3128  */
3129 static void ena_remove(struct pci_dev *pdev)
3130 {
3131 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
3132 	struct ena_com_dev *ena_dev;
3133 	struct net_device *netdev;
3134 
3135 	ena_dev = adapter->ena_dev;
3136 	netdev = adapter->netdev;
3137 
3138 #ifdef CONFIG_RFS_ACCEL
3139 	if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
3140 		free_irq_cpu_rmap(netdev->rx_cpu_rmap);
3141 		netdev->rx_cpu_rmap = NULL;
3142 	}
3143 #endif /* CONFIG_RFS_ACCEL */
3144 
3145 	unregister_netdev(netdev);
3146 	del_timer_sync(&adapter->timer_service);
3147 
3148 	cancel_work_sync(&adapter->reset_task);
3149 
3150 	cancel_work_sync(&adapter->suspend_io_task);
3151 
3152 	cancel_work_sync(&adapter->resume_io_task);
3153 
3154 	/* Reset the device only if the device is running. */
3155 	if (test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3156 		ena_com_dev_reset(ena_dev);
3157 
3158 	ena_free_mgmnt_irq(adapter);
3159 
3160 	pci_free_irq_vectors(adapter->pdev);
3161 
3162 	free_netdev(netdev);
3163 
3164 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3165 
3166 	ena_com_abort_admin_commands(ena_dev);
3167 
3168 	ena_com_wait_for_abort_completion(ena_dev);
3169 
3170 	ena_com_admin_destroy(ena_dev);
3171 
3172 	ena_com_rss_destroy(ena_dev);
3173 
3174 	ena_com_delete_debug_area(ena_dev);
3175 
3176 	ena_com_delete_host_info(ena_dev);
3177 
3178 	ena_release_bars(ena_dev, pdev);
3179 
3180 	pci_disable_device(pdev);
3181 
3182 	ena_com_destroy_interrupt_moderation(ena_dev);
3183 
3184 	vfree(ena_dev);
3185 }
3186 
3187 static struct pci_driver ena_pci_driver = {
3188 	.name		= DRV_MODULE_NAME,
3189 	.id_table	= ena_pci_tbl,
3190 	.probe		= ena_probe,
3191 	.remove		= ena_remove,
3192 	.sriov_configure = ena_sriov_configure,
3193 };
3194 
3195 static int __init ena_init(void)
3196 {
3197 	pr_info("%s", version);
3198 
3199 	ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
3200 	if (!ena_wq) {
3201 		pr_err("Failed to create workqueue\n");
3202 		return -ENOMEM;
3203 	}
3204 
3205 	return pci_register_driver(&ena_pci_driver);
3206 }
3207 
3208 static void __exit ena_cleanup(void)
3209 {
3210 	pci_unregister_driver(&ena_pci_driver);
3211 
3212 	if (ena_wq) {
3213 		destroy_workqueue(ena_wq);
3214 		ena_wq = NULL;
3215 	}
3216 }
3217 
3218 /******************************************************************************
3219  ******************************** AENQ Handlers *******************************
3220  *****************************************************************************/
3221 /* ena_update_on_link_change:
3222  * Notify the network interface about the change in link status
3223  */
3224 static void ena_update_on_link_change(void *adapter_data,
3225 				      struct ena_admin_aenq_entry *aenq_e)
3226 {
3227 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3228 	struct ena_admin_aenq_link_change_desc *aenq_desc =
3229 		(struct ena_admin_aenq_link_change_desc *)aenq_e;
3230 	int status = aenq_desc->flags &
3231 		ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3232 
3233 	if (status) {
3234 		netdev_dbg(adapter->netdev, "%s\n", __func__);
3235 		set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3236 		netif_carrier_on(adapter->netdev);
3237 	} else {
3238 		clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3239 		netif_carrier_off(adapter->netdev);
3240 	}
3241 }
3242 
3243 static void ena_keep_alive_wd(void *adapter_data,
3244 			      struct ena_admin_aenq_entry *aenq_e)
3245 {
3246 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3247 
3248 	adapter->last_keep_alive_jiffies = jiffies;
3249 }
3250 
3251 static void ena_notification(void *adapter_data,
3252 			     struct ena_admin_aenq_entry *aenq_e)
3253 {
3254 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3255 
3256 	WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3257 	     "Invalid group(%x) expected %x\n",
3258 	     aenq_e->aenq_common_desc.group,
3259 	     ENA_ADMIN_NOTIFICATION);
3260 
3261 	switch (aenq_e->aenq_common_desc.syndrom) {
3262 	case ENA_ADMIN_SUSPEND:
3263 		/* Suspend just the IO queues.
3264 		 * We deliberately don't suspend admin so the timer and
3265 		 * the keep_alive events should remain.
3266 		 */
3267 		queue_work(ena_wq, &adapter->suspend_io_task);
3268 		break;
3269 	case ENA_ADMIN_RESUME:
3270 		queue_work(ena_wq, &adapter->resume_io_task);
3271 		break;
3272 	default:
3273 		netif_err(adapter, drv, adapter->netdev,
3274 			  "Invalid aenq notification link state %d\n",
3275 			  aenq_e->aenq_common_desc.syndrom);
3276 	}
3277 }
3278 
3279 /* This handler will called for unknown event group or unimplemented handlers*/
3280 static void unimplemented_aenq_handler(void *data,
3281 				       struct ena_admin_aenq_entry *aenq_e)
3282 {
3283 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3284 
3285 	netif_err(adapter, drv, adapter->netdev,
3286 		  "Unknown event was received or event with unimplemented handler\n");
3287 }
3288 
3289 static struct ena_aenq_handlers aenq_handlers = {
3290 	.handlers = {
3291 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3292 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
3293 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3294 	},
3295 	.unimplemented_handler = unimplemented_aenq_handler
3296 };
3297 
3298 module_init(ena_init);
3299 module_exit(ena_cleanup);
3300