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/numa.h>
43 #include <linux/pci.h>
44 #include <linux/utsname.h>
45 #include <linux/version.h>
46 #include <linux/vmalloc.h>
47 #include <net/ip.h>
48 
49 #include "ena_netdev.h"
50 #include "ena_pci_id_tbl.h"
51 
52 static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
53 
54 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
55 MODULE_DESCRIPTION(DEVICE_NAME);
56 MODULE_LICENSE("GPL");
57 MODULE_VERSION(DRV_MODULE_VERSION);
58 
59 /* Time in jiffies before concluding the transmitter is hung. */
60 #define TX_TIMEOUT  (5 * HZ)
61 
62 #define ENA_NAPI_BUDGET 64
63 
64 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
65 		NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
66 static int debug = -1;
67 module_param(debug, int, 0);
68 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
69 
70 static struct ena_aenq_handlers aenq_handlers;
71 
72 static struct workqueue_struct *ena_wq;
73 
74 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
75 
76 static int ena_rss_init_default(struct ena_adapter *adapter);
77 static void check_for_admin_com_state(struct ena_adapter *adapter);
78 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
79 static int ena_restore_device(struct ena_adapter *adapter);
80 
81 static void ena_tx_timeout(struct net_device *dev)
82 {
83 	struct ena_adapter *adapter = netdev_priv(dev);
84 
85 	/* Change the state of the device to trigger reset
86 	 * Check that we are not in the middle or a trigger already
87 	 */
88 
89 	if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
90 		return;
91 
92 	adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
93 	u64_stats_update_begin(&adapter->syncp);
94 	adapter->dev_stats.tx_timeout++;
95 	u64_stats_update_end(&adapter->syncp);
96 
97 	netif_err(adapter, tx_err, dev, "Transmit time out\n");
98 }
99 
100 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
101 {
102 	int i;
103 
104 	for (i = 0; i < adapter->num_queues; i++)
105 		adapter->rx_ring[i].mtu = mtu;
106 }
107 
108 static int ena_change_mtu(struct net_device *dev, int new_mtu)
109 {
110 	struct ena_adapter *adapter = netdev_priv(dev);
111 	int ret;
112 
113 	ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
114 	if (!ret) {
115 		netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
116 		update_rx_ring_mtu(adapter, new_mtu);
117 		dev->mtu = new_mtu;
118 	} else {
119 		netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
120 			  new_mtu);
121 	}
122 
123 	return ret;
124 }
125 
126 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
127 {
128 #ifdef CONFIG_RFS_ACCEL
129 	u32 i;
130 	int rc;
131 
132 	adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_queues);
133 	if (!adapter->netdev->rx_cpu_rmap)
134 		return -ENOMEM;
135 	for (i = 0; i < adapter->num_queues; i++) {
136 		int irq_idx = ENA_IO_IRQ_IDX(i);
137 
138 		rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
139 				      pci_irq_vector(adapter->pdev, irq_idx));
140 		if (rc) {
141 			free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
142 			adapter->netdev->rx_cpu_rmap = NULL;
143 			return rc;
144 		}
145 	}
146 #endif /* CONFIG_RFS_ACCEL */
147 	return 0;
148 }
149 
150 static void ena_init_io_rings_common(struct ena_adapter *adapter,
151 				     struct ena_ring *ring, u16 qid)
152 {
153 	ring->qid = qid;
154 	ring->pdev = adapter->pdev;
155 	ring->dev = &adapter->pdev->dev;
156 	ring->netdev = adapter->netdev;
157 	ring->napi = &adapter->ena_napi[qid].napi;
158 	ring->adapter = adapter;
159 	ring->ena_dev = adapter->ena_dev;
160 	ring->per_napi_packets = 0;
161 	ring->per_napi_bytes = 0;
162 	ring->cpu = 0;
163 	ring->first_interrupt = false;
164 	ring->no_interrupt_event_cnt = 0;
165 	u64_stats_init(&ring->syncp);
166 }
167 
168 static void ena_init_io_rings(struct ena_adapter *adapter)
169 {
170 	struct ena_com_dev *ena_dev;
171 	struct ena_ring *txr, *rxr;
172 	int i;
173 
174 	ena_dev = adapter->ena_dev;
175 
176 	for (i = 0; i < adapter->num_queues; i++) {
177 		txr = &adapter->tx_ring[i];
178 		rxr = &adapter->rx_ring[i];
179 
180 		/* TX/RX common ring state */
181 		ena_init_io_rings_common(adapter, txr, i);
182 		ena_init_io_rings_common(adapter, rxr, i);
183 
184 		/* TX specific ring state */
185 		txr->ring_size = adapter->requested_tx_ring_size;
186 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
187 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
188 		txr->sgl_size = adapter->max_tx_sgl_size;
189 		txr->smoothed_interval =
190 			ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
191 
192 		/* RX specific ring state */
193 		rxr->ring_size = adapter->requested_rx_ring_size;
194 		rxr->rx_copybreak = adapter->rx_copybreak;
195 		rxr->sgl_size = adapter->max_rx_sgl_size;
196 		rxr->smoothed_interval =
197 			ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
198 		rxr->empty_rx_queue = 0;
199 	}
200 }
201 
202 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
203  * @adapter: network interface device structure
204  * @qid: queue index
205  *
206  * Return 0 on success, negative on failure
207  */
208 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
209 {
210 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
211 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
212 	int size, i, node;
213 
214 	if (tx_ring->tx_buffer_info) {
215 		netif_err(adapter, ifup,
216 			  adapter->netdev, "tx_buffer_info info is not NULL");
217 		return -EEXIST;
218 	}
219 
220 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
221 	node = cpu_to_node(ena_irq->cpu);
222 
223 	tx_ring->tx_buffer_info = vzalloc_node(size, node);
224 	if (!tx_ring->tx_buffer_info) {
225 		tx_ring->tx_buffer_info = vzalloc(size);
226 		if (!tx_ring->tx_buffer_info)
227 			goto err_tx_buffer_info;
228 	}
229 
230 	size = sizeof(u16) * tx_ring->ring_size;
231 	tx_ring->free_ids = vzalloc_node(size, node);
232 	if (!tx_ring->free_ids) {
233 		tx_ring->free_ids = vzalloc(size);
234 		if (!tx_ring->free_ids)
235 			goto err_tx_free_ids;
236 	}
237 
238 	size = tx_ring->tx_max_header_size;
239 	tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
240 	if (!tx_ring->push_buf_intermediate_buf) {
241 		tx_ring->push_buf_intermediate_buf = vzalloc(size);
242 		if (!tx_ring->push_buf_intermediate_buf)
243 			goto err_push_buf_intermediate_buf;
244 	}
245 
246 	/* Req id ring for TX out of order completions */
247 	for (i = 0; i < tx_ring->ring_size; i++)
248 		tx_ring->free_ids[i] = i;
249 
250 	/* Reset tx statistics */
251 	memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
252 
253 	tx_ring->next_to_use = 0;
254 	tx_ring->next_to_clean = 0;
255 	tx_ring->cpu = ena_irq->cpu;
256 	return 0;
257 
258 err_push_buf_intermediate_buf:
259 	vfree(tx_ring->free_ids);
260 	tx_ring->free_ids = NULL;
261 err_tx_free_ids:
262 	vfree(tx_ring->tx_buffer_info);
263 	tx_ring->tx_buffer_info = NULL;
264 err_tx_buffer_info:
265 	return -ENOMEM;
266 }
267 
268 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
269  * @adapter: network interface device structure
270  * @qid: queue index
271  *
272  * Free all transmit software resources
273  */
274 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
275 {
276 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
277 
278 	vfree(tx_ring->tx_buffer_info);
279 	tx_ring->tx_buffer_info = NULL;
280 
281 	vfree(tx_ring->free_ids);
282 	tx_ring->free_ids = NULL;
283 
284 	vfree(tx_ring->push_buf_intermediate_buf);
285 	tx_ring->push_buf_intermediate_buf = NULL;
286 }
287 
288 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
289  * @adapter: private structure
290  *
291  * Return 0 on success, negative on failure
292  */
293 static int ena_setup_all_tx_resources(struct ena_adapter *adapter)
294 {
295 	int i, rc = 0;
296 
297 	for (i = 0; i < adapter->num_queues; i++) {
298 		rc = ena_setup_tx_resources(adapter, i);
299 		if (rc)
300 			goto err_setup_tx;
301 	}
302 
303 	return 0;
304 
305 err_setup_tx:
306 
307 	netif_err(adapter, ifup, adapter->netdev,
308 		  "Tx queue %d: allocation failed\n", i);
309 
310 	/* rewind the index freeing the rings as we go */
311 	while (i--)
312 		ena_free_tx_resources(adapter, i);
313 	return rc;
314 }
315 
316 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
317  * @adapter: board private structure
318  *
319  * Free all transmit software resources
320  */
321 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
322 {
323 	int i;
324 
325 	for (i = 0; i < adapter->num_queues; i++)
326 		ena_free_tx_resources(adapter, i);
327 }
328 
329 static int validate_rx_req_id(struct ena_ring *rx_ring, u16 req_id)
330 {
331 	if (likely(req_id < rx_ring->ring_size))
332 		return 0;
333 
334 	netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
335 		  "Invalid rx req_id: %hu\n", req_id);
336 
337 	u64_stats_update_begin(&rx_ring->syncp);
338 	rx_ring->rx_stats.bad_req_id++;
339 	u64_stats_update_end(&rx_ring->syncp);
340 
341 	/* Trigger device reset */
342 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
343 	set_bit(ENA_FLAG_TRIGGER_RESET, &rx_ring->adapter->flags);
344 	return -EFAULT;
345 }
346 
347 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
348  * @adapter: network interface device structure
349  * @qid: queue index
350  *
351  * Returns 0 on success, negative on failure
352  */
353 static int ena_setup_rx_resources(struct ena_adapter *adapter,
354 				  u32 qid)
355 {
356 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
357 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
358 	int size, node, i;
359 
360 	if (rx_ring->rx_buffer_info) {
361 		netif_err(adapter, ifup, adapter->netdev,
362 			  "rx_buffer_info is not NULL");
363 		return -EEXIST;
364 	}
365 
366 	/* alloc extra element so in rx path
367 	 * we can always prefetch rx_info + 1
368 	 */
369 	size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
370 	node = cpu_to_node(ena_irq->cpu);
371 
372 	rx_ring->rx_buffer_info = vzalloc_node(size, node);
373 	if (!rx_ring->rx_buffer_info) {
374 		rx_ring->rx_buffer_info = vzalloc(size);
375 		if (!rx_ring->rx_buffer_info)
376 			return -ENOMEM;
377 	}
378 
379 	size = sizeof(u16) * rx_ring->ring_size;
380 	rx_ring->free_ids = vzalloc_node(size, node);
381 	if (!rx_ring->free_ids) {
382 		rx_ring->free_ids = vzalloc(size);
383 		if (!rx_ring->free_ids) {
384 			vfree(rx_ring->rx_buffer_info);
385 			rx_ring->rx_buffer_info = NULL;
386 			return -ENOMEM;
387 		}
388 	}
389 
390 	/* Req id ring for receiving RX pkts out of order */
391 	for (i = 0; i < rx_ring->ring_size; i++)
392 		rx_ring->free_ids[i] = i;
393 
394 	/* Reset rx statistics */
395 	memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
396 
397 	rx_ring->next_to_clean = 0;
398 	rx_ring->next_to_use = 0;
399 	rx_ring->cpu = ena_irq->cpu;
400 
401 	return 0;
402 }
403 
404 /* ena_free_rx_resources - Free I/O Rx Resources
405  * @adapter: network interface device structure
406  * @qid: queue index
407  *
408  * Free all receive software resources
409  */
410 static void ena_free_rx_resources(struct ena_adapter *adapter,
411 				  u32 qid)
412 {
413 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
414 
415 	vfree(rx_ring->rx_buffer_info);
416 	rx_ring->rx_buffer_info = NULL;
417 
418 	vfree(rx_ring->free_ids);
419 	rx_ring->free_ids = NULL;
420 }
421 
422 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
423  * @adapter: board private structure
424  *
425  * Return 0 on success, negative on failure
426  */
427 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
428 {
429 	int i, rc = 0;
430 
431 	for (i = 0; i < adapter->num_queues; i++) {
432 		rc = ena_setup_rx_resources(adapter, i);
433 		if (rc)
434 			goto err_setup_rx;
435 	}
436 
437 	return 0;
438 
439 err_setup_rx:
440 
441 	netif_err(adapter, ifup, adapter->netdev,
442 		  "Rx queue %d: allocation failed\n", i);
443 
444 	/* rewind the index freeing the rings as we go */
445 	while (i--)
446 		ena_free_rx_resources(adapter, i);
447 	return rc;
448 }
449 
450 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
451  * @adapter: board private structure
452  *
453  * Free all receive software resources
454  */
455 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
456 {
457 	int i;
458 
459 	for (i = 0; i < adapter->num_queues; i++)
460 		ena_free_rx_resources(adapter, i);
461 }
462 
463 static int ena_alloc_rx_page(struct ena_ring *rx_ring,
464 				    struct ena_rx_buffer *rx_info, gfp_t gfp)
465 {
466 	struct ena_com_buf *ena_buf;
467 	struct page *page;
468 	dma_addr_t dma;
469 
470 	/* if previous allocated page is not used */
471 	if (unlikely(rx_info->page))
472 		return 0;
473 
474 	page = alloc_page(gfp);
475 	if (unlikely(!page)) {
476 		u64_stats_update_begin(&rx_ring->syncp);
477 		rx_ring->rx_stats.page_alloc_fail++;
478 		u64_stats_update_end(&rx_ring->syncp);
479 		return -ENOMEM;
480 	}
481 
482 	dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
483 			   DMA_FROM_DEVICE);
484 	if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
485 		u64_stats_update_begin(&rx_ring->syncp);
486 		rx_ring->rx_stats.dma_mapping_err++;
487 		u64_stats_update_end(&rx_ring->syncp);
488 
489 		__free_page(page);
490 		return -EIO;
491 	}
492 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
493 		  "alloc page %p, rx_info %p\n", page, rx_info);
494 
495 	rx_info->page = page;
496 	rx_info->page_offset = 0;
497 	ena_buf = &rx_info->ena_buf;
498 	ena_buf->paddr = dma;
499 	ena_buf->len = ENA_PAGE_SIZE;
500 
501 	return 0;
502 }
503 
504 static void ena_free_rx_page(struct ena_ring *rx_ring,
505 			     struct ena_rx_buffer *rx_info)
506 {
507 	struct page *page = rx_info->page;
508 	struct ena_com_buf *ena_buf = &rx_info->ena_buf;
509 
510 	if (unlikely(!page)) {
511 		netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
512 			   "Trying to free unallocated buffer\n");
513 		return;
514 	}
515 
516 	dma_unmap_page(rx_ring->dev, ena_buf->paddr, ENA_PAGE_SIZE,
517 		       DMA_FROM_DEVICE);
518 
519 	__free_page(page);
520 	rx_info->page = NULL;
521 }
522 
523 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
524 {
525 	u16 next_to_use, req_id;
526 	u32 i;
527 	int rc;
528 
529 	next_to_use = rx_ring->next_to_use;
530 
531 	for (i = 0; i < num; i++) {
532 		struct ena_rx_buffer *rx_info;
533 
534 		req_id = rx_ring->free_ids[next_to_use];
535 		rc = validate_rx_req_id(rx_ring, req_id);
536 		if (unlikely(rc < 0))
537 			break;
538 
539 		rx_info = &rx_ring->rx_buffer_info[req_id];
540 
541 
542 		rc = ena_alloc_rx_page(rx_ring, rx_info,
543 				       GFP_ATOMIC | __GFP_COMP);
544 		if (unlikely(rc < 0)) {
545 			netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
546 				   "failed to alloc buffer for rx queue %d\n",
547 				   rx_ring->qid);
548 			break;
549 		}
550 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
551 						&rx_info->ena_buf,
552 						req_id);
553 		if (unlikely(rc)) {
554 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
555 				   "failed to add buffer for rx queue %d\n",
556 				   rx_ring->qid);
557 			break;
558 		}
559 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
560 						   rx_ring->ring_size);
561 	}
562 
563 	if (unlikely(i < num)) {
564 		u64_stats_update_begin(&rx_ring->syncp);
565 		rx_ring->rx_stats.refil_partial++;
566 		u64_stats_update_end(&rx_ring->syncp);
567 		netdev_warn(rx_ring->netdev,
568 			    "refilled rx qid %d with only %d buffers (from %d)\n",
569 			    rx_ring->qid, i, num);
570 	}
571 
572 	/* ena_com_write_sq_doorbell issues a wmb() */
573 	if (likely(i))
574 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
575 
576 	rx_ring->next_to_use = next_to_use;
577 
578 	return i;
579 }
580 
581 static void ena_free_rx_bufs(struct ena_adapter *adapter,
582 			     u32 qid)
583 {
584 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
585 	u32 i;
586 
587 	for (i = 0; i < rx_ring->ring_size; i++) {
588 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
589 
590 		if (rx_info->page)
591 			ena_free_rx_page(rx_ring, rx_info);
592 	}
593 }
594 
595 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
596  * @adapter: board private structure
597  */
598 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
599 {
600 	struct ena_ring *rx_ring;
601 	int i, rc, bufs_num;
602 
603 	for (i = 0; i < adapter->num_queues; i++) {
604 		rx_ring = &adapter->rx_ring[i];
605 		bufs_num = rx_ring->ring_size - 1;
606 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
607 
608 		if (unlikely(rc != bufs_num))
609 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
610 				   "refilling Queue %d failed. allocated %d buffers from: %d\n",
611 				   i, rc, bufs_num);
612 	}
613 }
614 
615 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
616 {
617 	int i;
618 
619 	for (i = 0; i < adapter->num_queues; i++)
620 		ena_free_rx_bufs(adapter, i);
621 }
622 
623 static void ena_unmap_tx_skb(struct ena_ring *tx_ring,
624 				    struct ena_tx_buffer *tx_info)
625 {
626 	struct ena_com_buf *ena_buf;
627 	u32 cnt;
628 	int i;
629 
630 	ena_buf = tx_info->bufs;
631 	cnt = tx_info->num_of_bufs;
632 
633 	if (unlikely(!cnt))
634 		return;
635 
636 	if (tx_info->map_linear_data) {
637 		dma_unmap_single(tx_ring->dev,
638 				 dma_unmap_addr(ena_buf, paddr),
639 				 dma_unmap_len(ena_buf, len),
640 				 DMA_TO_DEVICE);
641 		ena_buf++;
642 		cnt--;
643 	}
644 
645 	/* unmap remaining mapped pages */
646 	for (i = 0; i < cnt; i++) {
647 		dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
648 			       dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
649 		ena_buf++;
650 	}
651 }
652 
653 /* ena_free_tx_bufs - Free Tx Buffers per Queue
654  * @tx_ring: TX ring for which buffers be freed
655  */
656 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
657 {
658 	bool print_once = true;
659 	u32 i;
660 
661 	for (i = 0; i < tx_ring->ring_size; i++) {
662 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
663 
664 		if (!tx_info->skb)
665 			continue;
666 
667 		if (print_once) {
668 			netdev_notice(tx_ring->netdev,
669 				      "free uncompleted tx skb qid %d idx 0x%x\n",
670 				      tx_ring->qid, i);
671 			print_once = false;
672 		} else {
673 			netdev_dbg(tx_ring->netdev,
674 				   "free uncompleted tx skb qid %d idx 0x%x\n",
675 				   tx_ring->qid, i);
676 		}
677 
678 		ena_unmap_tx_skb(tx_ring, tx_info);
679 
680 		dev_kfree_skb_any(tx_info->skb);
681 	}
682 	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
683 						  tx_ring->qid));
684 }
685 
686 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
687 {
688 	struct ena_ring *tx_ring;
689 	int i;
690 
691 	for (i = 0; i < adapter->num_queues; i++) {
692 		tx_ring = &adapter->tx_ring[i];
693 		ena_free_tx_bufs(tx_ring);
694 	}
695 }
696 
697 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
698 {
699 	u16 ena_qid;
700 	int i;
701 
702 	for (i = 0; i < adapter->num_queues; i++) {
703 		ena_qid = ENA_IO_TXQ_IDX(i);
704 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
705 	}
706 }
707 
708 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
709 {
710 	u16 ena_qid;
711 	int i;
712 
713 	for (i = 0; i < adapter->num_queues; i++) {
714 		ena_qid = ENA_IO_RXQ_IDX(i);
715 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
716 	}
717 }
718 
719 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
720 {
721 	ena_destroy_all_tx_queues(adapter);
722 	ena_destroy_all_rx_queues(adapter);
723 }
724 
725 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
726 {
727 	struct ena_tx_buffer *tx_info = NULL;
728 
729 	if (likely(req_id < tx_ring->ring_size)) {
730 		tx_info = &tx_ring->tx_buffer_info[req_id];
731 		if (likely(tx_info->skb))
732 			return 0;
733 	}
734 
735 	if (tx_info)
736 		netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
737 			  "tx_info doesn't have valid skb\n");
738 	else
739 		netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
740 			  "Invalid req_id: %hu\n", req_id);
741 
742 	u64_stats_update_begin(&tx_ring->syncp);
743 	tx_ring->tx_stats.bad_req_id++;
744 	u64_stats_update_end(&tx_ring->syncp);
745 
746 	/* Trigger device reset */
747 	tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
748 	set_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags);
749 	return -EFAULT;
750 }
751 
752 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
753 {
754 	struct netdev_queue *txq;
755 	bool above_thresh;
756 	u32 tx_bytes = 0;
757 	u32 total_done = 0;
758 	u16 next_to_clean;
759 	u16 req_id;
760 	int tx_pkts = 0;
761 	int rc;
762 
763 	next_to_clean = tx_ring->next_to_clean;
764 	txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
765 
766 	while (tx_pkts < budget) {
767 		struct ena_tx_buffer *tx_info;
768 		struct sk_buff *skb;
769 
770 		rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
771 						&req_id);
772 		if (rc)
773 			break;
774 
775 		rc = validate_tx_req_id(tx_ring, req_id);
776 		if (rc)
777 			break;
778 
779 		tx_info = &tx_ring->tx_buffer_info[req_id];
780 		skb = tx_info->skb;
781 
782 		/* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
783 		prefetch(&skb->end);
784 
785 		tx_info->skb = NULL;
786 		tx_info->last_jiffies = 0;
787 
788 		ena_unmap_tx_skb(tx_ring, tx_info);
789 
790 		netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
791 			  "tx_poll: q %d skb %p completed\n", tx_ring->qid,
792 			  skb);
793 
794 		tx_bytes += skb->len;
795 		dev_kfree_skb(skb);
796 		tx_pkts++;
797 		total_done += tx_info->tx_descs;
798 
799 		tx_ring->free_ids[next_to_clean] = req_id;
800 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
801 						     tx_ring->ring_size);
802 	}
803 
804 	tx_ring->next_to_clean = next_to_clean;
805 	ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
806 	ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
807 
808 	netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
809 
810 	netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
811 		  "tx_poll: q %d done. total pkts: %d\n",
812 		  tx_ring->qid, tx_pkts);
813 
814 	/* need to make the rings circular update visible to
815 	 * ena_start_xmit() before checking for netif_queue_stopped().
816 	 */
817 	smp_mb();
818 
819 	above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
820 						    ENA_TX_WAKEUP_THRESH);
821 	if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
822 		__netif_tx_lock(txq, smp_processor_id());
823 		above_thresh =
824 			ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
825 						     ENA_TX_WAKEUP_THRESH);
826 		if (netif_tx_queue_stopped(txq) && above_thresh) {
827 			netif_tx_wake_queue(txq);
828 			u64_stats_update_begin(&tx_ring->syncp);
829 			tx_ring->tx_stats.queue_wakeup++;
830 			u64_stats_update_end(&tx_ring->syncp);
831 		}
832 		__netif_tx_unlock(txq);
833 	}
834 
835 	tx_ring->per_napi_bytes += tx_bytes;
836 	tx_ring->per_napi_packets += tx_pkts;
837 
838 	return tx_pkts;
839 }
840 
841 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
842 {
843 	struct sk_buff *skb;
844 
845 	if (frags)
846 		skb = napi_get_frags(rx_ring->napi);
847 	else
848 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
849 						rx_ring->rx_copybreak);
850 
851 	if (unlikely(!skb)) {
852 		u64_stats_update_begin(&rx_ring->syncp);
853 		rx_ring->rx_stats.skb_alloc_fail++;
854 		u64_stats_update_end(&rx_ring->syncp);
855 		netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
856 			  "Failed to allocate skb. frags: %d\n", frags);
857 		return NULL;
858 	}
859 
860 	return skb;
861 }
862 
863 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
864 				  struct ena_com_rx_buf_info *ena_bufs,
865 				  u32 descs,
866 				  u16 *next_to_clean)
867 {
868 	struct sk_buff *skb;
869 	struct ena_rx_buffer *rx_info;
870 	u16 len, req_id, buf = 0;
871 	void *va;
872 
873 	len = ena_bufs[buf].len;
874 	req_id = ena_bufs[buf].req_id;
875 	rx_info = &rx_ring->rx_buffer_info[req_id];
876 
877 	if (unlikely(!rx_info->page)) {
878 		netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
879 			  "Page is NULL\n");
880 		return NULL;
881 	}
882 
883 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
884 		  "rx_info %p page %p\n",
885 		  rx_info, rx_info->page);
886 
887 	/* save virt address of first buffer */
888 	va = page_address(rx_info->page) + rx_info->page_offset;
889 	prefetch(va + NET_IP_ALIGN);
890 
891 	if (len <= rx_ring->rx_copybreak) {
892 		skb = ena_alloc_skb(rx_ring, false);
893 		if (unlikely(!skb))
894 			return NULL;
895 
896 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
897 			  "rx allocated small packet. len %d. data_len %d\n",
898 			  skb->len, skb->data_len);
899 
900 		/* sync this buffer for CPU use */
901 		dma_sync_single_for_cpu(rx_ring->dev,
902 					dma_unmap_addr(&rx_info->ena_buf, paddr),
903 					len,
904 					DMA_FROM_DEVICE);
905 		skb_copy_to_linear_data(skb, va, len);
906 		dma_sync_single_for_device(rx_ring->dev,
907 					   dma_unmap_addr(&rx_info->ena_buf, paddr),
908 					   len,
909 					   DMA_FROM_DEVICE);
910 
911 		skb_put(skb, len);
912 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
913 		rx_ring->free_ids[*next_to_clean] = req_id;
914 		*next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
915 						     rx_ring->ring_size);
916 		return skb;
917 	}
918 
919 	skb = ena_alloc_skb(rx_ring, true);
920 	if (unlikely(!skb))
921 		return NULL;
922 
923 	do {
924 		dma_unmap_page(rx_ring->dev,
925 			       dma_unmap_addr(&rx_info->ena_buf, paddr),
926 			       ENA_PAGE_SIZE, DMA_FROM_DEVICE);
927 
928 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
929 				rx_info->page_offset, len, ENA_PAGE_SIZE);
930 
931 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
932 			  "rx skb updated. len %d. data_len %d\n",
933 			  skb->len, skb->data_len);
934 
935 		rx_info->page = NULL;
936 
937 		rx_ring->free_ids[*next_to_clean] = req_id;
938 		*next_to_clean =
939 			ENA_RX_RING_IDX_NEXT(*next_to_clean,
940 					     rx_ring->ring_size);
941 		if (likely(--descs == 0))
942 			break;
943 
944 		buf++;
945 		len = ena_bufs[buf].len;
946 		req_id = ena_bufs[buf].req_id;
947 		rx_info = &rx_ring->rx_buffer_info[req_id];
948 	} while (1);
949 
950 	return skb;
951 }
952 
953 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
954  * @adapter: structure containing adapter specific data
955  * @ena_rx_ctx: received packet context/metadata
956  * @skb: skb currently being received and modified
957  */
958 static void ena_rx_checksum(struct ena_ring *rx_ring,
959 				   struct ena_com_rx_ctx *ena_rx_ctx,
960 				   struct sk_buff *skb)
961 {
962 	/* Rx csum disabled */
963 	if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
964 		skb->ip_summed = CHECKSUM_NONE;
965 		return;
966 	}
967 
968 	/* For fragmented packets the checksum isn't valid */
969 	if (ena_rx_ctx->frag) {
970 		skb->ip_summed = CHECKSUM_NONE;
971 		return;
972 	}
973 
974 	/* if IP and error */
975 	if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
976 		     (ena_rx_ctx->l3_csum_err))) {
977 		/* ipv4 checksum error */
978 		skb->ip_summed = CHECKSUM_NONE;
979 		u64_stats_update_begin(&rx_ring->syncp);
980 		rx_ring->rx_stats.bad_csum++;
981 		u64_stats_update_end(&rx_ring->syncp);
982 		netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
983 			  "RX IPv4 header checksum error\n");
984 		return;
985 	}
986 
987 	/* if TCP/UDP */
988 	if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
989 		   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
990 		if (unlikely(ena_rx_ctx->l4_csum_err)) {
991 			/* TCP/UDP checksum error */
992 			u64_stats_update_begin(&rx_ring->syncp);
993 			rx_ring->rx_stats.bad_csum++;
994 			u64_stats_update_end(&rx_ring->syncp);
995 			netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
996 				  "RX L4 checksum error\n");
997 			skb->ip_summed = CHECKSUM_NONE;
998 			return;
999 		}
1000 
1001 		if (likely(ena_rx_ctx->l4_csum_checked)) {
1002 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1003 			u64_stats_update_begin(&rx_ring->syncp);
1004 			rx_ring->rx_stats.csum_good++;
1005 			u64_stats_update_end(&rx_ring->syncp);
1006 		} else {
1007 			u64_stats_update_begin(&rx_ring->syncp);
1008 			rx_ring->rx_stats.csum_unchecked++;
1009 			u64_stats_update_end(&rx_ring->syncp);
1010 			skb->ip_summed = CHECKSUM_NONE;
1011 		}
1012 	} else {
1013 		skb->ip_summed = CHECKSUM_NONE;
1014 		return;
1015 	}
1016 
1017 }
1018 
1019 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1020 			    struct ena_com_rx_ctx *ena_rx_ctx,
1021 			    struct sk_buff *skb)
1022 {
1023 	enum pkt_hash_types hash_type;
1024 
1025 	if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1026 		if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1027 			   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1028 
1029 			hash_type = PKT_HASH_TYPE_L4;
1030 		else
1031 			hash_type = PKT_HASH_TYPE_NONE;
1032 
1033 		/* Override hash type if the packet is fragmented */
1034 		if (ena_rx_ctx->frag)
1035 			hash_type = PKT_HASH_TYPE_NONE;
1036 
1037 		skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1038 	}
1039 }
1040 
1041 /* ena_clean_rx_irq - Cleanup RX irq
1042  * @rx_ring: RX ring to clean
1043  * @napi: napi handler
1044  * @budget: how many packets driver is allowed to clean
1045  *
1046  * Returns the number of cleaned buffers.
1047  */
1048 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1049 			    u32 budget)
1050 {
1051 	u16 next_to_clean = rx_ring->next_to_clean;
1052 	u32 res_budget, work_done;
1053 
1054 	struct ena_com_rx_ctx ena_rx_ctx;
1055 	struct ena_adapter *adapter;
1056 	struct sk_buff *skb;
1057 	int refill_required;
1058 	int refill_threshold;
1059 	int rc = 0;
1060 	int total_len = 0;
1061 	int rx_copybreak_pkt = 0;
1062 	int i;
1063 
1064 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1065 		  "%s qid %d\n", __func__, rx_ring->qid);
1066 	res_budget = budget;
1067 
1068 	do {
1069 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1070 		ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1071 		ena_rx_ctx.descs = 0;
1072 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1073 				    rx_ring->ena_com_io_sq,
1074 				    &ena_rx_ctx);
1075 		if (unlikely(rc))
1076 			goto error;
1077 
1078 		if (unlikely(ena_rx_ctx.descs == 0))
1079 			break;
1080 
1081 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1082 			  "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1083 			  rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1084 			  ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1085 
1086 		/* allocate skb and fill it */
1087 		skb = ena_rx_skb(rx_ring, rx_ring->ena_bufs, ena_rx_ctx.descs,
1088 				 &next_to_clean);
1089 
1090 		/* exit if we failed to retrieve a buffer */
1091 		if (unlikely(!skb)) {
1092 			for (i = 0; i < ena_rx_ctx.descs; i++) {
1093 				rx_ring->free_ids[next_to_clean] =
1094 					rx_ring->ena_bufs[i].req_id;
1095 				next_to_clean =
1096 					ENA_RX_RING_IDX_NEXT(next_to_clean,
1097 							     rx_ring->ring_size);
1098 			}
1099 			break;
1100 		}
1101 
1102 		ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1103 
1104 		ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1105 
1106 		skb_record_rx_queue(skb, rx_ring->qid);
1107 
1108 		if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1109 			total_len += rx_ring->ena_bufs[0].len;
1110 			rx_copybreak_pkt++;
1111 			napi_gro_receive(napi, skb);
1112 		} else {
1113 			total_len += skb->len;
1114 			napi_gro_frags(napi);
1115 		}
1116 
1117 		res_budget--;
1118 	} while (likely(res_budget));
1119 
1120 	work_done = budget - res_budget;
1121 	rx_ring->per_napi_bytes += total_len;
1122 	rx_ring->per_napi_packets += work_done;
1123 	u64_stats_update_begin(&rx_ring->syncp);
1124 	rx_ring->rx_stats.bytes += total_len;
1125 	rx_ring->rx_stats.cnt += work_done;
1126 	rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1127 	u64_stats_update_end(&rx_ring->syncp);
1128 
1129 	rx_ring->next_to_clean = next_to_clean;
1130 
1131 	refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
1132 	refill_threshold =
1133 		min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1134 		      ENA_RX_REFILL_THRESH_PACKET);
1135 
1136 	/* Optimization, try to batch new rx buffers */
1137 	if (refill_required > refill_threshold) {
1138 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1139 		ena_refill_rx_bufs(rx_ring, refill_required);
1140 	}
1141 
1142 	return work_done;
1143 
1144 error:
1145 	adapter = netdev_priv(rx_ring->netdev);
1146 
1147 	u64_stats_update_begin(&rx_ring->syncp);
1148 	rx_ring->rx_stats.bad_desc_num++;
1149 	u64_stats_update_end(&rx_ring->syncp);
1150 
1151 	/* Too many desc from the device. Trigger reset */
1152 	adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1153 	set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1154 
1155 	return 0;
1156 }
1157 
1158 void ena_adjust_intr_moderation(struct ena_ring *rx_ring,
1159 				       struct ena_ring *tx_ring)
1160 {
1161 	/* We apply adaptive moderation on Rx path only.
1162 	 * Tx uses static interrupt moderation.
1163 	 */
1164 	ena_com_calculate_interrupt_delay(rx_ring->ena_dev,
1165 					  rx_ring->per_napi_packets,
1166 					  rx_ring->per_napi_bytes,
1167 					  &rx_ring->smoothed_interval,
1168 					  &rx_ring->moder_tbl_idx);
1169 
1170 	/* Reset per napi packets/bytes */
1171 	tx_ring->per_napi_packets = 0;
1172 	tx_ring->per_napi_bytes = 0;
1173 	rx_ring->per_napi_packets = 0;
1174 	rx_ring->per_napi_bytes = 0;
1175 }
1176 
1177 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1178 					struct ena_ring *rx_ring)
1179 {
1180 	struct ena_eth_io_intr_reg intr_reg;
1181 
1182 	/* Update intr register: rx intr delay,
1183 	 * tx intr delay and interrupt unmask
1184 	 */
1185 	ena_com_update_intr_reg(&intr_reg,
1186 				rx_ring->smoothed_interval,
1187 				tx_ring->smoothed_interval,
1188 				true);
1189 
1190 	/* It is a shared MSI-X.
1191 	 * Tx and Rx CQ have pointer to it.
1192 	 * So we use one of them to reach the intr reg
1193 	 */
1194 	ena_com_unmask_intr(rx_ring->ena_com_io_cq, &intr_reg);
1195 }
1196 
1197 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1198 					     struct ena_ring *rx_ring)
1199 {
1200 	int cpu = get_cpu();
1201 	int numa_node;
1202 
1203 	/* Check only one ring since the 2 rings are running on the same cpu */
1204 	if (likely(tx_ring->cpu == cpu))
1205 		goto out;
1206 
1207 	numa_node = cpu_to_node(cpu);
1208 	put_cpu();
1209 
1210 	if (numa_node != NUMA_NO_NODE) {
1211 		ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1212 		ena_com_update_numa_node(rx_ring->ena_com_io_cq, numa_node);
1213 	}
1214 
1215 	tx_ring->cpu = cpu;
1216 	rx_ring->cpu = cpu;
1217 
1218 	return;
1219 out:
1220 	put_cpu();
1221 }
1222 
1223 static int ena_io_poll(struct napi_struct *napi, int budget)
1224 {
1225 	struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1226 	struct ena_ring *tx_ring, *rx_ring;
1227 
1228 	u32 tx_work_done;
1229 	u32 rx_work_done;
1230 	int tx_budget;
1231 	int napi_comp_call = 0;
1232 	int ret;
1233 
1234 	tx_ring = ena_napi->tx_ring;
1235 	rx_ring = ena_napi->rx_ring;
1236 
1237 	tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1238 
1239 	if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1240 	    test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1241 		napi_complete_done(napi, 0);
1242 		return 0;
1243 	}
1244 
1245 	tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1246 	rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1247 
1248 	/* If the device is about to reset or down, avoid unmask
1249 	 * the interrupt and return 0 so NAPI won't reschedule
1250 	 */
1251 	if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1252 		     test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1253 		napi_complete_done(napi, 0);
1254 		ret = 0;
1255 
1256 	} else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1257 		napi_comp_call = 1;
1258 
1259 		/* Update numa and unmask the interrupt only when schedule
1260 		 * from the interrupt context (vs from sk_busy_loop)
1261 		 */
1262 		if (napi_complete_done(napi, rx_work_done)) {
1263 			/* Tx and Rx share the same interrupt vector */
1264 			if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1265 				ena_adjust_intr_moderation(rx_ring, tx_ring);
1266 
1267 			ena_unmask_interrupt(tx_ring, rx_ring);
1268 		}
1269 
1270 		ena_update_ring_numa_node(tx_ring, rx_ring);
1271 
1272 		ret = rx_work_done;
1273 	} else {
1274 		ret = budget;
1275 	}
1276 
1277 	u64_stats_update_begin(&tx_ring->syncp);
1278 	tx_ring->tx_stats.napi_comp += napi_comp_call;
1279 	tx_ring->tx_stats.tx_poll++;
1280 	u64_stats_update_end(&tx_ring->syncp);
1281 
1282 	return ret;
1283 }
1284 
1285 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1286 {
1287 	struct ena_adapter *adapter = (struct ena_adapter *)data;
1288 
1289 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1290 
1291 	/* Don't call the aenq handler before probe is done */
1292 	if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1293 		ena_com_aenq_intr_handler(adapter->ena_dev, data);
1294 
1295 	return IRQ_HANDLED;
1296 }
1297 
1298 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1299  * @irq: interrupt number
1300  * @data: pointer to a network interface private napi device structure
1301  */
1302 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1303 {
1304 	struct ena_napi *ena_napi = data;
1305 
1306 	ena_napi->tx_ring->first_interrupt = true;
1307 	ena_napi->rx_ring->first_interrupt = true;
1308 
1309 	napi_schedule_irqoff(&ena_napi->napi);
1310 
1311 	return IRQ_HANDLED;
1312 }
1313 
1314 /* Reserve a single MSI-X vector for management (admin + aenq).
1315  * plus reserve one vector for each potential io queue.
1316  * the number of potential io queues is the minimum of what the device
1317  * supports and the number of vCPUs.
1318  */
1319 static int ena_enable_msix(struct ena_adapter *adapter, int num_queues)
1320 {
1321 	int msix_vecs, irq_cnt;
1322 
1323 	if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1324 		netif_err(adapter, probe, adapter->netdev,
1325 			  "Error, MSI-X is already enabled\n");
1326 		return -EPERM;
1327 	}
1328 
1329 	/* Reserved the max msix vectors we might need */
1330 	msix_vecs = ENA_MAX_MSIX_VEC(num_queues);
1331 	netif_dbg(adapter, probe, adapter->netdev,
1332 		  "trying to enable MSI-X, vectors %d\n", msix_vecs);
1333 
1334 	irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
1335 					msix_vecs, PCI_IRQ_MSIX);
1336 
1337 	if (irq_cnt < 0) {
1338 		netif_err(adapter, probe, adapter->netdev,
1339 			  "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
1340 		return -ENOSPC;
1341 	}
1342 
1343 	if (irq_cnt != msix_vecs) {
1344 		netif_notice(adapter, probe, adapter->netdev,
1345 			     "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1346 			     irq_cnt, msix_vecs);
1347 		adapter->num_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
1348 	}
1349 
1350 	if (ena_init_rx_cpu_rmap(adapter))
1351 		netif_warn(adapter, probe, adapter->netdev,
1352 			   "Failed to map IRQs to CPUs\n");
1353 
1354 	adapter->msix_vecs = irq_cnt;
1355 	set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
1356 
1357 	return 0;
1358 }
1359 
1360 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1361 {
1362 	u32 cpu;
1363 
1364 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1365 		 ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1366 		 pci_name(adapter->pdev));
1367 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
1368 		ena_intr_msix_mgmnt;
1369 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1370 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1371 		pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
1372 	cpu = cpumask_first(cpu_online_mask);
1373 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
1374 	cpumask_set_cpu(cpu,
1375 			&adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
1376 }
1377 
1378 static void ena_setup_io_intr(struct ena_adapter *adapter)
1379 {
1380 	struct net_device *netdev;
1381 	int irq_idx, i, cpu;
1382 
1383 	netdev = adapter->netdev;
1384 
1385 	for (i = 0; i < adapter->num_queues; i++) {
1386 		irq_idx = ENA_IO_IRQ_IDX(i);
1387 		cpu = i % num_online_cpus();
1388 
1389 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1390 			 "%s-Tx-Rx-%d", netdev->name, i);
1391 		adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
1392 		adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
1393 		adapter->irq_tbl[irq_idx].vector =
1394 			pci_irq_vector(adapter->pdev, irq_idx);
1395 		adapter->irq_tbl[irq_idx].cpu = cpu;
1396 
1397 		cpumask_set_cpu(cpu,
1398 				&adapter->irq_tbl[irq_idx].affinity_hint_mask);
1399 	}
1400 }
1401 
1402 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
1403 {
1404 	unsigned long flags = 0;
1405 	struct ena_irq *irq;
1406 	int rc;
1407 
1408 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1409 	rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1410 			 irq->data);
1411 	if (rc) {
1412 		netif_err(adapter, probe, adapter->netdev,
1413 			  "failed to request admin irq\n");
1414 		return rc;
1415 	}
1416 
1417 	netif_dbg(adapter, probe, adapter->netdev,
1418 		  "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1419 		  irq->affinity_hint_mask.bits[0], irq->vector);
1420 
1421 	irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1422 
1423 	return rc;
1424 }
1425 
1426 static int ena_request_io_irq(struct ena_adapter *adapter)
1427 {
1428 	unsigned long flags = 0;
1429 	struct ena_irq *irq;
1430 	int rc = 0, i, k;
1431 
1432 	if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1433 		netif_err(adapter, ifup, adapter->netdev,
1434 			  "Failed to request I/O IRQ: MSI-X is not enabled\n");
1435 		return -EINVAL;
1436 	}
1437 
1438 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1439 		irq = &adapter->irq_tbl[i];
1440 		rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1441 				 irq->data);
1442 		if (rc) {
1443 			netif_err(adapter, ifup, adapter->netdev,
1444 				  "Failed to request I/O IRQ. index %d rc %d\n",
1445 				   i, rc);
1446 			goto err;
1447 		}
1448 
1449 		netif_dbg(adapter, ifup, adapter->netdev,
1450 			  "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1451 			  i, irq->affinity_hint_mask.bits[0], irq->vector);
1452 
1453 		irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1454 	}
1455 
1456 	return rc;
1457 
1458 err:
1459 	for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
1460 		irq = &adapter->irq_tbl[k];
1461 		free_irq(irq->vector, irq->data);
1462 	}
1463 
1464 	return rc;
1465 }
1466 
1467 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
1468 {
1469 	struct ena_irq *irq;
1470 
1471 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1472 	synchronize_irq(irq->vector);
1473 	irq_set_affinity_hint(irq->vector, NULL);
1474 	free_irq(irq->vector, irq->data);
1475 }
1476 
1477 static void ena_free_io_irq(struct ena_adapter *adapter)
1478 {
1479 	struct ena_irq *irq;
1480 	int i;
1481 
1482 #ifdef CONFIG_RFS_ACCEL
1483 	if (adapter->msix_vecs >= 1) {
1484 		free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
1485 		adapter->netdev->rx_cpu_rmap = NULL;
1486 	}
1487 #endif /* CONFIG_RFS_ACCEL */
1488 
1489 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1490 		irq = &adapter->irq_tbl[i];
1491 		irq_set_affinity_hint(irq->vector, NULL);
1492 		free_irq(irq->vector, irq->data);
1493 	}
1494 }
1495 
1496 static void ena_disable_msix(struct ena_adapter *adapter)
1497 {
1498 	if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
1499 		pci_free_irq_vectors(adapter->pdev);
1500 }
1501 
1502 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
1503 {
1504 	int i;
1505 
1506 	if (!netif_running(adapter->netdev))
1507 		return;
1508 
1509 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++)
1510 		synchronize_irq(adapter->irq_tbl[i].vector);
1511 }
1512 
1513 static void ena_del_napi(struct ena_adapter *adapter)
1514 {
1515 	int i;
1516 
1517 	for (i = 0; i < adapter->num_queues; i++)
1518 		netif_napi_del(&adapter->ena_napi[i].napi);
1519 }
1520 
1521 static void ena_init_napi(struct ena_adapter *adapter)
1522 {
1523 	struct ena_napi *napi;
1524 	int i;
1525 
1526 	for (i = 0; i < adapter->num_queues; i++) {
1527 		napi = &adapter->ena_napi[i];
1528 
1529 		netif_napi_add(adapter->netdev,
1530 			       &adapter->ena_napi[i].napi,
1531 			       ena_io_poll,
1532 			       ENA_NAPI_BUDGET);
1533 		napi->rx_ring = &adapter->rx_ring[i];
1534 		napi->tx_ring = &adapter->tx_ring[i];
1535 		napi->qid = i;
1536 	}
1537 }
1538 
1539 static void ena_napi_disable_all(struct ena_adapter *adapter)
1540 {
1541 	int i;
1542 
1543 	for (i = 0; i < adapter->num_queues; i++)
1544 		napi_disable(&adapter->ena_napi[i].napi);
1545 }
1546 
1547 static void ena_napi_enable_all(struct ena_adapter *adapter)
1548 {
1549 	int i;
1550 
1551 	for (i = 0; i < adapter->num_queues; i++)
1552 		napi_enable(&adapter->ena_napi[i].napi);
1553 }
1554 
1555 static void ena_restore_ethtool_params(struct ena_adapter *adapter)
1556 {
1557 	adapter->tx_usecs = 0;
1558 	adapter->rx_usecs = 0;
1559 	adapter->tx_frames = 1;
1560 	adapter->rx_frames = 1;
1561 }
1562 
1563 /* Configure the Rx forwarding */
1564 static int ena_rss_configure(struct ena_adapter *adapter)
1565 {
1566 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1567 	int rc;
1568 
1569 	/* In case the RSS table wasn't initialized by probe */
1570 	if (!ena_dev->rss.tbl_log_size) {
1571 		rc = ena_rss_init_default(adapter);
1572 		if (rc && (rc != -EOPNOTSUPP)) {
1573 			netif_err(adapter, ifup, adapter->netdev,
1574 				  "Failed to init RSS rc: %d\n", rc);
1575 			return rc;
1576 		}
1577 	}
1578 
1579 	/* Set indirect table */
1580 	rc = ena_com_indirect_table_set(ena_dev);
1581 	if (unlikely(rc && rc != -EOPNOTSUPP))
1582 		return rc;
1583 
1584 	/* Configure hash function (if supported) */
1585 	rc = ena_com_set_hash_function(ena_dev);
1586 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
1587 		return rc;
1588 
1589 	/* Configure hash inputs (if supported) */
1590 	rc = ena_com_set_hash_ctrl(ena_dev);
1591 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
1592 		return rc;
1593 
1594 	return 0;
1595 }
1596 
1597 static int ena_up_complete(struct ena_adapter *adapter)
1598 {
1599 	int rc;
1600 
1601 	rc = ena_rss_configure(adapter);
1602 	if (rc)
1603 		return rc;
1604 
1605 	ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
1606 
1607 	ena_refill_all_rx_bufs(adapter);
1608 
1609 	/* enable transmits */
1610 	netif_tx_start_all_queues(adapter->netdev);
1611 
1612 	ena_restore_ethtool_params(adapter);
1613 
1614 	ena_napi_enable_all(adapter);
1615 
1616 	return 0;
1617 }
1618 
1619 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
1620 {
1621 	struct ena_com_create_io_ctx ctx;
1622 	struct ena_com_dev *ena_dev;
1623 	struct ena_ring *tx_ring;
1624 	u32 msix_vector;
1625 	u16 ena_qid;
1626 	int rc;
1627 
1628 	ena_dev = adapter->ena_dev;
1629 
1630 	tx_ring = &adapter->tx_ring[qid];
1631 	msix_vector = ENA_IO_IRQ_IDX(qid);
1632 	ena_qid = ENA_IO_TXQ_IDX(qid);
1633 
1634 	memset(&ctx, 0x0, sizeof(ctx));
1635 
1636 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1637 	ctx.qid = ena_qid;
1638 	ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1639 	ctx.msix_vector = msix_vector;
1640 	ctx.queue_size = tx_ring->ring_size;
1641 	ctx.numa_node = cpu_to_node(tx_ring->cpu);
1642 
1643 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1644 	if (rc) {
1645 		netif_err(adapter, ifup, adapter->netdev,
1646 			  "Failed to create I/O TX queue num %d rc: %d\n",
1647 			  qid, rc);
1648 		return rc;
1649 	}
1650 
1651 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1652 				     &tx_ring->ena_com_io_sq,
1653 				     &tx_ring->ena_com_io_cq);
1654 	if (rc) {
1655 		netif_err(adapter, ifup, adapter->netdev,
1656 			  "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1657 			  qid, rc);
1658 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1659 		return rc;
1660 	}
1661 
1662 	ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
1663 	return rc;
1664 }
1665 
1666 static int ena_create_all_io_tx_queues(struct ena_adapter *adapter)
1667 {
1668 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1669 	int rc, i;
1670 
1671 	for (i = 0; i < adapter->num_queues; i++) {
1672 		rc = ena_create_io_tx_queue(adapter, i);
1673 		if (rc)
1674 			goto create_err;
1675 	}
1676 
1677 	return 0;
1678 
1679 create_err:
1680 	while (i--)
1681 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1682 
1683 	return rc;
1684 }
1685 
1686 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
1687 {
1688 	struct ena_com_dev *ena_dev;
1689 	struct ena_com_create_io_ctx ctx;
1690 	struct ena_ring *rx_ring;
1691 	u32 msix_vector;
1692 	u16 ena_qid;
1693 	int rc;
1694 
1695 	ena_dev = adapter->ena_dev;
1696 
1697 	rx_ring = &adapter->rx_ring[qid];
1698 	msix_vector = ENA_IO_IRQ_IDX(qid);
1699 	ena_qid = ENA_IO_RXQ_IDX(qid);
1700 
1701 	memset(&ctx, 0x0, sizeof(ctx));
1702 
1703 	ctx.qid = ena_qid;
1704 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1705 	ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1706 	ctx.msix_vector = msix_vector;
1707 	ctx.queue_size = rx_ring->ring_size;
1708 	ctx.numa_node = cpu_to_node(rx_ring->cpu);
1709 
1710 	rc = ena_com_create_io_queue(ena_dev, &ctx);
1711 	if (rc) {
1712 		netif_err(adapter, ifup, adapter->netdev,
1713 			  "Failed to create I/O RX queue num %d rc: %d\n",
1714 			  qid, rc);
1715 		return rc;
1716 	}
1717 
1718 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1719 				     &rx_ring->ena_com_io_sq,
1720 				     &rx_ring->ena_com_io_cq);
1721 	if (rc) {
1722 		netif_err(adapter, ifup, adapter->netdev,
1723 			  "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1724 			  qid, rc);
1725 		ena_com_destroy_io_queue(ena_dev, ena_qid);
1726 		return rc;
1727 	}
1728 
1729 	ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
1730 
1731 	return rc;
1732 }
1733 
1734 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
1735 {
1736 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1737 	int rc, i;
1738 
1739 	for (i = 0; i < adapter->num_queues; i++) {
1740 		rc = ena_create_io_rx_queue(adapter, i);
1741 		if (rc)
1742 			goto create_err;
1743 	}
1744 
1745 	return 0;
1746 
1747 create_err:
1748 	while (i--)
1749 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1750 
1751 	return rc;
1752 }
1753 
1754 static void set_io_rings_size(struct ena_adapter *adapter,
1755 				     int new_tx_size, int new_rx_size)
1756 {
1757 	int i;
1758 
1759 	for (i = 0; i < adapter->num_queues; i++) {
1760 		adapter->tx_ring[i].ring_size = new_tx_size;
1761 		adapter->rx_ring[i].ring_size = new_rx_size;
1762 	}
1763 }
1764 
1765 /* This function allows queue allocation to backoff when the system is
1766  * low on memory. If there is not enough memory to allocate io queues
1767  * the driver will try to allocate smaller queues.
1768  *
1769  * The backoff algorithm is as follows:
1770  *  1. Try to allocate TX and RX and if successful.
1771  *  1.1. return success
1772  *
1773  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
1774  *
1775  *  3. If TX or RX is smaller than 256
1776  *  3.1. return failure.
1777  *  4. else
1778  *  4.1. go back to 1.
1779  */
1780 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
1781 {
1782 	int rc, cur_rx_ring_size, cur_tx_ring_size;
1783 	int new_rx_ring_size, new_tx_ring_size;
1784 
1785 	/* current queue sizes might be set to smaller than the requested
1786 	 * ones due to past queue allocation failures.
1787 	 */
1788 	set_io_rings_size(adapter, adapter->requested_tx_ring_size,
1789 			  adapter->requested_rx_ring_size);
1790 
1791 	while (1) {
1792 		rc = ena_setup_all_tx_resources(adapter);
1793 		if (rc)
1794 			goto err_setup_tx;
1795 
1796 		rc = ena_create_all_io_tx_queues(adapter);
1797 		if (rc)
1798 			goto err_create_tx_queues;
1799 
1800 		rc = ena_setup_all_rx_resources(adapter);
1801 		if (rc)
1802 			goto err_setup_rx;
1803 
1804 		rc = ena_create_all_io_rx_queues(adapter);
1805 		if (rc)
1806 			goto err_create_rx_queues;
1807 
1808 		return 0;
1809 
1810 err_create_rx_queues:
1811 		ena_free_all_io_rx_resources(adapter);
1812 err_setup_rx:
1813 		ena_destroy_all_tx_queues(adapter);
1814 err_create_tx_queues:
1815 		ena_free_all_io_tx_resources(adapter);
1816 err_setup_tx:
1817 		if (rc != -ENOMEM) {
1818 			netif_err(adapter, ifup, adapter->netdev,
1819 				  "Queue creation failed with error code %d\n",
1820 				  rc);
1821 			return rc;
1822 		}
1823 
1824 		cur_tx_ring_size = adapter->tx_ring[0].ring_size;
1825 		cur_rx_ring_size = adapter->rx_ring[0].ring_size;
1826 
1827 		netif_err(adapter, ifup, adapter->netdev,
1828 			  "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
1829 			  cur_tx_ring_size, cur_rx_ring_size);
1830 
1831 		new_tx_ring_size = cur_tx_ring_size;
1832 		new_rx_ring_size = cur_rx_ring_size;
1833 
1834 		/* Decrease the size of the larger queue, or
1835 		 * decrease both if they are the same size.
1836 		 */
1837 		if (cur_rx_ring_size <= cur_tx_ring_size)
1838 			new_tx_ring_size = cur_tx_ring_size / 2;
1839 		if (cur_rx_ring_size >= cur_tx_ring_size)
1840 			new_rx_ring_size = cur_rx_ring_size / 2;
1841 
1842 		if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
1843 		    new_rx_ring_size < ENA_MIN_RING_SIZE) {
1844 			netif_err(adapter, ifup, adapter->netdev,
1845 				  "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
1846 				  ENA_MIN_RING_SIZE);
1847 			return rc;
1848 		}
1849 
1850 		netif_err(adapter, ifup, adapter->netdev,
1851 			  "Retrying queue creation with sizes TX=%d, RX=%d\n",
1852 			  new_tx_ring_size,
1853 			  new_rx_ring_size);
1854 
1855 		set_io_rings_size(adapter, new_tx_ring_size,
1856 				  new_rx_ring_size);
1857 	}
1858 }
1859 
1860 static int ena_up(struct ena_adapter *adapter)
1861 {
1862 	int rc, i;
1863 
1864 	netdev_dbg(adapter->netdev, "%s\n", __func__);
1865 
1866 	ena_setup_io_intr(adapter);
1867 
1868 	/* napi poll functions should be initialized before running
1869 	 * request_irq(), to handle a rare condition where there is a pending
1870 	 * interrupt, causing the ISR to fire immediately while the poll
1871 	 * function wasn't set yet, causing a null dereference
1872 	 */
1873 	ena_init_napi(adapter);
1874 
1875 	rc = ena_request_io_irq(adapter);
1876 	if (rc)
1877 		goto err_req_irq;
1878 
1879 	rc = create_queues_with_size_backoff(adapter);
1880 	if (rc)
1881 		goto err_create_queues_with_backoff;
1882 
1883 	rc = ena_up_complete(adapter);
1884 	if (rc)
1885 		goto err_up;
1886 
1887 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
1888 		netif_carrier_on(adapter->netdev);
1889 
1890 	u64_stats_update_begin(&adapter->syncp);
1891 	adapter->dev_stats.interface_up++;
1892 	u64_stats_update_end(&adapter->syncp);
1893 
1894 	set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1895 
1896 	/* Enable completion queues interrupt */
1897 	for (i = 0; i < adapter->num_queues; i++)
1898 		ena_unmask_interrupt(&adapter->tx_ring[i],
1899 				     &adapter->rx_ring[i]);
1900 
1901 	/* schedule napi in case we had pending packets
1902 	 * from the last time we disable napi
1903 	 */
1904 	for (i = 0; i < adapter->num_queues; i++)
1905 		napi_schedule(&adapter->ena_napi[i].napi);
1906 
1907 	return rc;
1908 
1909 err_up:
1910 	ena_destroy_all_tx_queues(adapter);
1911 	ena_free_all_io_tx_resources(adapter);
1912 	ena_destroy_all_rx_queues(adapter);
1913 	ena_free_all_io_rx_resources(adapter);
1914 err_create_queues_with_backoff:
1915 	ena_free_io_irq(adapter);
1916 err_req_irq:
1917 	ena_del_napi(adapter);
1918 
1919 	return rc;
1920 }
1921 
1922 static void ena_down(struct ena_adapter *adapter)
1923 {
1924 	netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
1925 
1926 	clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1927 
1928 	u64_stats_update_begin(&adapter->syncp);
1929 	adapter->dev_stats.interface_down++;
1930 	u64_stats_update_end(&adapter->syncp);
1931 
1932 	netif_carrier_off(adapter->netdev);
1933 	netif_tx_disable(adapter->netdev);
1934 
1935 	/* After this point the napi handler won't enable the tx queue */
1936 	ena_napi_disable_all(adapter);
1937 
1938 	/* After destroy the queue there won't be any new interrupts */
1939 
1940 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
1941 		int rc;
1942 
1943 		rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
1944 		if (rc)
1945 			dev_err(&adapter->pdev->dev, "Device reset failed\n");
1946 		/* stop submitting admin commands on a device that was reset */
1947 		ena_com_set_admin_running_state(adapter->ena_dev, false);
1948 	}
1949 
1950 	ena_destroy_all_io_queues(adapter);
1951 
1952 	ena_disable_io_intr_sync(adapter);
1953 	ena_free_io_irq(adapter);
1954 	ena_del_napi(adapter);
1955 
1956 	ena_free_all_tx_bufs(adapter);
1957 	ena_free_all_rx_bufs(adapter);
1958 	ena_free_all_io_tx_resources(adapter);
1959 	ena_free_all_io_rx_resources(adapter);
1960 }
1961 
1962 /* ena_open - Called when a network interface is made active
1963  * @netdev: network interface device structure
1964  *
1965  * Returns 0 on success, negative value on failure
1966  *
1967  * The open entry point is called when a network interface is made
1968  * active by the system (IFF_UP).  At this point all resources needed
1969  * for transmit and receive operations are allocated, the interrupt
1970  * handler is registered with the OS, the watchdog timer is started,
1971  * and the stack is notified that the interface is ready.
1972  */
1973 static int ena_open(struct net_device *netdev)
1974 {
1975 	struct ena_adapter *adapter = netdev_priv(netdev);
1976 	int rc;
1977 
1978 	/* Notify the stack of the actual queue counts. */
1979 	rc = netif_set_real_num_tx_queues(netdev, adapter->num_queues);
1980 	if (rc) {
1981 		netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
1982 		return rc;
1983 	}
1984 
1985 	rc = netif_set_real_num_rx_queues(netdev, adapter->num_queues);
1986 	if (rc) {
1987 		netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
1988 		return rc;
1989 	}
1990 
1991 	rc = ena_up(adapter);
1992 	if (rc)
1993 		return rc;
1994 
1995 	return rc;
1996 }
1997 
1998 /* ena_close - Disables a network interface
1999  * @netdev: network interface device structure
2000  *
2001  * Returns 0, this is not allowed to fail
2002  *
2003  * The close entry point is called when an interface is de-activated
2004  * by the OS.  The hardware is still under the drivers control, but
2005  * needs to be disabled.  A global MAC reset is issued to stop the
2006  * hardware, and all transmit and receive resources are freed.
2007  */
2008 static int ena_close(struct net_device *netdev)
2009 {
2010 	struct ena_adapter *adapter = netdev_priv(netdev);
2011 
2012 	netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2013 
2014 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2015 		return 0;
2016 
2017 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2018 		ena_down(adapter);
2019 
2020 	/* Check for device status and issue reset if needed*/
2021 	check_for_admin_com_state(adapter);
2022 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2023 		netif_err(adapter, ifdown, adapter->netdev,
2024 			  "Destroy failure, restarting device\n");
2025 		ena_dump_stats_to_dmesg(adapter);
2026 		/* rtnl lock already obtained in dev_ioctl() layer */
2027 		ena_destroy_device(adapter, false);
2028 		ena_restore_device(adapter);
2029 	}
2030 
2031 	return 0;
2032 }
2033 
2034 int ena_update_queue_sizes(struct ena_adapter *adapter,
2035 			   u32 new_tx_size,
2036 			   u32 new_rx_size)
2037 {
2038 	bool dev_up;
2039 
2040 	dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2041 	ena_close(adapter->netdev);
2042 	adapter->requested_tx_ring_size = new_tx_size;
2043 	adapter->requested_rx_ring_size = new_rx_size;
2044 	ena_init_io_rings(adapter);
2045 	return dev_up ? ena_up(adapter) : 0;
2046 }
2047 
2048 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
2049 {
2050 	u32 mss = skb_shinfo(skb)->gso_size;
2051 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2052 	u8 l4_protocol = 0;
2053 
2054 	if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2055 		ena_tx_ctx->l4_csum_enable = 1;
2056 		if (mss) {
2057 			ena_tx_ctx->tso_enable = 1;
2058 			ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2059 			ena_tx_ctx->l4_csum_partial = 0;
2060 		} else {
2061 			ena_tx_ctx->tso_enable = 0;
2062 			ena_meta->l4_hdr_len = 0;
2063 			ena_tx_ctx->l4_csum_partial = 1;
2064 		}
2065 
2066 		switch (ip_hdr(skb)->version) {
2067 		case IPVERSION:
2068 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2069 			if (ip_hdr(skb)->frag_off & htons(IP_DF))
2070 				ena_tx_ctx->df = 1;
2071 			if (mss)
2072 				ena_tx_ctx->l3_csum_enable = 1;
2073 			l4_protocol = ip_hdr(skb)->protocol;
2074 			break;
2075 		case 6:
2076 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2077 			l4_protocol = ipv6_hdr(skb)->nexthdr;
2078 			break;
2079 		default:
2080 			break;
2081 		}
2082 
2083 		if (l4_protocol == IPPROTO_TCP)
2084 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2085 		else
2086 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2087 
2088 		ena_meta->mss = mss;
2089 		ena_meta->l3_hdr_len = skb_network_header_len(skb);
2090 		ena_meta->l3_hdr_offset = skb_network_offset(skb);
2091 		ena_tx_ctx->meta_valid = 1;
2092 
2093 	} else {
2094 		ena_tx_ctx->meta_valid = 0;
2095 	}
2096 }
2097 
2098 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2099 				       struct sk_buff *skb)
2100 {
2101 	int num_frags, header_len, rc;
2102 
2103 	num_frags = skb_shinfo(skb)->nr_frags;
2104 	header_len = skb_headlen(skb);
2105 
2106 	if (num_frags < tx_ring->sgl_size)
2107 		return 0;
2108 
2109 	if ((num_frags == tx_ring->sgl_size) &&
2110 	    (header_len < tx_ring->tx_max_header_size))
2111 		return 0;
2112 
2113 	u64_stats_update_begin(&tx_ring->syncp);
2114 	tx_ring->tx_stats.linearize++;
2115 	u64_stats_update_end(&tx_ring->syncp);
2116 
2117 	rc = skb_linearize(skb);
2118 	if (unlikely(rc)) {
2119 		u64_stats_update_begin(&tx_ring->syncp);
2120 		tx_ring->tx_stats.linearize_failed++;
2121 		u64_stats_update_end(&tx_ring->syncp);
2122 	}
2123 
2124 	return rc;
2125 }
2126 
2127 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2128 			  struct ena_tx_buffer *tx_info,
2129 			  struct sk_buff *skb,
2130 			  void **push_hdr,
2131 			  u16 *header_len)
2132 {
2133 	struct ena_adapter *adapter = tx_ring->adapter;
2134 	struct ena_com_buf *ena_buf;
2135 	dma_addr_t dma;
2136 	u32 skb_head_len, frag_len, last_frag;
2137 	u16 push_len = 0;
2138 	u16 delta = 0;
2139 	int i = 0;
2140 
2141 	skb_head_len = skb_headlen(skb);
2142 	tx_info->skb = skb;
2143 	ena_buf = tx_info->bufs;
2144 
2145 	if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2146 		/* When the device is LLQ mode, the driver will copy
2147 		 * the header into the device memory space.
2148 		 * the ena_com layer assume the header is in a linear
2149 		 * memory space.
2150 		 * This assumption might be wrong since part of the header
2151 		 * can be in the fragmented buffers.
2152 		 * Use skb_header_pointer to make sure the header is in a
2153 		 * linear memory space.
2154 		 */
2155 
2156 		push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2157 		*push_hdr = skb_header_pointer(skb, 0, push_len,
2158 					       tx_ring->push_buf_intermediate_buf);
2159 		*header_len = push_len;
2160 		if (unlikely(skb->data != *push_hdr)) {
2161 			u64_stats_update_begin(&tx_ring->syncp);
2162 			tx_ring->tx_stats.llq_buffer_copy++;
2163 			u64_stats_update_end(&tx_ring->syncp);
2164 
2165 			delta = push_len - skb_head_len;
2166 		}
2167 	} else {
2168 		*push_hdr = NULL;
2169 		*header_len = min_t(u32, skb_head_len,
2170 				    tx_ring->tx_max_header_size);
2171 	}
2172 
2173 	netif_dbg(adapter, tx_queued, adapter->netdev,
2174 		  "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2175 		  *push_hdr, push_len);
2176 
2177 	if (skb_head_len > push_len) {
2178 		dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2179 				     skb_head_len - push_len, DMA_TO_DEVICE);
2180 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2181 			goto error_report_dma_error;
2182 
2183 		ena_buf->paddr = dma;
2184 		ena_buf->len = skb_head_len - push_len;
2185 
2186 		ena_buf++;
2187 		tx_info->num_of_bufs++;
2188 		tx_info->map_linear_data = 1;
2189 	} else {
2190 		tx_info->map_linear_data = 0;
2191 	}
2192 
2193 	last_frag = skb_shinfo(skb)->nr_frags;
2194 
2195 	for (i = 0; i < last_frag; i++) {
2196 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2197 
2198 		frag_len = skb_frag_size(frag);
2199 
2200 		if (unlikely(delta >= frag_len)) {
2201 			delta -= frag_len;
2202 			continue;
2203 		}
2204 
2205 		dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2206 				       frag_len - delta, DMA_TO_DEVICE);
2207 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2208 			goto error_report_dma_error;
2209 
2210 		ena_buf->paddr = dma;
2211 		ena_buf->len = frag_len - delta;
2212 		ena_buf++;
2213 		tx_info->num_of_bufs++;
2214 		delta = 0;
2215 	}
2216 
2217 	return 0;
2218 
2219 error_report_dma_error:
2220 	u64_stats_update_begin(&tx_ring->syncp);
2221 	tx_ring->tx_stats.dma_mapping_err++;
2222 	u64_stats_update_end(&tx_ring->syncp);
2223 	netdev_warn(adapter->netdev, "failed to map skb\n");
2224 
2225 	tx_info->skb = NULL;
2226 
2227 	tx_info->num_of_bufs += i;
2228 	ena_unmap_tx_skb(tx_ring, tx_info);
2229 
2230 	return -EINVAL;
2231 }
2232 
2233 /* Called with netif_tx_lock. */
2234 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
2235 {
2236 	struct ena_adapter *adapter = netdev_priv(dev);
2237 	struct ena_tx_buffer *tx_info;
2238 	struct ena_com_tx_ctx ena_tx_ctx;
2239 	struct ena_ring *tx_ring;
2240 	struct netdev_queue *txq;
2241 	void *push_hdr;
2242 	u16 next_to_use, req_id, header_len;
2243 	int qid, rc, nb_hw_desc;
2244 
2245 	netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
2246 	/*  Determine which tx ring we will be placed on */
2247 	qid = skb_get_queue_mapping(skb);
2248 	tx_ring = &adapter->tx_ring[qid];
2249 	txq = netdev_get_tx_queue(dev, qid);
2250 
2251 	rc = ena_check_and_linearize_skb(tx_ring, skb);
2252 	if (unlikely(rc))
2253 		goto error_drop_packet;
2254 
2255 	skb_tx_timestamp(skb);
2256 
2257 	next_to_use = tx_ring->next_to_use;
2258 	req_id = tx_ring->free_ids[next_to_use];
2259 	tx_info = &tx_ring->tx_buffer_info[req_id];
2260 	tx_info->num_of_bufs = 0;
2261 
2262 	WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
2263 
2264 	rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
2265 	if (unlikely(rc))
2266 		goto error_drop_packet;
2267 
2268 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2269 	ena_tx_ctx.ena_bufs = tx_info->bufs;
2270 	ena_tx_ctx.push_header = push_hdr;
2271 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2272 	ena_tx_ctx.req_id = req_id;
2273 	ena_tx_ctx.header_len = header_len;
2274 
2275 	/* set flags and meta data */
2276 	ena_tx_csum(&ena_tx_ctx, skb);
2277 
2278 	if (unlikely(ena_com_is_doorbell_needed(tx_ring->ena_com_io_sq, &ena_tx_ctx))) {
2279 		netif_dbg(adapter, tx_queued, dev,
2280 			  "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
2281 			  qid);
2282 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2283 	}
2284 
2285 	/* prepare the packet's descriptors to dma engine */
2286 	rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
2287 				&nb_hw_desc);
2288 
2289 	/* ena_com_prepare_tx() can't fail due to overflow of tx queue,
2290 	 * since the number of free descriptors in the queue is checked
2291 	 * after sending the previous packet. In case there isn't enough
2292 	 * space in the queue for the next packet, it is stopped
2293 	 * until there is again enough available space in the queue.
2294 	 * All other failure reasons of ena_com_prepare_tx() are fatal
2295 	 * and therefore require a device reset.
2296 	 */
2297 	if (unlikely(rc)) {
2298 		netif_err(adapter, tx_queued, dev,
2299 			  "failed to prepare tx bufs\n");
2300 		u64_stats_update_begin(&tx_ring->syncp);
2301 		tx_ring->tx_stats.prepare_ctx_err++;
2302 		u64_stats_update_end(&tx_ring->syncp);
2303 		adapter->reset_reason = ENA_REGS_RESET_DRIVER_INVALID_STATE;
2304 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2305 		goto error_unmap_dma;
2306 	}
2307 
2308 	netdev_tx_sent_queue(txq, skb->len);
2309 
2310 	u64_stats_update_begin(&tx_ring->syncp);
2311 	tx_ring->tx_stats.cnt++;
2312 	tx_ring->tx_stats.bytes += skb->len;
2313 	u64_stats_update_end(&tx_ring->syncp);
2314 
2315 	tx_info->tx_descs = nb_hw_desc;
2316 	tx_info->last_jiffies = jiffies;
2317 	tx_info->print_once = 0;
2318 
2319 	tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2320 		tx_ring->ring_size);
2321 
2322 	/* stop the queue when no more space available, the packet can have up
2323 	 * to sgl_size + 2. one for the meta descriptor and one for header
2324 	 * (if the header is larger than tx_max_header_size).
2325 	 */
2326 	if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
2327 						   tx_ring->sgl_size + 2))) {
2328 		netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
2329 			  __func__, qid);
2330 
2331 		netif_tx_stop_queue(txq);
2332 		u64_stats_update_begin(&tx_ring->syncp);
2333 		tx_ring->tx_stats.queue_stop++;
2334 		u64_stats_update_end(&tx_ring->syncp);
2335 
2336 		/* There is a rare condition where this function decide to
2337 		 * stop the queue but meanwhile clean_tx_irq updates
2338 		 * next_to_completion and terminates.
2339 		 * The queue will remain stopped forever.
2340 		 * To solve this issue add a mb() to make sure that
2341 		 * netif_tx_stop_queue() write is vissible before checking if
2342 		 * there is additional space in the queue.
2343 		 */
2344 		smp_mb();
2345 
2346 		if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
2347 						 ENA_TX_WAKEUP_THRESH)) {
2348 			netif_tx_wake_queue(txq);
2349 			u64_stats_update_begin(&tx_ring->syncp);
2350 			tx_ring->tx_stats.queue_wakeup++;
2351 			u64_stats_update_end(&tx_ring->syncp);
2352 		}
2353 	}
2354 
2355 	if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
2356 		/* trigger the dma engine. ena_com_write_sq_doorbell()
2357 		 * has a mb
2358 		 */
2359 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2360 		u64_stats_update_begin(&tx_ring->syncp);
2361 		tx_ring->tx_stats.doorbells++;
2362 		u64_stats_update_end(&tx_ring->syncp);
2363 	}
2364 
2365 	return NETDEV_TX_OK;
2366 
2367 error_unmap_dma:
2368 	ena_unmap_tx_skb(tx_ring, tx_info);
2369 	tx_info->skb = NULL;
2370 
2371 error_drop_packet:
2372 	dev_kfree_skb(skb);
2373 	return NETDEV_TX_OK;
2374 }
2375 
2376 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
2377 			    struct net_device *sb_dev)
2378 {
2379 	u16 qid;
2380 	/* we suspect that this is good for in--kernel network services that
2381 	 * want to loop incoming skb rx to tx in normal user generated traffic,
2382 	 * most probably we will not get to this
2383 	 */
2384 	if (skb_rx_queue_recorded(skb))
2385 		qid = skb_get_rx_queue(skb);
2386 	else
2387 		qid = netdev_pick_tx(dev, skb, NULL);
2388 
2389 	return qid;
2390 }
2391 
2392 static void ena_config_host_info(struct ena_com_dev *ena_dev,
2393 				 struct pci_dev *pdev)
2394 {
2395 	struct ena_admin_host_info *host_info;
2396 	int rc;
2397 
2398 	/* Allocate only the host info */
2399 	rc = ena_com_allocate_host_info(ena_dev);
2400 	if (rc) {
2401 		pr_err("Cannot allocate host info\n");
2402 		return;
2403 	}
2404 
2405 	host_info = ena_dev->host_attr.host_info;
2406 
2407 	host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
2408 	host_info->os_type = ENA_ADMIN_OS_LINUX;
2409 	host_info->kernel_ver = LINUX_VERSION_CODE;
2410 	strlcpy(host_info->kernel_ver_str, utsname()->version,
2411 		sizeof(host_info->kernel_ver_str) - 1);
2412 	host_info->os_dist = 0;
2413 	strncpy(host_info->os_dist_str, utsname()->release,
2414 		sizeof(host_info->os_dist_str) - 1);
2415 	host_info->driver_version =
2416 		(DRV_MODULE_VER_MAJOR) |
2417 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2418 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
2419 		("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
2420 	host_info->num_cpus = num_online_cpus();
2421 
2422 	rc = ena_com_set_host_attributes(ena_dev);
2423 	if (rc) {
2424 		if (rc == -EOPNOTSUPP)
2425 			pr_warn("Cannot set host attributes\n");
2426 		else
2427 			pr_err("Cannot set host attributes\n");
2428 
2429 		goto err;
2430 	}
2431 
2432 	return;
2433 
2434 err:
2435 	ena_com_delete_host_info(ena_dev);
2436 }
2437 
2438 static void ena_config_debug_area(struct ena_adapter *adapter)
2439 {
2440 	u32 debug_area_size;
2441 	int rc, ss_count;
2442 
2443 	ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
2444 	if (ss_count <= 0) {
2445 		netif_err(adapter, drv, adapter->netdev,
2446 			  "SS count is negative\n");
2447 		return;
2448 	}
2449 
2450 	/* allocate 32 bytes for each string and 64bit for the value */
2451 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
2452 
2453 	rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
2454 	if (rc) {
2455 		pr_err("Cannot allocate debug area\n");
2456 		return;
2457 	}
2458 
2459 	rc = ena_com_set_host_attributes(adapter->ena_dev);
2460 	if (rc) {
2461 		if (rc == -EOPNOTSUPP)
2462 			netif_warn(adapter, drv, adapter->netdev,
2463 				   "Cannot set host attributes\n");
2464 		else
2465 			netif_err(adapter, drv, adapter->netdev,
2466 				  "Cannot set host attributes\n");
2467 		goto err;
2468 	}
2469 
2470 	return;
2471 err:
2472 	ena_com_delete_debug_area(adapter->ena_dev);
2473 }
2474 
2475 static void ena_get_stats64(struct net_device *netdev,
2476 			    struct rtnl_link_stats64 *stats)
2477 {
2478 	struct ena_adapter *adapter = netdev_priv(netdev);
2479 	struct ena_ring *rx_ring, *tx_ring;
2480 	unsigned int start;
2481 	u64 rx_drops;
2482 	int i;
2483 
2484 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2485 		return;
2486 
2487 	for (i = 0; i < adapter->num_queues; i++) {
2488 		u64 bytes, packets;
2489 
2490 		tx_ring = &adapter->tx_ring[i];
2491 
2492 		do {
2493 			start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
2494 			packets = tx_ring->tx_stats.cnt;
2495 			bytes = tx_ring->tx_stats.bytes;
2496 		} while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
2497 
2498 		stats->tx_packets += packets;
2499 		stats->tx_bytes += bytes;
2500 
2501 		rx_ring = &adapter->rx_ring[i];
2502 
2503 		do {
2504 			start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
2505 			packets = rx_ring->rx_stats.cnt;
2506 			bytes = rx_ring->rx_stats.bytes;
2507 		} while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
2508 
2509 		stats->rx_packets += packets;
2510 		stats->rx_bytes += bytes;
2511 	}
2512 
2513 	do {
2514 		start = u64_stats_fetch_begin_irq(&adapter->syncp);
2515 		rx_drops = adapter->dev_stats.rx_drops;
2516 	} while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
2517 
2518 	stats->rx_dropped = rx_drops;
2519 
2520 	stats->multicast = 0;
2521 	stats->collisions = 0;
2522 
2523 	stats->rx_length_errors = 0;
2524 	stats->rx_crc_errors = 0;
2525 	stats->rx_frame_errors = 0;
2526 	stats->rx_fifo_errors = 0;
2527 	stats->rx_missed_errors = 0;
2528 	stats->tx_window_errors = 0;
2529 
2530 	stats->rx_errors = 0;
2531 	stats->tx_errors = 0;
2532 }
2533 
2534 static const struct net_device_ops ena_netdev_ops = {
2535 	.ndo_open		= ena_open,
2536 	.ndo_stop		= ena_close,
2537 	.ndo_start_xmit		= ena_start_xmit,
2538 	.ndo_select_queue	= ena_select_queue,
2539 	.ndo_get_stats64	= ena_get_stats64,
2540 	.ndo_tx_timeout		= ena_tx_timeout,
2541 	.ndo_change_mtu		= ena_change_mtu,
2542 	.ndo_set_mac_address	= NULL,
2543 	.ndo_validate_addr	= eth_validate_addr,
2544 };
2545 
2546 static int ena_device_validate_params(struct ena_adapter *adapter,
2547 				      struct ena_com_dev_get_features_ctx *get_feat_ctx)
2548 {
2549 	struct net_device *netdev = adapter->netdev;
2550 	int rc;
2551 
2552 	rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
2553 			      adapter->mac_addr);
2554 	if (!rc) {
2555 		netif_err(adapter, drv, netdev,
2556 			  "Error, mac address are different\n");
2557 		return -EINVAL;
2558 	}
2559 
2560 	if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
2561 		netif_err(adapter, drv, netdev,
2562 			  "Error, device max mtu is smaller than netdev MTU\n");
2563 		return -EINVAL;
2564 	}
2565 
2566 	return 0;
2567 }
2568 
2569 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
2570 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
2571 			   bool *wd_state)
2572 {
2573 	struct device *dev = &pdev->dev;
2574 	bool readless_supported;
2575 	u32 aenq_groups;
2576 	int dma_width;
2577 	int rc;
2578 
2579 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
2580 	if (rc) {
2581 		dev_err(dev, "failed to init mmio read less\n");
2582 		return rc;
2583 	}
2584 
2585 	/* The PCIe configuration space revision id indicate if mmio reg
2586 	 * read is disabled
2587 	 */
2588 	readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
2589 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2590 
2591 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2592 	if (rc) {
2593 		dev_err(dev, "Can not reset device\n");
2594 		goto err_mmio_read_less;
2595 	}
2596 
2597 	rc = ena_com_validate_version(ena_dev);
2598 	if (rc) {
2599 		dev_err(dev, "device version is too low\n");
2600 		goto err_mmio_read_less;
2601 	}
2602 
2603 	dma_width = ena_com_get_dma_width(ena_dev);
2604 	if (dma_width < 0) {
2605 		dev_err(dev, "Invalid dma width value %d", dma_width);
2606 		rc = dma_width;
2607 		goto err_mmio_read_less;
2608 	}
2609 
2610 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2611 	if (rc) {
2612 		dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
2613 		goto err_mmio_read_less;
2614 	}
2615 
2616 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2617 	if (rc) {
2618 		dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2619 			rc);
2620 		goto err_mmio_read_less;
2621 	}
2622 
2623 	/* ENA admin level init */
2624 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
2625 	if (rc) {
2626 		dev_err(dev,
2627 			"Can not initialize ena admin queue with device\n");
2628 		goto err_mmio_read_less;
2629 	}
2630 
2631 	/* To enable the msix interrupts the driver needs to know the number
2632 	 * of queues. So the driver uses polling mode to retrieve this
2633 	 * information
2634 	 */
2635 	ena_com_set_admin_polling_mode(ena_dev, true);
2636 
2637 	ena_config_host_info(ena_dev, pdev);
2638 
2639 	/* Get Device Attributes*/
2640 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2641 	if (rc) {
2642 		dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
2643 		goto err_admin_init;
2644 	}
2645 
2646 	/* Try to turn all the available aenq groups */
2647 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2648 		BIT(ENA_ADMIN_FATAL_ERROR) |
2649 		BIT(ENA_ADMIN_WARNING) |
2650 		BIT(ENA_ADMIN_NOTIFICATION) |
2651 		BIT(ENA_ADMIN_KEEP_ALIVE);
2652 
2653 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
2654 
2655 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2656 	if (rc) {
2657 		dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
2658 		goto err_admin_init;
2659 	}
2660 
2661 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2662 
2663 	return 0;
2664 
2665 err_admin_init:
2666 	ena_com_delete_host_info(ena_dev);
2667 	ena_com_admin_destroy(ena_dev);
2668 err_mmio_read_less:
2669 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2670 
2671 	return rc;
2672 }
2673 
2674 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
2675 						    int io_vectors)
2676 {
2677 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2678 	struct device *dev = &adapter->pdev->dev;
2679 	int rc;
2680 
2681 	rc = ena_enable_msix(adapter, io_vectors);
2682 	if (rc) {
2683 		dev_err(dev, "Can not reserve msix vectors\n");
2684 		return rc;
2685 	}
2686 
2687 	ena_setup_mgmnt_intr(adapter);
2688 
2689 	rc = ena_request_mgmnt_irq(adapter);
2690 	if (rc) {
2691 		dev_err(dev, "Can not setup management interrupts\n");
2692 		goto err_disable_msix;
2693 	}
2694 
2695 	ena_com_set_admin_polling_mode(ena_dev, false);
2696 
2697 	ena_com_admin_aenq_enable(ena_dev);
2698 
2699 	return 0;
2700 
2701 err_disable_msix:
2702 	ena_disable_msix(adapter);
2703 
2704 	return rc;
2705 }
2706 
2707 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
2708 {
2709 	struct net_device *netdev = adapter->netdev;
2710 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2711 	bool dev_up;
2712 
2713 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2714 		return;
2715 
2716 	netif_carrier_off(netdev);
2717 
2718 	del_timer_sync(&adapter->timer_service);
2719 
2720 	dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2721 	adapter->dev_up_before_reset = dev_up;
2722 	if (!graceful)
2723 		ena_com_set_admin_running_state(ena_dev, false);
2724 
2725 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2726 		ena_down(adapter);
2727 
2728 	/* Stop the device from sending AENQ events (in case reset flag is set
2729 	 *  and device is up, ena_down() already reset the device.
2730 	 */
2731 	if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
2732 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2733 
2734 	ena_free_mgmnt_irq(adapter);
2735 
2736 	ena_disable_msix(adapter);
2737 
2738 	ena_com_abort_admin_commands(ena_dev);
2739 
2740 	ena_com_wait_for_abort_completion(ena_dev);
2741 
2742 	ena_com_admin_destroy(ena_dev);
2743 
2744 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2745 
2746 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
2747 
2748 	clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2749 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2750 }
2751 
2752 static int ena_restore_device(struct ena_adapter *adapter)
2753 {
2754 	struct ena_com_dev_get_features_ctx get_feat_ctx;
2755 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2756 	struct pci_dev *pdev = adapter->pdev;
2757 	bool wd_state;
2758 	int rc;
2759 
2760 	set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2761 	rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
2762 	if (rc) {
2763 		dev_err(&pdev->dev, "Can not initialize device\n");
2764 		goto err;
2765 	}
2766 	adapter->wd_state = wd_state;
2767 
2768 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
2769 	if (rc) {
2770 		dev_err(&pdev->dev, "Validation of device parameters failed\n");
2771 		goto err_device_destroy;
2772 	}
2773 
2774 	rc = ena_enable_msix_and_set_admin_interrupts(adapter,
2775 						      adapter->num_queues);
2776 	if (rc) {
2777 		dev_err(&pdev->dev, "Enable MSI-X failed\n");
2778 		goto err_device_destroy;
2779 	}
2780 	/* If the interface was up before the reset bring it up */
2781 	if (adapter->dev_up_before_reset) {
2782 		rc = ena_up(adapter);
2783 		if (rc) {
2784 			dev_err(&pdev->dev, "Failed to create I/O queues\n");
2785 			goto err_disable_msix;
2786 		}
2787 	}
2788 
2789 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2790 
2791 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2792 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2793 		netif_carrier_on(adapter->netdev);
2794 
2795 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
2796 	dev_err(&pdev->dev,
2797 		"Device reset completed successfully, Driver info: %s\n",
2798 		version);
2799 
2800 	return rc;
2801 err_disable_msix:
2802 	ena_free_mgmnt_irq(adapter);
2803 	ena_disable_msix(adapter);
2804 err_device_destroy:
2805 	ena_com_abort_admin_commands(ena_dev);
2806 	ena_com_wait_for_abort_completion(ena_dev);
2807 	ena_com_admin_destroy(ena_dev);
2808 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
2809 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2810 err:
2811 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2812 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2813 	dev_err(&pdev->dev,
2814 		"Reset attempt failed. Can not reset the device\n");
2815 
2816 	return rc;
2817 }
2818 
2819 static void ena_fw_reset_device(struct work_struct *work)
2820 {
2821 	struct ena_adapter *adapter =
2822 		container_of(work, struct ena_adapter, reset_task);
2823 	struct pci_dev *pdev = adapter->pdev;
2824 
2825 	if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2826 		dev_err(&pdev->dev,
2827 			"device reset schedule while reset bit is off\n");
2828 		return;
2829 	}
2830 	rtnl_lock();
2831 	ena_destroy_device(adapter, false);
2832 	ena_restore_device(adapter);
2833 	rtnl_unlock();
2834 }
2835 
2836 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
2837 					struct ena_ring *rx_ring)
2838 {
2839 	if (likely(rx_ring->first_interrupt))
2840 		return 0;
2841 
2842 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
2843 		return 0;
2844 
2845 	rx_ring->no_interrupt_event_cnt++;
2846 
2847 	if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
2848 		netif_err(adapter, rx_err, adapter->netdev,
2849 			  "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
2850 			  rx_ring->qid);
2851 		adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2852 		smp_mb__before_atomic();
2853 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2854 		return -EIO;
2855 	}
2856 
2857 	return 0;
2858 }
2859 
2860 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
2861 					  struct ena_ring *tx_ring)
2862 {
2863 	struct ena_tx_buffer *tx_buf;
2864 	unsigned long last_jiffies;
2865 	u32 missed_tx = 0;
2866 	int i, rc = 0;
2867 
2868 	for (i = 0; i < tx_ring->ring_size; i++) {
2869 		tx_buf = &tx_ring->tx_buffer_info[i];
2870 		last_jiffies = tx_buf->last_jiffies;
2871 
2872 		if (last_jiffies == 0)
2873 			/* no pending Tx at this location */
2874 			continue;
2875 
2876 		if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
2877 			     2 * adapter->missing_tx_completion_to))) {
2878 			/* If after graceful period interrupt is still not
2879 			 * received, we schedule a reset
2880 			 */
2881 			netif_err(adapter, tx_err, adapter->netdev,
2882 				  "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
2883 				  tx_ring->qid);
2884 			adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2885 			smp_mb__before_atomic();
2886 			set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2887 			return -EIO;
2888 		}
2889 
2890 		if (unlikely(time_is_before_jiffies(last_jiffies +
2891 				adapter->missing_tx_completion_to))) {
2892 			if (!tx_buf->print_once)
2893 				netif_notice(adapter, tx_err, adapter->netdev,
2894 					     "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2895 					     tx_ring->qid, i);
2896 
2897 			tx_buf->print_once = 1;
2898 			missed_tx++;
2899 		}
2900 	}
2901 
2902 	if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
2903 		netif_err(adapter, tx_err, adapter->netdev,
2904 			  "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
2905 			  missed_tx,
2906 			  adapter->missing_tx_completion_threshold);
2907 		adapter->reset_reason =
2908 			ENA_REGS_RESET_MISS_TX_CMPL;
2909 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2910 		rc = -EIO;
2911 	}
2912 
2913 	u64_stats_update_begin(&tx_ring->syncp);
2914 	tx_ring->tx_stats.missed_tx = missed_tx;
2915 	u64_stats_update_end(&tx_ring->syncp);
2916 
2917 	return rc;
2918 }
2919 
2920 static void check_for_missing_completions(struct ena_adapter *adapter)
2921 {
2922 	struct ena_ring *tx_ring;
2923 	struct ena_ring *rx_ring;
2924 	int i, budget, rc;
2925 
2926 	/* Make sure the driver doesn't turn the device in other process */
2927 	smp_rmb();
2928 
2929 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2930 		return;
2931 
2932 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2933 		return;
2934 
2935 	if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
2936 		return;
2937 
2938 	budget = ENA_MONITORED_TX_QUEUES;
2939 
2940 	for (i = adapter->last_monitored_tx_qid; i < adapter->num_queues; i++) {
2941 		tx_ring = &adapter->tx_ring[i];
2942 		rx_ring = &adapter->rx_ring[i];
2943 
2944 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
2945 		if (unlikely(rc))
2946 			return;
2947 
2948 		rc = check_for_rx_interrupt_queue(adapter, rx_ring);
2949 		if (unlikely(rc))
2950 			return;
2951 
2952 		budget--;
2953 		if (!budget)
2954 			break;
2955 	}
2956 
2957 	adapter->last_monitored_tx_qid = i % adapter->num_queues;
2958 }
2959 
2960 /* trigger napi schedule after 2 consecutive detections */
2961 #define EMPTY_RX_REFILL 2
2962 /* For the rare case where the device runs out of Rx descriptors and the
2963  * napi handler failed to refill new Rx descriptors (due to a lack of memory
2964  * for example).
2965  * This case will lead to a deadlock:
2966  * The device won't send interrupts since all the new Rx packets will be dropped
2967  * The napi handler won't allocate new Rx descriptors so the device will be
2968  * able to send new packets.
2969  *
2970  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
2971  * It is recommended to have at least 512MB, with a minimum of 128MB for
2972  * constrained environment).
2973  *
2974  * When such a situation is detected - Reschedule napi
2975  */
2976 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
2977 {
2978 	struct ena_ring *rx_ring;
2979 	int i, refill_required;
2980 
2981 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2982 		return;
2983 
2984 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2985 		return;
2986 
2987 	for (i = 0; i < adapter->num_queues; i++) {
2988 		rx_ring = &adapter->rx_ring[i];
2989 
2990 		refill_required =
2991 			ena_com_free_desc(rx_ring->ena_com_io_sq);
2992 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
2993 			rx_ring->empty_rx_queue++;
2994 
2995 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
2996 				u64_stats_update_begin(&rx_ring->syncp);
2997 				rx_ring->rx_stats.empty_rx_ring++;
2998 				u64_stats_update_end(&rx_ring->syncp);
2999 
3000 				netif_err(adapter, drv, adapter->netdev,
3001 					  "trigger refill for ring %d\n", i);
3002 
3003 				napi_schedule(rx_ring->napi);
3004 				rx_ring->empty_rx_queue = 0;
3005 			}
3006 		} else {
3007 			rx_ring->empty_rx_queue = 0;
3008 		}
3009 	}
3010 }
3011 
3012 /* Check for keep alive expiration */
3013 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3014 {
3015 	unsigned long keep_alive_expired;
3016 
3017 	if (!adapter->wd_state)
3018 		return;
3019 
3020 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3021 		return;
3022 
3023 	keep_alive_expired = round_jiffies(adapter->last_keep_alive_jiffies +
3024 					   adapter->keep_alive_timeout);
3025 	if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3026 		netif_err(adapter, drv, adapter->netdev,
3027 			  "Keep alive watchdog timeout.\n");
3028 		u64_stats_update_begin(&adapter->syncp);
3029 		adapter->dev_stats.wd_expired++;
3030 		u64_stats_update_end(&adapter->syncp);
3031 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3032 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3033 	}
3034 }
3035 
3036 static void check_for_admin_com_state(struct ena_adapter *adapter)
3037 {
3038 	if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3039 		netif_err(adapter, drv, adapter->netdev,
3040 			  "ENA admin queue is not in running state!\n");
3041 		u64_stats_update_begin(&adapter->syncp);
3042 		adapter->dev_stats.admin_q_pause++;
3043 		u64_stats_update_end(&adapter->syncp);
3044 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3045 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3046 	}
3047 }
3048 
3049 static void ena_update_hints(struct ena_adapter *adapter,
3050 			     struct ena_admin_ena_hw_hints *hints)
3051 {
3052 	struct net_device *netdev = adapter->netdev;
3053 
3054 	if (hints->admin_completion_tx_timeout)
3055 		adapter->ena_dev->admin_queue.completion_timeout =
3056 			hints->admin_completion_tx_timeout * 1000;
3057 
3058 	if (hints->mmio_read_timeout)
3059 		/* convert to usec */
3060 		adapter->ena_dev->mmio_read.reg_read_to =
3061 			hints->mmio_read_timeout * 1000;
3062 
3063 	if (hints->missed_tx_completion_count_threshold_to_reset)
3064 		adapter->missing_tx_completion_threshold =
3065 			hints->missed_tx_completion_count_threshold_to_reset;
3066 
3067 	if (hints->missing_tx_completion_timeout) {
3068 		if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3069 			adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3070 		else
3071 			adapter->missing_tx_completion_to =
3072 				msecs_to_jiffies(hints->missing_tx_completion_timeout);
3073 	}
3074 
3075 	if (hints->netdev_wd_timeout)
3076 		netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3077 
3078 	if (hints->driver_watchdog_timeout) {
3079 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3080 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3081 		else
3082 			adapter->keep_alive_timeout =
3083 				msecs_to_jiffies(hints->driver_watchdog_timeout);
3084 	}
3085 }
3086 
3087 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3088 				 struct net_device *netdev)
3089 {
3090 	host_info->supported_network_features[0] =
3091 		netdev->features & GENMASK_ULL(31, 0);
3092 	host_info->supported_network_features[1] =
3093 		(netdev->features & GENMASK_ULL(63, 32)) >> 32;
3094 }
3095 
3096 static void ena_timer_service(struct timer_list *t)
3097 {
3098 	struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3099 	u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3100 	struct ena_admin_host_info *host_info =
3101 		adapter->ena_dev->host_attr.host_info;
3102 
3103 	check_for_missing_keep_alive(adapter);
3104 
3105 	check_for_admin_com_state(adapter);
3106 
3107 	check_for_missing_completions(adapter);
3108 
3109 	check_for_empty_rx_ring(adapter);
3110 
3111 	if (debug_area)
3112 		ena_dump_stats_to_buf(adapter, debug_area);
3113 
3114 	if (host_info)
3115 		ena_update_host_info(host_info, adapter->netdev);
3116 
3117 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3118 		netif_err(adapter, drv, adapter->netdev,
3119 			  "Trigger reset is on\n");
3120 		ena_dump_stats_to_dmesg(adapter);
3121 		queue_work(ena_wq, &adapter->reset_task);
3122 		return;
3123 	}
3124 
3125 	/* Reset the timer */
3126 	mod_timer(&adapter->timer_service, jiffies + HZ);
3127 }
3128 
3129 static int ena_calc_io_queue_num(struct pci_dev *pdev,
3130 				 struct ena_com_dev *ena_dev,
3131 				 struct ena_com_dev_get_features_ctx *get_feat_ctx)
3132 {
3133 	int io_tx_sq_num, io_tx_cq_num, io_rx_num, io_queue_num;
3134 
3135 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3136 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3137 			&get_feat_ctx->max_queue_ext.max_queue_ext;
3138 		io_rx_num = min_t(int, max_queue_ext->max_rx_sq_num,
3139 				  max_queue_ext->max_rx_cq_num);
3140 
3141 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3142 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3143 	} else {
3144 		struct ena_admin_queue_feature_desc *max_queues =
3145 			&get_feat_ctx->max_queues;
3146 		io_tx_sq_num = max_queues->max_sq_num;
3147 		io_tx_cq_num = max_queues->max_cq_num;
3148 		io_rx_num = min_t(int, io_tx_sq_num, io_tx_cq_num);
3149 	}
3150 
3151 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
3152 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3153 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3154 
3155 	io_queue_num = min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3156 	io_queue_num = min_t(int, io_queue_num, io_rx_num);
3157 	io_queue_num = min_t(int, io_queue_num, io_tx_sq_num);
3158 	io_queue_num = min_t(int, io_queue_num, io_tx_cq_num);
3159 	/* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
3160 	io_queue_num = min_t(int, io_queue_num, pci_msix_vec_count(pdev) - 1);
3161 	if (unlikely(!io_queue_num)) {
3162 		dev_err(&pdev->dev, "The device doesn't have io queues\n");
3163 		return -EFAULT;
3164 	}
3165 
3166 	return io_queue_num;
3167 }
3168 
3169 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3170 					   struct ena_com_dev *ena_dev,
3171 					   struct ena_admin_feature_llq_desc *llq,
3172 					   struct ena_llq_configurations *llq_default_configurations)
3173 {
3174 	bool has_mem_bar;
3175 	int rc;
3176 	u32 llq_feature_mask;
3177 
3178 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3179 	if (!(ena_dev->supported_features & llq_feature_mask)) {
3180 		dev_err(&pdev->dev,
3181 			"LLQ is not supported Fallback to host mode policy.\n");
3182 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3183 		return 0;
3184 	}
3185 
3186 	has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
3187 
3188 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3189 	if (unlikely(rc)) {
3190 		dev_err(&pdev->dev,
3191 			"Failed to configure the device mode.  Fallback to host mode policy.\n");
3192 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3193 		return 0;
3194 	}
3195 
3196 	/* Nothing to config, exit */
3197 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3198 		return 0;
3199 
3200 	if (!has_mem_bar) {
3201 		dev_err(&pdev->dev,
3202 			"ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3203 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3204 		return 0;
3205 	}
3206 
3207 	ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3208 					   pci_resource_start(pdev, ENA_MEM_BAR),
3209 					   pci_resource_len(pdev, ENA_MEM_BAR));
3210 
3211 	if (!ena_dev->mem_bar)
3212 		return -EFAULT;
3213 
3214 	return 0;
3215 }
3216 
3217 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3218 				 struct net_device *netdev)
3219 {
3220 	netdev_features_t dev_features = 0;
3221 
3222 	/* Set offload features */
3223 	if (feat->offload.tx &
3224 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3225 		dev_features |= NETIF_F_IP_CSUM;
3226 
3227 	if (feat->offload.tx &
3228 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
3229 		dev_features |= NETIF_F_IPV6_CSUM;
3230 
3231 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
3232 		dev_features |= NETIF_F_TSO;
3233 
3234 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
3235 		dev_features |= NETIF_F_TSO6;
3236 
3237 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
3238 		dev_features |= NETIF_F_TSO_ECN;
3239 
3240 	if (feat->offload.rx_supported &
3241 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
3242 		dev_features |= NETIF_F_RXCSUM;
3243 
3244 	if (feat->offload.rx_supported &
3245 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
3246 		dev_features |= NETIF_F_RXCSUM;
3247 
3248 	netdev->features =
3249 		dev_features |
3250 		NETIF_F_SG |
3251 		NETIF_F_RXHASH |
3252 		NETIF_F_HIGHDMA;
3253 
3254 	netdev->hw_features |= netdev->features;
3255 	netdev->vlan_features |= netdev->features;
3256 }
3257 
3258 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
3259 				     struct ena_com_dev_get_features_ctx *feat)
3260 {
3261 	struct net_device *netdev = adapter->netdev;
3262 
3263 	/* Copy mac address */
3264 	if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
3265 		eth_hw_addr_random(netdev);
3266 		ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
3267 	} else {
3268 		ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
3269 		ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
3270 	}
3271 
3272 	/* Set offload features */
3273 	ena_set_dev_offloads(feat, netdev);
3274 
3275 	adapter->max_mtu = feat->dev_attr.max_mtu;
3276 	netdev->max_mtu = adapter->max_mtu;
3277 	netdev->min_mtu = ENA_MIN_MTU;
3278 }
3279 
3280 static int ena_rss_init_default(struct ena_adapter *adapter)
3281 {
3282 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3283 	struct device *dev = &adapter->pdev->dev;
3284 	int rc, i;
3285 	u32 val;
3286 
3287 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3288 	if (unlikely(rc)) {
3289 		dev_err(dev, "Cannot init indirect table\n");
3290 		goto err_rss_init;
3291 	}
3292 
3293 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3294 		val = ethtool_rxfh_indir_default(i, adapter->num_queues);
3295 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3296 						       ENA_IO_RXQ_IDX(val));
3297 		if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3298 			dev_err(dev, "Cannot fill indirect table\n");
3299 			goto err_fill_indir;
3300 		}
3301 	}
3302 
3303 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3304 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3305 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3306 		dev_err(dev, "Cannot fill hash function\n");
3307 		goto err_fill_indir;
3308 	}
3309 
3310 	rc = ena_com_set_default_hash_ctrl(ena_dev);
3311 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3312 		dev_err(dev, "Cannot fill hash control\n");
3313 		goto err_fill_indir;
3314 	}
3315 
3316 	return 0;
3317 
3318 err_fill_indir:
3319 	ena_com_rss_destroy(ena_dev);
3320 err_rss_init:
3321 
3322 	return rc;
3323 }
3324 
3325 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3326 {
3327 	int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3328 
3329 	pci_release_selected_regions(pdev, release_bars);
3330 }
3331 
3332 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3333 {
3334 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3335 	llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3336 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3337 	llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3338 	llq_config->llq_ring_entry_size_value = 128;
3339 }
3340 
3341 static int ena_calc_queue_size(struct ena_calc_queue_size_ctx *ctx)
3342 {
3343 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
3344 	struct ena_com_dev *ena_dev = ctx->ena_dev;
3345 	u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
3346 	u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
3347 	u32 max_tx_queue_size;
3348 	u32 max_rx_queue_size;
3349 
3350 	if (ctx->ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3351 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3352 			&ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
3353 		max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
3354 					  max_queue_ext->max_rx_sq_depth);
3355 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
3356 
3357 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3358 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3359 						  llq->max_llq_depth);
3360 		else
3361 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3362 						  max_queue_ext->max_tx_sq_depth);
3363 
3364 		ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3365 					     max_queue_ext->max_per_packet_tx_descs);
3366 		ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3367 					     max_queue_ext->max_per_packet_rx_descs);
3368 	} else {
3369 		struct ena_admin_queue_feature_desc *max_queues =
3370 			&ctx->get_feat_ctx->max_queues;
3371 		max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
3372 					  max_queues->max_sq_depth);
3373 		max_tx_queue_size = max_queues->max_cq_depth;
3374 
3375 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3376 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3377 						  llq->max_llq_depth);
3378 		else
3379 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3380 						  max_queues->max_sq_depth);
3381 
3382 		ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3383 					     max_queues->max_packet_tx_descs);
3384 		ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3385 					     max_queues->max_packet_rx_descs);
3386 	}
3387 
3388 	max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
3389 	max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
3390 
3391 	tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
3392 				  max_tx_queue_size);
3393 	rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
3394 				  max_rx_queue_size);
3395 
3396 	tx_queue_size = rounddown_pow_of_two(tx_queue_size);
3397 	rx_queue_size = rounddown_pow_of_two(rx_queue_size);
3398 
3399 	ctx->max_tx_queue_size = max_tx_queue_size;
3400 	ctx->max_rx_queue_size = max_rx_queue_size;
3401 	ctx->tx_queue_size = tx_queue_size;
3402 	ctx->rx_queue_size = rx_queue_size;
3403 
3404 	return 0;
3405 }
3406 
3407 /* ena_probe - Device Initialization Routine
3408  * @pdev: PCI device information struct
3409  * @ent: entry in ena_pci_tbl
3410  *
3411  * Returns 0 on success, negative on failure
3412  *
3413  * ena_probe initializes an adapter identified by a pci_dev structure.
3414  * The OS initialization, configuring of the adapter private structure,
3415  * and a hardware reset occur.
3416  */
3417 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3418 {
3419 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3420 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
3421 	struct ena_llq_configurations llq_config;
3422 	struct ena_com_dev *ena_dev = NULL;
3423 	struct ena_adapter *adapter;
3424 	int io_queue_num, bars, rc;
3425 	struct net_device *netdev;
3426 	static int adapters_found;
3427 	char *queue_type_str;
3428 	bool wd_state;
3429 
3430 	dev_dbg(&pdev->dev, "%s\n", __func__);
3431 
3432 	dev_info_once(&pdev->dev, "%s", version);
3433 
3434 	rc = pci_enable_device_mem(pdev);
3435 	if (rc) {
3436 		dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
3437 		return rc;
3438 	}
3439 
3440 	pci_set_master(pdev);
3441 
3442 	ena_dev = vzalloc(sizeof(*ena_dev));
3443 	if (!ena_dev) {
3444 		rc = -ENOMEM;
3445 		goto err_disable_device;
3446 	}
3447 
3448 	bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3449 	rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
3450 	if (rc) {
3451 		dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
3452 			rc);
3453 		goto err_free_ena_dev;
3454 	}
3455 
3456 	ena_dev->reg_bar = devm_ioremap(&pdev->dev,
3457 					pci_resource_start(pdev, ENA_REG_BAR),
3458 					pci_resource_len(pdev, ENA_REG_BAR));
3459 	if (!ena_dev->reg_bar) {
3460 		dev_err(&pdev->dev, "failed to remap regs bar\n");
3461 		rc = -EFAULT;
3462 		goto err_free_region;
3463 	}
3464 
3465 	ena_dev->dmadev = &pdev->dev;
3466 
3467 	rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
3468 	if (rc) {
3469 		dev_err(&pdev->dev, "ena device init failed\n");
3470 		if (rc == -ETIME)
3471 			rc = -EPROBE_DEFER;
3472 		goto err_free_region;
3473 	}
3474 
3475 	set_default_llq_configurations(&llq_config);
3476 
3477 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
3478 					     &llq_config);
3479 	if (rc) {
3480 		dev_err(&pdev->dev, "ena device init failed\n");
3481 		goto err_device_destroy;
3482 	}
3483 
3484 	calc_queue_ctx.ena_dev = ena_dev;
3485 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
3486 	calc_queue_ctx.pdev = pdev;
3487 
3488 	/* initial Tx interrupt delay, Assumes 1 usec granularity.
3489 	* Updated during device initialization with the real granularity
3490 	*/
3491 	ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
3492 	io_queue_num = ena_calc_io_queue_num(pdev, ena_dev, &get_feat_ctx);
3493 	rc = ena_calc_queue_size(&calc_queue_ctx);
3494 	if (rc || io_queue_num <= 0) {
3495 		rc = -EFAULT;
3496 		goto err_device_destroy;
3497 	}
3498 
3499 	dev_info(&pdev->dev, "creating %d io queues. rx queue size: %d tx queue size. %d LLQ is %s\n",
3500 		 io_queue_num,
3501 		 calc_queue_ctx.rx_queue_size,
3502 		 calc_queue_ctx.tx_queue_size,
3503 		 (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) ?
3504 		 "ENABLED" : "DISABLED");
3505 
3506 	/* dev zeroed in init_etherdev */
3507 	netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), io_queue_num);
3508 	if (!netdev) {
3509 		dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
3510 		rc = -ENOMEM;
3511 		goto err_device_destroy;
3512 	}
3513 
3514 	SET_NETDEV_DEV(netdev, &pdev->dev);
3515 
3516 	adapter = netdev_priv(netdev);
3517 	pci_set_drvdata(pdev, adapter);
3518 
3519 	adapter->ena_dev = ena_dev;
3520 	adapter->netdev = netdev;
3521 	adapter->pdev = pdev;
3522 
3523 	ena_set_conf_feat_params(adapter, &get_feat_ctx);
3524 
3525 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3526 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3527 
3528 	adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
3529 	adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
3530 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
3531 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
3532 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
3533 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
3534 
3535 	adapter->num_queues = io_queue_num;
3536 	adapter->last_monitored_tx_qid = 0;
3537 
3538 	adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
3539 	adapter->wd_state = wd_state;
3540 
3541 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
3542 
3543 	rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
3544 	if (rc) {
3545 		dev_err(&pdev->dev,
3546 			"Failed to query interrupt moderation feature\n");
3547 		goto err_netdev_destroy;
3548 	}
3549 	ena_init_io_rings(adapter);
3550 
3551 	netdev->netdev_ops = &ena_netdev_ops;
3552 	netdev->watchdog_timeo = TX_TIMEOUT;
3553 	ena_set_ethtool_ops(netdev);
3554 
3555 	netdev->priv_flags |= IFF_UNICAST_FLT;
3556 
3557 	u64_stats_init(&adapter->syncp);
3558 
3559 	rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3560 	if (rc) {
3561 		dev_err(&pdev->dev,
3562 			"Failed to enable and set the admin interrupts\n");
3563 		goto err_worker_destroy;
3564 	}
3565 	rc = ena_rss_init_default(adapter);
3566 	if (rc && (rc != -EOPNOTSUPP)) {
3567 		dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
3568 		goto err_free_msix;
3569 	}
3570 
3571 	ena_config_debug_area(adapter);
3572 
3573 	memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
3574 
3575 	netif_carrier_off(netdev);
3576 
3577 	rc = register_netdev(netdev);
3578 	if (rc) {
3579 		dev_err(&pdev->dev, "Cannot register net device\n");
3580 		goto err_rss;
3581 	}
3582 
3583 	INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
3584 
3585 	adapter->last_keep_alive_jiffies = jiffies;
3586 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
3587 	adapter->missing_tx_completion_to = TX_TIMEOUT;
3588 	adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
3589 
3590 	ena_update_hints(adapter, &get_feat_ctx.hw_hints);
3591 
3592 	timer_setup(&adapter->timer_service, ena_timer_service, 0);
3593 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3594 
3595 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3596 		queue_type_str = "Regular";
3597 	else
3598 		queue_type_str = "Low Latency";
3599 
3600 	dev_info(&pdev->dev,
3601 		 "%s found at mem %lx, mac addr %pM Queues %d, Placement policy: %s\n",
3602 		 DEVICE_NAME, (long)pci_resource_start(pdev, 0),
3603 		 netdev->dev_addr, io_queue_num, queue_type_str);
3604 
3605 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3606 
3607 	adapters_found++;
3608 
3609 	return 0;
3610 
3611 err_rss:
3612 	ena_com_delete_debug_area(ena_dev);
3613 	ena_com_rss_destroy(ena_dev);
3614 err_free_msix:
3615 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
3616 	/* stop submitting admin commands on a device that was reset */
3617 	ena_com_set_admin_running_state(ena_dev, false);
3618 	ena_free_mgmnt_irq(adapter);
3619 	ena_disable_msix(adapter);
3620 err_worker_destroy:
3621 	ena_com_destroy_interrupt_moderation(ena_dev);
3622 	del_timer(&adapter->timer_service);
3623 err_netdev_destroy:
3624 	free_netdev(netdev);
3625 err_device_destroy:
3626 	ena_com_delete_host_info(ena_dev);
3627 	ena_com_admin_destroy(ena_dev);
3628 err_free_region:
3629 	ena_release_bars(ena_dev, pdev);
3630 err_free_ena_dev:
3631 	vfree(ena_dev);
3632 err_disable_device:
3633 	pci_disable_device(pdev);
3634 	return rc;
3635 }
3636 
3637 /*****************************************************************************/
3638 
3639 /* ena_remove - Device Removal Routine
3640  * @pdev: PCI device information struct
3641  *
3642  * ena_remove is called by the PCI subsystem to alert the driver
3643  * that it should release a PCI device.
3644  */
3645 static void ena_remove(struct pci_dev *pdev)
3646 {
3647 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
3648 	struct ena_com_dev *ena_dev;
3649 	struct net_device *netdev;
3650 
3651 	ena_dev = adapter->ena_dev;
3652 	netdev = adapter->netdev;
3653 
3654 #ifdef CONFIG_RFS_ACCEL
3655 	if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
3656 		free_irq_cpu_rmap(netdev->rx_cpu_rmap);
3657 		netdev->rx_cpu_rmap = NULL;
3658 	}
3659 #endif /* CONFIG_RFS_ACCEL */
3660 	del_timer_sync(&adapter->timer_service);
3661 
3662 	cancel_work_sync(&adapter->reset_task);
3663 
3664 	rtnl_lock();
3665 	ena_destroy_device(adapter, true);
3666 	rtnl_unlock();
3667 
3668 	unregister_netdev(netdev);
3669 
3670 	free_netdev(netdev);
3671 
3672 	ena_com_rss_destroy(ena_dev);
3673 
3674 	ena_com_delete_debug_area(ena_dev);
3675 
3676 	ena_com_delete_host_info(ena_dev);
3677 
3678 	ena_release_bars(ena_dev, pdev);
3679 
3680 	pci_disable_device(pdev);
3681 
3682 	ena_com_destroy_interrupt_moderation(ena_dev);
3683 
3684 	vfree(ena_dev);
3685 }
3686 
3687 #ifdef CONFIG_PM
3688 /* ena_suspend - PM suspend callback
3689  * @pdev: PCI device information struct
3690  * @state:power state
3691  */
3692 static int ena_suspend(struct pci_dev *pdev,  pm_message_t state)
3693 {
3694 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
3695 
3696 	u64_stats_update_begin(&adapter->syncp);
3697 	adapter->dev_stats.suspend++;
3698 	u64_stats_update_end(&adapter->syncp);
3699 
3700 	rtnl_lock();
3701 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3702 		dev_err(&pdev->dev,
3703 			"ignoring device reset request as the device is being suspended\n");
3704 		clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3705 	}
3706 	ena_destroy_device(adapter, true);
3707 	rtnl_unlock();
3708 	return 0;
3709 }
3710 
3711 /* ena_resume - PM resume callback
3712  * @pdev: PCI device information struct
3713  *
3714  */
3715 static int ena_resume(struct pci_dev *pdev)
3716 {
3717 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
3718 	int rc;
3719 
3720 	u64_stats_update_begin(&adapter->syncp);
3721 	adapter->dev_stats.resume++;
3722 	u64_stats_update_end(&adapter->syncp);
3723 
3724 	rtnl_lock();
3725 	rc = ena_restore_device(adapter);
3726 	rtnl_unlock();
3727 	return rc;
3728 }
3729 #endif
3730 
3731 static struct pci_driver ena_pci_driver = {
3732 	.name		= DRV_MODULE_NAME,
3733 	.id_table	= ena_pci_tbl,
3734 	.probe		= ena_probe,
3735 	.remove		= ena_remove,
3736 #ifdef CONFIG_PM
3737 	.suspend    = ena_suspend,
3738 	.resume     = ena_resume,
3739 #endif
3740 	.sriov_configure = pci_sriov_configure_simple,
3741 };
3742 
3743 static int __init ena_init(void)
3744 {
3745 	pr_info("%s", version);
3746 
3747 	ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
3748 	if (!ena_wq) {
3749 		pr_err("Failed to create workqueue\n");
3750 		return -ENOMEM;
3751 	}
3752 
3753 	return pci_register_driver(&ena_pci_driver);
3754 }
3755 
3756 static void __exit ena_cleanup(void)
3757 {
3758 	pci_unregister_driver(&ena_pci_driver);
3759 
3760 	if (ena_wq) {
3761 		destroy_workqueue(ena_wq);
3762 		ena_wq = NULL;
3763 	}
3764 }
3765 
3766 /******************************************************************************
3767  ******************************** AENQ Handlers *******************************
3768  *****************************************************************************/
3769 /* ena_update_on_link_change:
3770  * Notify the network interface about the change in link status
3771  */
3772 static void ena_update_on_link_change(void *adapter_data,
3773 				      struct ena_admin_aenq_entry *aenq_e)
3774 {
3775 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3776 	struct ena_admin_aenq_link_change_desc *aenq_desc =
3777 		(struct ena_admin_aenq_link_change_desc *)aenq_e;
3778 	int status = aenq_desc->flags &
3779 		ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3780 
3781 	if (status) {
3782 		netdev_dbg(adapter->netdev, "%s\n", __func__);
3783 		set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3784 		if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
3785 			netif_carrier_on(adapter->netdev);
3786 	} else {
3787 		clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3788 		netif_carrier_off(adapter->netdev);
3789 	}
3790 }
3791 
3792 static void ena_keep_alive_wd(void *adapter_data,
3793 			      struct ena_admin_aenq_entry *aenq_e)
3794 {
3795 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3796 	struct ena_admin_aenq_keep_alive_desc *desc;
3797 	u64 rx_drops;
3798 
3799 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3800 	adapter->last_keep_alive_jiffies = jiffies;
3801 
3802 	rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
3803 
3804 	u64_stats_update_begin(&adapter->syncp);
3805 	adapter->dev_stats.rx_drops = rx_drops;
3806 	u64_stats_update_end(&adapter->syncp);
3807 }
3808 
3809 static void ena_notification(void *adapter_data,
3810 			     struct ena_admin_aenq_entry *aenq_e)
3811 {
3812 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3813 	struct ena_admin_ena_hw_hints *hints;
3814 
3815 	WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3816 	     "Invalid group(%x) expected %x\n",
3817 	     aenq_e->aenq_common_desc.group,
3818 	     ENA_ADMIN_NOTIFICATION);
3819 
3820 	switch (aenq_e->aenq_common_desc.syndrom) {
3821 	case ENA_ADMIN_UPDATE_HINTS:
3822 		hints = (struct ena_admin_ena_hw_hints *)
3823 			(&aenq_e->inline_data_w4);
3824 		ena_update_hints(adapter, hints);
3825 		break;
3826 	default:
3827 		netif_err(adapter, drv, adapter->netdev,
3828 			  "Invalid aenq notification link state %d\n",
3829 			  aenq_e->aenq_common_desc.syndrom);
3830 	}
3831 }
3832 
3833 /* This handler will called for unknown event group or unimplemented handlers*/
3834 static void unimplemented_aenq_handler(void *data,
3835 				       struct ena_admin_aenq_entry *aenq_e)
3836 {
3837 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3838 
3839 	netif_err(adapter, drv, adapter->netdev,
3840 		  "Unknown event was received or event with unimplemented handler\n");
3841 }
3842 
3843 static struct ena_aenq_handlers aenq_handlers = {
3844 	.handlers = {
3845 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3846 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
3847 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3848 	},
3849 	.unimplemented_handler = unimplemented_aenq_handler
3850 };
3851 
3852 module_init(ena_init);
3853 module_exit(ena_cleanup);
3854