xref: /linux/drivers/net/ethernet/amazon/ena/ena_netdev.c (revision 8ccd54fe45713cd458015b5b08d6098545e70543)
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3  * Copyright 2015-2020 Amazon.com, Inc. or its affiliates. All rights reserved.
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #ifdef CONFIG_RFS_ACCEL
9 #include <linux/cpu_rmap.h>
10 #endif /* CONFIG_RFS_ACCEL */
11 #include <linux/ethtool.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/numa.h>
15 #include <linux/pci.h>
16 #include <linux/utsname.h>
17 #include <linux/version.h>
18 #include <linux/vmalloc.h>
19 #include <net/ip.h>
20 
21 #include "ena_netdev.h"
22 #include <linux/bpf_trace.h>
23 #include "ena_pci_id_tbl.h"
24 
25 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
26 MODULE_DESCRIPTION(DEVICE_NAME);
27 MODULE_LICENSE("GPL");
28 
29 /* Time in jiffies before concluding the transmitter is hung. */
30 #define TX_TIMEOUT  (5 * HZ)
31 
32 #define ENA_MAX_RINGS min_t(unsigned int, ENA_MAX_NUM_IO_QUEUES, num_possible_cpus())
33 
34 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
35 		NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
36 
37 static struct ena_aenq_handlers aenq_handlers;
38 
39 static struct workqueue_struct *ena_wq;
40 
41 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
42 
43 static int ena_rss_init_default(struct ena_adapter *adapter);
44 static void check_for_admin_com_state(struct ena_adapter *adapter);
45 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
46 static int ena_restore_device(struct ena_adapter *adapter);
47 
48 static void ena_init_io_rings(struct ena_adapter *adapter,
49 			      int first_index, int count);
50 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
51 				   int count);
52 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
53 				  int count);
54 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
55 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
56 					   int first_index,
57 					   int count);
58 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
59 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
60 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
61 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
62 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
63 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
64 				      int first_index, int count);
65 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
66 				     int first_index, int count);
67 static int ena_up(struct ena_adapter *adapter);
68 static void ena_down(struct ena_adapter *adapter);
69 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
70 				 struct ena_ring *rx_ring);
71 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
72 				      struct ena_ring *rx_ring);
73 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
74 			      struct ena_tx_buffer *tx_info);
75 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
76 					    int first_index, int count);
77 
78 /* Increase a stat by cnt while holding syncp seqlock on 32bit machines */
79 static void ena_increase_stat(u64 *statp, u64 cnt,
80 			      struct u64_stats_sync *syncp)
81 {
82 	u64_stats_update_begin(syncp);
83 	(*statp) += cnt;
84 	u64_stats_update_end(syncp);
85 }
86 
87 static void ena_ring_tx_doorbell(struct ena_ring *tx_ring)
88 {
89 	ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
90 	ena_increase_stat(&tx_ring->tx_stats.doorbells, 1, &tx_ring->syncp);
91 }
92 
93 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
94 {
95 	struct ena_adapter *adapter = netdev_priv(dev);
96 
97 	/* Change the state of the device to trigger reset
98 	 * Check that we are not in the middle or a trigger already
99 	 */
100 
101 	if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
102 		return;
103 
104 	ena_reset_device(adapter, ENA_REGS_RESET_OS_NETDEV_WD);
105 	ena_increase_stat(&adapter->dev_stats.tx_timeout, 1, &adapter->syncp);
106 
107 	netif_err(adapter, tx_err, dev, "Transmit time out\n");
108 }
109 
110 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
111 {
112 	int i;
113 
114 	for (i = 0; i < adapter->num_io_queues; i++)
115 		adapter->rx_ring[i].mtu = mtu;
116 }
117 
118 static int ena_change_mtu(struct net_device *dev, int new_mtu)
119 {
120 	struct ena_adapter *adapter = netdev_priv(dev);
121 	int ret;
122 
123 	ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
124 	if (!ret) {
125 		netif_dbg(adapter, drv, dev, "Set MTU to %d\n", new_mtu);
126 		update_rx_ring_mtu(adapter, new_mtu);
127 		dev->mtu = new_mtu;
128 	} else {
129 		netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
130 			  new_mtu);
131 	}
132 
133 	return ret;
134 }
135 
136 static int ena_xmit_common(struct net_device *dev,
137 			   struct ena_ring *ring,
138 			   struct ena_tx_buffer *tx_info,
139 			   struct ena_com_tx_ctx *ena_tx_ctx,
140 			   u16 next_to_use,
141 			   u32 bytes)
142 {
143 	struct ena_adapter *adapter = netdev_priv(dev);
144 	int rc, nb_hw_desc;
145 
146 	if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
147 						ena_tx_ctx))) {
148 		netif_dbg(adapter, tx_queued, dev,
149 			  "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
150 			  ring->qid);
151 		ena_ring_tx_doorbell(ring);
152 	}
153 
154 	/* prepare the packet's descriptors to dma engine */
155 	rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
156 				&nb_hw_desc);
157 
158 	/* In case there isn't enough space in the queue for the packet,
159 	 * we simply drop it. All other failure reasons of
160 	 * ena_com_prepare_tx() are fatal and therefore require a device reset.
161 	 */
162 	if (unlikely(rc)) {
163 		netif_err(adapter, tx_queued, dev,
164 			  "Failed to prepare tx bufs\n");
165 		ena_increase_stat(&ring->tx_stats.prepare_ctx_err, 1,
166 				  &ring->syncp);
167 		if (rc != -ENOMEM)
168 			ena_reset_device(adapter,
169 					 ENA_REGS_RESET_DRIVER_INVALID_STATE);
170 		return rc;
171 	}
172 
173 	u64_stats_update_begin(&ring->syncp);
174 	ring->tx_stats.cnt++;
175 	ring->tx_stats.bytes += bytes;
176 	u64_stats_update_end(&ring->syncp);
177 
178 	tx_info->tx_descs = nb_hw_desc;
179 	tx_info->last_jiffies = jiffies;
180 	tx_info->print_once = 0;
181 
182 	ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
183 						 ring->ring_size);
184 	return 0;
185 }
186 
187 /* This is the XDP napi callback. XDP queues use a separate napi callback
188  * than Rx/Tx queues.
189  */
190 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
191 {
192 	struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
193 	u32 xdp_work_done, xdp_budget;
194 	struct ena_ring *xdp_ring;
195 	int napi_comp_call = 0;
196 	int ret;
197 
198 	xdp_ring = ena_napi->xdp_ring;
199 
200 	xdp_budget = budget;
201 
202 	if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
203 	    test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
204 		napi_complete_done(napi, 0);
205 		return 0;
206 	}
207 
208 	xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
209 
210 	/* If the device is about to reset or down, avoid unmask
211 	 * the interrupt and return 0 so NAPI won't reschedule
212 	 */
213 	if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
214 		napi_complete_done(napi, 0);
215 		ret = 0;
216 	} else if (xdp_budget > xdp_work_done) {
217 		napi_comp_call = 1;
218 		if (napi_complete_done(napi, xdp_work_done))
219 			ena_unmask_interrupt(xdp_ring, NULL);
220 		ena_update_ring_numa_node(xdp_ring, NULL);
221 		ret = xdp_work_done;
222 	} else {
223 		ret = xdp_budget;
224 	}
225 
226 	u64_stats_update_begin(&xdp_ring->syncp);
227 	xdp_ring->tx_stats.napi_comp += napi_comp_call;
228 	xdp_ring->tx_stats.tx_poll++;
229 	u64_stats_update_end(&xdp_ring->syncp);
230 	xdp_ring->tx_stats.last_napi_jiffies = jiffies;
231 
232 	return ret;
233 }
234 
235 static int ena_xdp_tx_map_frame(struct ena_ring *xdp_ring,
236 				struct ena_tx_buffer *tx_info,
237 				struct xdp_frame *xdpf,
238 				struct ena_com_tx_ctx *ena_tx_ctx)
239 {
240 	struct ena_adapter *adapter = xdp_ring->adapter;
241 	struct ena_com_buf *ena_buf;
242 	int push_len = 0;
243 	dma_addr_t dma;
244 	void *data;
245 	u32 size;
246 
247 	tx_info->xdpf = xdpf;
248 	data = tx_info->xdpf->data;
249 	size = tx_info->xdpf->len;
250 
251 	if (xdp_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
252 		/* Designate part of the packet for LLQ */
253 		push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
254 
255 		ena_tx_ctx->push_header = data;
256 
257 		size -= push_len;
258 		data += push_len;
259 	}
260 
261 	ena_tx_ctx->header_len = push_len;
262 
263 	if (size > 0) {
264 		dma = dma_map_single(xdp_ring->dev,
265 				     data,
266 				     size,
267 				     DMA_TO_DEVICE);
268 		if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
269 			goto error_report_dma_error;
270 
271 		tx_info->map_linear_data = 0;
272 
273 		ena_buf = tx_info->bufs;
274 		ena_buf->paddr = dma;
275 		ena_buf->len = size;
276 
277 		ena_tx_ctx->ena_bufs = ena_buf;
278 		ena_tx_ctx->num_bufs = tx_info->num_of_bufs = 1;
279 	}
280 
281 	return 0;
282 
283 error_report_dma_error:
284 	ena_increase_stat(&xdp_ring->tx_stats.dma_mapping_err, 1,
285 			  &xdp_ring->syncp);
286 	netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map xdp buff\n");
287 
288 	return -EINVAL;
289 }
290 
291 static int ena_xdp_xmit_frame(struct ena_ring *xdp_ring,
292 			      struct net_device *dev,
293 			      struct xdp_frame *xdpf,
294 			      int flags)
295 {
296 	struct ena_com_tx_ctx ena_tx_ctx = {};
297 	struct ena_tx_buffer *tx_info;
298 	u16 next_to_use, req_id;
299 	int rc;
300 
301 	next_to_use = xdp_ring->next_to_use;
302 	req_id = xdp_ring->free_ids[next_to_use];
303 	tx_info = &xdp_ring->tx_buffer_info[req_id];
304 	tx_info->num_of_bufs = 0;
305 
306 	rc = ena_xdp_tx_map_frame(xdp_ring, tx_info, xdpf, &ena_tx_ctx);
307 	if (unlikely(rc))
308 		return rc;
309 
310 	ena_tx_ctx.req_id = req_id;
311 
312 	rc = ena_xmit_common(dev,
313 			     xdp_ring,
314 			     tx_info,
315 			     &ena_tx_ctx,
316 			     next_to_use,
317 			     xdpf->len);
318 	if (rc)
319 		goto error_unmap_dma;
320 
321 	/* trigger the dma engine. ena_ring_tx_doorbell()
322 	 * calls a memory barrier inside it.
323 	 */
324 	if (flags & XDP_XMIT_FLUSH)
325 		ena_ring_tx_doorbell(xdp_ring);
326 
327 	return rc;
328 
329 error_unmap_dma:
330 	ena_unmap_tx_buff(xdp_ring, tx_info);
331 	tx_info->xdpf = NULL;
332 	return rc;
333 }
334 
335 static int ena_xdp_xmit(struct net_device *dev, int n,
336 			struct xdp_frame **frames, u32 flags)
337 {
338 	struct ena_adapter *adapter = netdev_priv(dev);
339 	struct ena_ring *xdp_ring;
340 	int qid, i, nxmit = 0;
341 
342 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
343 		return -EINVAL;
344 
345 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
346 		return -ENETDOWN;
347 
348 	/* We assume that all rings have the same XDP program */
349 	if (!READ_ONCE(adapter->rx_ring->xdp_bpf_prog))
350 		return -ENXIO;
351 
352 	qid = smp_processor_id() % adapter->xdp_num_queues;
353 	qid += adapter->xdp_first_ring;
354 	xdp_ring = &adapter->tx_ring[qid];
355 
356 	/* Other CPU ids might try to send thorugh this queue */
357 	spin_lock(&xdp_ring->xdp_tx_lock);
358 
359 	for (i = 0; i < n; i++) {
360 		if (ena_xdp_xmit_frame(xdp_ring, dev, frames[i], 0))
361 			break;
362 		nxmit++;
363 	}
364 
365 	/* Ring doorbell to make device aware of the packets */
366 	if (flags & XDP_XMIT_FLUSH)
367 		ena_ring_tx_doorbell(xdp_ring);
368 
369 	spin_unlock(&xdp_ring->xdp_tx_lock);
370 
371 	/* Return number of packets sent */
372 	return nxmit;
373 }
374 
375 static int ena_xdp_execute(struct ena_ring *rx_ring, struct xdp_buff *xdp)
376 {
377 	u32 verdict = ENA_XDP_PASS;
378 	struct bpf_prog *xdp_prog;
379 	struct ena_ring *xdp_ring;
380 	struct xdp_frame *xdpf;
381 	u64 *xdp_stat;
382 
383 	xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
384 
385 	if (!xdp_prog)
386 		goto out;
387 
388 	verdict = bpf_prog_run_xdp(xdp_prog, xdp);
389 
390 	switch (verdict) {
391 	case XDP_TX:
392 		xdpf = xdp_convert_buff_to_frame(xdp);
393 		if (unlikely(!xdpf)) {
394 			trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
395 			xdp_stat = &rx_ring->rx_stats.xdp_aborted;
396 			verdict = ENA_XDP_DROP;
397 			break;
398 		}
399 
400 		/* Find xmit queue */
401 		xdp_ring = rx_ring->xdp_ring;
402 
403 		/* The XDP queues are shared between XDP_TX and XDP_REDIRECT */
404 		spin_lock(&xdp_ring->xdp_tx_lock);
405 
406 		if (ena_xdp_xmit_frame(xdp_ring, rx_ring->netdev, xdpf,
407 				       XDP_XMIT_FLUSH))
408 			xdp_return_frame(xdpf);
409 
410 		spin_unlock(&xdp_ring->xdp_tx_lock);
411 		xdp_stat = &rx_ring->rx_stats.xdp_tx;
412 		verdict = ENA_XDP_TX;
413 		break;
414 	case XDP_REDIRECT:
415 		if (likely(!xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog))) {
416 			xdp_stat = &rx_ring->rx_stats.xdp_redirect;
417 			verdict = ENA_XDP_REDIRECT;
418 			break;
419 		}
420 		trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
421 		xdp_stat = &rx_ring->rx_stats.xdp_aborted;
422 		verdict = ENA_XDP_DROP;
423 		break;
424 	case XDP_ABORTED:
425 		trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
426 		xdp_stat = &rx_ring->rx_stats.xdp_aborted;
427 		verdict = ENA_XDP_DROP;
428 		break;
429 	case XDP_DROP:
430 		xdp_stat = &rx_ring->rx_stats.xdp_drop;
431 		verdict = ENA_XDP_DROP;
432 		break;
433 	case XDP_PASS:
434 		xdp_stat = &rx_ring->rx_stats.xdp_pass;
435 		verdict = ENA_XDP_PASS;
436 		break;
437 	default:
438 		bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, verdict);
439 		xdp_stat = &rx_ring->rx_stats.xdp_invalid;
440 		verdict = ENA_XDP_DROP;
441 	}
442 
443 	ena_increase_stat(xdp_stat, 1, &rx_ring->syncp);
444 out:
445 	return verdict;
446 }
447 
448 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
449 {
450 	adapter->xdp_first_ring = adapter->num_io_queues;
451 	adapter->xdp_num_queues = adapter->num_io_queues;
452 
453 	ena_init_io_rings(adapter,
454 			  adapter->xdp_first_ring,
455 			  adapter->xdp_num_queues);
456 }
457 
458 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
459 {
460 	int rc = 0;
461 
462 	rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
463 					     adapter->xdp_num_queues);
464 	if (rc)
465 		goto setup_err;
466 
467 	rc = ena_create_io_tx_queues_in_range(adapter,
468 					      adapter->xdp_first_ring,
469 					      adapter->xdp_num_queues);
470 	if (rc)
471 		goto create_err;
472 
473 	return 0;
474 
475 create_err:
476 	ena_free_all_io_tx_resources(adapter);
477 setup_err:
478 	return rc;
479 }
480 
481 /* Provides a way for both kernel and bpf-prog to know
482  * more about the RX-queue a given XDP frame arrived on.
483  */
484 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
485 {
486 	int rc;
487 
488 	rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid, 0);
489 
490 	if (rc) {
491 		netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
492 			  "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
493 			  rx_ring->qid, rc);
494 		goto err;
495 	}
496 
497 	rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
498 					NULL);
499 
500 	if (rc) {
501 		netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
502 			  "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
503 			  rx_ring->qid, rc);
504 		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
505 	}
506 
507 err:
508 	return rc;
509 }
510 
511 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
512 {
513 	xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
514 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
515 }
516 
517 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
518 						 struct bpf_prog *prog,
519 						 int first, int count)
520 {
521 	struct bpf_prog *old_bpf_prog;
522 	struct ena_ring *rx_ring;
523 	int i = 0;
524 
525 	for (i = first; i < count; i++) {
526 		rx_ring = &adapter->rx_ring[i];
527 		old_bpf_prog = xchg(&rx_ring->xdp_bpf_prog, prog);
528 
529 		if (!old_bpf_prog && prog) {
530 			ena_xdp_register_rxq_info(rx_ring);
531 			rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
532 		} else if (old_bpf_prog && !prog) {
533 			ena_xdp_unregister_rxq_info(rx_ring);
534 			rx_ring->rx_headroom = NET_SKB_PAD;
535 		}
536 	}
537 }
538 
539 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
540 				     struct bpf_prog *prog)
541 {
542 	struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
543 
544 	ena_xdp_exchange_program_rx_in_range(adapter,
545 					     prog,
546 					     0,
547 					     adapter->num_io_queues);
548 
549 	if (old_bpf_prog)
550 		bpf_prog_put(old_bpf_prog);
551 }
552 
553 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
554 {
555 	bool was_up;
556 	int rc;
557 
558 	was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
559 
560 	if (was_up)
561 		ena_down(adapter);
562 
563 	adapter->xdp_first_ring = 0;
564 	adapter->xdp_num_queues = 0;
565 	ena_xdp_exchange_program(adapter, NULL);
566 	if (was_up) {
567 		rc = ena_up(adapter);
568 		if (rc)
569 			return rc;
570 	}
571 	return 0;
572 }
573 
574 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
575 {
576 	struct ena_adapter *adapter = netdev_priv(netdev);
577 	struct bpf_prog *prog = bpf->prog;
578 	struct bpf_prog *old_bpf_prog;
579 	int rc, prev_mtu;
580 	bool is_up;
581 
582 	is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
583 	rc = ena_xdp_allowed(adapter);
584 	if (rc == ENA_XDP_ALLOWED) {
585 		old_bpf_prog = adapter->xdp_bpf_prog;
586 		if (prog) {
587 			if (!is_up) {
588 				ena_init_all_xdp_queues(adapter);
589 			} else if (!old_bpf_prog) {
590 				ena_down(adapter);
591 				ena_init_all_xdp_queues(adapter);
592 			}
593 			ena_xdp_exchange_program(adapter, prog);
594 
595 			if (is_up && !old_bpf_prog) {
596 				rc = ena_up(adapter);
597 				if (rc)
598 					return rc;
599 			}
600 			xdp_features_set_redirect_target(netdev, false);
601 		} else if (old_bpf_prog) {
602 			xdp_features_clear_redirect_target(netdev);
603 			rc = ena_destroy_and_free_all_xdp_queues(adapter);
604 			if (rc)
605 				return rc;
606 		}
607 
608 		prev_mtu = netdev->max_mtu;
609 		netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
610 
611 		if (!old_bpf_prog)
612 			netif_info(adapter, drv, adapter->netdev,
613 				   "XDP program is set, changing the max_mtu from %d to %d",
614 				   prev_mtu, netdev->max_mtu);
615 
616 	} else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
617 		netif_err(adapter, drv, adapter->netdev,
618 			  "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
619 			  netdev->mtu, ENA_XDP_MAX_MTU);
620 		NL_SET_ERR_MSG_MOD(bpf->extack,
621 				   "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
622 		return -EINVAL;
623 	} else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
624 		netif_err(adapter, drv, adapter->netdev,
625 			  "Failed to set xdp program, the Rx/Tx channel count should be at most half of the maximum allowed channel count. The current queue count (%d), the maximal queue count (%d)\n",
626 			  adapter->num_io_queues, adapter->max_num_io_queues);
627 		NL_SET_ERR_MSG_MOD(bpf->extack,
628 				   "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
629 		return -EINVAL;
630 	}
631 
632 	return 0;
633 }
634 
635 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
636  * program as well as to query the current xdp program id.
637  */
638 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
639 {
640 	switch (bpf->command) {
641 	case XDP_SETUP_PROG:
642 		return ena_xdp_set(netdev, bpf);
643 	default:
644 		return -EINVAL;
645 	}
646 	return 0;
647 }
648 
649 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
650 {
651 #ifdef CONFIG_RFS_ACCEL
652 	u32 i;
653 	int rc;
654 
655 	adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
656 	if (!adapter->netdev->rx_cpu_rmap)
657 		return -ENOMEM;
658 	for (i = 0; i < adapter->num_io_queues; i++) {
659 		int irq_idx = ENA_IO_IRQ_IDX(i);
660 
661 		rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
662 				      pci_irq_vector(adapter->pdev, irq_idx));
663 		if (rc) {
664 			free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
665 			adapter->netdev->rx_cpu_rmap = NULL;
666 			return rc;
667 		}
668 	}
669 #endif /* CONFIG_RFS_ACCEL */
670 	return 0;
671 }
672 
673 static void ena_init_io_rings_common(struct ena_adapter *adapter,
674 				     struct ena_ring *ring, u16 qid)
675 {
676 	ring->qid = qid;
677 	ring->pdev = adapter->pdev;
678 	ring->dev = &adapter->pdev->dev;
679 	ring->netdev = adapter->netdev;
680 	ring->napi = &adapter->ena_napi[qid].napi;
681 	ring->adapter = adapter;
682 	ring->ena_dev = adapter->ena_dev;
683 	ring->per_napi_packets = 0;
684 	ring->cpu = 0;
685 	ring->numa_node = 0;
686 	ring->no_interrupt_event_cnt = 0;
687 	u64_stats_init(&ring->syncp);
688 }
689 
690 static void ena_init_io_rings(struct ena_adapter *adapter,
691 			      int first_index, int count)
692 {
693 	struct ena_com_dev *ena_dev;
694 	struct ena_ring *txr, *rxr;
695 	int i;
696 
697 	ena_dev = adapter->ena_dev;
698 
699 	for (i = first_index; i < first_index + count; i++) {
700 		txr = &adapter->tx_ring[i];
701 		rxr = &adapter->rx_ring[i];
702 
703 		/* TX common ring state */
704 		ena_init_io_rings_common(adapter, txr, i);
705 
706 		/* TX specific ring state */
707 		txr->ring_size = adapter->requested_tx_ring_size;
708 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
709 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
710 		txr->sgl_size = adapter->max_tx_sgl_size;
711 		txr->smoothed_interval =
712 			ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
713 		txr->disable_meta_caching = adapter->disable_meta_caching;
714 		spin_lock_init(&txr->xdp_tx_lock);
715 
716 		/* Don't init RX queues for xdp queues */
717 		if (!ENA_IS_XDP_INDEX(adapter, i)) {
718 			/* RX common ring state */
719 			ena_init_io_rings_common(adapter, rxr, i);
720 
721 			/* RX specific ring state */
722 			rxr->ring_size = adapter->requested_rx_ring_size;
723 			rxr->rx_copybreak = adapter->rx_copybreak;
724 			rxr->sgl_size = adapter->max_rx_sgl_size;
725 			rxr->smoothed_interval =
726 				ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
727 			rxr->empty_rx_queue = 0;
728 			rxr->rx_headroom = NET_SKB_PAD;
729 			adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
730 			rxr->xdp_ring = &adapter->tx_ring[i + adapter->num_io_queues];
731 		}
732 	}
733 }
734 
735 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
736  * @adapter: network interface device structure
737  * @qid: queue index
738  *
739  * Return 0 on success, negative on failure
740  */
741 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
742 {
743 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
744 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
745 	int size, i, node;
746 
747 	if (tx_ring->tx_buffer_info) {
748 		netif_err(adapter, ifup,
749 			  adapter->netdev, "tx_buffer_info info is not NULL");
750 		return -EEXIST;
751 	}
752 
753 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
754 	node = cpu_to_node(ena_irq->cpu);
755 
756 	tx_ring->tx_buffer_info = vzalloc_node(size, node);
757 	if (!tx_ring->tx_buffer_info) {
758 		tx_ring->tx_buffer_info = vzalloc(size);
759 		if (!tx_ring->tx_buffer_info)
760 			goto err_tx_buffer_info;
761 	}
762 
763 	size = sizeof(u16) * tx_ring->ring_size;
764 	tx_ring->free_ids = vzalloc_node(size, node);
765 	if (!tx_ring->free_ids) {
766 		tx_ring->free_ids = vzalloc(size);
767 		if (!tx_ring->free_ids)
768 			goto err_tx_free_ids;
769 	}
770 
771 	size = tx_ring->tx_max_header_size;
772 	tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
773 	if (!tx_ring->push_buf_intermediate_buf) {
774 		tx_ring->push_buf_intermediate_buf = vzalloc(size);
775 		if (!tx_ring->push_buf_intermediate_buf)
776 			goto err_push_buf_intermediate_buf;
777 	}
778 
779 	/* Req id ring for TX out of order completions */
780 	for (i = 0; i < tx_ring->ring_size; i++)
781 		tx_ring->free_ids[i] = i;
782 
783 	/* Reset tx statistics */
784 	memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
785 
786 	tx_ring->next_to_use = 0;
787 	tx_ring->next_to_clean = 0;
788 	tx_ring->cpu = ena_irq->cpu;
789 	tx_ring->numa_node = node;
790 	return 0;
791 
792 err_push_buf_intermediate_buf:
793 	vfree(tx_ring->free_ids);
794 	tx_ring->free_ids = NULL;
795 err_tx_free_ids:
796 	vfree(tx_ring->tx_buffer_info);
797 	tx_ring->tx_buffer_info = NULL;
798 err_tx_buffer_info:
799 	return -ENOMEM;
800 }
801 
802 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
803  * @adapter: network interface device structure
804  * @qid: queue index
805  *
806  * Free all transmit software resources
807  */
808 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
809 {
810 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
811 
812 	vfree(tx_ring->tx_buffer_info);
813 	tx_ring->tx_buffer_info = NULL;
814 
815 	vfree(tx_ring->free_ids);
816 	tx_ring->free_ids = NULL;
817 
818 	vfree(tx_ring->push_buf_intermediate_buf);
819 	tx_ring->push_buf_intermediate_buf = NULL;
820 }
821 
822 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
823 					   int first_index,
824 					   int count)
825 {
826 	int i, rc = 0;
827 
828 	for (i = first_index; i < first_index + count; i++) {
829 		rc = ena_setup_tx_resources(adapter, i);
830 		if (rc)
831 			goto err_setup_tx;
832 	}
833 
834 	return 0;
835 
836 err_setup_tx:
837 
838 	netif_err(adapter, ifup, adapter->netdev,
839 		  "Tx queue %d: allocation failed\n", i);
840 
841 	/* rewind the index freeing the rings as we go */
842 	while (first_index < i--)
843 		ena_free_tx_resources(adapter, i);
844 	return rc;
845 }
846 
847 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
848 						  int first_index, int count)
849 {
850 	int i;
851 
852 	for (i = first_index; i < first_index + count; i++)
853 		ena_free_tx_resources(adapter, i);
854 }
855 
856 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
857  * @adapter: board private structure
858  *
859  * Free all transmit software resources
860  */
861 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
862 {
863 	ena_free_all_io_tx_resources_in_range(adapter,
864 					      0,
865 					      adapter->xdp_num_queues +
866 					      adapter->num_io_queues);
867 }
868 
869 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
870  * @adapter: network interface device structure
871  * @qid: queue index
872  *
873  * Returns 0 on success, negative on failure
874  */
875 static int ena_setup_rx_resources(struct ena_adapter *adapter,
876 				  u32 qid)
877 {
878 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
879 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
880 	int size, node, i;
881 
882 	if (rx_ring->rx_buffer_info) {
883 		netif_err(adapter, ifup, adapter->netdev,
884 			  "rx_buffer_info is not NULL");
885 		return -EEXIST;
886 	}
887 
888 	/* alloc extra element so in rx path
889 	 * we can always prefetch rx_info + 1
890 	 */
891 	size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
892 	node = cpu_to_node(ena_irq->cpu);
893 
894 	rx_ring->rx_buffer_info = vzalloc_node(size, node);
895 	if (!rx_ring->rx_buffer_info) {
896 		rx_ring->rx_buffer_info = vzalloc(size);
897 		if (!rx_ring->rx_buffer_info)
898 			return -ENOMEM;
899 	}
900 
901 	size = sizeof(u16) * rx_ring->ring_size;
902 	rx_ring->free_ids = vzalloc_node(size, node);
903 	if (!rx_ring->free_ids) {
904 		rx_ring->free_ids = vzalloc(size);
905 		if (!rx_ring->free_ids) {
906 			vfree(rx_ring->rx_buffer_info);
907 			rx_ring->rx_buffer_info = NULL;
908 			return -ENOMEM;
909 		}
910 	}
911 
912 	/* Req id ring for receiving RX pkts out of order */
913 	for (i = 0; i < rx_ring->ring_size; i++)
914 		rx_ring->free_ids[i] = i;
915 
916 	/* Reset rx statistics */
917 	memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
918 
919 	rx_ring->next_to_clean = 0;
920 	rx_ring->next_to_use = 0;
921 	rx_ring->cpu = ena_irq->cpu;
922 	rx_ring->numa_node = node;
923 
924 	return 0;
925 }
926 
927 /* ena_free_rx_resources - Free I/O Rx Resources
928  * @adapter: network interface device structure
929  * @qid: queue index
930  *
931  * Free all receive software resources
932  */
933 static void ena_free_rx_resources(struct ena_adapter *adapter,
934 				  u32 qid)
935 {
936 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
937 
938 	vfree(rx_ring->rx_buffer_info);
939 	rx_ring->rx_buffer_info = NULL;
940 
941 	vfree(rx_ring->free_ids);
942 	rx_ring->free_ids = NULL;
943 }
944 
945 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
946  * @adapter: board private structure
947  *
948  * Return 0 on success, negative on failure
949  */
950 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
951 {
952 	int i, rc = 0;
953 
954 	for (i = 0; i < adapter->num_io_queues; i++) {
955 		rc = ena_setup_rx_resources(adapter, i);
956 		if (rc)
957 			goto err_setup_rx;
958 	}
959 
960 	return 0;
961 
962 err_setup_rx:
963 
964 	netif_err(adapter, ifup, adapter->netdev,
965 		  "Rx queue %d: allocation failed\n", i);
966 
967 	/* rewind the index freeing the rings as we go */
968 	while (i--)
969 		ena_free_rx_resources(adapter, i);
970 	return rc;
971 }
972 
973 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
974  * @adapter: board private structure
975  *
976  * Free all receive software resources
977  */
978 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
979 {
980 	int i;
981 
982 	for (i = 0; i < adapter->num_io_queues; i++)
983 		ena_free_rx_resources(adapter, i);
984 }
985 
986 static struct page *ena_alloc_map_page(struct ena_ring *rx_ring,
987 				       dma_addr_t *dma)
988 {
989 	struct page *page;
990 
991 	/* This would allocate the page on the same NUMA node the executing code
992 	 * is running on.
993 	 */
994 	page = dev_alloc_page();
995 	if (!page) {
996 		ena_increase_stat(&rx_ring->rx_stats.page_alloc_fail, 1,
997 				  &rx_ring->syncp);
998 		return ERR_PTR(-ENOSPC);
999 	}
1000 
1001 	/* To enable NIC-side port-mirroring, AKA SPAN port,
1002 	 * we make the buffer readable from the nic as well
1003 	 */
1004 	*dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
1005 			    DMA_BIDIRECTIONAL);
1006 	if (unlikely(dma_mapping_error(rx_ring->dev, *dma))) {
1007 		ena_increase_stat(&rx_ring->rx_stats.dma_mapping_err, 1,
1008 				  &rx_ring->syncp);
1009 		__free_page(page);
1010 		return ERR_PTR(-EIO);
1011 	}
1012 
1013 	return page;
1014 }
1015 
1016 static int ena_alloc_rx_buffer(struct ena_ring *rx_ring,
1017 			       struct ena_rx_buffer *rx_info)
1018 {
1019 	int headroom = rx_ring->rx_headroom;
1020 	struct ena_com_buf *ena_buf;
1021 	struct page *page;
1022 	dma_addr_t dma;
1023 	int tailroom;
1024 
1025 	/* restore page offset value in case it has been changed by device */
1026 	rx_info->page_offset = headroom;
1027 
1028 	/* if previous allocated page is not used */
1029 	if (unlikely(rx_info->page))
1030 		return 0;
1031 
1032 	/* We handle DMA here */
1033 	page = ena_alloc_map_page(rx_ring, &dma);
1034 	if (unlikely(IS_ERR(page)))
1035 		return PTR_ERR(page);
1036 
1037 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1038 		  "Allocate page %p, rx_info %p\n", page, rx_info);
1039 
1040 	tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1041 
1042 	rx_info->page = page;
1043 	ena_buf = &rx_info->ena_buf;
1044 	ena_buf->paddr = dma + headroom;
1045 	ena_buf->len = ENA_PAGE_SIZE - headroom - tailroom;
1046 
1047 	return 0;
1048 }
1049 
1050 static void ena_unmap_rx_buff(struct ena_ring *rx_ring,
1051 			      struct ena_rx_buffer *rx_info)
1052 {
1053 	struct ena_com_buf *ena_buf = &rx_info->ena_buf;
1054 
1055 	dma_unmap_page(rx_ring->dev, ena_buf->paddr - rx_ring->rx_headroom,
1056 		       ENA_PAGE_SIZE,
1057 		       DMA_BIDIRECTIONAL);
1058 }
1059 
1060 static void ena_free_rx_page(struct ena_ring *rx_ring,
1061 			     struct ena_rx_buffer *rx_info)
1062 {
1063 	struct page *page = rx_info->page;
1064 
1065 	if (unlikely(!page)) {
1066 		netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1067 			   "Trying to free unallocated buffer\n");
1068 		return;
1069 	}
1070 
1071 	ena_unmap_rx_buff(rx_ring, rx_info);
1072 
1073 	__free_page(page);
1074 	rx_info->page = NULL;
1075 }
1076 
1077 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1078 {
1079 	u16 next_to_use, req_id;
1080 	u32 i;
1081 	int rc;
1082 
1083 	next_to_use = rx_ring->next_to_use;
1084 
1085 	for (i = 0; i < num; i++) {
1086 		struct ena_rx_buffer *rx_info;
1087 
1088 		req_id = rx_ring->free_ids[next_to_use];
1089 
1090 		rx_info = &rx_ring->rx_buffer_info[req_id];
1091 
1092 		rc = ena_alloc_rx_buffer(rx_ring, rx_info);
1093 		if (unlikely(rc < 0)) {
1094 			netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1095 				   "Failed to allocate buffer for rx queue %d\n",
1096 				   rx_ring->qid);
1097 			break;
1098 		}
1099 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1100 						&rx_info->ena_buf,
1101 						req_id);
1102 		if (unlikely(rc)) {
1103 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1104 				   "Failed to add buffer for rx queue %d\n",
1105 				   rx_ring->qid);
1106 			break;
1107 		}
1108 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1109 						   rx_ring->ring_size);
1110 	}
1111 
1112 	if (unlikely(i < num)) {
1113 		ena_increase_stat(&rx_ring->rx_stats.refil_partial, 1,
1114 				  &rx_ring->syncp);
1115 		netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1116 			   "Refilled rx qid %d with only %d buffers (from %d)\n",
1117 			   rx_ring->qid, i, num);
1118 	}
1119 
1120 	/* ena_com_write_sq_doorbell issues a wmb() */
1121 	if (likely(i))
1122 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1123 
1124 	rx_ring->next_to_use = next_to_use;
1125 
1126 	return i;
1127 }
1128 
1129 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1130 			     u32 qid)
1131 {
1132 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1133 	u32 i;
1134 
1135 	for (i = 0; i < rx_ring->ring_size; i++) {
1136 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1137 
1138 		if (rx_info->page)
1139 			ena_free_rx_page(rx_ring, rx_info);
1140 	}
1141 }
1142 
1143 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1144  * @adapter: board private structure
1145  */
1146 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1147 {
1148 	struct ena_ring *rx_ring;
1149 	int i, rc, bufs_num;
1150 
1151 	for (i = 0; i < adapter->num_io_queues; i++) {
1152 		rx_ring = &adapter->rx_ring[i];
1153 		bufs_num = rx_ring->ring_size - 1;
1154 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1155 
1156 		if (unlikely(rc != bufs_num))
1157 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1158 				   "Refilling Queue %d failed. allocated %d buffers from: %d\n",
1159 				   i, rc, bufs_num);
1160 	}
1161 }
1162 
1163 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1164 {
1165 	int i;
1166 
1167 	for (i = 0; i < adapter->num_io_queues; i++)
1168 		ena_free_rx_bufs(adapter, i);
1169 }
1170 
1171 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1172 			      struct ena_tx_buffer *tx_info)
1173 {
1174 	struct ena_com_buf *ena_buf;
1175 	u32 cnt;
1176 	int i;
1177 
1178 	ena_buf = tx_info->bufs;
1179 	cnt = tx_info->num_of_bufs;
1180 
1181 	if (unlikely(!cnt))
1182 		return;
1183 
1184 	if (tx_info->map_linear_data) {
1185 		dma_unmap_single(tx_ring->dev,
1186 				 dma_unmap_addr(ena_buf, paddr),
1187 				 dma_unmap_len(ena_buf, len),
1188 				 DMA_TO_DEVICE);
1189 		ena_buf++;
1190 		cnt--;
1191 	}
1192 
1193 	/* unmap remaining mapped pages */
1194 	for (i = 0; i < cnt; i++) {
1195 		dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1196 			       dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1197 		ena_buf++;
1198 	}
1199 }
1200 
1201 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1202  * @tx_ring: TX ring for which buffers be freed
1203  */
1204 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1205 {
1206 	bool print_once = true;
1207 	u32 i;
1208 
1209 	for (i = 0; i < tx_ring->ring_size; i++) {
1210 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1211 
1212 		if (!tx_info->skb)
1213 			continue;
1214 
1215 		if (print_once) {
1216 			netif_notice(tx_ring->adapter, ifdown, tx_ring->netdev,
1217 				     "Free uncompleted tx skb qid %d idx 0x%x\n",
1218 				     tx_ring->qid, i);
1219 			print_once = false;
1220 		} else {
1221 			netif_dbg(tx_ring->adapter, ifdown, tx_ring->netdev,
1222 				  "Free uncompleted tx skb qid %d idx 0x%x\n",
1223 				  tx_ring->qid, i);
1224 		}
1225 
1226 		ena_unmap_tx_buff(tx_ring, tx_info);
1227 
1228 		dev_kfree_skb_any(tx_info->skb);
1229 	}
1230 	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1231 						  tx_ring->qid));
1232 }
1233 
1234 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1235 {
1236 	struct ena_ring *tx_ring;
1237 	int i;
1238 
1239 	for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1240 		tx_ring = &adapter->tx_ring[i];
1241 		ena_free_tx_bufs(tx_ring);
1242 	}
1243 }
1244 
1245 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1246 {
1247 	u16 ena_qid;
1248 	int i;
1249 
1250 	for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1251 		ena_qid = ENA_IO_TXQ_IDX(i);
1252 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1253 	}
1254 }
1255 
1256 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1257 {
1258 	u16 ena_qid;
1259 	int i;
1260 
1261 	for (i = 0; i < adapter->num_io_queues; i++) {
1262 		ena_qid = ENA_IO_RXQ_IDX(i);
1263 		cancel_work_sync(&adapter->ena_napi[i].dim.work);
1264 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1265 	}
1266 }
1267 
1268 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1269 {
1270 	ena_destroy_all_tx_queues(adapter);
1271 	ena_destroy_all_rx_queues(adapter);
1272 }
1273 
1274 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1275 				 struct ena_tx_buffer *tx_info, bool is_xdp)
1276 {
1277 	if (tx_info)
1278 		netif_err(ring->adapter,
1279 			  tx_done,
1280 			  ring->netdev,
1281 			  "tx_info doesn't have valid %s. qid %u req_id %u",
1282 			   is_xdp ? "xdp frame" : "skb", ring->qid, req_id);
1283 	else
1284 		netif_err(ring->adapter,
1285 			  tx_done,
1286 			  ring->netdev,
1287 			  "Invalid req_id %u in qid %u\n",
1288 			  req_id, ring->qid);
1289 
1290 	ena_increase_stat(&ring->tx_stats.bad_req_id, 1, &ring->syncp);
1291 	ena_reset_device(ring->adapter, ENA_REGS_RESET_INV_TX_REQ_ID);
1292 
1293 	return -EFAULT;
1294 }
1295 
1296 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1297 {
1298 	struct ena_tx_buffer *tx_info;
1299 
1300 	tx_info = &tx_ring->tx_buffer_info[req_id];
1301 	if (likely(tx_info->skb))
1302 		return 0;
1303 
1304 	return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1305 }
1306 
1307 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1308 {
1309 	struct ena_tx_buffer *tx_info;
1310 
1311 	tx_info = &xdp_ring->tx_buffer_info[req_id];
1312 	if (likely(tx_info->xdpf))
1313 		return 0;
1314 
1315 	return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1316 }
1317 
1318 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1319 {
1320 	struct netdev_queue *txq;
1321 	bool above_thresh;
1322 	u32 tx_bytes = 0;
1323 	u32 total_done = 0;
1324 	u16 next_to_clean;
1325 	u16 req_id;
1326 	int tx_pkts = 0;
1327 	int rc;
1328 
1329 	next_to_clean = tx_ring->next_to_clean;
1330 	txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1331 
1332 	while (tx_pkts < budget) {
1333 		struct ena_tx_buffer *tx_info;
1334 		struct sk_buff *skb;
1335 
1336 		rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1337 						&req_id);
1338 		if (rc) {
1339 			if (unlikely(rc == -EINVAL))
1340 				handle_invalid_req_id(tx_ring, req_id, NULL,
1341 						      false);
1342 			break;
1343 		}
1344 
1345 		/* validate that the request id points to a valid skb */
1346 		rc = validate_tx_req_id(tx_ring, req_id);
1347 		if (rc)
1348 			break;
1349 
1350 		tx_info = &tx_ring->tx_buffer_info[req_id];
1351 		skb = tx_info->skb;
1352 
1353 		/* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1354 		prefetch(&skb->end);
1355 
1356 		tx_info->skb = NULL;
1357 		tx_info->last_jiffies = 0;
1358 
1359 		ena_unmap_tx_buff(tx_ring, tx_info);
1360 
1361 		netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1362 			  "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1363 			  skb);
1364 
1365 		tx_bytes += skb->len;
1366 		dev_kfree_skb(skb);
1367 		tx_pkts++;
1368 		total_done += tx_info->tx_descs;
1369 
1370 		tx_ring->free_ids[next_to_clean] = req_id;
1371 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1372 						     tx_ring->ring_size);
1373 	}
1374 
1375 	tx_ring->next_to_clean = next_to_clean;
1376 	ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1377 	ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1378 
1379 	netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1380 
1381 	netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1382 		  "tx_poll: q %d done. total pkts: %d\n",
1383 		  tx_ring->qid, tx_pkts);
1384 
1385 	/* need to make the rings circular update visible to
1386 	 * ena_start_xmit() before checking for netif_queue_stopped().
1387 	 */
1388 	smp_mb();
1389 
1390 	above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1391 						    ENA_TX_WAKEUP_THRESH);
1392 	if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1393 		__netif_tx_lock(txq, smp_processor_id());
1394 		above_thresh =
1395 			ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1396 						     ENA_TX_WAKEUP_THRESH);
1397 		if (netif_tx_queue_stopped(txq) && above_thresh &&
1398 		    test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1399 			netif_tx_wake_queue(txq);
1400 			ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1401 					  &tx_ring->syncp);
1402 		}
1403 		__netif_tx_unlock(txq);
1404 	}
1405 
1406 	return tx_pkts;
1407 }
1408 
1409 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, void *first_frag)
1410 {
1411 	struct sk_buff *skb;
1412 
1413 	if (!first_frag)
1414 		skb = napi_alloc_skb(rx_ring->napi, rx_ring->rx_copybreak);
1415 	else
1416 		skb = napi_build_skb(first_frag, ENA_PAGE_SIZE);
1417 
1418 	if (unlikely(!skb)) {
1419 		ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1420 				  &rx_ring->syncp);
1421 
1422 		netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1423 			  "Failed to allocate skb. first_frag %s\n",
1424 			  first_frag ? "provided" : "not provided");
1425 		return NULL;
1426 	}
1427 
1428 	return skb;
1429 }
1430 
1431 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1432 				  struct ena_com_rx_buf_info *ena_bufs,
1433 				  u32 descs,
1434 				  u16 *next_to_clean)
1435 {
1436 	struct ena_rx_buffer *rx_info;
1437 	struct ena_adapter *adapter;
1438 	u16 len, req_id, buf = 0;
1439 	struct sk_buff *skb;
1440 	void *page_addr;
1441 	u32 page_offset;
1442 	void *data_addr;
1443 
1444 	len = ena_bufs[buf].len;
1445 	req_id = ena_bufs[buf].req_id;
1446 
1447 	rx_info = &rx_ring->rx_buffer_info[req_id];
1448 
1449 	if (unlikely(!rx_info->page)) {
1450 		adapter = rx_ring->adapter;
1451 		netif_err(adapter, rx_err, rx_ring->netdev,
1452 			  "Page is NULL. qid %u req_id %u\n", rx_ring->qid, req_id);
1453 		ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1, &rx_ring->syncp);
1454 		ena_reset_device(adapter, ENA_REGS_RESET_INV_RX_REQ_ID);
1455 		return NULL;
1456 	}
1457 
1458 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1459 		  "rx_info %p page %p\n",
1460 		  rx_info, rx_info->page);
1461 
1462 	/* save virt address of first buffer */
1463 	page_addr = page_address(rx_info->page);
1464 	page_offset = rx_info->page_offset;
1465 	data_addr = page_addr + page_offset;
1466 
1467 	prefetch(data_addr);
1468 
1469 	if (len <= rx_ring->rx_copybreak) {
1470 		skb = ena_alloc_skb(rx_ring, NULL);
1471 		if (unlikely(!skb))
1472 			return NULL;
1473 
1474 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1475 			  "RX allocated small packet. len %d. data_len %d\n",
1476 			  skb->len, skb->data_len);
1477 
1478 		/* sync this buffer for CPU use */
1479 		dma_sync_single_for_cpu(rx_ring->dev,
1480 					dma_unmap_addr(&rx_info->ena_buf, paddr),
1481 					len,
1482 					DMA_FROM_DEVICE);
1483 		skb_copy_to_linear_data(skb, data_addr, len);
1484 		dma_sync_single_for_device(rx_ring->dev,
1485 					   dma_unmap_addr(&rx_info->ena_buf, paddr),
1486 					   len,
1487 					   DMA_FROM_DEVICE);
1488 
1489 		skb_put(skb, len);
1490 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1491 		rx_ring->free_ids[*next_to_clean] = req_id;
1492 		*next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1493 						     rx_ring->ring_size);
1494 		return skb;
1495 	}
1496 
1497 	ena_unmap_rx_buff(rx_ring, rx_info);
1498 
1499 	skb = ena_alloc_skb(rx_ring, page_addr);
1500 	if (unlikely(!skb))
1501 		return NULL;
1502 
1503 	/* Populate skb's linear part */
1504 	skb_reserve(skb, page_offset);
1505 	skb_put(skb, len);
1506 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1507 
1508 	do {
1509 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1510 			  "RX skb updated. len %d. data_len %d\n",
1511 			  skb->len, skb->data_len);
1512 
1513 		rx_info->page = NULL;
1514 
1515 		rx_ring->free_ids[*next_to_clean] = req_id;
1516 		*next_to_clean =
1517 			ENA_RX_RING_IDX_NEXT(*next_to_clean,
1518 					     rx_ring->ring_size);
1519 		if (likely(--descs == 0))
1520 			break;
1521 
1522 		buf++;
1523 		len = ena_bufs[buf].len;
1524 		req_id = ena_bufs[buf].req_id;
1525 
1526 		rx_info = &rx_ring->rx_buffer_info[req_id];
1527 
1528 		ena_unmap_rx_buff(rx_ring, rx_info);
1529 
1530 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1531 				rx_info->page_offset, len, ENA_PAGE_SIZE);
1532 
1533 	} while (1);
1534 
1535 	return skb;
1536 }
1537 
1538 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1539  * @adapter: structure containing adapter specific data
1540  * @ena_rx_ctx: received packet context/metadata
1541  * @skb: skb currently being received and modified
1542  */
1543 static void ena_rx_checksum(struct ena_ring *rx_ring,
1544 				   struct ena_com_rx_ctx *ena_rx_ctx,
1545 				   struct sk_buff *skb)
1546 {
1547 	/* Rx csum disabled */
1548 	if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1549 		skb->ip_summed = CHECKSUM_NONE;
1550 		return;
1551 	}
1552 
1553 	/* For fragmented packets the checksum isn't valid */
1554 	if (ena_rx_ctx->frag) {
1555 		skb->ip_summed = CHECKSUM_NONE;
1556 		return;
1557 	}
1558 
1559 	/* if IP and error */
1560 	if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1561 		     (ena_rx_ctx->l3_csum_err))) {
1562 		/* ipv4 checksum error */
1563 		skb->ip_summed = CHECKSUM_NONE;
1564 		ena_increase_stat(&rx_ring->rx_stats.csum_bad, 1,
1565 				  &rx_ring->syncp);
1566 		netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1567 			  "RX IPv4 header checksum error\n");
1568 		return;
1569 	}
1570 
1571 	/* if TCP/UDP */
1572 	if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1573 		   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1574 		if (unlikely(ena_rx_ctx->l4_csum_err)) {
1575 			/* TCP/UDP checksum error */
1576 			ena_increase_stat(&rx_ring->rx_stats.csum_bad, 1,
1577 					  &rx_ring->syncp);
1578 			netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1579 				  "RX L4 checksum error\n");
1580 			skb->ip_summed = CHECKSUM_NONE;
1581 			return;
1582 		}
1583 
1584 		if (likely(ena_rx_ctx->l4_csum_checked)) {
1585 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1586 			ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1587 					  &rx_ring->syncp);
1588 		} else {
1589 			ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1590 					  &rx_ring->syncp);
1591 			skb->ip_summed = CHECKSUM_NONE;
1592 		}
1593 	} else {
1594 		skb->ip_summed = CHECKSUM_NONE;
1595 		return;
1596 	}
1597 
1598 }
1599 
1600 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1601 			    struct ena_com_rx_ctx *ena_rx_ctx,
1602 			    struct sk_buff *skb)
1603 {
1604 	enum pkt_hash_types hash_type;
1605 
1606 	if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1607 		if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1608 			   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1609 
1610 			hash_type = PKT_HASH_TYPE_L4;
1611 		else
1612 			hash_type = PKT_HASH_TYPE_NONE;
1613 
1614 		/* Override hash type if the packet is fragmented */
1615 		if (ena_rx_ctx->frag)
1616 			hash_type = PKT_HASH_TYPE_NONE;
1617 
1618 		skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1619 	}
1620 }
1621 
1622 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1623 {
1624 	struct ena_rx_buffer *rx_info;
1625 	int ret;
1626 
1627 	rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1628 	xdp_prepare_buff(xdp, page_address(rx_info->page),
1629 			 rx_info->page_offset,
1630 			 rx_ring->ena_bufs[0].len, false);
1631 	/* If for some reason we received a bigger packet than
1632 	 * we expect, then we simply drop it
1633 	 */
1634 	if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1635 		return ENA_XDP_DROP;
1636 
1637 	ret = ena_xdp_execute(rx_ring, xdp);
1638 
1639 	/* The xdp program might expand the headers */
1640 	if (ret == ENA_XDP_PASS) {
1641 		rx_info->page_offset = xdp->data - xdp->data_hard_start;
1642 		rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1643 	}
1644 
1645 	return ret;
1646 }
1647 /* ena_clean_rx_irq - Cleanup RX irq
1648  * @rx_ring: RX ring to clean
1649  * @napi: napi handler
1650  * @budget: how many packets driver is allowed to clean
1651  *
1652  * Returns the number of cleaned buffers.
1653  */
1654 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1655 			    u32 budget)
1656 {
1657 	u16 next_to_clean = rx_ring->next_to_clean;
1658 	struct ena_com_rx_ctx ena_rx_ctx;
1659 	struct ena_rx_buffer *rx_info;
1660 	struct ena_adapter *adapter;
1661 	u32 res_budget, work_done;
1662 	int rx_copybreak_pkt = 0;
1663 	int refill_threshold;
1664 	struct sk_buff *skb;
1665 	int refill_required;
1666 	struct xdp_buff xdp;
1667 	int xdp_flags = 0;
1668 	int total_len = 0;
1669 	int xdp_verdict;
1670 	int rc = 0;
1671 	int i;
1672 
1673 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1674 		  "%s qid %d\n", __func__, rx_ring->qid);
1675 	res_budget = budget;
1676 	xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1677 
1678 	do {
1679 		xdp_verdict = ENA_XDP_PASS;
1680 		skb = NULL;
1681 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1682 		ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1683 		ena_rx_ctx.descs = 0;
1684 		ena_rx_ctx.pkt_offset = 0;
1685 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1686 				    rx_ring->ena_com_io_sq,
1687 				    &ena_rx_ctx);
1688 		if (unlikely(rc))
1689 			goto error;
1690 
1691 		if (unlikely(ena_rx_ctx.descs == 0))
1692 			break;
1693 
1694 		/* First descriptor might have an offset set by the device */
1695 		rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1696 		rx_info->page_offset += ena_rx_ctx.pkt_offset;
1697 
1698 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1699 			  "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1700 			  rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1701 			  ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1702 
1703 		if (ena_xdp_present_ring(rx_ring))
1704 			xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1705 
1706 		/* allocate skb and fill it */
1707 		if (xdp_verdict == ENA_XDP_PASS)
1708 			skb = ena_rx_skb(rx_ring,
1709 					 rx_ring->ena_bufs,
1710 					 ena_rx_ctx.descs,
1711 					 &next_to_clean);
1712 
1713 		if (unlikely(!skb)) {
1714 			for (i = 0; i < ena_rx_ctx.descs; i++) {
1715 				int req_id = rx_ring->ena_bufs[i].req_id;
1716 
1717 				rx_ring->free_ids[next_to_clean] = req_id;
1718 				next_to_clean =
1719 					ENA_RX_RING_IDX_NEXT(next_to_clean,
1720 							     rx_ring->ring_size);
1721 
1722 				/* Packets was passed for transmission, unmap it
1723 				 * from RX side.
1724 				 */
1725 				if (xdp_verdict & ENA_XDP_FORWARDED) {
1726 					ena_unmap_rx_buff(rx_ring,
1727 							  &rx_ring->rx_buffer_info[req_id]);
1728 					rx_ring->rx_buffer_info[req_id].page = NULL;
1729 				}
1730 			}
1731 			if (xdp_verdict != ENA_XDP_PASS) {
1732 				xdp_flags |= xdp_verdict;
1733 				total_len += ena_rx_ctx.ena_bufs[0].len;
1734 				res_budget--;
1735 				continue;
1736 			}
1737 			break;
1738 		}
1739 
1740 		ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1741 
1742 		ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1743 
1744 		skb_record_rx_queue(skb, rx_ring->qid);
1745 
1746 		if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak)
1747 			rx_copybreak_pkt++;
1748 
1749 		total_len += skb->len;
1750 
1751 		napi_gro_receive(napi, skb);
1752 
1753 		res_budget--;
1754 	} while (likely(res_budget));
1755 
1756 	work_done = budget - res_budget;
1757 	rx_ring->per_napi_packets += work_done;
1758 	u64_stats_update_begin(&rx_ring->syncp);
1759 	rx_ring->rx_stats.bytes += total_len;
1760 	rx_ring->rx_stats.cnt += work_done;
1761 	rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1762 	u64_stats_update_end(&rx_ring->syncp);
1763 
1764 	rx_ring->next_to_clean = next_to_clean;
1765 
1766 	refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1767 	refill_threshold =
1768 		min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1769 		      ENA_RX_REFILL_THRESH_PACKET);
1770 
1771 	/* Optimization, try to batch new rx buffers */
1772 	if (refill_required > refill_threshold) {
1773 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1774 		ena_refill_rx_bufs(rx_ring, refill_required);
1775 	}
1776 
1777 	if (xdp_flags & ENA_XDP_REDIRECT)
1778 		xdp_do_flush_map();
1779 
1780 	return work_done;
1781 
1782 error:
1783 	adapter = netdev_priv(rx_ring->netdev);
1784 
1785 	if (rc == -ENOSPC) {
1786 		ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1787 				  &rx_ring->syncp);
1788 		ena_reset_device(adapter, ENA_REGS_RESET_TOO_MANY_RX_DESCS);
1789 	} else {
1790 		ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1791 				  &rx_ring->syncp);
1792 		ena_reset_device(adapter, ENA_REGS_RESET_INV_RX_REQ_ID);
1793 	}
1794 	return 0;
1795 }
1796 
1797 static void ena_dim_work(struct work_struct *w)
1798 {
1799 	struct dim *dim = container_of(w, struct dim, work);
1800 	struct dim_cq_moder cur_moder =
1801 		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1802 	struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1803 
1804 	ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1805 	dim->state = DIM_START_MEASURE;
1806 }
1807 
1808 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1809 {
1810 	struct dim_sample dim_sample;
1811 	struct ena_ring *rx_ring = ena_napi->rx_ring;
1812 
1813 	if (!rx_ring->per_napi_packets)
1814 		return;
1815 
1816 	rx_ring->non_empty_napi_events++;
1817 
1818 	dim_update_sample(rx_ring->non_empty_napi_events,
1819 			  rx_ring->rx_stats.cnt,
1820 			  rx_ring->rx_stats.bytes,
1821 			  &dim_sample);
1822 
1823 	net_dim(&ena_napi->dim, dim_sample);
1824 
1825 	rx_ring->per_napi_packets = 0;
1826 }
1827 
1828 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1829 					struct ena_ring *rx_ring)
1830 {
1831 	u32 rx_interval = tx_ring->smoothed_interval;
1832 	struct ena_eth_io_intr_reg intr_reg;
1833 
1834 	/* Rx ring can be NULL when for XDP tx queues which don't have an
1835 	 * accompanying rx_ring pair.
1836 	 */
1837 	if (rx_ring)
1838 		rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1839 			rx_ring->smoothed_interval :
1840 			ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1841 
1842 	/* Update intr register: rx intr delay,
1843 	 * tx intr delay and interrupt unmask
1844 	 */
1845 	ena_com_update_intr_reg(&intr_reg,
1846 				rx_interval,
1847 				tx_ring->smoothed_interval,
1848 				true);
1849 
1850 	ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1851 			  &tx_ring->syncp);
1852 
1853 	/* It is a shared MSI-X.
1854 	 * Tx and Rx CQ have pointer to it.
1855 	 * So we use one of them to reach the intr reg
1856 	 * The Tx ring is used because the rx_ring is NULL for XDP queues
1857 	 */
1858 	ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1859 }
1860 
1861 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1862 					     struct ena_ring *rx_ring)
1863 {
1864 	int cpu = get_cpu();
1865 	int numa_node;
1866 
1867 	/* Check only one ring since the 2 rings are running on the same cpu */
1868 	if (likely(tx_ring->cpu == cpu))
1869 		goto out;
1870 
1871 	tx_ring->cpu = cpu;
1872 	if (rx_ring)
1873 		rx_ring->cpu = cpu;
1874 
1875 	numa_node = cpu_to_node(cpu);
1876 
1877 	if (likely(tx_ring->numa_node == numa_node))
1878 		goto out;
1879 
1880 	put_cpu();
1881 
1882 	if (numa_node != NUMA_NO_NODE) {
1883 		ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1884 		tx_ring->numa_node = numa_node;
1885 		if (rx_ring) {
1886 			rx_ring->numa_node = numa_node;
1887 			ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1888 						 numa_node);
1889 		}
1890 	}
1891 
1892 	return;
1893 out:
1894 	put_cpu();
1895 }
1896 
1897 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1898 {
1899 	u32 total_done = 0;
1900 	u16 next_to_clean;
1901 	int tx_pkts = 0;
1902 	u16 req_id;
1903 	int rc;
1904 
1905 	if (unlikely(!xdp_ring))
1906 		return 0;
1907 	next_to_clean = xdp_ring->next_to_clean;
1908 
1909 	while (tx_pkts < budget) {
1910 		struct ena_tx_buffer *tx_info;
1911 		struct xdp_frame *xdpf;
1912 
1913 		rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1914 						&req_id);
1915 		if (rc) {
1916 			if (unlikely(rc == -EINVAL))
1917 				handle_invalid_req_id(xdp_ring, req_id, NULL,
1918 						      true);
1919 			break;
1920 		}
1921 
1922 		/* validate that the request id points to a valid xdp_frame */
1923 		rc = validate_xdp_req_id(xdp_ring, req_id);
1924 		if (rc)
1925 			break;
1926 
1927 		tx_info = &xdp_ring->tx_buffer_info[req_id];
1928 		xdpf = tx_info->xdpf;
1929 
1930 		tx_info->xdpf = NULL;
1931 		tx_info->last_jiffies = 0;
1932 		ena_unmap_tx_buff(xdp_ring, tx_info);
1933 
1934 		netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1935 			  "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1936 			  xdpf);
1937 
1938 		tx_pkts++;
1939 		total_done += tx_info->tx_descs;
1940 
1941 		xdp_return_frame(xdpf);
1942 		xdp_ring->free_ids[next_to_clean] = req_id;
1943 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1944 						     xdp_ring->ring_size);
1945 	}
1946 
1947 	xdp_ring->next_to_clean = next_to_clean;
1948 	ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1949 	ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1950 
1951 	netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1952 		  "tx_poll: q %d done. total pkts: %d\n",
1953 		  xdp_ring->qid, tx_pkts);
1954 
1955 	return tx_pkts;
1956 }
1957 
1958 static int ena_io_poll(struct napi_struct *napi, int budget)
1959 {
1960 	struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1961 	struct ena_ring *tx_ring, *rx_ring;
1962 	int tx_work_done;
1963 	int rx_work_done = 0;
1964 	int tx_budget;
1965 	int napi_comp_call = 0;
1966 	int ret;
1967 
1968 	tx_ring = ena_napi->tx_ring;
1969 	rx_ring = ena_napi->rx_ring;
1970 
1971 	tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1972 
1973 	if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1974 	    test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1975 		napi_complete_done(napi, 0);
1976 		return 0;
1977 	}
1978 
1979 	tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1980 	/* On netpoll the budget is zero and the handler should only clean the
1981 	 * tx completions.
1982 	 */
1983 	if (likely(budget))
1984 		rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1985 
1986 	/* If the device is about to reset or down, avoid unmask
1987 	 * the interrupt and return 0 so NAPI won't reschedule
1988 	 */
1989 	if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1990 		     test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1991 		napi_complete_done(napi, 0);
1992 		ret = 0;
1993 
1994 	} else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1995 		napi_comp_call = 1;
1996 
1997 		/* Update numa and unmask the interrupt only when schedule
1998 		 * from the interrupt context (vs from sk_busy_loop)
1999 		 */
2000 		if (napi_complete_done(napi, rx_work_done) &&
2001 		    READ_ONCE(ena_napi->interrupts_masked)) {
2002 			smp_rmb(); /* make sure interrupts_masked is read */
2003 			WRITE_ONCE(ena_napi->interrupts_masked, false);
2004 			/* We apply adaptive moderation on Rx path only.
2005 			 * Tx uses static interrupt moderation.
2006 			 */
2007 			if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
2008 				ena_adjust_adaptive_rx_intr_moderation(ena_napi);
2009 
2010 			ena_update_ring_numa_node(tx_ring, rx_ring);
2011 			ena_unmask_interrupt(tx_ring, rx_ring);
2012 		}
2013 
2014 		ret = rx_work_done;
2015 	} else {
2016 		ret = budget;
2017 	}
2018 
2019 	u64_stats_update_begin(&tx_ring->syncp);
2020 	tx_ring->tx_stats.napi_comp += napi_comp_call;
2021 	tx_ring->tx_stats.tx_poll++;
2022 	u64_stats_update_end(&tx_ring->syncp);
2023 
2024 	tx_ring->tx_stats.last_napi_jiffies = jiffies;
2025 
2026 	return ret;
2027 }
2028 
2029 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
2030 {
2031 	struct ena_adapter *adapter = (struct ena_adapter *)data;
2032 
2033 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
2034 
2035 	/* Don't call the aenq handler before probe is done */
2036 	if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
2037 		ena_com_aenq_intr_handler(adapter->ena_dev, data);
2038 
2039 	return IRQ_HANDLED;
2040 }
2041 
2042 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
2043  * @irq: interrupt number
2044  * @data: pointer to a network interface private napi device structure
2045  */
2046 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2047 {
2048 	struct ena_napi *ena_napi = data;
2049 
2050 	/* Used to check HW health */
2051 	WRITE_ONCE(ena_napi->first_interrupt, true);
2052 
2053 	WRITE_ONCE(ena_napi->interrupts_masked, true);
2054 	smp_wmb(); /* write interrupts_masked before calling napi */
2055 
2056 	napi_schedule_irqoff(&ena_napi->napi);
2057 
2058 	return IRQ_HANDLED;
2059 }
2060 
2061 /* Reserve a single MSI-X vector for management (admin + aenq).
2062  * plus reserve one vector for each potential io queue.
2063  * the number of potential io queues is the minimum of what the device
2064  * supports and the number of vCPUs.
2065  */
2066 static int ena_enable_msix(struct ena_adapter *adapter)
2067 {
2068 	int msix_vecs, irq_cnt;
2069 
2070 	if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2071 		netif_err(adapter, probe, adapter->netdev,
2072 			  "Error, MSI-X is already enabled\n");
2073 		return -EPERM;
2074 	}
2075 
2076 	/* Reserved the max msix vectors we might need */
2077 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2078 	netif_dbg(adapter, probe, adapter->netdev,
2079 		  "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2080 
2081 	irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2082 					msix_vecs, PCI_IRQ_MSIX);
2083 
2084 	if (irq_cnt < 0) {
2085 		netif_err(adapter, probe, adapter->netdev,
2086 			  "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2087 		return -ENOSPC;
2088 	}
2089 
2090 	if (irq_cnt != msix_vecs) {
2091 		netif_notice(adapter, probe, adapter->netdev,
2092 			     "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2093 			     irq_cnt, msix_vecs);
2094 		adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2095 	}
2096 
2097 	if (ena_init_rx_cpu_rmap(adapter))
2098 		netif_warn(adapter, probe, adapter->netdev,
2099 			   "Failed to map IRQs to CPUs\n");
2100 
2101 	adapter->msix_vecs = irq_cnt;
2102 	set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2103 
2104 	return 0;
2105 }
2106 
2107 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2108 {
2109 	u32 cpu;
2110 
2111 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2112 		 ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2113 		 pci_name(adapter->pdev));
2114 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2115 		ena_intr_msix_mgmnt;
2116 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2117 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2118 		pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2119 	cpu = cpumask_first(cpu_online_mask);
2120 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2121 	cpumask_set_cpu(cpu,
2122 			&adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2123 }
2124 
2125 static void ena_setup_io_intr(struct ena_adapter *adapter)
2126 {
2127 	struct net_device *netdev;
2128 	int irq_idx, i, cpu;
2129 	int io_queue_count;
2130 
2131 	netdev = adapter->netdev;
2132 	io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2133 
2134 	for (i = 0; i < io_queue_count; i++) {
2135 		irq_idx = ENA_IO_IRQ_IDX(i);
2136 		cpu = i % num_online_cpus();
2137 
2138 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2139 			 "%s-Tx-Rx-%d", netdev->name, i);
2140 		adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2141 		adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2142 		adapter->irq_tbl[irq_idx].vector =
2143 			pci_irq_vector(adapter->pdev, irq_idx);
2144 		adapter->irq_tbl[irq_idx].cpu = cpu;
2145 
2146 		cpumask_set_cpu(cpu,
2147 				&adapter->irq_tbl[irq_idx].affinity_hint_mask);
2148 	}
2149 }
2150 
2151 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2152 {
2153 	unsigned long flags = 0;
2154 	struct ena_irq *irq;
2155 	int rc;
2156 
2157 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2158 	rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2159 			 irq->data);
2160 	if (rc) {
2161 		netif_err(adapter, probe, adapter->netdev,
2162 			  "Failed to request admin irq\n");
2163 		return rc;
2164 	}
2165 
2166 	netif_dbg(adapter, probe, adapter->netdev,
2167 		  "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2168 		  irq->affinity_hint_mask.bits[0], irq->vector);
2169 
2170 	irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2171 
2172 	return rc;
2173 }
2174 
2175 static int ena_request_io_irq(struct ena_adapter *adapter)
2176 {
2177 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2178 	unsigned long flags = 0;
2179 	struct ena_irq *irq;
2180 	int rc = 0, i, k;
2181 
2182 	if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2183 		netif_err(adapter, ifup, adapter->netdev,
2184 			  "Failed to request I/O IRQ: MSI-X is not enabled\n");
2185 		return -EINVAL;
2186 	}
2187 
2188 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2189 		irq = &adapter->irq_tbl[i];
2190 		rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2191 				 irq->data);
2192 		if (rc) {
2193 			netif_err(adapter, ifup, adapter->netdev,
2194 				  "Failed to request I/O IRQ. index %d rc %d\n",
2195 				   i, rc);
2196 			goto err;
2197 		}
2198 
2199 		netif_dbg(adapter, ifup, adapter->netdev,
2200 			  "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2201 			  i, irq->affinity_hint_mask.bits[0], irq->vector);
2202 
2203 		irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2204 	}
2205 
2206 	return rc;
2207 
2208 err:
2209 	for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2210 		irq = &adapter->irq_tbl[k];
2211 		free_irq(irq->vector, irq->data);
2212 	}
2213 
2214 	return rc;
2215 }
2216 
2217 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2218 {
2219 	struct ena_irq *irq;
2220 
2221 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2222 	synchronize_irq(irq->vector);
2223 	irq_set_affinity_hint(irq->vector, NULL);
2224 	free_irq(irq->vector, irq->data);
2225 }
2226 
2227 static void ena_free_io_irq(struct ena_adapter *adapter)
2228 {
2229 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2230 	struct ena_irq *irq;
2231 	int i;
2232 
2233 #ifdef CONFIG_RFS_ACCEL
2234 	if (adapter->msix_vecs >= 1) {
2235 		free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2236 		adapter->netdev->rx_cpu_rmap = NULL;
2237 	}
2238 #endif /* CONFIG_RFS_ACCEL */
2239 
2240 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2241 		irq = &adapter->irq_tbl[i];
2242 		irq_set_affinity_hint(irq->vector, NULL);
2243 		free_irq(irq->vector, irq->data);
2244 	}
2245 }
2246 
2247 static void ena_disable_msix(struct ena_adapter *adapter)
2248 {
2249 	if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2250 		pci_free_irq_vectors(adapter->pdev);
2251 }
2252 
2253 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2254 {
2255 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2256 	int i;
2257 
2258 	if (!netif_running(adapter->netdev))
2259 		return;
2260 
2261 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2262 		synchronize_irq(adapter->irq_tbl[i].vector);
2263 }
2264 
2265 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2266 				  int first_index,
2267 				  int count)
2268 {
2269 	int i;
2270 
2271 	for (i = first_index; i < first_index + count; i++) {
2272 		netif_napi_del(&adapter->ena_napi[i].napi);
2273 
2274 		WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2275 			adapter->ena_napi[i].xdp_ring);
2276 	}
2277 }
2278 
2279 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2280 				   int first_index, int count)
2281 {
2282 	int i;
2283 
2284 	for (i = first_index; i < first_index + count; i++) {
2285 		struct ena_napi *napi = &adapter->ena_napi[i];
2286 
2287 		netif_napi_add(adapter->netdev, &napi->napi,
2288 			       ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll);
2289 
2290 		if (!ENA_IS_XDP_INDEX(adapter, i)) {
2291 			napi->rx_ring = &adapter->rx_ring[i];
2292 			napi->tx_ring = &adapter->tx_ring[i];
2293 		} else {
2294 			napi->xdp_ring = &adapter->tx_ring[i];
2295 		}
2296 		napi->qid = i;
2297 	}
2298 }
2299 
2300 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2301 				      int first_index,
2302 				      int count)
2303 {
2304 	int i;
2305 
2306 	for (i = first_index; i < first_index + count; i++)
2307 		napi_disable(&adapter->ena_napi[i].napi);
2308 }
2309 
2310 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2311 				     int first_index,
2312 				     int count)
2313 {
2314 	int i;
2315 
2316 	for (i = first_index; i < first_index + count; i++)
2317 		napi_enable(&adapter->ena_napi[i].napi);
2318 }
2319 
2320 /* Configure the Rx forwarding */
2321 static int ena_rss_configure(struct ena_adapter *adapter)
2322 {
2323 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2324 	int rc;
2325 
2326 	/* In case the RSS table wasn't initialized by probe */
2327 	if (!ena_dev->rss.tbl_log_size) {
2328 		rc = ena_rss_init_default(adapter);
2329 		if (rc && (rc != -EOPNOTSUPP)) {
2330 			netif_err(adapter, ifup, adapter->netdev,
2331 				  "Failed to init RSS rc: %d\n", rc);
2332 			return rc;
2333 		}
2334 	}
2335 
2336 	/* Set indirect table */
2337 	rc = ena_com_indirect_table_set(ena_dev);
2338 	if (unlikely(rc && rc != -EOPNOTSUPP))
2339 		return rc;
2340 
2341 	/* Configure hash function (if supported) */
2342 	rc = ena_com_set_hash_function(ena_dev);
2343 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
2344 		return rc;
2345 
2346 	/* Configure hash inputs (if supported) */
2347 	rc = ena_com_set_hash_ctrl(ena_dev);
2348 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
2349 		return rc;
2350 
2351 	return 0;
2352 }
2353 
2354 static int ena_up_complete(struct ena_adapter *adapter)
2355 {
2356 	int rc;
2357 
2358 	rc = ena_rss_configure(adapter);
2359 	if (rc)
2360 		return rc;
2361 
2362 	ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2363 
2364 	ena_refill_all_rx_bufs(adapter);
2365 
2366 	/* enable transmits */
2367 	netif_tx_start_all_queues(adapter->netdev);
2368 
2369 	ena_napi_enable_in_range(adapter,
2370 				 0,
2371 				 adapter->xdp_num_queues + adapter->num_io_queues);
2372 
2373 	return 0;
2374 }
2375 
2376 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2377 {
2378 	struct ena_com_create_io_ctx ctx;
2379 	struct ena_com_dev *ena_dev;
2380 	struct ena_ring *tx_ring;
2381 	u32 msix_vector;
2382 	u16 ena_qid;
2383 	int rc;
2384 
2385 	ena_dev = adapter->ena_dev;
2386 
2387 	tx_ring = &adapter->tx_ring[qid];
2388 	msix_vector = ENA_IO_IRQ_IDX(qid);
2389 	ena_qid = ENA_IO_TXQ_IDX(qid);
2390 
2391 	memset(&ctx, 0x0, sizeof(ctx));
2392 
2393 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2394 	ctx.qid = ena_qid;
2395 	ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2396 	ctx.msix_vector = msix_vector;
2397 	ctx.queue_size = tx_ring->ring_size;
2398 	ctx.numa_node = tx_ring->numa_node;
2399 
2400 	rc = ena_com_create_io_queue(ena_dev, &ctx);
2401 	if (rc) {
2402 		netif_err(adapter, ifup, adapter->netdev,
2403 			  "Failed to create I/O TX queue num %d rc: %d\n",
2404 			  qid, rc);
2405 		return rc;
2406 	}
2407 
2408 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2409 				     &tx_ring->ena_com_io_sq,
2410 				     &tx_ring->ena_com_io_cq);
2411 	if (rc) {
2412 		netif_err(adapter, ifup, adapter->netdev,
2413 			  "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2414 			  qid, rc);
2415 		ena_com_destroy_io_queue(ena_dev, ena_qid);
2416 		return rc;
2417 	}
2418 
2419 	ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2420 	return rc;
2421 }
2422 
2423 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2424 					    int first_index, int count)
2425 {
2426 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2427 	int rc, i;
2428 
2429 	for (i = first_index; i < first_index + count; i++) {
2430 		rc = ena_create_io_tx_queue(adapter, i);
2431 		if (rc)
2432 			goto create_err;
2433 	}
2434 
2435 	return 0;
2436 
2437 create_err:
2438 	while (i-- > first_index)
2439 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2440 
2441 	return rc;
2442 }
2443 
2444 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2445 {
2446 	struct ena_com_dev *ena_dev;
2447 	struct ena_com_create_io_ctx ctx;
2448 	struct ena_ring *rx_ring;
2449 	u32 msix_vector;
2450 	u16 ena_qid;
2451 	int rc;
2452 
2453 	ena_dev = adapter->ena_dev;
2454 
2455 	rx_ring = &adapter->rx_ring[qid];
2456 	msix_vector = ENA_IO_IRQ_IDX(qid);
2457 	ena_qid = ENA_IO_RXQ_IDX(qid);
2458 
2459 	memset(&ctx, 0x0, sizeof(ctx));
2460 
2461 	ctx.qid = ena_qid;
2462 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2463 	ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2464 	ctx.msix_vector = msix_vector;
2465 	ctx.queue_size = rx_ring->ring_size;
2466 	ctx.numa_node = rx_ring->numa_node;
2467 
2468 	rc = ena_com_create_io_queue(ena_dev, &ctx);
2469 	if (rc) {
2470 		netif_err(adapter, ifup, adapter->netdev,
2471 			  "Failed to create I/O RX queue num %d rc: %d\n",
2472 			  qid, rc);
2473 		return rc;
2474 	}
2475 
2476 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2477 				     &rx_ring->ena_com_io_sq,
2478 				     &rx_ring->ena_com_io_cq);
2479 	if (rc) {
2480 		netif_err(adapter, ifup, adapter->netdev,
2481 			  "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2482 			  qid, rc);
2483 		goto err;
2484 	}
2485 
2486 	ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2487 
2488 	return rc;
2489 err:
2490 	ena_com_destroy_io_queue(ena_dev, ena_qid);
2491 	return rc;
2492 }
2493 
2494 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2495 {
2496 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2497 	int rc, i;
2498 
2499 	for (i = 0; i < adapter->num_io_queues; i++) {
2500 		rc = ena_create_io_rx_queue(adapter, i);
2501 		if (rc)
2502 			goto create_err;
2503 		INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2504 	}
2505 
2506 	return 0;
2507 
2508 create_err:
2509 	while (i--) {
2510 		cancel_work_sync(&adapter->ena_napi[i].dim.work);
2511 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2512 	}
2513 
2514 	return rc;
2515 }
2516 
2517 static void set_io_rings_size(struct ena_adapter *adapter,
2518 			      int new_tx_size,
2519 			      int new_rx_size)
2520 {
2521 	int i;
2522 
2523 	for (i = 0; i < adapter->num_io_queues; i++) {
2524 		adapter->tx_ring[i].ring_size = new_tx_size;
2525 		adapter->rx_ring[i].ring_size = new_rx_size;
2526 	}
2527 }
2528 
2529 /* This function allows queue allocation to backoff when the system is
2530  * low on memory. If there is not enough memory to allocate io queues
2531  * the driver will try to allocate smaller queues.
2532  *
2533  * The backoff algorithm is as follows:
2534  *  1. Try to allocate TX and RX and if successful.
2535  *  1.1. return success
2536  *
2537  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2538  *
2539  *  3. If TX or RX is smaller than 256
2540  *  3.1. return failure.
2541  *  4. else
2542  *  4.1. go back to 1.
2543  */
2544 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2545 {
2546 	int rc, cur_rx_ring_size, cur_tx_ring_size;
2547 	int new_rx_ring_size, new_tx_ring_size;
2548 
2549 	/* current queue sizes might be set to smaller than the requested
2550 	 * ones due to past queue allocation failures.
2551 	 */
2552 	set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2553 			  adapter->requested_rx_ring_size);
2554 
2555 	while (1) {
2556 		if (ena_xdp_present(adapter)) {
2557 			rc = ena_setup_and_create_all_xdp_queues(adapter);
2558 
2559 			if (rc)
2560 				goto err_setup_tx;
2561 		}
2562 		rc = ena_setup_tx_resources_in_range(adapter,
2563 						     0,
2564 						     adapter->num_io_queues);
2565 		if (rc)
2566 			goto err_setup_tx;
2567 
2568 		rc = ena_create_io_tx_queues_in_range(adapter,
2569 						      0,
2570 						      adapter->num_io_queues);
2571 		if (rc)
2572 			goto err_create_tx_queues;
2573 
2574 		rc = ena_setup_all_rx_resources(adapter);
2575 		if (rc)
2576 			goto err_setup_rx;
2577 
2578 		rc = ena_create_all_io_rx_queues(adapter);
2579 		if (rc)
2580 			goto err_create_rx_queues;
2581 
2582 		return 0;
2583 
2584 err_create_rx_queues:
2585 		ena_free_all_io_rx_resources(adapter);
2586 err_setup_rx:
2587 		ena_destroy_all_tx_queues(adapter);
2588 err_create_tx_queues:
2589 		ena_free_all_io_tx_resources(adapter);
2590 err_setup_tx:
2591 		if (rc != -ENOMEM) {
2592 			netif_err(adapter, ifup, adapter->netdev,
2593 				  "Queue creation failed with error code %d\n",
2594 				  rc);
2595 			return rc;
2596 		}
2597 
2598 		cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2599 		cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2600 
2601 		netif_err(adapter, ifup, adapter->netdev,
2602 			  "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2603 			  cur_tx_ring_size, cur_rx_ring_size);
2604 
2605 		new_tx_ring_size = cur_tx_ring_size;
2606 		new_rx_ring_size = cur_rx_ring_size;
2607 
2608 		/* Decrease the size of the larger queue, or
2609 		 * decrease both if they are the same size.
2610 		 */
2611 		if (cur_rx_ring_size <= cur_tx_ring_size)
2612 			new_tx_ring_size = cur_tx_ring_size / 2;
2613 		if (cur_rx_ring_size >= cur_tx_ring_size)
2614 			new_rx_ring_size = cur_rx_ring_size / 2;
2615 
2616 		if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2617 		    new_rx_ring_size < ENA_MIN_RING_SIZE) {
2618 			netif_err(adapter, ifup, adapter->netdev,
2619 				  "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2620 				  ENA_MIN_RING_SIZE);
2621 			return rc;
2622 		}
2623 
2624 		netif_err(adapter, ifup, adapter->netdev,
2625 			  "Retrying queue creation with sizes TX=%d, RX=%d\n",
2626 			  new_tx_ring_size,
2627 			  new_rx_ring_size);
2628 
2629 		set_io_rings_size(adapter, new_tx_ring_size,
2630 				  new_rx_ring_size);
2631 	}
2632 }
2633 
2634 static int ena_up(struct ena_adapter *adapter)
2635 {
2636 	int io_queue_count, rc, i;
2637 
2638 	netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2639 
2640 	io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2641 	ena_setup_io_intr(adapter);
2642 
2643 	/* napi poll functions should be initialized before running
2644 	 * request_irq(), to handle a rare condition where there is a pending
2645 	 * interrupt, causing the ISR to fire immediately while the poll
2646 	 * function wasn't set yet, causing a null dereference
2647 	 */
2648 	ena_init_napi_in_range(adapter, 0, io_queue_count);
2649 
2650 	rc = ena_request_io_irq(adapter);
2651 	if (rc)
2652 		goto err_req_irq;
2653 
2654 	rc = create_queues_with_size_backoff(adapter);
2655 	if (rc)
2656 		goto err_create_queues_with_backoff;
2657 
2658 	rc = ena_up_complete(adapter);
2659 	if (rc)
2660 		goto err_up;
2661 
2662 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2663 		netif_carrier_on(adapter->netdev);
2664 
2665 	ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2666 			  &adapter->syncp);
2667 
2668 	set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2669 
2670 	/* Enable completion queues interrupt */
2671 	for (i = 0; i < adapter->num_io_queues; i++)
2672 		ena_unmask_interrupt(&adapter->tx_ring[i],
2673 				     &adapter->rx_ring[i]);
2674 
2675 	/* schedule napi in case we had pending packets
2676 	 * from the last time we disable napi
2677 	 */
2678 	for (i = 0; i < io_queue_count; i++)
2679 		napi_schedule(&adapter->ena_napi[i].napi);
2680 
2681 	return rc;
2682 
2683 err_up:
2684 	ena_destroy_all_tx_queues(adapter);
2685 	ena_free_all_io_tx_resources(adapter);
2686 	ena_destroy_all_rx_queues(adapter);
2687 	ena_free_all_io_rx_resources(adapter);
2688 err_create_queues_with_backoff:
2689 	ena_free_io_irq(adapter);
2690 err_req_irq:
2691 	ena_del_napi_in_range(adapter, 0, io_queue_count);
2692 
2693 	return rc;
2694 }
2695 
2696 static void ena_down(struct ena_adapter *adapter)
2697 {
2698 	int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2699 
2700 	netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2701 
2702 	clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2703 
2704 	ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2705 			  &adapter->syncp);
2706 
2707 	netif_carrier_off(adapter->netdev);
2708 	netif_tx_disable(adapter->netdev);
2709 
2710 	/* After this point the napi handler won't enable the tx queue */
2711 	ena_napi_disable_in_range(adapter, 0, io_queue_count);
2712 
2713 	/* After destroy the queue there won't be any new interrupts */
2714 
2715 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2716 		int rc;
2717 
2718 		rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2719 		if (rc)
2720 			netif_err(adapter, ifdown, adapter->netdev,
2721 				  "Device reset failed\n");
2722 		/* stop submitting admin commands on a device that was reset */
2723 		ena_com_set_admin_running_state(adapter->ena_dev, false);
2724 	}
2725 
2726 	ena_destroy_all_io_queues(adapter);
2727 
2728 	ena_disable_io_intr_sync(adapter);
2729 	ena_free_io_irq(adapter);
2730 	ena_del_napi_in_range(adapter, 0, io_queue_count);
2731 
2732 	ena_free_all_tx_bufs(adapter);
2733 	ena_free_all_rx_bufs(adapter);
2734 	ena_free_all_io_tx_resources(adapter);
2735 	ena_free_all_io_rx_resources(adapter);
2736 }
2737 
2738 /* ena_open - Called when a network interface is made active
2739  * @netdev: network interface device structure
2740  *
2741  * Returns 0 on success, negative value on failure
2742  *
2743  * The open entry point is called when a network interface is made
2744  * active by the system (IFF_UP).  At this point all resources needed
2745  * for transmit and receive operations are allocated, the interrupt
2746  * handler is registered with the OS, the watchdog timer is started,
2747  * and the stack is notified that the interface is ready.
2748  */
2749 static int ena_open(struct net_device *netdev)
2750 {
2751 	struct ena_adapter *adapter = netdev_priv(netdev);
2752 	int rc;
2753 
2754 	/* Notify the stack of the actual queue counts. */
2755 	rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2756 	if (rc) {
2757 		netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2758 		return rc;
2759 	}
2760 
2761 	rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2762 	if (rc) {
2763 		netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2764 		return rc;
2765 	}
2766 
2767 	rc = ena_up(adapter);
2768 	if (rc)
2769 		return rc;
2770 
2771 	return rc;
2772 }
2773 
2774 /* ena_close - Disables a network interface
2775  * @netdev: network interface device structure
2776  *
2777  * Returns 0, this is not allowed to fail
2778  *
2779  * The close entry point is called when an interface is de-activated
2780  * by the OS.  The hardware is still under the drivers control, but
2781  * needs to be disabled.  A global MAC reset is issued to stop the
2782  * hardware, and all transmit and receive resources are freed.
2783  */
2784 static int ena_close(struct net_device *netdev)
2785 {
2786 	struct ena_adapter *adapter = netdev_priv(netdev);
2787 
2788 	netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2789 
2790 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2791 		return 0;
2792 
2793 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2794 		ena_down(adapter);
2795 
2796 	/* Check for device status and issue reset if needed*/
2797 	check_for_admin_com_state(adapter);
2798 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2799 		netif_err(adapter, ifdown, adapter->netdev,
2800 			  "Destroy failure, restarting device\n");
2801 		ena_dump_stats_to_dmesg(adapter);
2802 		/* rtnl lock already obtained in dev_ioctl() layer */
2803 		ena_destroy_device(adapter, false);
2804 		ena_restore_device(adapter);
2805 	}
2806 
2807 	return 0;
2808 }
2809 
2810 int ena_update_queue_params(struct ena_adapter *adapter,
2811 			    u32 new_tx_size,
2812 			    u32 new_rx_size,
2813 			    u32 new_llq_header_len)
2814 {
2815 	bool dev_was_up, large_llq_changed = false;
2816 	int rc = 0;
2817 
2818 	dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2819 	ena_close(adapter->netdev);
2820 	adapter->requested_tx_ring_size = new_tx_size;
2821 	adapter->requested_rx_ring_size = new_rx_size;
2822 	ena_init_io_rings(adapter,
2823 			  0,
2824 			  adapter->xdp_num_queues +
2825 			  adapter->num_io_queues);
2826 
2827 	large_llq_changed = adapter->ena_dev->tx_mem_queue_type ==
2828 			    ENA_ADMIN_PLACEMENT_POLICY_DEV;
2829 	large_llq_changed &=
2830 		new_llq_header_len != adapter->ena_dev->tx_max_header_size;
2831 
2832 	/* a check that the configuration is valid is done by caller */
2833 	if (large_llq_changed) {
2834 		adapter->large_llq_header_enabled = !adapter->large_llq_header_enabled;
2835 
2836 		ena_destroy_device(adapter, false);
2837 		rc = ena_restore_device(adapter);
2838 	}
2839 
2840 	return dev_was_up && !rc ? ena_up(adapter) : rc;
2841 }
2842 
2843 int ena_set_rx_copybreak(struct ena_adapter *adapter, u32 rx_copybreak)
2844 {
2845 	struct ena_ring *rx_ring;
2846 	int i;
2847 
2848 	if (rx_copybreak > min_t(u16, adapter->netdev->mtu, ENA_PAGE_SIZE))
2849 		return -EINVAL;
2850 
2851 	adapter->rx_copybreak = rx_copybreak;
2852 
2853 	for (i = 0; i < adapter->num_io_queues; i++) {
2854 		rx_ring = &adapter->rx_ring[i];
2855 		rx_ring->rx_copybreak = rx_copybreak;
2856 	}
2857 
2858 	return 0;
2859 }
2860 
2861 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2862 {
2863 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2864 	int prev_channel_count;
2865 	bool dev_was_up;
2866 
2867 	dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2868 	ena_close(adapter->netdev);
2869 	prev_channel_count = adapter->num_io_queues;
2870 	adapter->num_io_queues = new_channel_count;
2871 	if (ena_xdp_present(adapter) &&
2872 	    ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2873 		adapter->xdp_first_ring = new_channel_count;
2874 		adapter->xdp_num_queues = new_channel_count;
2875 		if (prev_channel_count > new_channel_count)
2876 			ena_xdp_exchange_program_rx_in_range(adapter,
2877 							     NULL,
2878 							     new_channel_count,
2879 							     prev_channel_count);
2880 		else
2881 			ena_xdp_exchange_program_rx_in_range(adapter,
2882 							     adapter->xdp_bpf_prog,
2883 							     prev_channel_count,
2884 							     new_channel_count);
2885 	}
2886 
2887 	/* We need to destroy the rss table so that the indirection
2888 	 * table will be reinitialized by ena_up()
2889 	 */
2890 	ena_com_rss_destroy(ena_dev);
2891 	ena_init_io_rings(adapter,
2892 			  0,
2893 			  adapter->xdp_num_queues +
2894 			  adapter->num_io_queues);
2895 	return dev_was_up ? ena_open(adapter->netdev) : 0;
2896 }
2897 
2898 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2899 			struct sk_buff *skb,
2900 			bool disable_meta_caching)
2901 {
2902 	u32 mss = skb_shinfo(skb)->gso_size;
2903 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2904 	u8 l4_protocol = 0;
2905 
2906 	if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2907 		ena_tx_ctx->l4_csum_enable = 1;
2908 		if (mss) {
2909 			ena_tx_ctx->tso_enable = 1;
2910 			ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2911 			ena_tx_ctx->l4_csum_partial = 0;
2912 		} else {
2913 			ena_tx_ctx->tso_enable = 0;
2914 			ena_meta->l4_hdr_len = 0;
2915 			ena_tx_ctx->l4_csum_partial = 1;
2916 		}
2917 
2918 		switch (ip_hdr(skb)->version) {
2919 		case IPVERSION:
2920 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2921 			if (ip_hdr(skb)->frag_off & htons(IP_DF))
2922 				ena_tx_ctx->df = 1;
2923 			if (mss)
2924 				ena_tx_ctx->l3_csum_enable = 1;
2925 			l4_protocol = ip_hdr(skb)->protocol;
2926 			break;
2927 		case 6:
2928 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2929 			l4_protocol = ipv6_hdr(skb)->nexthdr;
2930 			break;
2931 		default:
2932 			break;
2933 		}
2934 
2935 		if (l4_protocol == IPPROTO_TCP)
2936 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2937 		else
2938 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2939 
2940 		ena_meta->mss = mss;
2941 		ena_meta->l3_hdr_len = skb_network_header_len(skb);
2942 		ena_meta->l3_hdr_offset = skb_network_offset(skb);
2943 		ena_tx_ctx->meta_valid = 1;
2944 	} else if (disable_meta_caching) {
2945 		memset(ena_meta, 0, sizeof(*ena_meta));
2946 		ena_tx_ctx->meta_valid = 1;
2947 	} else {
2948 		ena_tx_ctx->meta_valid = 0;
2949 	}
2950 }
2951 
2952 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2953 				       struct sk_buff *skb)
2954 {
2955 	int num_frags, header_len, rc;
2956 
2957 	num_frags = skb_shinfo(skb)->nr_frags;
2958 	header_len = skb_headlen(skb);
2959 
2960 	if (num_frags < tx_ring->sgl_size)
2961 		return 0;
2962 
2963 	if ((num_frags == tx_ring->sgl_size) &&
2964 	    (header_len < tx_ring->tx_max_header_size))
2965 		return 0;
2966 
2967 	ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
2968 
2969 	rc = skb_linearize(skb);
2970 	if (unlikely(rc)) {
2971 		ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
2972 				  &tx_ring->syncp);
2973 	}
2974 
2975 	return rc;
2976 }
2977 
2978 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2979 			  struct ena_tx_buffer *tx_info,
2980 			  struct sk_buff *skb,
2981 			  void **push_hdr,
2982 			  u16 *header_len)
2983 {
2984 	struct ena_adapter *adapter = tx_ring->adapter;
2985 	struct ena_com_buf *ena_buf;
2986 	dma_addr_t dma;
2987 	u32 skb_head_len, frag_len, last_frag;
2988 	u16 push_len = 0;
2989 	u16 delta = 0;
2990 	int i = 0;
2991 
2992 	skb_head_len = skb_headlen(skb);
2993 	tx_info->skb = skb;
2994 	ena_buf = tx_info->bufs;
2995 
2996 	if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2997 		/* When the device is LLQ mode, the driver will copy
2998 		 * the header into the device memory space.
2999 		 * the ena_com layer assume the header is in a linear
3000 		 * memory space.
3001 		 * This assumption might be wrong since part of the header
3002 		 * can be in the fragmented buffers.
3003 		 * Use skb_header_pointer to make sure the header is in a
3004 		 * linear memory space.
3005 		 */
3006 
3007 		push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
3008 		*push_hdr = skb_header_pointer(skb, 0, push_len,
3009 					       tx_ring->push_buf_intermediate_buf);
3010 		*header_len = push_len;
3011 		if (unlikely(skb->data != *push_hdr)) {
3012 			ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
3013 					  &tx_ring->syncp);
3014 
3015 			delta = push_len - skb_head_len;
3016 		}
3017 	} else {
3018 		*push_hdr = NULL;
3019 		*header_len = min_t(u32, skb_head_len,
3020 				    tx_ring->tx_max_header_size);
3021 	}
3022 
3023 	netif_dbg(adapter, tx_queued, adapter->netdev,
3024 		  "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
3025 		  *push_hdr, push_len);
3026 
3027 	if (skb_head_len > push_len) {
3028 		dma = dma_map_single(tx_ring->dev, skb->data + push_len,
3029 				     skb_head_len - push_len, DMA_TO_DEVICE);
3030 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
3031 			goto error_report_dma_error;
3032 
3033 		ena_buf->paddr = dma;
3034 		ena_buf->len = skb_head_len - push_len;
3035 
3036 		ena_buf++;
3037 		tx_info->num_of_bufs++;
3038 		tx_info->map_linear_data = 1;
3039 	} else {
3040 		tx_info->map_linear_data = 0;
3041 	}
3042 
3043 	last_frag = skb_shinfo(skb)->nr_frags;
3044 
3045 	for (i = 0; i < last_frag; i++) {
3046 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3047 
3048 		frag_len = skb_frag_size(frag);
3049 
3050 		if (unlikely(delta >= frag_len)) {
3051 			delta -= frag_len;
3052 			continue;
3053 		}
3054 
3055 		dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
3056 				       frag_len - delta, DMA_TO_DEVICE);
3057 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
3058 			goto error_report_dma_error;
3059 
3060 		ena_buf->paddr = dma;
3061 		ena_buf->len = frag_len - delta;
3062 		ena_buf++;
3063 		tx_info->num_of_bufs++;
3064 		delta = 0;
3065 	}
3066 
3067 	return 0;
3068 
3069 error_report_dma_error:
3070 	ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
3071 			  &tx_ring->syncp);
3072 	netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
3073 
3074 	tx_info->skb = NULL;
3075 
3076 	tx_info->num_of_bufs += i;
3077 	ena_unmap_tx_buff(tx_ring, tx_info);
3078 
3079 	return -EINVAL;
3080 }
3081 
3082 /* Called with netif_tx_lock. */
3083 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3084 {
3085 	struct ena_adapter *adapter = netdev_priv(dev);
3086 	struct ena_tx_buffer *tx_info;
3087 	struct ena_com_tx_ctx ena_tx_ctx;
3088 	struct ena_ring *tx_ring;
3089 	struct netdev_queue *txq;
3090 	void *push_hdr;
3091 	u16 next_to_use, req_id, header_len;
3092 	int qid, rc;
3093 
3094 	netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3095 	/*  Determine which tx ring we will be placed on */
3096 	qid = skb_get_queue_mapping(skb);
3097 	tx_ring = &adapter->tx_ring[qid];
3098 	txq = netdev_get_tx_queue(dev, qid);
3099 
3100 	rc = ena_check_and_linearize_skb(tx_ring, skb);
3101 	if (unlikely(rc))
3102 		goto error_drop_packet;
3103 
3104 	skb_tx_timestamp(skb);
3105 
3106 	next_to_use = tx_ring->next_to_use;
3107 	req_id = tx_ring->free_ids[next_to_use];
3108 	tx_info = &tx_ring->tx_buffer_info[req_id];
3109 	tx_info->num_of_bufs = 0;
3110 
3111 	WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3112 
3113 	rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3114 	if (unlikely(rc))
3115 		goto error_drop_packet;
3116 
3117 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3118 	ena_tx_ctx.ena_bufs = tx_info->bufs;
3119 	ena_tx_ctx.push_header = push_hdr;
3120 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3121 	ena_tx_ctx.req_id = req_id;
3122 	ena_tx_ctx.header_len = header_len;
3123 
3124 	/* set flags and meta data */
3125 	ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3126 
3127 	rc = ena_xmit_common(dev,
3128 			     tx_ring,
3129 			     tx_info,
3130 			     &ena_tx_ctx,
3131 			     next_to_use,
3132 			     skb->len);
3133 	if (rc)
3134 		goto error_unmap_dma;
3135 
3136 	netdev_tx_sent_queue(txq, skb->len);
3137 
3138 	/* stop the queue when no more space available, the packet can have up
3139 	 * to sgl_size + 2. one for the meta descriptor and one for header
3140 	 * (if the header is larger than tx_max_header_size).
3141 	 */
3142 	if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3143 						   tx_ring->sgl_size + 2))) {
3144 		netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3145 			  __func__, qid);
3146 
3147 		netif_tx_stop_queue(txq);
3148 		ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3149 				  &tx_ring->syncp);
3150 
3151 		/* There is a rare condition where this function decide to
3152 		 * stop the queue but meanwhile clean_tx_irq updates
3153 		 * next_to_completion and terminates.
3154 		 * The queue will remain stopped forever.
3155 		 * To solve this issue add a mb() to make sure that
3156 		 * netif_tx_stop_queue() write is vissible before checking if
3157 		 * there is additional space in the queue.
3158 		 */
3159 		smp_mb();
3160 
3161 		if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3162 						 ENA_TX_WAKEUP_THRESH)) {
3163 			netif_tx_wake_queue(txq);
3164 			ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3165 					  &tx_ring->syncp);
3166 		}
3167 	}
3168 
3169 	if (netif_xmit_stopped(txq) || !netdev_xmit_more())
3170 		/* trigger the dma engine. ena_ring_tx_doorbell()
3171 		 * calls a memory barrier inside it.
3172 		 */
3173 		ena_ring_tx_doorbell(tx_ring);
3174 
3175 	return NETDEV_TX_OK;
3176 
3177 error_unmap_dma:
3178 	ena_unmap_tx_buff(tx_ring, tx_info);
3179 	tx_info->skb = NULL;
3180 
3181 error_drop_packet:
3182 	dev_kfree_skb(skb);
3183 	return NETDEV_TX_OK;
3184 }
3185 
3186 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3187 			    struct net_device *sb_dev)
3188 {
3189 	u16 qid;
3190 	/* we suspect that this is good for in--kernel network services that
3191 	 * want to loop incoming skb rx to tx in normal user generated traffic,
3192 	 * most probably we will not get to this
3193 	 */
3194 	if (skb_rx_queue_recorded(skb))
3195 		qid = skb_get_rx_queue(skb);
3196 	else
3197 		qid = netdev_pick_tx(dev, skb, NULL);
3198 
3199 	return qid;
3200 }
3201 
3202 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3203 {
3204 	struct device *dev = &pdev->dev;
3205 	struct ena_admin_host_info *host_info;
3206 	int rc;
3207 
3208 	/* Allocate only the host info */
3209 	rc = ena_com_allocate_host_info(ena_dev);
3210 	if (rc) {
3211 		dev_err(dev, "Cannot allocate host info\n");
3212 		return;
3213 	}
3214 
3215 	host_info = ena_dev->host_attr.host_info;
3216 
3217 	host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3218 	host_info->os_type = ENA_ADMIN_OS_LINUX;
3219 	host_info->kernel_ver = LINUX_VERSION_CODE;
3220 	strscpy(host_info->kernel_ver_str, utsname()->version,
3221 		sizeof(host_info->kernel_ver_str) - 1);
3222 	host_info->os_dist = 0;
3223 	strncpy(host_info->os_dist_str, utsname()->release,
3224 		sizeof(host_info->os_dist_str) - 1);
3225 	host_info->driver_version =
3226 		(DRV_MODULE_GEN_MAJOR) |
3227 		(DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3228 		(DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3229 		("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3230 	host_info->num_cpus = num_online_cpus();
3231 
3232 	host_info->driver_supported_features =
3233 		ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3234 		ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3235 		ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3236 		ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
3237 
3238 	rc = ena_com_set_host_attributes(ena_dev);
3239 	if (rc) {
3240 		if (rc == -EOPNOTSUPP)
3241 			dev_warn(dev, "Cannot set host attributes\n");
3242 		else
3243 			dev_err(dev, "Cannot set host attributes\n");
3244 
3245 		goto err;
3246 	}
3247 
3248 	return;
3249 
3250 err:
3251 	ena_com_delete_host_info(ena_dev);
3252 }
3253 
3254 static void ena_config_debug_area(struct ena_adapter *adapter)
3255 {
3256 	u32 debug_area_size;
3257 	int rc, ss_count;
3258 
3259 	ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3260 	if (ss_count <= 0) {
3261 		netif_err(adapter, drv, adapter->netdev,
3262 			  "SS count is negative\n");
3263 		return;
3264 	}
3265 
3266 	/* allocate 32 bytes for each string and 64bit for the value */
3267 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3268 
3269 	rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3270 	if (rc) {
3271 		netif_err(adapter, drv, adapter->netdev,
3272 			  "Cannot allocate debug area\n");
3273 		return;
3274 	}
3275 
3276 	rc = ena_com_set_host_attributes(adapter->ena_dev);
3277 	if (rc) {
3278 		if (rc == -EOPNOTSUPP)
3279 			netif_warn(adapter, drv, adapter->netdev,
3280 				   "Cannot set host attributes\n");
3281 		else
3282 			netif_err(adapter, drv, adapter->netdev,
3283 				  "Cannot set host attributes\n");
3284 		goto err;
3285 	}
3286 
3287 	return;
3288 err:
3289 	ena_com_delete_debug_area(adapter->ena_dev);
3290 }
3291 
3292 int ena_update_hw_stats(struct ena_adapter *adapter)
3293 {
3294 	int rc;
3295 
3296 	rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3297 	if (rc) {
3298 		netdev_err(adapter->netdev, "Failed to get ENI stats\n");
3299 		return rc;
3300 	}
3301 
3302 	return 0;
3303 }
3304 
3305 static void ena_get_stats64(struct net_device *netdev,
3306 			    struct rtnl_link_stats64 *stats)
3307 {
3308 	struct ena_adapter *adapter = netdev_priv(netdev);
3309 	struct ena_ring *rx_ring, *tx_ring;
3310 	unsigned int start;
3311 	u64 rx_drops;
3312 	u64 tx_drops;
3313 	int i;
3314 
3315 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3316 		return;
3317 
3318 	for (i = 0; i < adapter->num_io_queues; i++) {
3319 		u64 bytes, packets;
3320 
3321 		tx_ring = &adapter->tx_ring[i];
3322 
3323 		do {
3324 			start = u64_stats_fetch_begin(&tx_ring->syncp);
3325 			packets = tx_ring->tx_stats.cnt;
3326 			bytes = tx_ring->tx_stats.bytes;
3327 		} while (u64_stats_fetch_retry(&tx_ring->syncp, start));
3328 
3329 		stats->tx_packets += packets;
3330 		stats->tx_bytes += bytes;
3331 
3332 		rx_ring = &adapter->rx_ring[i];
3333 
3334 		do {
3335 			start = u64_stats_fetch_begin(&rx_ring->syncp);
3336 			packets = rx_ring->rx_stats.cnt;
3337 			bytes = rx_ring->rx_stats.bytes;
3338 		} while (u64_stats_fetch_retry(&rx_ring->syncp, start));
3339 
3340 		stats->rx_packets += packets;
3341 		stats->rx_bytes += bytes;
3342 	}
3343 
3344 	do {
3345 		start = u64_stats_fetch_begin(&adapter->syncp);
3346 		rx_drops = adapter->dev_stats.rx_drops;
3347 		tx_drops = adapter->dev_stats.tx_drops;
3348 	} while (u64_stats_fetch_retry(&adapter->syncp, start));
3349 
3350 	stats->rx_dropped = rx_drops;
3351 	stats->tx_dropped = tx_drops;
3352 
3353 	stats->multicast = 0;
3354 	stats->collisions = 0;
3355 
3356 	stats->rx_length_errors = 0;
3357 	stats->rx_crc_errors = 0;
3358 	stats->rx_frame_errors = 0;
3359 	stats->rx_fifo_errors = 0;
3360 	stats->rx_missed_errors = 0;
3361 	stats->tx_window_errors = 0;
3362 
3363 	stats->rx_errors = 0;
3364 	stats->tx_errors = 0;
3365 }
3366 
3367 static const struct net_device_ops ena_netdev_ops = {
3368 	.ndo_open		= ena_open,
3369 	.ndo_stop		= ena_close,
3370 	.ndo_start_xmit		= ena_start_xmit,
3371 	.ndo_select_queue	= ena_select_queue,
3372 	.ndo_get_stats64	= ena_get_stats64,
3373 	.ndo_tx_timeout		= ena_tx_timeout,
3374 	.ndo_change_mtu		= ena_change_mtu,
3375 	.ndo_set_mac_address	= NULL,
3376 	.ndo_validate_addr	= eth_validate_addr,
3377 	.ndo_bpf		= ena_xdp,
3378 	.ndo_xdp_xmit		= ena_xdp_xmit,
3379 };
3380 
3381 static void ena_calc_io_queue_size(struct ena_adapter *adapter,
3382 				   struct ena_com_dev_get_features_ctx *get_feat_ctx)
3383 {
3384 	struct ena_admin_feature_llq_desc *llq = &get_feat_ctx->llq;
3385 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3386 	u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
3387 	u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
3388 	u32 max_tx_queue_size;
3389 	u32 max_rx_queue_size;
3390 
3391 	/* If this function is called after driver load, the ring sizes have already
3392 	 * been configured. Take it into account when recalculating ring size.
3393 	 */
3394 	if (adapter->tx_ring->ring_size)
3395 		tx_queue_size = adapter->tx_ring->ring_size;
3396 
3397 	if (adapter->rx_ring->ring_size)
3398 		rx_queue_size = adapter->rx_ring->ring_size;
3399 
3400 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3401 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3402 			&get_feat_ctx->max_queue_ext.max_queue_ext;
3403 		max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
3404 					  max_queue_ext->max_rx_sq_depth);
3405 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
3406 
3407 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3408 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3409 						  llq->max_llq_depth);
3410 		else
3411 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3412 						  max_queue_ext->max_tx_sq_depth);
3413 
3414 		adapter->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3415 						 max_queue_ext->max_per_packet_tx_descs);
3416 		adapter->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3417 						 max_queue_ext->max_per_packet_rx_descs);
3418 	} else {
3419 		struct ena_admin_queue_feature_desc *max_queues =
3420 			&get_feat_ctx->max_queues;
3421 		max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
3422 					  max_queues->max_sq_depth);
3423 		max_tx_queue_size = max_queues->max_cq_depth;
3424 
3425 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3426 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3427 						  llq->max_llq_depth);
3428 		else
3429 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
3430 						  max_queues->max_sq_depth);
3431 
3432 		adapter->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3433 						 max_queues->max_packet_tx_descs);
3434 		adapter->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3435 						 max_queues->max_packet_rx_descs);
3436 	}
3437 
3438 	max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
3439 	max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
3440 
3441 	/* When forcing large headers, we multiply the entry size by 2, and therefore divide
3442 	 * the queue size by 2, leaving the amount of memory used by the queues unchanged.
3443 	 */
3444 	if (adapter->large_llq_header_enabled) {
3445 		if ((llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
3446 		    ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3447 			max_tx_queue_size /= 2;
3448 			dev_info(&adapter->pdev->dev,
3449 				 "Forcing large headers and decreasing maximum TX queue size to %d\n",
3450 				 max_tx_queue_size);
3451 		} else {
3452 			dev_err(&adapter->pdev->dev,
3453 				"Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
3454 
3455 			adapter->large_llq_header_enabled = false;
3456 		}
3457 	}
3458 
3459 	tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
3460 				  max_tx_queue_size);
3461 	rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
3462 				  max_rx_queue_size);
3463 
3464 	tx_queue_size = rounddown_pow_of_two(tx_queue_size);
3465 	rx_queue_size = rounddown_pow_of_two(rx_queue_size);
3466 
3467 	adapter->max_tx_ring_size  = max_tx_queue_size;
3468 	adapter->max_rx_ring_size = max_rx_queue_size;
3469 	adapter->requested_tx_ring_size = tx_queue_size;
3470 	adapter->requested_rx_ring_size = rx_queue_size;
3471 }
3472 
3473 static int ena_device_validate_params(struct ena_adapter *adapter,
3474 				      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3475 {
3476 	struct net_device *netdev = adapter->netdev;
3477 	int rc;
3478 
3479 	rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3480 			      adapter->mac_addr);
3481 	if (!rc) {
3482 		netif_err(adapter, drv, netdev,
3483 			  "Error, mac address are different\n");
3484 		return -EINVAL;
3485 	}
3486 
3487 	if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3488 		netif_err(adapter, drv, netdev,
3489 			  "Error, device max mtu is smaller than netdev MTU\n");
3490 		return -EINVAL;
3491 	}
3492 
3493 	return 0;
3494 }
3495 
3496 static void set_default_llq_configurations(struct ena_adapter *adapter,
3497 					   struct ena_llq_configurations *llq_config,
3498 					   struct ena_admin_feature_llq_desc *llq)
3499 {
3500 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3501 
3502 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3503 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3504 	llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3505 
3506 	adapter->large_llq_header_supported =
3507 		!!(ena_dev->supported_features & BIT(ENA_ADMIN_LLQ));
3508 	adapter->large_llq_header_supported &=
3509 		!!(llq->entry_size_ctrl_supported &
3510 			ENA_ADMIN_LIST_ENTRY_SIZE_256B);
3511 
3512 	if ((llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
3513 	    adapter->large_llq_header_enabled) {
3514 		llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_256B;
3515 		llq_config->llq_ring_entry_size_value = 256;
3516 	} else {
3517 		llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3518 		llq_config->llq_ring_entry_size_value = 128;
3519 	}
3520 }
3521 
3522 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3523 					   struct ena_com_dev *ena_dev,
3524 					   struct ena_admin_feature_llq_desc *llq,
3525 					   struct ena_llq_configurations *llq_default_configurations)
3526 {
3527 	int rc;
3528 	u32 llq_feature_mask;
3529 
3530 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3531 	if (!(ena_dev->supported_features & llq_feature_mask)) {
3532 		dev_warn(&pdev->dev,
3533 			"LLQ is not supported Fallback to host mode policy.\n");
3534 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3535 		return 0;
3536 	}
3537 
3538 	if (!ena_dev->mem_bar) {
3539 		netdev_err(ena_dev->net_device,
3540 			   "LLQ is advertised as supported but device doesn't expose mem bar\n");
3541 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3542 		return 0;
3543 	}
3544 
3545 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3546 	if (unlikely(rc)) {
3547 		dev_err(&pdev->dev,
3548 			"Failed to configure the device mode.  Fallback to host mode policy.\n");
3549 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3550 	}
3551 
3552 	return 0;
3553 }
3554 
3555 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3556 			       int bars)
3557 {
3558 	bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3559 
3560 	if (!has_mem_bar)
3561 		return 0;
3562 
3563 	ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3564 					   pci_resource_start(pdev, ENA_MEM_BAR),
3565 					   pci_resource_len(pdev, ENA_MEM_BAR));
3566 
3567 	if (!ena_dev->mem_bar)
3568 		return -EFAULT;
3569 
3570 	return 0;
3571 }
3572 
3573 static int ena_device_init(struct ena_adapter *adapter, struct pci_dev *pdev,
3574 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
3575 			   bool *wd_state)
3576 {
3577 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3578 	struct ena_llq_configurations llq_config;
3579 	struct device *dev = &pdev->dev;
3580 	bool readless_supported;
3581 	u32 aenq_groups;
3582 	int dma_width;
3583 	int rc;
3584 
3585 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
3586 	if (rc) {
3587 		dev_err(dev, "Failed to init mmio read less\n");
3588 		return rc;
3589 	}
3590 
3591 	/* The PCIe configuration space revision id indicate if mmio reg
3592 	 * read is disabled
3593 	 */
3594 	readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3595 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3596 
3597 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3598 	if (rc) {
3599 		dev_err(dev, "Can not reset device\n");
3600 		goto err_mmio_read_less;
3601 	}
3602 
3603 	rc = ena_com_validate_version(ena_dev);
3604 	if (rc) {
3605 		dev_err(dev, "Device version is too low\n");
3606 		goto err_mmio_read_less;
3607 	}
3608 
3609 	dma_width = ena_com_get_dma_width(ena_dev);
3610 	if (dma_width < 0) {
3611 		dev_err(dev, "Invalid dma width value %d", dma_width);
3612 		rc = dma_width;
3613 		goto err_mmio_read_less;
3614 	}
3615 
3616 	rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3617 	if (rc) {
3618 		dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3619 		goto err_mmio_read_less;
3620 	}
3621 
3622 	/* ENA admin level init */
3623 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3624 	if (rc) {
3625 		dev_err(dev,
3626 			"Can not initialize ena admin queue with device\n");
3627 		goto err_mmio_read_less;
3628 	}
3629 
3630 	/* To enable the msix interrupts the driver needs to know the number
3631 	 * of queues. So the driver uses polling mode to retrieve this
3632 	 * information
3633 	 */
3634 	ena_com_set_admin_polling_mode(ena_dev, true);
3635 
3636 	ena_config_host_info(ena_dev, pdev);
3637 
3638 	/* Get Device Attributes*/
3639 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3640 	if (rc) {
3641 		dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3642 		goto err_admin_init;
3643 	}
3644 
3645 	/* Try to turn all the available aenq groups */
3646 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3647 		BIT(ENA_ADMIN_FATAL_ERROR) |
3648 		BIT(ENA_ADMIN_WARNING) |
3649 		BIT(ENA_ADMIN_NOTIFICATION) |
3650 		BIT(ENA_ADMIN_KEEP_ALIVE);
3651 
3652 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
3653 
3654 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3655 	if (rc) {
3656 		dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3657 		goto err_admin_init;
3658 	}
3659 
3660 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3661 
3662 	set_default_llq_configurations(adapter, &llq_config, &get_feat_ctx->llq);
3663 
3664 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3665 					     &llq_config);
3666 	if (rc) {
3667 		dev_err(dev, "ENA device init failed\n");
3668 		goto err_admin_init;
3669 	}
3670 
3671 	ena_calc_io_queue_size(adapter, get_feat_ctx);
3672 
3673 	return 0;
3674 
3675 err_admin_init:
3676 	ena_com_delete_host_info(ena_dev);
3677 	ena_com_admin_destroy(ena_dev);
3678 err_mmio_read_less:
3679 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3680 
3681 	return rc;
3682 }
3683 
3684 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3685 {
3686 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3687 	struct device *dev = &adapter->pdev->dev;
3688 	int rc;
3689 
3690 	rc = ena_enable_msix(adapter);
3691 	if (rc) {
3692 		dev_err(dev, "Can not reserve msix vectors\n");
3693 		return rc;
3694 	}
3695 
3696 	ena_setup_mgmnt_intr(adapter);
3697 
3698 	rc = ena_request_mgmnt_irq(adapter);
3699 	if (rc) {
3700 		dev_err(dev, "Can not setup management interrupts\n");
3701 		goto err_disable_msix;
3702 	}
3703 
3704 	ena_com_set_admin_polling_mode(ena_dev, false);
3705 
3706 	ena_com_admin_aenq_enable(ena_dev);
3707 
3708 	return 0;
3709 
3710 err_disable_msix:
3711 	ena_disable_msix(adapter);
3712 
3713 	return rc;
3714 }
3715 
3716 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3717 {
3718 	struct net_device *netdev = adapter->netdev;
3719 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3720 	bool dev_up;
3721 
3722 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3723 		return;
3724 
3725 	netif_carrier_off(netdev);
3726 
3727 	del_timer_sync(&adapter->timer_service);
3728 
3729 	dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3730 	adapter->dev_up_before_reset = dev_up;
3731 	if (!graceful)
3732 		ena_com_set_admin_running_state(ena_dev, false);
3733 
3734 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3735 		ena_down(adapter);
3736 
3737 	/* Stop the device from sending AENQ events (in case reset flag is set
3738 	 *  and device is up, ena_down() already reset the device.
3739 	 */
3740 	if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3741 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3742 
3743 	ena_free_mgmnt_irq(adapter);
3744 
3745 	ena_disable_msix(adapter);
3746 
3747 	ena_com_abort_admin_commands(ena_dev);
3748 
3749 	ena_com_wait_for_abort_completion(ena_dev);
3750 
3751 	ena_com_admin_destroy(ena_dev);
3752 
3753 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3754 
3755 	/* return reset reason to default value */
3756 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3757 
3758 	clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3759 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3760 }
3761 
3762 static int ena_restore_device(struct ena_adapter *adapter)
3763 {
3764 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3765 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3766 	struct pci_dev *pdev = adapter->pdev;
3767 	struct ena_ring *txr;
3768 	int rc, count, i;
3769 	bool wd_state;
3770 
3771 	set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3772 	rc = ena_device_init(adapter, adapter->pdev, &get_feat_ctx, &wd_state);
3773 	if (rc) {
3774 		dev_err(&pdev->dev, "Can not initialize device\n");
3775 		goto err;
3776 	}
3777 	adapter->wd_state = wd_state;
3778 
3779 	count =  adapter->xdp_num_queues + adapter->num_io_queues;
3780 	for (i = 0 ; i < count; i++) {
3781 		txr = &adapter->tx_ring[i];
3782 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
3783 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
3784 	}
3785 
3786 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
3787 	if (rc) {
3788 		dev_err(&pdev->dev, "Validation of device parameters failed\n");
3789 		goto err_device_destroy;
3790 	}
3791 
3792 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3793 	if (rc) {
3794 		dev_err(&pdev->dev, "Enable MSI-X failed\n");
3795 		goto err_device_destroy;
3796 	}
3797 	/* If the interface was up before the reset bring it up */
3798 	if (adapter->dev_up_before_reset) {
3799 		rc = ena_up(adapter);
3800 		if (rc) {
3801 			dev_err(&pdev->dev, "Failed to create I/O queues\n");
3802 			goto err_disable_msix;
3803 		}
3804 	}
3805 
3806 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3807 
3808 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3809 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3810 		netif_carrier_on(adapter->netdev);
3811 
3812 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3813 	adapter->last_keep_alive_jiffies = jiffies;
3814 
3815 	return rc;
3816 err_disable_msix:
3817 	ena_free_mgmnt_irq(adapter);
3818 	ena_disable_msix(adapter);
3819 err_device_destroy:
3820 	ena_com_abort_admin_commands(ena_dev);
3821 	ena_com_wait_for_abort_completion(ena_dev);
3822 	ena_com_admin_destroy(ena_dev);
3823 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3824 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3825 err:
3826 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3827 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3828 	dev_err(&pdev->dev,
3829 		"Reset attempt failed. Can not reset the device\n");
3830 
3831 	return rc;
3832 }
3833 
3834 static void ena_fw_reset_device(struct work_struct *work)
3835 {
3836 	struct ena_adapter *adapter =
3837 		container_of(work, struct ena_adapter, reset_task);
3838 
3839 	rtnl_lock();
3840 
3841 	if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3842 		ena_destroy_device(adapter, false);
3843 		ena_restore_device(adapter);
3844 
3845 		dev_err(&adapter->pdev->dev, "Device reset completed successfully\n");
3846 	}
3847 
3848 	rtnl_unlock();
3849 }
3850 
3851 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3852 					struct ena_ring *rx_ring)
3853 {
3854 	struct ena_napi *ena_napi = container_of(rx_ring->napi, struct ena_napi, napi);
3855 
3856 	if (likely(READ_ONCE(ena_napi->first_interrupt)))
3857 		return 0;
3858 
3859 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3860 		return 0;
3861 
3862 	rx_ring->no_interrupt_event_cnt++;
3863 
3864 	if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3865 		netif_err(adapter, rx_err, adapter->netdev,
3866 			  "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3867 			  rx_ring->qid);
3868 
3869 		ena_reset_device(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
3870 		return -EIO;
3871 	}
3872 
3873 	return 0;
3874 }
3875 
3876 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3877 					  struct ena_ring *tx_ring)
3878 {
3879 	struct ena_napi *ena_napi = container_of(tx_ring->napi, struct ena_napi, napi);
3880 	unsigned int time_since_last_napi;
3881 	unsigned int missing_tx_comp_to;
3882 	bool is_tx_comp_time_expired;
3883 	struct ena_tx_buffer *tx_buf;
3884 	unsigned long last_jiffies;
3885 	u32 missed_tx = 0;
3886 	int i, rc = 0;
3887 
3888 	for (i = 0; i < tx_ring->ring_size; i++) {
3889 		tx_buf = &tx_ring->tx_buffer_info[i];
3890 		last_jiffies = tx_buf->last_jiffies;
3891 
3892 		if (last_jiffies == 0)
3893 			/* no pending Tx at this location */
3894 			continue;
3895 
3896 		is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3897 			 2 * adapter->missing_tx_completion_to);
3898 
3899 		if (unlikely(!READ_ONCE(ena_napi->first_interrupt) && is_tx_comp_time_expired)) {
3900 			/* If after graceful period interrupt is still not
3901 			 * received, we schedule a reset
3902 			 */
3903 			netif_err(adapter, tx_err, adapter->netdev,
3904 				  "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3905 				  tx_ring->qid);
3906 			ena_reset_device(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
3907 			return -EIO;
3908 		}
3909 
3910 		is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3911 			adapter->missing_tx_completion_to);
3912 
3913 		if (unlikely(is_tx_comp_time_expired)) {
3914 			if (!tx_buf->print_once) {
3915 				time_since_last_napi = jiffies_to_usecs(jiffies - tx_ring->tx_stats.last_napi_jiffies);
3916 				missing_tx_comp_to = jiffies_to_msecs(adapter->missing_tx_completion_to);
3917 				netif_notice(adapter, tx_err, adapter->netdev,
3918 					     "Found a Tx that wasn't completed on time, qid %d, index %d. %u usecs have passed since last napi execution. Missing Tx timeout value %u msecs\n",
3919 					     tx_ring->qid, i, time_since_last_napi, missing_tx_comp_to);
3920 			}
3921 
3922 			tx_buf->print_once = 1;
3923 			missed_tx++;
3924 		}
3925 	}
3926 
3927 	if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3928 		netif_err(adapter, tx_err, adapter->netdev,
3929 			  "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3930 			  missed_tx,
3931 			  adapter->missing_tx_completion_threshold);
3932 		ena_reset_device(adapter, ENA_REGS_RESET_MISS_TX_CMPL);
3933 		rc = -EIO;
3934 	}
3935 
3936 	ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3937 			  &tx_ring->syncp);
3938 
3939 	return rc;
3940 }
3941 
3942 static void check_for_missing_completions(struct ena_adapter *adapter)
3943 {
3944 	struct ena_ring *tx_ring;
3945 	struct ena_ring *rx_ring;
3946 	int i, budget, rc;
3947 	int io_queue_count;
3948 
3949 	io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3950 	/* Make sure the driver doesn't turn the device in other process */
3951 	smp_rmb();
3952 
3953 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3954 		return;
3955 
3956 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3957 		return;
3958 
3959 	if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3960 		return;
3961 
3962 	budget = ENA_MONITORED_TX_QUEUES;
3963 
3964 	for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3965 		tx_ring = &adapter->tx_ring[i];
3966 		rx_ring = &adapter->rx_ring[i];
3967 
3968 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3969 		if (unlikely(rc))
3970 			return;
3971 
3972 		rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3973 			check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3974 		if (unlikely(rc))
3975 			return;
3976 
3977 		budget--;
3978 		if (!budget)
3979 			break;
3980 	}
3981 
3982 	adapter->last_monitored_tx_qid = i % io_queue_count;
3983 }
3984 
3985 /* trigger napi schedule after 2 consecutive detections */
3986 #define EMPTY_RX_REFILL 2
3987 /* For the rare case where the device runs out of Rx descriptors and the
3988  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3989  * for example).
3990  * This case will lead to a deadlock:
3991  * The device won't send interrupts since all the new Rx packets will be dropped
3992  * The napi handler won't allocate new Rx descriptors so the device will be
3993  * able to send new packets.
3994  *
3995  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3996  * It is recommended to have at least 512MB, with a minimum of 128MB for
3997  * constrained environment).
3998  *
3999  * When such a situation is detected - Reschedule napi
4000  */
4001 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
4002 {
4003 	struct ena_ring *rx_ring;
4004 	int i, refill_required;
4005 
4006 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
4007 		return;
4008 
4009 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
4010 		return;
4011 
4012 	for (i = 0; i < adapter->num_io_queues; i++) {
4013 		rx_ring = &adapter->rx_ring[i];
4014 
4015 		refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
4016 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
4017 			rx_ring->empty_rx_queue++;
4018 
4019 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
4020 				ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
4021 						  &rx_ring->syncp);
4022 
4023 				netif_err(adapter, drv, adapter->netdev,
4024 					  "Trigger refill for ring %d\n", i);
4025 
4026 				napi_schedule(rx_ring->napi);
4027 				rx_ring->empty_rx_queue = 0;
4028 			}
4029 		} else {
4030 			rx_ring->empty_rx_queue = 0;
4031 		}
4032 	}
4033 }
4034 
4035 /* Check for keep alive expiration */
4036 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
4037 {
4038 	unsigned long keep_alive_expired;
4039 
4040 	if (!adapter->wd_state)
4041 		return;
4042 
4043 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
4044 		return;
4045 
4046 	keep_alive_expired = adapter->last_keep_alive_jiffies +
4047 			     adapter->keep_alive_timeout;
4048 	if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
4049 		netif_err(adapter, drv, adapter->netdev,
4050 			  "Keep alive watchdog timeout.\n");
4051 		ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
4052 				  &adapter->syncp);
4053 		ena_reset_device(adapter, ENA_REGS_RESET_KEEP_ALIVE_TO);
4054 	}
4055 }
4056 
4057 static void check_for_admin_com_state(struct ena_adapter *adapter)
4058 {
4059 	if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
4060 		netif_err(adapter, drv, adapter->netdev,
4061 			  "ENA admin queue is not in running state!\n");
4062 		ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
4063 				  &adapter->syncp);
4064 		ena_reset_device(adapter, ENA_REGS_RESET_ADMIN_TO);
4065 	}
4066 }
4067 
4068 static void ena_update_hints(struct ena_adapter *adapter,
4069 			     struct ena_admin_ena_hw_hints *hints)
4070 {
4071 	struct net_device *netdev = adapter->netdev;
4072 
4073 	if (hints->admin_completion_tx_timeout)
4074 		adapter->ena_dev->admin_queue.completion_timeout =
4075 			hints->admin_completion_tx_timeout * 1000;
4076 
4077 	if (hints->mmio_read_timeout)
4078 		/* convert to usec */
4079 		adapter->ena_dev->mmio_read.reg_read_to =
4080 			hints->mmio_read_timeout * 1000;
4081 
4082 	if (hints->missed_tx_completion_count_threshold_to_reset)
4083 		adapter->missing_tx_completion_threshold =
4084 			hints->missed_tx_completion_count_threshold_to_reset;
4085 
4086 	if (hints->missing_tx_completion_timeout) {
4087 		if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
4088 			adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
4089 		else
4090 			adapter->missing_tx_completion_to =
4091 				msecs_to_jiffies(hints->missing_tx_completion_timeout);
4092 	}
4093 
4094 	if (hints->netdev_wd_timeout)
4095 		netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
4096 
4097 	if (hints->driver_watchdog_timeout) {
4098 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
4099 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
4100 		else
4101 			adapter->keep_alive_timeout =
4102 				msecs_to_jiffies(hints->driver_watchdog_timeout);
4103 	}
4104 }
4105 
4106 static void ena_update_host_info(struct ena_admin_host_info *host_info,
4107 				 struct net_device *netdev)
4108 {
4109 	host_info->supported_network_features[0] =
4110 		netdev->features & GENMASK_ULL(31, 0);
4111 	host_info->supported_network_features[1] =
4112 		(netdev->features & GENMASK_ULL(63, 32)) >> 32;
4113 }
4114 
4115 static void ena_timer_service(struct timer_list *t)
4116 {
4117 	struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
4118 	u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
4119 	struct ena_admin_host_info *host_info =
4120 		adapter->ena_dev->host_attr.host_info;
4121 
4122 	check_for_missing_keep_alive(adapter);
4123 
4124 	check_for_admin_com_state(adapter);
4125 
4126 	check_for_missing_completions(adapter);
4127 
4128 	check_for_empty_rx_ring(adapter);
4129 
4130 	if (debug_area)
4131 		ena_dump_stats_to_buf(adapter, debug_area);
4132 
4133 	if (host_info)
4134 		ena_update_host_info(host_info, adapter->netdev);
4135 
4136 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4137 		netif_err(adapter, drv, adapter->netdev,
4138 			  "Trigger reset is on\n");
4139 		ena_dump_stats_to_dmesg(adapter);
4140 		queue_work(ena_wq, &adapter->reset_task);
4141 		return;
4142 	}
4143 
4144 	/* Reset the timer */
4145 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4146 }
4147 
4148 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
4149 				     struct ena_com_dev *ena_dev,
4150 				     struct ena_com_dev_get_features_ctx *get_feat_ctx)
4151 {
4152 	u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
4153 
4154 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4155 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4156 			&get_feat_ctx->max_queue_ext.max_queue_ext;
4157 		io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
4158 				  max_queue_ext->max_rx_cq_num);
4159 
4160 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
4161 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
4162 	} else {
4163 		struct ena_admin_queue_feature_desc *max_queues =
4164 			&get_feat_ctx->max_queues;
4165 		io_tx_sq_num = max_queues->max_sq_num;
4166 		io_tx_cq_num = max_queues->max_cq_num;
4167 		io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
4168 	}
4169 
4170 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
4171 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4172 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
4173 
4174 	max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
4175 	max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
4176 	max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
4177 	max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
4178 	/* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
4179 	max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
4180 
4181 	return max_num_io_queues;
4182 }
4183 
4184 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
4185 				 struct net_device *netdev)
4186 {
4187 	netdev_features_t dev_features = 0;
4188 
4189 	/* Set offload features */
4190 	if (feat->offload.tx &
4191 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
4192 		dev_features |= NETIF_F_IP_CSUM;
4193 
4194 	if (feat->offload.tx &
4195 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4196 		dev_features |= NETIF_F_IPV6_CSUM;
4197 
4198 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4199 		dev_features |= NETIF_F_TSO;
4200 
4201 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4202 		dev_features |= NETIF_F_TSO6;
4203 
4204 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4205 		dev_features |= NETIF_F_TSO_ECN;
4206 
4207 	if (feat->offload.rx_supported &
4208 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4209 		dev_features |= NETIF_F_RXCSUM;
4210 
4211 	if (feat->offload.rx_supported &
4212 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4213 		dev_features |= NETIF_F_RXCSUM;
4214 
4215 	netdev->features =
4216 		dev_features |
4217 		NETIF_F_SG |
4218 		NETIF_F_RXHASH |
4219 		NETIF_F_HIGHDMA;
4220 
4221 	netdev->hw_features |= netdev->features;
4222 	netdev->vlan_features |= netdev->features;
4223 }
4224 
4225 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4226 				     struct ena_com_dev_get_features_ctx *feat)
4227 {
4228 	struct net_device *netdev = adapter->netdev;
4229 
4230 	/* Copy mac address */
4231 	if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4232 		eth_hw_addr_random(netdev);
4233 		ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4234 	} else {
4235 		ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4236 		eth_hw_addr_set(netdev, adapter->mac_addr);
4237 	}
4238 
4239 	/* Set offload features */
4240 	ena_set_dev_offloads(feat, netdev);
4241 
4242 	adapter->max_mtu = feat->dev_attr.max_mtu;
4243 	netdev->max_mtu = adapter->max_mtu;
4244 	netdev->min_mtu = ENA_MIN_MTU;
4245 }
4246 
4247 static int ena_rss_init_default(struct ena_adapter *adapter)
4248 {
4249 	struct ena_com_dev *ena_dev = adapter->ena_dev;
4250 	struct device *dev = &adapter->pdev->dev;
4251 	int rc, i;
4252 	u32 val;
4253 
4254 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4255 	if (unlikely(rc)) {
4256 		dev_err(dev, "Cannot init indirect table\n");
4257 		goto err_rss_init;
4258 	}
4259 
4260 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4261 		val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4262 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4263 						       ENA_IO_RXQ_IDX(val));
4264 		if (unlikely(rc)) {
4265 			dev_err(dev, "Cannot fill indirect table\n");
4266 			goto err_fill_indir;
4267 		}
4268 	}
4269 
4270 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4271 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4272 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4273 		dev_err(dev, "Cannot fill hash function\n");
4274 		goto err_fill_indir;
4275 	}
4276 
4277 	rc = ena_com_set_default_hash_ctrl(ena_dev);
4278 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4279 		dev_err(dev, "Cannot fill hash control\n");
4280 		goto err_fill_indir;
4281 	}
4282 
4283 	return 0;
4284 
4285 err_fill_indir:
4286 	ena_com_rss_destroy(ena_dev);
4287 err_rss_init:
4288 
4289 	return rc;
4290 }
4291 
4292 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4293 {
4294 	int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4295 
4296 	pci_release_selected_regions(pdev, release_bars);
4297 }
4298 
4299 /* ena_probe - Device Initialization Routine
4300  * @pdev: PCI device information struct
4301  * @ent: entry in ena_pci_tbl
4302  *
4303  * Returns 0 on success, negative on failure
4304  *
4305  * ena_probe initializes an adapter identified by a pci_dev structure.
4306  * The OS initialization, configuring of the adapter private structure,
4307  * and a hardware reset occur.
4308  */
4309 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4310 {
4311 	struct ena_com_dev_get_features_ctx get_feat_ctx;
4312 	struct ena_com_dev *ena_dev = NULL;
4313 	struct ena_adapter *adapter;
4314 	struct net_device *netdev;
4315 	static int adapters_found;
4316 	u32 max_num_io_queues;
4317 	bool wd_state;
4318 	int bars, rc;
4319 
4320 	dev_dbg(&pdev->dev, "%s\n", __func__);
4321 
4322 	rc = pci_enable_device_mem(pdev);
4323 	if (rc) {
4324 		dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4325 		return rc;
4326 	}
4327 
4328 	rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4329 	if (rc) {
4330 		dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4331 		goto err_disable_device;
4332 	}
4333 
4334 	pci_set_master(pdev);
4335 
4336 	ena_dev = vzalloc(sizeof(*ena_dev));
4337 	if (!ena_dev) {
4338 		rc = -ENOMEM;
4339 		goto err_disable_device;
4340 	}
4341 
4342 	bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4343 	rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4344 	if (rc) {
4345 		dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4346 			rc);
4347 		goto err_free_ena_dev;
4348 	}
4349 
4350 	ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4351 					pci_resource_start(pdev, ENA_REG_BAR),
4352 					pci_resource_len(pdev, ENA_REG_BAR));
4353 	if (!ena_dev->reg_bar) {
4354 		dev_err(&pdev->dev, "Failed to remap regs bar\n");
4355 		rc = -EFAULT;
4356 		goto err_free_region;
4357 	}
4358 
4359 	ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4360 
4361 	ena_dev->dmadev = &pdev->dev;
4362 
4363 	netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4364 	if (!netdev) {
4365 		dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4366 		rc = -ENOMEM;
4367 		goto err_free_region;
4368 	}
4369 
4370 	SET_NETDEV_DEV(netdev, &pdev->dev);
4371 	adapter = netdev_priv(netdev);
4372 	adapter->ena_dev = ena_dev;
4373 	adapter->netdev = netdev;
4374 	adapter->pdev = pdev;
4375 	adapter->msg_enable = DEFAULT_MSG_ENABLE;
4376 
4377 	ena_dev->net_device = netdev;
4378 
4379 	pci_set_drvdata(pdev, adapter);
4380 
4381 	rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4382 	if (rc) {
4383 		dev_err(&pdev->dev, "ENA LLQ bar mapping failed\n");
4384 		goto err_netdev_destroy;
4385 	}
4386 
4387 	rc = ena_device_init(adapter, pdev, &get_feat_ctx, &wd_state);
4388 	if (rc) {
4389 		dev_err(&pdev->dev, "ENA device init failed\n");
4390 		if (rc == -ETIME)
4391 			rc = -EPROBE_DEFER;
4392 		goto err_netdev_destroy;
4393 	}
4394 
4395 	/* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4396 	 * Updated during device initialization with the real granularity
4397 	 */
4398 	ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4399 	ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4400 	ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4401 	max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4402 	if (unlikely(!max_num_io_queues)) {
4403 		rc = -EFAULT;
4404 		goto err_device_destroy;
4405 	}
4406 
4407 	ena_set_conf_feat_params(adapter, &get_feat_ctx);
4408 
4409 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4410 
4411 	adapter->num_io_queues = max_num_io_queues;
4412 	adapter->max_num_io_queues = max_num_io_queues;
4413 	adapter->last_monitored_tx_qid = 0;
4414 
4415 	adapter->xdp_first_ring = 0;
4416 	adapter->xdp_num_queues = 0;
4417 
4418 	adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4419 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4420 		adapter->disable_meta_caching =
4421 			!!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4422 			   BIT(ENA_ADMIN_DISABLE_META_CACHING));
4423 
4424 	adapter->wd_state = wd_state;
4425 
4426 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4427 
4428 	rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4429 	if (rc) {
4430 		dev_err(&pdev->dev,
4431 			"Failed to query interrupt moderation feature\n");
4432 		goto err_device_destroy;
4433 	}
4434 
4435 	ena_init_io_rings(adapter,
4436 			  0,
4437 			  adapter->xdp_num_queues +
4438 			  adapter->num_io_queues);
4439 
4440 	netdev->netdev_ops = &ena_netdev_ops;
4441 	netdev->watchdog_timeo = TX_TIMEOUT;
4442 	ena_set_ethtool_ops(netdev);
4443 
4444 	netdev->priv_flags |= IFF_UNICAST_FLT;
4445 
4446 	u64_stats_init(&adapter->syncp);
4447 
4448 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4449 	if (rc) {
4450 		dev_err(&pdev->dev,
4451 			"Failed to enable and set the admin interrupts\n");
4452 		goto err_worker_destroy;
4453 	}
4454 	rc = ena_rss_init_default(adapter);
4455 	if (rc && (rc != -EOPNOTSUPP)) {
4456 		dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4457 		goto err_free_msix;
4458 	}
4459 
4460 	ena_config_debug_area(adapter);
4461 
4462 	if (ena_xdp_legal_queue_count(adapter, adapter->num_io_queues))
4463 		netdev->xdp_features = NETDEV_XDP_ACT_BASIC |
4464 				       NETDEV_XDP_ACT_REDIRECT;
4465 
4466 	memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4467 
4468 	netif_carrier_off(netdev);
4469 
4470 	rc = register_netdev(netdev);
4471 	if (rc) {
4472 		dev_err(&pdev->dev, "Cannot register net device\n");
4473 		goto err_rss;
4474 	}
4475 
4476 	INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4477 
4478 	adapter->last_keep_alive_jiffies = jiffies;
4479 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4480 	adapter->missing_tx_completion_to = TX_TIMEOUT;
4481 	adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4482 
4483 	ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4484 
4485 	timer_setup(&adapter->timer_service, ena_timer_service, 0);
4486 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4487 
4488 	dev_info(&pdev->dev,
4489 		 "%s found at mem %lx, mac addr %pM\n",
4490 		 DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4491 		 netdev->dev_addr);
4492 
4493 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4494 
4495 	adapters_found++;
4496 
4497 	return 0;
4498 
4499 err_rss:
4500 	ena_com_delete_debug_area(ena_dev);
4501 	ena_com_rss_destroy(ena_dev);
4502 err_free_msix:
4503 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4504 	/* stop submitting admin commands on a device that was reset */
4505 	ena_com_set_admin_running_state(ena_dev, false);
4506 	ena_free_mgmnt_irq(adapter);
4507 	ena_disable_msix(adapter);
4508 err_worker_destroy:
4509 	del_timer(&adapter->timer_service);
4510 err_device_destroy:
4511 	ena_com_delete_host_info(ena_dev);
4512 	ena_com_admin_destroy(ena_dev);
4513 err_netdev_destroy:
4514 	free_netdev(netdev);
4515 err_free_region:
4516 	ena_release_bars(ena_dev, pdev);
4517 err_free_ena_dev:
4518 	vfree(ena_dev);
4519 err_disable_device:
4520 	pci_disable_device(pdev);
4521 	return rc;
4522 }
4523 
4524 /*****************************************************************************/
4525 
4526 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4527  * @pdev: PCI device information struct
4528  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4529  *
4530  * __ena_shutoff is a helper routine that does the real work on shutdown and
4531  * removal paths; the difference between those paths is with regards to whether
4532  * dettach or unregister the netdevice.
4533  */
4534 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4535 {
4536 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4537 	struct ena_com_dev *ena_dev;
4538 	struct net_device *netdev;
4539 
4540 	ena_dev = adapter->ena_dev;
4541 	netdev = adapter->netdev;
4542 
4543 #ifdef CONFIG_RFS_ACCEL
4544 	if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4545 		free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4546 		netdev->rx_cpu_rmap = NULL;
4547 	}
4548 #endif /* CONFIG_RFS_ACCEL */
4549 
4550 	/* Make sure timer and reset routine won't be called after
4551 	 * freeing device resources.
4552 	 */
4553 	del_timer_sync(&adapter->timer_service);
4554 	cancel_work_sync(&adapter->reset_task);
4555 
4556 	rtnl_lock(); /* lock released inside the below if-else block */
4557 	adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4558 	ena_destroy_device(adapter, true);
4559 
4560 	if (shutdown) {
4561 		netif_device_detach(netdev);
4562 		dev_close(netdev);
4563 		rtnl_unlock();
4564 	} else {
4565 		rtnl_unlock();
4566 		unregister_netdev(netdev);
4567 		free_netdev(netdev);
4568 	}
4569 
4570 	ena_com_rss_destroy(ena_dev);
4571 
4572 	ena_com_delete_debug_area(ena_dev);
4573 
4574 	ena_com_delete_host_info(ena_dev);
4575 
4576 	ena_release_bars(ena_dev, pdev);
4577 
4578 	pci_disable_device(pdev);
4579 
4580 	vfree(ena_dev);
4581 }
4582 
4583 /* ena_remove - Device Removal Routine
4584  * @pdev: PCI device information struct
4585  *
4586  * ena_remove is called by the PCI subsystem to alert the driver
4587  * that it should release a PCI device.
4588  */
4589 
4590 static void ena_remove(struct pci_dev *pdev)
4591 {
4592 	__ena_shutoff(pdev, false);
4593 }
4594 
4595 /* ena_shutdown - Device Shutdown Routine
4596  * @pdev: PCI device information struct
4597  *
4598  * ena_shutdown is called by the PCI subsystem to alert the driver that
4599  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4600  */
4601 
4602 static void ena_shutdown(struct pci_dev *pdev)
4603 {
4604 	__ena_shutoff(pdev, true);
4605 }
4606 
4607 /* ena_suspend - PM suspend callback
4608  * @dev_d: Device information struct
4609  */
4610 static int __maybe_unused ena_suspend(struct device *dev_d)
4611 {
4612 	struct pci_dev *pdev = to_pci_dev(dev_d);
4613 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4614 
4615 	ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4616 
4617 	rtnl_lock();
4618 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4619 		dev_err(&pdev->dev,
4620 			"Ignoring device reset request as the device is being suspended\n");
4621 		clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4622 	}
4623 	ena_destroy_device(adapter, true);
4624 	rtnl_unlock();
4625 	return 0;
4626 }
4627 
4628 /* ena_resume - PM resume callback
4629  * @dev_d: Device information struct
4630  */
4631 static int __maybe_unused ena_resume(struct device *dev_d)
4632 {
4633 	struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4634 	int rc;
4635 
4636 	ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4637 
4638 	rtnl_lock();
4639 	rc = ena_restore_device(adapter);
4640 	rtnl_unlock();
4641 	return rc;
4642 }
4643 
4644 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4645 
4646 static struct pci_driver ena_pci_driver = {
4647 	.name		= DRV_MODULE_NAME,
4648 	.id_table	= ena_pci_tbl,
4649 	.probe		= ena_probe,
4650 	.remove		= ena_remove,
4651 	.shutdown	= ena_shutdown,
4652 	.driver.pm	= &ena_pm_ops,
4653 	.sriov_configure = pci_sriov_configure_simple,
4654 };
4655 
4656 static int __init ena_init(void)
4657 {
4658 	int ret;
4659 
4660 	ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4661 	if (!ena_wq) {
4662 		pr_err("Failed to create workqueue\n");
4663 		return -ENOMEM;
4664 	}
4665 
4666 	ret = pci_register_driver(&ena_pci_driver);
4667 	if (ret)
4668 		destroy_workqueue(ena_wq);
4669 
4670 	return ret;
4671 }
4672 
4673 static void __exit ena_cleanup(void)
4674 {
4675 	pci_unregister_driver(&ena_pci_driver);
4676 
4677 	if (ena_wq) {
4678 		destroy_workqueue(ena_wq);
4679 		ena_wq = NULL;
4680 	}
4681 }
4682 
4683 /******************************************************************************
4684  ******************************** AENQ Handlers *******************************
4685  *****************************************************************************/
4686 /* ena_update_on_link_change:
4687  * Notify the network interface about the change in link status
4688  */
4689 static void ena_update_on_link_change(void *adapter_data,
4690 				      struct ena_admin_aenq_entry *aenq_e)
4691 {
4692 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4693 	struct ena_admin_aenq_link_change_desc *aenq_desc =
4694 		(struct ena_admin_aenq_link_change_desc *)aenq_e;
4695 	int status = aenq_desc->flags &
4696 		ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4697 
4698 	if (status) {
4699 		netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4700 		set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4701 		if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4702 			netif_carrier_on(adapter->netdev);
4703 	} else {
4704 		clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4705 		netif_carrier_off(adapter->netdev);
4706 	}
4707 }
4708 
4709 static void ena_keep_alive_wd(void *adapter_data,
4710 			      struct ena_admin_aenq_entry *aenq_e)
4711 {
4712 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4713 	struct ena_admin_aenq_keep_alive_desc *desc;
4714 	u64 rx_drops;
4715 	u64 tx_drops;
4716 
4717 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4718 	adapter->last_keep_alive_jiffies = jiffies;
4719 
4720 	rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4721 	tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4722 
4723 	u64_stats_update_begin(&adapter->syncp);
4724 	/* These stats are accumulated by the device, so the counters indicate
4725 	 * all drops since last reset.
4726 	 */
4727 	adapter->dev_stats.rx_drops = rx_drops;
4728 	adapter->dev_stats.tx_drops = tx_drops;
4729 	u64_stats_update_end(&adapter->syncp);
4730 }
4731 
4732 static void ena_notification(void *adapter_data,
4733 			     struct ena_admin_aenq_entry *aenq_e)
4734 {
4735 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4736 	struct ena_admin_ena_hw_hints *hints;
4737 
4738 	WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4739 	     "Invalid group(%x) expected %x\n",
4740 	     aenq_e->aenq_common_desc.group,
4741 	     ENA_ADMIN_NOTIFICATION);
4742 
4743 	switch (aenq_e->aenq_common_desc.syndrome) {
4744 	case ENA_ADMIN_UPDATE_HINTS:
4745 		hints = (struct ena_admin_ena_hw_hints *)
4746 			(&aenq_e->inline_data_w4);
4747 		ena_update_hints(adapter, hints);
4748 		break;
4749 	default:
4750 		netif_err(adapter, drv, adapter->netdev,
4751 			  "Invalid aenq notification link state %d\n",
4752 			  aenq_e->aenq_common_desc.syndrome);
4753 	}
4754 }
4755 
4756 /* This handler will called for unknown event group or unimplemented handlers*/
4757 static void unimplemented_aenq_handler(void *data,
4758 				       struct ena_admin_aenq_entry *aenq_e)
4759 {
4760 	struct ena_adapter *adapter = (struct ena_adapter *)data;
4761 
4762 	netif_err(adapter, drv, adapter->netdev,
4763 		  "Unknown event was received or event with unimplemented handler\n");
4764 }
4765 
4766 static struct ena_aenq_handlers aenq_handlers = {
4767 	.handlers = {
4768 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4769 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
4770 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4771 	},
4772 	.unimplemented_handler = unimplemented_aenq_handler
4773 };
4774 
4775 module_init(ena_init);
4776 module_exit(ena_cleanup);
4777