xref: /linux/drivers/net/ethernet/amazon/ena/ena_netdev.c (revision 7bb377107c72a40ab7505341f8626c8eb79a0cb7)
1 /*
2  * Copyright 2015 Amazon.com, Inc. or its affiliates.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 
35 #ifdef CONFIG_RFS_ACCEL
36 #include <linux/cpu_rmap.h>
37 #endif /* CONFIG_RFS_ACCEL */
38 #include <linux/ethtool.h>
39 #include <linux/kernel.h>
40 #include <linux/module.h>
41 #include <linux/numa.h>
42 #include <linux/pci.h>
43 #include <linux/utsname.h>
44 #include <linux/version.h>
45 #include <linux/vmalloc.h>
46 #include <net/ip.h>
47 
48 #include "ena_netdev.h"
49 #include <linux/bpf_trace.h>
50 #include "ena_pci_id_tbl.h"
51 
52 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
53 MODULE_DESCRIPTION(DEVICE_NAME);
54 MODULE_LICENSE("GPL");
55 
56 /* Time in jiffies before concluding the transmitter is hung. */
57 #define TX_TIMEOUT  (5 * HZ)
58 
59 #define ENA_NAPI_BUDGET 64
60 
61 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
62 		NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
63 static int debug = -1;
64 module_param(debug, int, 0);
65 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
66 
67 static struct ena_aenq_handlers aenq_handlers;
68 
69 static struct workqueue_struct *ena_wq;
70 
71 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
72 
73 static int ena_rss_init_default(struct ena_adapter *adapter);
74 static void check_for_admin_com_state(struct ena_adapter *adapter);
75 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
76 static int ena_restore_device(struct ena_adapter *adapter);
77 
78 static void ena_init_io_rings(struct ena_adapter *adapter,
79 			      int first_index, int count);
80 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
81 				   int count);
82 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
83 				  int count);
84 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
85 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
86 					   int first_index,
87 					   int count);
88 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
89 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
90 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
91 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
92 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
93 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
94 				      int first_index, int count);
95 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
96 				     int first_index, int count);
97 static int ena_up(struct ena_adapter *adapter);
98 static void ena_down(struct ena_adapter *adapter);
99 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
100 				 struct ena_ring *rx_ring);
101 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
102 				      struct ena_ring *rx_ring);
103 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
104 			      struct ena_tx_buffer *tx_info);
105 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
106 					    int first_index, int count);
107 
108 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
109 {
110 	struct ena_adapter *adapter = netdev_priv(dev);
111 
112 	/* Change the state of the device to trigger reset
113 	 * Check that we are not in the middle or a trigger already
114 	 */
115 
116 	if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
117 		return;
118 
119 	adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
120 	u64_stats_update_begin(&adapter->syncp);
121 	adapter->dev_stats.tx_timeout++;
122 	u64_stats_update_end(&adapter->syncp);
123 
124 	netif_err(adapter, tx_err, dev, "Transmit time out\n");
125 }
126 
127 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
128 {
129 	int i;
130 
131 	for (i = 0; i < adapter->num_io_queues; i++)
132 		adapter->rx_ring[i].mtu = mtu;
133 }
134 
135 static int ena_change_mtu(struct net_device *dev, int new_mtu)
136 {
137 	struct ena_adapter *adapter = netdev_priv(dev);
138 	int ret;
139 
140 	ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
141 	if (!ret) {
142 		netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
143 		update_rx_ring_mtu(adapter, new_mtu);
144 		dev->mtu = new_mtu;
145 	} else {
146 		netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
147 			  new_mtu);
148 	}
149 
150 	return ret;
151 }
152 
153 static int ena_xmit_common(struct net_device *dev,
154 			   struct ena_ring *ring,
155 			   struct ena_tx_buffer *tx_info,
156 			   struct ena_com_tx_ctx *ena_tx_ctx,
157 			   u16 next_to_use,
158 			   u32 bytes)
159 {
160 	struct ena_adapter *adapter = netdev_priv(dev);
161 	int rc, nb_hw_desc;
162 
163 	if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
164 						ena_tx_ctx))) {
165 		netif_dbg(adapter, tx_queued, dev,
166 			  "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
167 			  ring->qid);
168 		ena_com_write_sq_doorbell(ring->ena_com_io_sq);
169 	}
170 
171 	/* prepare the packet's descriptors to dma engine */
172 	rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
173 				&nb_hw_desc);
174 
175 	/* In case there isn't enough space in the queue for the packet,
176 	 * we simply drop it. All other failure reasons of
177 	 * ena_com_prepare_tx() are fatal and therefore require a device reset.
178 	 */
179 	if (unlikely(rc)) {
180 		netif_err(adapter, tx_queued, dev,
181 			  "failed to prepare tx bufs\n");
182 		u64_stats_update_begin(&ring->syncp);
183 		ring->tx_stats.prepare_ctx_err++;
184 		u64_stats_update_end(&ring->syncp);
185 		if (rc != -ENOMEM) {
186 			adapter->reset_reason =
187 				ENA_REGS_RESET_DRIVER_INVALID_STATE;
188 			set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
189 		}
190 		return rc;
191 	}
192 
193 	u64_stats_update_begin(&ring->syncp);
194 	ring->tx_stats.cnt++;
195 	ring->tx_stats.bytes += bytes;
196 	u64_stats_update_end(&ring->syncp);
197 
198 	tx_info->tx_descs = nb_hw_desc;
199 	tx_info->last_jiffies = jiffies;
200 	tx_info->print_once = 0;
201 
202 	ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
203 						 ring->ring_size);
204 	return 0;
205 }
206 
207 /* This is the XDP napi callback. XDP queues use a separate napi callback
208  * than Rx/Tx queues.
209  */
210 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
211 {
212 	struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
213 	u32 xdp_work_done, xdp_budget;
214 	struct ena_ring *xdp_ring;
215 	int napi_comp_call = 0;
216 	int ret;
217 
218 	xdp_ring = ena_napi->xdp_ring;
219 	xdp_ring->first_interrupt = ena_napi->first_interrupt;
220 
221 	xdp_budget = budget;
222 
223 	if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
224 	    test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
225 		napi_complete_done(napi, 0);
226 		return 0;
227 	}
228 
229 	xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
230 
231 	/* If the device is about to reset or down, avoid unmask
232 	 * the interrupt and return 0 so NAPI won't reschedule
233 	 */
234 	if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
235 		napi_complete_done(napi, 0);
236 		ret = 0;
237 	} else if (xdp_budget > xdp_work_done) {
238 		napi_comp_call = 1;
239 		if (napi_complete_done(napi, xdp_work_done))
240 			ena_unmask_interrupt(xdp_ring, NULL);
241 		ena_update_ring_numa_node(xdp_ring, NULL);
242 		ret = xdp_work_done;
243 	} else {
244 		ret = xdp_budget;
245 	}
246 
247 	u64_stats_update_begin(&xdp_ring->syncp);
248 	xdp_ring->tx_stats.napi_comp += napi_comp_call;
249 	xdp_ring->tx_stats.tx_poll++;
250 	u64_stats_update_end(&xdp_ring->syncp);
251 
252 	return ret;
253 }
254 
255 static int ena_xdp_tx_map_buff(struct ena_ring *xdp_ring,
256 			       struct ena_tx_buffer *tx_info,
257 			       struct xdp_buff *xdp,
258 			       void **push_hdr,
259 			       u32 *push_len)
260 {
261 	struct ena_adapter *adapter = xdp_ring->adapter;
262 	struct ena_com_buf *ena_buf;
263 	dma_addr_t dma = 0;
264 	u32 size;
265 
266 	tx_info->xdpf = convert_to_xdp_frame(xdp);
267 	size = tx_info->xdpf->len;
268 	ena_buf = tx_info->bufs;
269 
270 	/* llq push buffer */
271 	*push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
272 	*push_hdr = tx_info->xdpf->data;
273 
274 	if (size - *push_len > 0) {
275 		dma = dma_map_single(xdp_ring->dev,
276 				     *push_hdr + *push_len,
277 				     size - *push_len,
278 				     DMA_TO_DEVICE);
279 		if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
280 			goto error_report_dma_error;
281 
282 		tx_info->map_linear_data = 1;
283 		tx_info->num_of_bufs = 1;
284 	}
285 
286 	ena_buf->paddr = dma;
287 	ena_buf->len = size;
288 
289 	return 0;
290 
291 error_report_dma_error:
292 	u64_stats_update_begin(&xdp_ring->syncp);
293 	xdp_ring->tx_stats.dma_mapping_err++;
294 	u64_stats_update_end(&xdp_ring->syncp);
295 	netdev_warn(adapter->netdev, "failed to map xdp buff\n");
296 
297 	xdp_return_frame_rx_napi(tx_info->xdpf);
298 	tx_info->xdpf = NULL;
299 	tx_info->num_of_bufs = 0;
300 
301 	return -EINVAL;
302 }
303 
304 static int ena_xdp_xmit_buff(struct net_device *dev,
305 			     struct xdp_buff *xdp,
306 			     int qid,
307 			     struct ena_rx_buffer *rx_info)
308 {
309 	struct ena_adapter *adapter = netdev_priv(dev);
310 	struct ena_com_tx_ctx ena_tx_ctx = {0};
311 	struct ena_tx_buffer *tx_info;
312 	struct ena_ring *xdp_ring;
313 	u16 next_to_use, req_id;
314 	int rc;
315 	void *push_hdr;
316 	u32 push_len;
317 
318 	xdp_ring = &adapter->tx_ring[qid];
319 	next_to_use = xdp_ring->next_to_use;
320 	req_id = xdp_ring->free_ids[next_to_use];
321 	tx_info = &xdp_ring->tx_buffer_info[req_id];
322 	tx_info->num_of_bufs = 0;
323 	page_ref_inc(rx_info->page);
324 	tx_info->xdp_rx_page = rx_info->page;
325 
326 	rc = ena_xdp_tx_map_buff(xdp_ring, tx_info, xdp, &push_hdr, &push_len);
327 	if (unlikely(rc))
328 		goto error_drop_packet;
329 
330 	ena_tx_ctx.ena_bufs = tx_info->bufs;
331 	ena_tx_ctx.push_header = push_hdr;
332 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
333 	ena_tx_ctx.req_id = req_id;
334 	ena_tx_ctx.header_len = push_len;
335 
336 	rc = ena_xmit_common(dev,
337 			     xdp_ring,
338 			     tx_info,
339 			     &ena_tx_ctx,
340 			     next_to_use,
341 			     xdp->data_end - xdp->data);
342 	if (rc)
343 		goto error_unmap_dma;
344 	/* trigger the dma engine. ena_com_write_sq_doorbell()
345 	 * has a mb
346 	 */
347 	ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
348 	u64_stats_update_begin(&xdp_ring->syncp);
349 	xdp_ring->tx_stats.doorbells++;
350 	u64_stats_update_end(&xdp_ring->syncp);
351 
352 	return NETDEV_TX_OK;
353 
354 error_unmap_dma:
355 	ena_unmap_tx_buff(xdp_ring, tx_info);
356 	tx_info->xdpf = NULL;
357 error_drop_packet:
358 
359 	return NETDEV_TX_OK;
360 }
361 
362 static int ena_xdp_execute(struct ena_ring *rx_ring,
363 			   struct xdp_buff *xdp,
364 			   struct ena_rx_buffer *rx_info)
365 {
366 	struct bpf_prog *xdp_prog;
367 	u32 verdict = XDP_PASS;
368 
369 	rcu_read_lock();
370 	xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
371 
372 	if (!xdp_prog)
373 		goto out;
374 
375 	verdict = bpf_prog_run_xdp(xdp_prog, xdp);
376 
377 	if (verdict == XDP_TX)
378 		ena_xdp_xmit_buff(rx_ring->netdev,
379 				  xdp,
380 				  rx_ring->qid + rx_ring->adapter->num_io_queues,
381 				  rx_info);
382 	else if (unlikely(verdict == XDP_ABORTED))
383 		trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
384 	else if (unlikely(verdict > XDP_TX))
385 		bpf_warn_invalid_xdp_action(verdict);
386 out:
387 	rcu_read_unlock();
388 	return verdict;
389 }
390 
391 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
392 {
393 	adapter->xdp_first_ring = adapter->num_io_queues;
394 	adapter->xdp_num_queues = adapter->num_io_queues;
395 
396 	ena_init_io_rings(adapter,
397 			  adapter->xdp_first_ring,
398 			  adapter->xdp_num_queues);
399 }
400 
401 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
402 {
403 	int rc = 0;
404 
405 	rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
406 					     adapter->xdp_num_queues);
407 	if (rc)
408 		goto setup_err;
409 
410 	rc = ena_create_io_tx_queues_in_range(adapter,
411 					      adapter->xdp_first_ring,
412 					      adapter->xdp_num_queues);
413 	if (rc)
414 		goto create_err;
415 
416 	return 0;
417 
418 create_err:
419 	ena_free_all_io_tx_resources(adapter);
420 setup_err:
421 	return rc;
422 }
423 
424 /* Provides a way for both kernel and bpf-prog to know
425  * more about the RX-queue a given XDP frame arrived on.
426  */
427 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
428 {
429 	int rc;
430 
431 	rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid);
432 
433 	if (rc) {
434 		netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
435 			  "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
436 			  rx_ring->qid, rc);
437 		goto err;
438 	}
439 
440 	rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
441 					NULL);
442 
443 	if (rc) {
444 		netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
445 			  "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
446 			  rx_ring->qid, rc);
447 		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
448 	}
449 
450 err:
451 	return rc;
452 }
453 
454 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
455 {
456 	xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
457 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
458 }
459 
460 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
461 						 struct bpf_prog *prog,
462 						 int first, int count)
463 {
464 	struct ena_ring *rx_ring;
465 	int i = 0;
466 
467 	for (i = first; i < count; i++) {
468 		rx_ring = &adapter->rx_ring[i];
469 		xchg(&rx_ring->xdp_bpf_prog, prog);
470 		if (prog) {
471 			ena_xdp_register_rxq_info(rx_ring);
472 			rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
473 		} else {
474 			ena_xdp_unregister_rxq_info(rx_ring);
475 			rx_ring->rx_headroom = 0;
476 		}
477 	}
478 }
479 
480 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
481 				     struct bpf_prog *prog)
482 {
483 	struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
484 
485 	ena_xdp_exchange_program_rx_in_range(adapter,
486 					     prog,
487 					     0,
488 					     adapter->num_io_queues);
489 
490 	if (old_bpf_prog)
491 		bpf_prog_put(old_bpf_prog);
492 }
493 
494 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
495 {
496 	bool was_up;
497 	int rc;
498 
499 	was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
500 
501 	if (was_up)
502 		ena_down(adapter);
503 
504 	adapter->xdp_first_ring = 0;
505 	adapter->xdp_num_queues = 0;
506 	ena_xdp_exchange_program(adapter, NULL);
507 	if (was_up) {
508 		rc = ena_up(adapter);
509 		if (rc)
510 			return rc;
511 	}
512 	return 0;
513 }
514 
515 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
516 {
517 	struct ena_adapter *adapter = netdev_priv(netdev);
518 	struct bpf_prog *prog = bpf->prog;
519 	struct bpf_prog *old_bpf_prog;
520 	int rc, prev_mtu;
521 	bool is_up;
522 
523 	is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
524 	rc = ena_xdp_allowed(adapter);
525 	if (rc == ENA_XDP_ALLOWED) {
526 		old_bpf_prog = adapter->xdp_bpf_prog;
527 		if (prog) {
528 			if (!is_up) {
529 				ena_init_all_xdp_queues(adapter);
530 			} else if (!old_bpf_prog) {
531 				ena_down(adapter);
532 				ena_init_all_xdp_queues(adapter);
533 			}
534 			ena_xdp_exchange_program(adapter, prog);
535 
536 			if (is_up && !old_bpf_prog) {
537 				rc = ena_up(adapter);
538 				if (rc)
539 					return rc;
540 			}
541 		} else if (old_bpf_prog) {
542 			rc = ena_destroy_and_free_all_xdp_queues(adapter);
543 			if (rc)
544 				return rc;
545 		}
546 
547 		prev_mtu = netdev->max_mtu;
548 		netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
549 
550 		if (!old_bpf_prog)
551 			netif_info(adapter, drv, adapter->netdev,
552 				   "xdp program set, changing the max_mtu from %d to %d",
553 				   prev_mtu, netdev->max_mtu);
554 
555 	} else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
556 		netif_err(adapter, drv, adapter->netdev,
557 			  "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
558 			  netdev->mtu, ENA_XDP_MAX_MTU);
559 		NL_SET_ERR_MSG_MOD(bpf->extack,
560 				   "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
561 		return -EINVAL;
562 	} else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
563 		netif_err(adapter, drv, adapter->netdev,
564 			  "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",
565 			  adapter->num_io_queues, adapter->max_num_io_queues);
566 		NL_SET_ERR_MSG_MOD(bpf->extack,
567 				   "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
568 		return -EINVAL;
569 	}
570 
571 	return 0;
572 }
573 
574 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
575  * program as well as to query the current xdp program id.
576  */
577 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
578 {
579 	struct ena_adapter *adapter = netdev_priv(netdev);
580 
581 	switch (bpf->command) {
582 	case XDP_SETUP_PROG:
583 		return ena_xdp_set(netdev, bpf);
584 	case XDP_QUERY_PROG:
585 		bpf->prog_id = adapter->xdp_bpf_prog ?
586 			adapter->xdp_bpf_prog->aux->id : 0;
587 		break;
588 	default:
589 		return -EINVAL;
590 	}
591 	return 0;
592 }
593 
594 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
595 {
596 #ifdef CONFIG_RFS_ACCEL
597 	u32 i;
598 	int rc;
599 
600 	adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
601 	if (!adapter->netdev->rx_cpu_rmap)
602 		return -ENOMEM;
603 	for (i = 0; i < adapter->num_io_queues; i++) {
604 		int irq_idx = ENA_IO_IRQ_IDX(i);
605 
606 		rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
607 				      pci_irq_vector(adapter->pdev, irq_idx));
608 		if (rc) {
609 			free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
610 			adapter->netdev->rx_cpu_rmap = NULL;
611 			return rc;
612 		}
613 	}
614 #endif /* CONFIG_RFS_ACCEL */
615 	return 0;
616 }
617 
618 static void ena_init_io_rings_common(struct ena_adapter *adapter,
619 				     struct ena_ring *ring, u16 qid)
620 {
621 	ring->qid = qid;
622 	ring->pdev = adapter->pdev;
623 	ring->dev = &adapter->pdev->dev;
624 	ring->netdev = adapter->netdev;
625 	ring->napi = &adapter->ena_napi[qid].napi;
626 	ring->adapter = adapter;
627 	ring->ena_dev = adapter->ena_dev;
628 	ring->per_napi_packets = 0;
629 	ring->cpu = 0;
630 	ring->first_interrupt = false;
631 	ring->no_interrupt_event_cnt = 0;
632 	u64_stats_init(&ring->syncp);
633 }
634 
635 static void ena_init_io_rings(struct ena_adapter *adapter,
636 			      int first_index, int count)
637 {
638 	struct ena_com_dev *ena_dev;
639 	struct ena_ring *txr, *rxr;
640 	int i;
641 
642 	ena_dev = adapter->ena_dev;
643 
644 	for (i = first_index; i < first_index + count; i++) {
645 		txr = &adapter->tx_ring[i];
646 		rxr = &adapter->rx_ring[i];
647 
648 		/* TX common ring state */
649 		ena_init_io_rings_common(adapter, txr, i);
650 
651 		/* TX specific ring state */
652 		txr->ring_size = adapter->requested_tx_ring_size;
653 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
654 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
655 		txr->sgl_size = adapter->max_tx_sgl_size;
656 		txr->smoothed_interval =
657 			ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
658 
659 		/* Don't init RX queues for xdp queues */
660 		if (!ENA_IS_XDP_INDEX(adapter, i)) {
661 			/* RX common ring state */
662 			ena_init_io_rings_common(adapter, rxr, i);
663 
664 			/* RX specific ring state */
665 			rxr->ring_size = adapter->requested_rx_ring_size;
666 			rxr->rx_copybreak = adapter->rx_copybreak;
667 			rxr->sgl_size = adapter->max_rx_sgl_size;
668 			rxr->smoothed_interval =
669 				ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
670 			rxr->empty_rx_queue = 0;
671 			adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
672 		}
673 	}
674 }
675 
676 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
677  * @adapter: network interface device structure
678  * @qid: queue index
679  *
680  * Return 0 on success, negative on failure
681  */
682 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
683 {
684 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
685 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
686 	int size, i, node;
687 
688 	if (tx_ring->tx_buffer_info) {
689 		netif_err(adapter, ifup,
690 			  adapter->netdev, "tx_buffer_info info is not NULL");
691 		return -EEXIST;
692 	}
693 
694 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
695 	node = cpu_to_node(ena_irq->cpu);
696 
697 	tx_ring->tx_buffer_info = vzalloc_node(size, node);
698 	if (!tx_ring->tx_buffer_info) {
699 		tx_ring->tx_buffer_info = vzalloc(size);
700 		if (!tx_ring->tx_buffer_info)
701 			goto err_tx_buffer_info;
702 	}
703 
704 	size = sizeof(u16) * tx_ring->ring_size;
705 	tx_ring->free_ids = vzalloc_node(size, node);
706 	if (!tx_ring->free_ids) {
707 		tx_ring->free_ids = vzalloc(size);
708 		if (!tx_ring->free_ids)
709 			goto err_tx_free_ids;
710 	}
711 
712 	size = tx_ring->tx_max_header_size;
713 	tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
714 	if (!tx_ring->push_buf_intermediate_buf) {
715 		tx_ring->push_buf_intermediate_buf = vzalloc(size);
716 		if (!tx_ring->push_buf_intermediate_buf)
717 			goto err_push_buf_intermediate_buf;
718 	}
719 
720 	/* Req id ring for TX out of order completions */
721 	for (i = 0; i < tx_ring->ring_size; i++)
722 		tx_ring->free_ids[i] = i;
723 
724 	/* Reset tx statistics */
725 	memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
726 
727 	tx_ring->next_to_use = 0;
728 	tx_ring->next_to_clean = 0;
729 	tx_ring->cpu = ena_irq->cpu;
730 	return 0;
731 
732 err_push_buf_intermediate_buf:
733 	vfree(tx_ring->free_ids);
734 	tx_ring->free_ids = NULL;
735 err_tx_free_ids:
736 	vfree(tx_ring->tx_buffer_info);
737 	tx_ring->tx_buffer_info = NULL;
738 err_tx_buffer_info:
739 	return -ENOMEM;
740 }
741 
742 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
743  * @adapter: network interface device structure
744  * @qid: queue index
745  *
746  * Free all transmit software resources
747  */
748 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
749 {
750 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
751 
752 	vfree(tx_ring->tx_buffer_info);
753 	tx_ring->tx_buffer_info = NULL;
754 
755 	vfree(tx_ring->free_ids);
756 	tx_ring->free_ids = NULL;
757 
758 	vfree(tx_ring->push_buf_intermediate_buf);
759 	tx_ring->push_buf_intermediate_buf = NULL;
760 }
761 
762 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
763 					   int first_index,
764 					   int count)
765 {
766 	int i, rc = 0;
767 
768 	for (i = first_index; i < first_index + count; i++) {
769 		rc = ena_setup_tx_resources(adapter, i);
770 		if (rc)
771 			goto err_setup_tx;
772 	}
773 
774 	return 0;
775 
776 err_setup_tx:
777 
778 	netif_err(adapter, ifup, adapter->netdev,
779 		  "Tx queue %d: allocation failed\n", i);
780 
781 	/* rewind the index freeing the rings as we go */
782 	while (first_index < i--)
783 		ena_free_tx_resources(adapter, i);
784 	return rc;
785 }
786 
787 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
788 						  int first_index, int count)
789 {
790 	int i;
791 
792 	for (i = first_index; i < first_index + count; i++)
793 		ena_free_tx_resources(adapter, i);
794 }
795 
796 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
797  * @adapter: board private structure
798  *
799  * Free all transmit software resources
800  */
801 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
802 {
803 	ena_free_all_io_tx_resources_in_range(adapter,
804 					      0,
805 					      adapter->xdp_num_queues +
806 					      adapter->num_io_queues);
807 }
808 
809 static int validate_rx_req_id(struct ena_ring *rx_ring, u16 req_id)
810 {
811 	if (likely(req_id < rx_ring->ring_size))
812 		return 0;
813 
814 	netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
815 		  "Invalid rx req_id: %hu\n", req_id);
816 
817 	u64_stats_update_begin(&rx_ring->syncp);
818 	rx_ring->rx_stats.bad_req_id++;
819 	u64_stats_update_end(&rx_ring->syncp);
820 
821 	/* Trigger device reset */
822 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
823 	set_bit(ENA_FLAG_TRIGGER_RESET, &rx_ring->adapter->flags);
824 	return -EFAULT;
825 }
826 
827 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
828  * @adapter: network interface device structure
829  * @qid: queue index
830  *
831  * Returns 0 on success, negative on failure
832  */
833 static int ena_setup_rx_resources(struct ena_adapter *adapter,
834 				  u32 qid)
835 {
836 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
837 	struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
838 	int size, node, i;
839 
840 	if (rx_ring->rx_buffer_info) {
841 		netif_err(adapter, ifup, adapter->netdev,
842 			  "rx_buffer_info is not NULL");
843 		return -EEXIST;
844 	}
845 
846 	/* alloc extra element so in rx path
847 	 * we can always prefetch rx_info + 1
848 	 */
849 	size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
850 	node = cpu_to_node(ena_irq->cpu);
851 
852 	rx_ring->rx_buffer_info = vzalloc_node(size, node);
853 	if (!rx_ring->rx_buffer_info) {
854 		rx_ring->rx_buffer_info = vzalloc(size);
855 		if (!rx_ring->rx_buffer_info)
856 			return -ENOMEM;
857 	}
858 
859 	size = sizeof(u16) * rx_ring->ring_size;
860 	rx_ring->free_ids = vzalloc_node(size, node);
861 	if (!rx_ring->free_ids) {
862 		rx_ring->free_ids = vzalloc(size);
863 		if (!rx_ring->free_ids) {
864 			vfree(rx_ring->rx_buffer_info);
865 			rx_ring->rx_buffer_info = NULL;
866 			return -ENOMEM;
867 		}
868 	}
869 
870 	/* Req id ring for receiving RX pkts out of order */
871 	for (i = 0; i < rx_ring->ring_size; i++)
872 		rx_ring->free_ids[i] = i;
873 
874 	/* Reset rx statistics */
875 	memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
876 
877 	rx_ring->next_to_clean = 0;
878 	rx_ring->next_to_use = 0;
879 	rx_ring->cpu = ena_irq->cpu;
880 
881 	return 0;
882 }
883 
884 /* ena_free_rx_resources - Free I/O Rx Resources
885  * @adapter: network interface device structure
886  * @qid: queue index
887  *
888  * Free all receive software resources
889  */
890 static void ena_free_rx_resources(struct ena_adapter *adapter,
891 				  u32 qid)
892 {
893 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
894 
895 	vfree(rx_ring->rx_buffer_info);
896 	rx_ring->rx_buffer_info = NULL;
897 
898 	vfree(rx_ring->free_ids);
899 	rx_ring->free_ids = NULL;
900 }
901 
902 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
903  * @adapter: board private structure
904  *
905  * Return 0 on success, negative on failure
906  */
907 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
908 {
909 	int i, rc = 0;
910 
911 	for (i = 0; i < adapter->num_io_queues; i++) {
912 		rc = ena_setup_rx_resources(adapter, i);
913 		if (rc)
914 			goto err_setup_rx;
915 	}
916 
917 	return 0;
918 
919 err_setup_rx:
920 
921 	netif_err(adapter, ifup, adapter->netdev,
922 		  "Rx queue %d: allocation failed\n", i);
923 
924 	/* rewind the index freeing the rings as we go */
925 	while (i--)
926 		ena_free_rx_resources(adapter, i);
927 	return rc;
928 }
929 
930 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
931  * @adapter: board private structure
932  *
933  * Free all receive software resources
934  */
935 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
936 {
937 	int i;
938 
939 	for (i = 0; i < adapter->num_io_queues; i++)
940 		ena_free_rx_resources(adapter, i);
941 }
942 
943 static int ena_alloc_rx_page(struct ena_ring *rx_ring,
944 				    struct ena_rx_buffer *rx_info, gfp_t gfp)
945 {
946 	struct ena_com_buf *ena_buf;
947 	struct page *page;
948 	dma_addr_t dma;
949 
950 	/* if previous allocated page is not used */
951 	if (unlikely(rx_info->page))
952 		return 0;
953 
954 	page = alloc_page(gfp);
955 	if (unlikely(!page)) {
956 		u64_stats_update_begin(&rx_ring->syncp);
957 		rx_ring->rx_stats.page_alloc_fail++;
958 		u64_stats_update_end(&rx_ring->syncp);
959 		return -ENOMEM;
960 	}
961 
962 	dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
963 			   DMA_FROM_DEVICE);
964 	if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
965 		u64_stats_update_begin(&rx_ring->syncp);
966 		rx_ring->rx_stats.dma_mapping_err++;
967 		u64_stats_update_end(&rx_ring->syncp);
968 
969 		__free_page(page);
970 		return -EIO;
971 	}
972 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
973 		  "alloc page %p, rx_info %p\n", page, rx_info);
974 
975 	rx_info->page = page;
976 	rx_info->page_offset = 0;
977 	ena_buf = &rx_info->ena_buf;
978 	ena_buf->paddr = dma + rx_ring->rx_headroom;
979 	ena_buf->len = ENA_PAGE_SIZE - rx_ring->rx_headroom;
980 
981 	return 0;
982 }
983 
984 static void ena_free_rx_page(struct ena_ring *rx_ring,
985 			     struct ena_rx_buffer *rx_info)
986 {
987 	struct page *page = rx_info->page;
988 	struct ena_com_buf *ena_buf = &rx_info->ena_buf;
989 
990 	if (unlikely(!page)) {
991 		netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
992 			   "Trying to free unallocated buffer\n");
993 		return;
994 	}
995 
996 	dma_unmap_page(rx_ring->dev,
997 		       ena_buf->paddr - rx_ring->rx_headroom,
998 		       ENA_PAGE_SIZE,
999 		       DMA_FROM_DEVICE);
1000 
1001 	__free_page(page);
1002 	rx_info->page = NULL;
1003 }
1004 
1005 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1006 {
1007 	u16 next_to_use, req_id;
1008 	u32 i;
1009 	int rc;
1010 
1011 	next_to_use = rx_ring->next_to_use;
1012 
1013 	for (i = 0; i < num; i++) {
1014 		struct ena_rx_buffer *rx_info;
1015 
1016 		req_id = rx_ring->free_ids[next_to_use];
1017 
1018 		rx_info = &rx_ring->rx_buffer_info[req_id];
1019 
1020 		rc = ena_alloc_rx_page(rx_ring, rx_info,
1021 				       GFP_ATOMIC | __GFP_COMP);
1022 		if (unlikely(rc < 0)) {
1023 			netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1024 				   "failed to alloc buffer for rx queue %d\n",
1025 				   rx_ring->qid);
1026 			break;
1027 		}
1028 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1029 						&rx_info->ena_buf,
1030 						req_id);
1031 		if (unlikely(rc)) {
1032 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1033 				   "failed to add buffer for rx queue %d\n",
1034 				   rx_ring->qid);
1035 			break;
1036 		}
1037 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1038 						   rx_ring->ring_size);
1039 	}
1040 
1041 	if (unlikely(i < num)) {
1042 		u64_stats_update_begin(&rx_ring->syncp);
1043 		rx_ring->rx_stats.refil_partial++;
1044 		u64_stats_update_end(&rx_ring->syncp);
1045 		netdev_warn(rx_ring->netdev,
1046 			    "refilled rx qid %d with only %d buffers (from %d)\n",
1047 			    rx_ring->qid, i, num);
1048 	}
1049 
1050 	/* ena_com_write_sq_doorbell issues a wmb() */
1051 	if (likely(i))
1052 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1053 
1054 	rx_ring->next_to_use = next_to_use;
1055 
1056 	return i;
1057 }
1058 
1059 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1060 			     u32 qid)
1061 {
1062 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1063 	u32 i;
1064 
1065 	for (i = 0; i < rx_ring->ring_size; i++) {
1066 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1067 
1068 		if (rx_info->page)
1069 			ena_free_rx_page(rx_ring, rx_info);
1070 	}
1071 }
1072 
1073 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1074  * @adapter: board private structure
1075  */
1076 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1077 {
1078 	struct ena_ring *rx_ring;
1079 	int i, rc, bufs_num;
1080 
1081 	for (i = 0; i < adapter->num_io_queues; i++) {
1082 		rx_ring = &adapter->rx_ring[i];
1083 		bufs_num = rx_ring->ring_size - 1;
1084 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1085 
1086 		if (unlikely(rc != bufs_num))
1087 			netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1088 				   "refilling Queue %d failed. allocated %d buffers from: %d\n",
1089 				   i, rc, bufs_num);
1090 	}
1091 }
1092 
1093 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1094 {
1095 	int i;
1096 
1097 	for (i = 0; i < adapter->num_io_queues; i++)
1098 		ena_free_rx_bufs(adapter, i);
1099 }
1100 
1101 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1102 			      struct ena_tx_buffer *tx_info)
1103 {
1104 	struct ena_com_buf *ena_buf;
1105 	u32 cnt;
1106 	int i;
1107 
1108 	ena_buf = tx_info->bufs;
1109 	cnt = tx_info->num_of_bufs;
1110 
1111 	if (unlikely(!cnt))
1112 		return;
1113 
1114 	if (tx_info->map_linear_data) {
1115 		dma_unmap_single(tx_ring->dev,
1116 				 dma_unmap_addr(ena_buf, paddr),
1117 				 dma_unmap_len(ena_buf, len),
1118 				 DMA_TO_DEVICE);
1119 		ena_buf++;
1120 		cnt--;
1121 	}
1122 
1123 	/* unmap remaining mapped pages */
1124 	for (i = 0; i < cnt; i++) {
1125 		dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1126 			       dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1127 		ena_buf++;
1128 	}
1129 }
1130 
1131 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1132  * @tx_ring: TX ring for which buffers be freed
1133  */
1134 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1135 {
1136 	bool print_once = true;
1137 	u32 i;
1138 
1139 	for (i = 0; i < tx_ring->ring_size; i++) {
1140 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1141 
1142 		if (!tx_info->skb)
1143 			continue;
1144 
1145 		if (print_once) {
1146 			netdev_notice(tx_ring->netdev,
1147 				      "free uncompleted tx skb qid %d idx 0x%x\n",
1148 				      tx_ring->qid, i);
1149 			print_once = false;
1150 		} else {
1151 			netdev_dbg(tx_ring->netdev,
1152 				   "free uncompleted tx skb qid %d idx 0x%x\n",
1153 				   tx_ring->qid, i);
1154 		}
1155 
1156 		ena_unmap_tx_buff(tx_ring, tx_info);
1157 
1158 		dev_kfree_skb_any(tx_info->skb);
1159 	}
1160 	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1161 						  tx_ring->qid));
1162 }
1163 
1164 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1165 {
1166 	struct ena_ring *tx_ring;
1167 	int i;
1168 
1169 	for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1170 		tx_ring = &adapter->tx_ring[i];
1171 		ena_free_tx_bufs(tx_ring);
1172 	}
1173 }
1174 
1175 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1176 {
1177 	u16 ena_qid;
1178 	int i;
1179 
1180 	for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1181 		ena_qid = ENA_IO_TXQ_IDX(i);
1182 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1183 	}
1184 }
1185 
1186 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1187 {
1188 	u16 ena_qid;
1189 	int i;
1190 
1191 	for (i = 0; i < adapter->num_io_queues; i++) {
1192 		ena_qid = ENA_IO_RXQ_IDX(i);
1193 		cancel_work_sync(&adapter->ena_napi[i].dim.work);
1194 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1195 	}
1196 }
1197 
1198 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1199 {
1200 	ena_destroy_all_tx_queues(adapter);
1201 	ena_destroy_all_rx_queues(adapter);
1202 }
1203 
1204 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1205 				 struct ena_tx_buffer *tx_info, bool is_xdp)
1206 {
1207 	if (tx_info)
1208 		netif_err(ring->adapter,
1209 			  tx_done,
1210 			  ring->netdev,
1211 			  "tx_info doesn't have valid %s",
1212 			   is_xdp ? "xdp frame" : "skb");
1213 	else
1214 		netif_err(ring->adapter,
1215 			  tx_done,
1216 			  ring->netdev,
1217 			  "Invalid req_id: %hu\n",
1218 			  req_id);
1219 
1220 	u64_stats_update_begin(&ring->syncp);
1221 	ring->tx_stats.bad_req_id++;
1222 	u64_stats_update_end(&ring->syncp);
1223 
1224 	/* Trigger device reset */
1225 	ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
1226 	set_bit(ENA_FLAG_TRIGGER_RESET, &ring->adapter->flags);
1227 	return -EFAULT;
1228 }
1229 
1230 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1231 {
1232 	struct ena_tx_buffer *tx_info = NULL;
1233 
1234 	if (likely(req_id < tx_ring->ring_size)) {
1235 		tx_info = &tx_ring->tx_buffer_info[req_id];
1236 		if (likely(tx_info->skb))
1237 			return 0;
1238 	}
1239 
1240 	return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1241 }
1242 
1243 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1244 {
1245 	struct ena_tx_buffer *tx_info = NULL;
1246 
1247 	if (likely(req_id < xdp_ring->ring_size)) {
1248 		tx_info = &xdp_ring->tx_buffer_info[req_id];
1249 		if (likely(tx_info->xdpf))
1250 			return 0;
1251 	}
1252 
1253 	return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1254 }
1255 
1256 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1257 {
1258 	struct netdev_queue *txq;
1259 	bool above_thresh;
1260 	u32 tx_bytes = 0;
1261 	u32 total_done = 0;
1262 	u16 next_to_clean;
1263 	u16 req_id;
1264 	int tx_pkts = 0;
1265 	int rc;
1266 
1267 	next_to_clean = tx_ring->next_to_clean;
1268 	txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1269 
1270 	while (tx_pkts < budget) {
1271 		struct ena_tx_buffer *tx_info;
1272 		struct sk_buff *skb;
1273 
1274 		rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1275 						&req_id);
1276 		if (rc)
1277 			break;
1278 
1279 		rc = validate_tx_req_id(tx_ring, req_id);
1280 		if (rc)
1281 			break;
1282 
1283 		tx_info = &tx_ring->tx_buffer_info[req_id];
1284 		skb = tx_info->skb;
1285 
1286 		/* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1287 		prefetch(&skb->end);
1288 
1289 		tx_info->skb = NULL;
1290 		tx_info->last_jiffies = 0;
1291 
1292 		ena_unmap_tx_buff(tx_ring, tx_info);
1293 
1294 		netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1295 			  "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1296 			  skb);
1297 
1298 		tx_bytes += skb->len;
1299 		dev_kfree_skb(skb);
1300 		tx_pkts++;
1301 		total_done += tx_info->tx_descs;
1302 
1303 		tx_ring->free_ids[next_to_clean] = req_id;
1304 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1305 						     tx_ring->ring_size);
1306 	}
1307 
1308 	tx_ring->next_to_clean = next_to_clean;
1309 	ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1310 	ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1311 
1312 	netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1313 
1314 	netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1315 		  "tx_poll: q %d done. total pkts: %d\n",
1316 		  tx_ring->qid, tx_pkts);
1317 
1318 	/* need to make the rings circular update visible to
1319 	 * ena_start_xmit() before checking for netif_queue_stopped().
1320 	 */
1321 	smp_mb();
1322 
1323 	above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1324 						    ENA_TX_WAKEUP_THRESH);
1325 	if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1326 		__netif_tx_lock(txq, smp_processor_id());
1327 		above_thresh =
1328 			ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1329 						     ENA_TX_WAKEUP_THRESH);
1330 		if (netif_tx_queue_stopped(txq) && above_thresh &&
1331 		    test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1332 			netif_tx_wake_queue(txq);
1333 			u64_stats_update_begin(&tx_ring->syncp);
1334 			tx_ring->tx_stats.queue_wakeup++;
1335 			u64_stats_update_end(&tx_ring->syncp);
1336 		}
1337 		__netif_tx_unlock(txq);
1338 	}
1339 
1340 	return tx_pkts;
1341 }
1342 
1343 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
1344 {
1345 	struct sk_buff *skb;
1346 
1347 	if (frags)
1348 		skb = napi_get_frags(rx_ring->napi);
1349 	else
1350 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1351 						rx_ring->rx_copybreak);
1352 
1353 	if (unlikely(!skb)) {
1354 		u64_stats_update_begin(&rx_ring->syncp);
1355 		rx_ring->rx_stats.skb_alloc_fail++;
1356 		u64_stats_update_end(&rx_ring->syncp);
1357 		netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1358 			  "Failed to allocate skb. frags: %d\n", frags);
1359 		return NULL;
1360 	}
1361 
1362 	return skb;
1363 }
1364 
1365 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1366 				  struct ena_com_rx_buf_info *ena_bufs,
1367 				  u32 descs,
1368 				  u16 *next_to_clean)
1369 {
1370 	struct sk_buff *skb;
1371 	struct ena_rx_buffer *rx_info;
1372 	u16 len, req_id, buf = 0;
1373 	void *va;
1374 	int rc;
1375 
1376 	len = ena_bufs[buf].len;
1377 	req_id = ena_bufs[buf].req_id;
1378 
1379 	rc = validate_rx_req_id(rx_ring, req_id);
1380 	if (unlikely(rc < 0))
1381 		return NULL;
1382 
1383 	rx_info = &rx_ring->rx_buffer_info[req_id];
1384 
1385 	if (unlikely(!rx_info->page)) {
1386 		netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
1387 			  "Page is NULL\n");
1388 		return NULL;
1389 	}
1390 
1391 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1392 		  "rx_info %p page %p\n",
1393 		  rx_info, rx_info->page);
1394 
1395 	/* save virt address of first buffer */
1396 	va = page_address(rx_info->page) + rx_info->page_offset;
1397 	prefetch(va + NET_IP_ALIGN);
1398 
1399 	if (len <= rx_ring->rx_copybreak) {
1400 		skb = ena_alloc_skb(rx_ring, false);
1401 		if (unlikely(!skb))
1402 			return NULL;
1403 
1404 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1405 			  "rx allocated small packet. len %d. data_len %d\n",
1406 			  skb->len, skb->data_len);
1407 
1408 		/* sync this buffer for CPU use */
1409 		dma_sync_single_for_cpu(rx_ring->dev,
1410 					dma_unmap_addr(&rx_info->ena_buf, paddr),
1411 					len,
1412 					DMA_FROM_DEVICE);
1413 		skb_copy_to_linear_data(skb, va, len);
1414 		dma_sync_single_for_device(rx_ring->dev,
1415 					   dma_unmap_addr(&rx_info->ena_buf, paddr),
1416 					   len,
1417 					   DMA_FROM_DEVICE);
1418 
1419 		skb_put(skb, len);
1420 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1421 		rx_ring->free_ids[*next_to_clean] = req_id;
1422 		*next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1423 						     rx_ring->ring_size);
1424 		return skb;
1425 	}
1426 
1427 	skb = ena_alloc_skb(rx_ring, true);
1428 	if (unlikely(!skb))
1429 		return NULL;
1430 
1431 	do {
1432 		dma_unmap_page(rx_ring->dev,
1433 			       dma_unmap_addr(&rx_info->ena_buf, paddr),
1434 			       ENA_PAGE_SIZE, DMA_FROM_DEVICE);
1435 
1436 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1437 				rx_info->page_offset, len, ENA_PAGE_SIZE);
1438 
1439 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1440 			  "rx skb updated. len %d. data_len %d\n",
1441 			  skb->len, skb->data_len);
1442 
1443 		rx_info->page = NULL;
1444 
1445 		rx_ring->free_ids[*next_to_clean] = req_id;
1446 		*next_to_clean =
1447 			ENA_RX_RING_IDX_NEXT(*next_to_clean,
1448 					     rx_ring->ring_size);
1449 		if (likely(--descs == 0))
1450 			break;
1451 
1452 		buf++;
1453 		len = ena_bufs[buf].len;
1454 		req_id = ena_bufs[buf].req_id;
1455 
1456 		rc = validate_rx_req_id(rx_ring, req_id);
1457 		if (unlikely(rc < 0))
1458 			return NULL;
1459 
1460 		rx_info = &rx_ring->rx_buffer_info[req_id];
1461 	} while (1);
1462 
1463 	return skb;
1464 }
1465 
1466 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1467  * @adapter: structure containing adapter specific data
1468  * @ena_rx_ctx: received packet context/metadata
1469  * @skb: skb currently being received and modified
1470  */
1471 static void ena_rx_checksum(struct ena_ring *rx_ring,
1472 				   struct ena_com_rx_ctx *ena_rx_ctx,
1473 				   struct sk_buff *skb)
1474 {
1475 	/* Rx csum disabled */
1476 	if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1477 		skb->ip_summed = CHECKSUM_NONE;
1478 		return;
1479 	}
1480 
1481 	/* For fragmented packets the checksum isn't valid */
1482 	if (ena_rx_ctx->frag) {
1483 		skb->ip_summed = CHECKSUM_NONE;
1484 		return;
1485 	}
1486 
1487 	/* if IP and error */
1488 	if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1489 		     (ena_rx_ctx->l3_csum_err))) {
1490 		/* ipv4 checksum error */
1491 		skb->ip_summed = CHECKSUM_NONE;
1492 		u64_stats_update_begin(&rx_ring->syncp);
1493 		rx_ring->rx_stats.bad_csum++;
1494 		u64_stats_update_end(&rx_ring->syncp);
1495 		netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1496 			  "RX IPv4 header checksum error\n");
1497 		return;
1498 	}
1499 
1500 	/* if TCP/UDP */
1501 	if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1502 		   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1503 		if (unlikely(ena_rx_ctx->l4_csum_err)) {
1504 			/* TCP/UDP checksum error */
1505 			u64_stats_update_begin(&rx_ring->syncp);
1506 			rx_ring->rx_stats.bad_csum++;
1507 			u64_stats_update_end(&rx_ring->syncp);
1508 			netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1509 				  "RX L4 checksum error\n");
1510 			skb->ip_summed = CHECKSUM_NONE;
1511 			return;
1512 		}
1513 
1514 		if (likely(ena_rx_ctx->l4_csum_checked)) {
1515 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1516 			u64_stats_update_begin(&rx_ring->syncp);
1517 			rx_ring->rx_stats.csum_good++;
1518 			u64_stats_update_end(&rx_ring->syncp);
1519 		} else {
1520 			u64_stats_update_begin(&rx_ring->syncp);
1521 			rx_ring->rx_stats.csum_unchecked++;
1522 			u64_stats_update_end(&rx_ring->syncp);
1523 			skb->ip_summed = CHECKSUM_NONE;
1524 		}
1525 	} else {
1526 		skb->ip_summed = CHECKSUM_NONE;
1527 		return;
1528 	}
1529 
1530 }
1531 
1532 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1533 			    struct ena_com_rx_ctx *ena_rx_ctx,
1534 			    struct sk_buff *skb)
1535 {
1536 	enum pkt_hash_types hash_type;
1537 
1538 	if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1539 		if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1540 			   (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1541 
1542 			hash_type = PKT_HASH_TYPE_L4;
1543 		else
1544 			hash_type = PKT_HASH_TYPE_NONE;
1545 
1546 		/* Override hash type if the packet is fragmented */
1547 		if (ena_rx_ctx->frag)
1548 			hash_type = PKT_HASH_TYPE_NONE;
1549 
1550 		skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1551 	}
1552 }
1553 
1554 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1555 {
1556 	struct ena_rx_buffer *rx_info;
1557 	int ret;
1558 
1559 	rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1560 	xdp->data = page_address(rx_info->page) +
1561 		rx_info->page_offset + rx_ring->rx_headroom;
1562 	xdp_set_data_meta_invalid(xdp);
1563 	xdp->data_hard_start = page_address(rx_info->page);
1564 	xdp->data_end = xdp->data + rx_ring->ena_bufs[0].len;
1565 	/* If for some reason we received a bigger packet than
1566 	 * we expect, then we simply drop it
1567 	 */
1568 	if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1569 		return XDP_DROP;
1570 
1571 	ret = ena_xdp_execute(rx_ring, xdp, rx_info);
1572 
1573 	/* The xdp program might expand the headers */
1574 	if (ret == XDP_PASS) {
1575 		rx_info->page_offset = xdp->data - xdp->data_hard_start;
1576 		rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1577 	}
1578 
1579 	return ret;
1580 }
1581 /* ena_clean_rx_irq - Cleanup RX irq
1582  * @rx_ring: RX ring to clean
1583  * @napi: napi handler
1584  * @budget: how many packets driver is allowed to clean
1585  *
1586  * Returns the number of cleaned buffers.
1587  */
1588 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1589 			    u32 budget)
1590 {
1591 	u16 next_to_clean = rx_ring->next_to_clean;
1592 	struct ena_com_rx_ctx ena_rx_ctx;
1593 	struct ena_adapter *adapter;
1594 	u32 res_budget, work_done;
1595 	int rx_copybreak_pkt = 0;
1596 	int refill_threshold;
1597 	struct sk_buff *skb;
1598 	int refill_required;
1599 	struct xdp_buff xdp;
1600 	int total_len = 0;
1601 	int xdp_verdict;
1602 	int rc = 0;
1603 	int i;
1604 
1605 	netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1606 		  "%s qid %d\n", __func__, rx_ring->qid);
1607 	res_budget = budget;
1608 	xdp.rxq = &rx_ring->xdp_rxq;
1609 
1610 	do {
1611 		xdp_verdict = XDP_PASS;
1612 		skb = NULL;
1613 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1614 		ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1615 		ena_rx_ctx.descs = 0;
1616 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1617 				    rx_ring->ena_com_io_sq,
1618 				    &ena_rx_ctx);
1619 		if (unlikely(rc))
1620 			goto error;
1621 
1622 		if (unlikely(ena_rx_ctx.descs == 0))
1623 			break;
1624 
1625 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1626 			  "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1627 			  rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1628 			  ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1629 
1630 		if (ena_xdp_present_ring(rx_ring))
1631 			xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1632 
1633 		/* allocate skb and fill it */
1634 		if (xdp_verdict == XDP_PASS)
1635 			skb = ena_rx_skb(rx_ring,
1636 					 rx_ring->ena_bufs,
1637 					 ena_rx_ctx.descs,
1638 					 &next_to_clean);
1639 
1640 		if (unlikely(!skb)) {
1641 			if (xdp_verdict == XDP_TX) {
1642 				ena_free_rx_page(rx_ring,
1643 						 &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id]);
1644 				res_budget--;
1645 			}
1646 			for (i = 0; i < ena_rx_ctx.descs; i++) {
1647 				rx_ring->free_ids[next_to_clean] =
1648 					rx_ring->ena_bufs[i].req_id;
1649 				next_to_clean =
1650 					ENA_RX_RING_IDX_NEXT(next_to_clean,
1651 							     rx_ring->ring_size);
1652 			}
1653 			if (xdp_verdict == XDP_TX || xdp_verdict == XDP_DROP)
1654 				continue;
1655 			break;
1656 		}
1657 
1658 		ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1659 
1660 		ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1661 
1662 		skb_record_rx_queue(skb, rx_ring->qid);
1663 
1664 		if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1665 			total_len += rx_ring->ena_bufs[0].len;
1666 			rx_copybreak_pkt++;
1667 			napi_gro_receive(napi, skb);
1668 		} else {
1669 			total_len += skb->len;
1670 			napi_gro_frags(napi);
1671 		}
1672 
1673 		res_budget--;
1674 	} while (likely(res_budget));
1675 
1676 	work_done = budget - res_budget;
1677 	rx_ring->per_napi_packets += work_done;
1678 	u64_stats_update_begin(&rx_ring->syncp);
1679 	rx_ring->rx_stats.bytes += total_len;
1680 	rx_ring->rx_stats.cnt += work_done;
1681 	rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1682 	u64_stats_update_end(&rx_ring->syncp);
1683 
1684 	rx_ring->next_to_clean = next_to_clean;
1685 
1686 	refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
1687 	refill_threshold =
1688 		min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1689 		      ENA_RX_REFILL_THRESH_PACKET);
1690 
1691 	/* Optimization, try to batch new rx buffers */
1692 	if (refill_required > refill_threshold) {
1693 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1694 		ena_refill_rx_bufs(rx_ring, refill_required);
1695 	}
1696 
1697 	return work_done;
1698 
1699 error:
1700 	adapter = netdev_priv(rx_ring->netdev);
1701 
1702 	u64_stats_update_begin(&rx_ring->syncp);
1703 	rx_ring->rx_stats.bad_desc_num++;
1704 	u64_stats_update_end(&rx_ring->syncp);
1705 
1706 	/* Too many desc from the device. Trigger reset */
1707 	adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1708 	set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1709 
1710 	return 0;
1711 }
1712 
1713 static void ena_dim_work(struct work_struct *w)
1714 {
1715 	struct dim *dim = container_of(w, struct dim, work);
1716 	struct dim_cq_moder cur_moder =
1717 		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1718 	struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1719 
1720 	ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1721 	dim->state = DIM_START_MEASURE;
1722 }
1723 
1724 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1725 {
1726 	struct dim_sample dim_sample;
1727 	struct ena_ring *rx_ring = ena_napi->rx_ring;
1728 
1729 	if (!rx_ring->per_napi_packets)
1730 		return;
1731 
1732 	rx_ring->non_empty_napi_events++;
1733 
1734 	dim_update_sample(rx_ring->non_empty_napi_events,
1735 			  rx_ring->rx_stats.cnt,
1736 			  rx_ring->rx_stats.bytes,
1737 			  &dim_sample);
1738 
1739 	net_dim(&ena_napi->dim, dim_sample);
1740 
1741 	rx_ring->per_napi_packets = 0;
1742 }
1743 
1744 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1745 					struct ena_ring *rx_ring)
1746 {
1747 	struct ena_eth_io_intr_reg intr_reg;
1748 	u32 rx_interval = 0;
1749 	/* Rx ring can be NULL when for XDP tx queues which don't have an
1750 	 * accompanying rx_ring pair.
1751 	 */
1752 	if (rx_ring)
1753 		rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1754 			rx_ring->smoothed_interval :
1755 			ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1756 
1757 	/* Update intr register: rx intr delay,
1758 	 * tx intr delay and interrupt unmask
1759 	 */
1760 	ena_com_update_intr_reg(&intr_reg,
1761 				rx_interval,
1762 				tx_ring->smoothed_interval,
1763 				true);
1764 
1765 	u64_stats_update_begin(&tx_ring->syncp);
1766 	tx_ring->tx_stats.unmask_interrupt++;
1767 	u64_stats_update_end(&tx_ring->syncp);
1768 	/* It is a shared MSI-X.
1769 	 * Tx and Rx CQ have pointer to it.
1770 	 * So we use one of them to reach the intr reg
1771 	 * The Tx ring is used because the rx_ring is NULL for XDP queues
1772 	 */
1773 	ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1774 }
1775 
1776 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1777 					     struct ena_ring *rx_ring)
1778 {
1779 	int cpu = get_cpu();
1780 	int numa_node;
1781 
1782 	/* Check only one ring since the 2 rings are running on the same cpu */
1783 	if (likely(tx_ring->cpu == cpu))
1784 		goto out;
1785 
1786 	numa_node = cpu_to_node(cpu);
1787 	put_cpu();
1788 
1789 	if (numa_node != NUMA_NO_NODE) {
1790 		ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1791 		if (rx_ring)
1792 			ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1793 						 numa_node);
1794 	}
1795 
1796 	tx_ring->cpu = cpu;
1797 	if (rx_ring)
1798 		rx_ring->cpu = cpu;
1799 
1800 	return;
1801 out:
1802 	put_cpu();
1803 }
1804 
1805 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1806 {
1807 	u32 total_done = 0;
1808 	u16 next_to_clean;
1809 	u32 tx_bytes = 0;
1810 	int tx_pkts = 0;
1811 	u16 req_id;
1812 	int rc;
1813 
1814 	if (unlikely(!xdp_ring))
1815 		return 0;
1816 	next_to_clean = xdp_ring->next_to_clean;
1817 
1818 	while (tx_pkts < budget) {
1819 		struct ena_tx_buffer *tx_info;
1820 		struct xdp_frame *xdpf;
1821 
1822 		rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1823 						&req_id);
1824 		if (rc)
1825 			break;
1826 
1827 		rc = validate_xdp_req_id(xdp_ring, req_id);
1828 		if (rc)
1829 			break;
1830 
1831 		tx_info = &xdp_ring->tx_buffer_info[req_id];
1832 		xdpf = tx_info->xdpf;
1833 
1834 		tx_info->xdpf = NULL;
1835 		tx_info->last_jiffies = 0;
1836 		ena_unmap_tx_buff(xdp_ring, tx_info);
1837 
1838 		netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1839 			  "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1840 			  xdpf);
1841 
1842 		tx_bytes += xdpf->len;
1843 		tx_pkts++;
1844 		total_done += tx_info->tx_descs;
1845 
1846 		__free_page(tx_info->xdp_rx_page);
1847 		xdp_ring->free_ids[next_to_clean] = req_id;
1848 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1849 						     xdp_ring->ring_size);
1850 	}
1851 
1852 	xdp_ring->next_to_clean = next_to_clean;
1853 	ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1854 	ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1855 
1856 	netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1857 		  "tx_poll: q %d done. total pkts: %d\n",
1858 		  xdp_ring->qid, tx_pkts);
1859 
1860 	return tx_pkts;
1861 }
1862 
1863 static int ena_io_poll(struct napi_struct *napi, int budget)
1864 {
1865 	struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1866 	struct ena_ring *tx_ring, *rx_ring;
1867 	int tx_work_done;
1868 	int rx_work_done = 0;
1869 	int tx_budget;
1870 	int napi_comp_call = 0;
1871 	int ret;
1872 
1873 	tx_ring = ena_napi->tx_ring;
1874 	rx_ring = ena_napi->rx_ring;
1875 
1876 	tx_ring->first_interrupt = ena_napi->first_interrupt;
1877 	rx_ring->first_interrupt = ena_napi->first_interrupt;
1878 
1879 	tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1880 
1881 	if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1882 	    test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1883 		napi_complete_done(napi, 0);
1884 		return 0;
1885 	}
1886 
1887 	tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1888 	/* On netpoll the budget is zero and the handler should only clean the
1889 	 * tx completions.
1890 	 */
1891 	if (likely(budget))
1892 		rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1893 
1894 	/* If the device is about to reset or down, avoid unmask
1895 	 * the interrupt and return 0 so NAPI won't reschedule
1896 	 */
1897 	if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1898 		     test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1899 		napi_complete_done(napi, 0);
1900 		ret = 0;
1901 
1902 	} else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1903 		napi_comp_call = 1;
1904 
1905 		/* Update numa and unmask the interrupt only when schedule
1906 		 * from the interrupt context (vs from sk_busy_loop)
1907 		 */
1908 		if (napi_complete_done(napi, rx_work_done)) {
1909 			/* We apply adaptive moderation on Rx path only.
1910 			 * Tx uses static interrupt moderation.
1911 			 */
1912 			if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1913 				ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1914 
1915 			ena_unmask_interrupt(tx_ring, rx_ring);
1916 		}
1917 
1918 		ena_update_ring_numa_node(tx_ring, rx_ring);
1919 
1920 		ret = rx_work_done;
1921 	} else {
1922 		ret = budget;
1923 	}
1924 
1925 	u64_stats_update_begin(&tx_ring->syncp);
1926 	tx_ring->tx_stats.napi_comp += napi_comp_call;
1927 	tx_ring->tx_stats.tx_poll++;
1928 	u64_stats_update_end(&tx_ring->syncp);
1929 
1930 	return ret;
1931 }
1932 
1933 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1934 {
1935 	struct ena_adapter *adapter = (struct ena_adapter *)data;
1936 
1937 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1938 
1939 	/* Don't call the aenq handler before probe is done */
1940 	if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1941 		ena_com_aenq_intr_handler(adapter->ena_dev, data);
1942 
1943 	return IRQ_HANDLED;
1944 }
1945 
1946 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1947  * @irq: interrupt number
1948  * @data: pointer to a network interface private napi device structure
1949  */
1950 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1951 {
1952 	struct ena_napi *ena_napi = data;
1953 
1954 	ena_napi->first_interrupt = true;
1955 
1956 	napi_schedule_irqoff(&ena_napi->napi);
1957 
1958 	return IRQ_HANDLED;
1959 }
1960 
1961 /* Reserve a single MSI-X vector for management (admin + aenq).
1962  * plus reserve one vector for each potential io queue.
1963  * the number of potential io queues is the minimum of what the device
1964  * supports and the number of vCPUs.
1965  */
1966 static int ena_enable_msix(struct ena_adapter *adapter)
1967 {
1968 	int msix_vecs, irq_cnt;
1969 
1970 	if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1971 		netif_err(adapter, probe, adapter->netdev,
1972 			  "Error, MSI-X is already enabled\n");
1973 		return -EPERM;
1974 	}
1975 
1976 	/* Reserved the max msix vectors we might need */
1977 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
1978 	netif_dbg(adapter, probe, adapter->netdev,
1979 		  "trying to enable MSI-X, vectors %d\n", msix_vecs);
1980 
1981 	irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
1982 					msix_vecs, PCI_IRQ_MSIX);
1983 
1984 	if (irq_cnt < 0) {
1985 		netif_err(adapter, probe, adapter->netdev,
1986 			  "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
1987 		return -ENOSPC;
1988 	}
1989 
1990 	if (irq_cnt != msix_vecs) {
1991 		netif_notice(adapter, probe, adapter->netdev,
1992 			     "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1993 			     irq_cnt, msix_vecs);
1994 		adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
1995 	}
1996 
1997 	if (ena_init_rx_cpu_rmap(adapter))
1998 		netif_warn(adapter, probe, adapter->netdev,
1999 			   "Failed to map IRQs to CPUs\n");
2000 
2001 	adapter->msix_vecs = irq_cnt;
2002 	set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2003 
2004 	return 0;
2005 }
2006 
2007 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2008 {
2009 	u32 cpu;
2010 
2011 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2012 		 ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2013 		 pci_name(adapter->pdev));
2014 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2015 		ena_intr_msix_mgmnt;
2016 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2017 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2018 		pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2019 	cpu = cpumask_first(cpu_online_mask);
2020 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2021 	cpumask_set_cpu(cpu,
2022 			&adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2023 }
2024 
2025 static void ena_setup_io_intr(struct ena_adapter *adapter)
2026 {
2027 	struct net_device *netdev;
2028 	int irq_idx, i, cpu;
2029 	int io_queue_count;
2030 
2031 	netdev = adapter->netdev;
2032 	io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2033 
2034 	for (i = 0; i < io_queue_count; i++) {
2035 		irq_idx = ENA_IO_IRQ_IDX(i);
2036 		cpu = i % num_online_cpus();
2037 
2038 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2039 			 "%s-Tx-Rx-%d", netdev->name, i);
2040 		adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2041 		adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2042 		adapter->irq_tbl[irq_idx].vector =
2043 			pci_irq_vector(adapter->pdev, irq_idx);
2044 		adapter->irq_tbl[irq_idx].cpu = cpu;
2045 
2046 		cpumask_set_cpu(cpu,
2047 				&adapter->irq_tbl[irq_idx].affinity_hint_mask);
2048 	}
2049 }
2050 
2051 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2052 {
2053 	unsigned long flags = 0;
2054 	struct ena_irq *irq;
2055 	int rc;
2056 
2057 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2058 	rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2059 			 irq->data);
2060 	if (rc) {
2061 		netif_err(adapter, probe, adapter->netdev,
2062 			  "failed to request admin irq\n");
2063 		return rc;
2064 	}
2065 
2066 	netif_dbg(adapter, probe, adapter->netdev,
2067 		  "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2068 		  irq->affinity_hint_mask.bits[0], irq->vector);
2069 
2070 	irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2071 
2072 	return rc;
2073 }
2074 
2075 static int ena_request_io_irq(struct ena_adapter *adapter)
2076 {
2077 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2078 	unsigned long flags = 0;
2079 	struct ena_irq *irq;
2080 	int rc = 0, i, k;
2081 
2082 	if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2083 		netif_err(adapter, ifup, adapter->netdev,
2084 			  "Failed to request I/O IRQ: MSI-X is not enabled\n");
2085 		return -EINVAL;
2086 	}
2087 
2088 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2089 		irq = &adapter->irq_tbl[i];
2090 		rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2091 				 irq->data);
2092 		if (rc) {
2093 			netif_err(adapter, ifup, adapter->netdev,
2094 				  "Failed to request I/O IRQ. index %d rc %d\n",
2095 				   i, rc);
2096 			goto err;
2097 		}
2098 
2099 		netif_dbg(adapter, ifup, adapter->netdev,
2100 			  "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2101 			  i, irq->affinity_hint_mask.bits[0], irq->vector);
2102 
2103 		irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2104 	}
2105 
2106 	return rc;
2107 
2108 err:
2109 	for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2110 		irq = &adapter->irq_tbl[k];
2111 		free_irq(irq->vector, irq->data);
2112 	}
2113 
2114 	return rc;
2115 }
2116 
2117 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2118 {
2119 	struct ena_irq *irq;
2120 
2121 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2122 	synchronize_irq(irq->vector);
2123 	irq_set_affinity_hint(irq->vector, NULL);
2124 	free_irq(irq->vector, irq->data);
2125 }
2126 
2127 static void ena_free_io_irq(struct ena_adapter *adapter)
2128 {
2129 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2130 	struct ena_irq *irq;
2131 	int i;
2132 
2133 #ifdef CONFIG_RFS_ACCEL
2134 	if (adapter->msix_vecs >= 1) {
2135 		free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2136 		adapter->netdev->rx_cpu_rmap = NULL;
2137 	}
2138 #endif /* CONFIG_RFS_ACCEL */
2139 
2140 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2141 		irq = &adapter->irq_tbl[i];
2142 		irq_set_affinity_hint(irq->vector, NULL);
2143 		free_irq(irq->vector, irq->data);
2144 	}
2145 }
2146 
2147 static void ena_disable_msix(struct ena_adapter *adapter)
2148 {
2149 	if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2150 		pci_free_irq_vectors(adapter->pdev);
2151 }
2152 
2153 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2154 {
2155 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2156 	int i;
2157 
2158 	if (!netif_running(adapter->netdev))
2159 		return;
2160 
2161 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2162 		synchronize_irq(adapter->irq_tbl[i].vector);
2163 }
2164 
2165 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2166 				  int first_index,
2167 				  int count)
2168 {
2169 	int i;
2170 
2171 	for (i = first_index; i < first_index + count; i++) {
2172 		/* Check if napi was initialized before */
2173 		if (!ENA_IS_XDP_INDEX(adapter, i) ||
2174 		    adapter->ena_napi[i].xdp_ring)
2175 			netif_napi_del(&adapter->ena_napi[i].napi);
2176 		else
2177 			WARN_ON(ENA_IS_XDP_INDEX(adapter, i) &&
2178 				adapter->ena_napi[i].xdp_ring);
2179 	}
2180 }
2181 
2182 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2183 				   int first_index, int count)
2184 {
2185 	struct ena_napi *napi = {0};
2186 	int i;
2187 
2188 	for (i = first_index; i < first_index + count; i++) {
2189 		napi = &adapter->ena_napi[i];
2190 
2191 		netif_napi_add(adapter->netdev,
2192 			       &adapter->ena_napi[i].napi,
2193 			       ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2194 			       ENA_NAPI_BUDGET);
2195 
2196 		if (!ENA_IS_XDP_INDEX(adapter, i)) {
2197 			napi->rx_ring = &adapter->rx_ring[i];
2198 			napi->tx_ring = &adapter->tx_ring[i];
2199 		} else {
2200 			napi->xdp_ring = &adapter->tx_ring[i];
2201 		}
2202 		napi->qid = i;
2203 	}
2204 }
2205 
2206 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2207 				      int first_index,
2208 				      int count)
2209 {
2210 	int i;
2211 
2212 	for (i = first_index; i < first_index + count; i++)
2213 		napi_disable(&adapter->ena_napi[i].napi);
2214 }
2215 
2216 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2217 				     int first_index,
2218 				     int count)
2219 {
2220 	int i;
2221 
2222 	for (i = first_index; i < first_index + count; i++)
2223 		napi_enable(&adapter->ena_napi[i].napi);
2224 }
2225 
2226 /* Configure the Rx forwarding */
2227 static int ena_rss_configure(struct ena_adapter *adapter)
2228 {
2229 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2230 	int rc;
2231 
2232 	/* In case the RSS table wasn't initialized by probe */
2233 	if (!ena_dev->rss.tbl_log_size) {
2234 		rc = ena_rss_init_default(adapter);
2235 		if (rc && (rc != -EOPNOTSUPP)) {
2236 			netif_err(adapter, ifup, adapter->netdev,
2237 					"Failed to init RSS rc: %d\n", rc);
2238 			return rc;
2239 		}
2240 	}
2241 
2242 	/* Set indirect table */
2243 	rc = ena_com_indirect_table_set(ena_dev);
2244 	if (unlikely(rc && rc != -EOPNOTSUPP))
2245 		return rc;
2246 
2247 	/* Configure hash function (if supported) */
2248 	rc = ena_com_set_hash_function(ena_dev);
2249 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
2250 		return rc;
2251 
2252 	/* Configure hash inputs (if supported) */
2253 	rc = ena_com_set_hash_ctrl(ena_dev);
2254 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
2255 		return rc;
2256 
2257 	return 0;
2258 }
2259 
2260 static int ena_up_complete(struct ena_adapter *adapter)
2261 {
2262 	int rc;
2263 
2264 	rc = ena_rss_configure(adapter);
2265 	if (rc)
2266 		return rc;
2267 
2268 	ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2269 
2270 	ena_refill_all_rx_bufs(adapter);
2271 
2272 	/* enable transmits */
2273 	netif_tx_start_all_queues(adapter->netdev);
2274 
2275 	ena_napi_enable_in_range(adapter,
2276 				 0,
2277 				 adapter->xdp_num_queues + adapter->num_io_queues);
2278 
2279 	return 0;
2280 }
2281 
2282 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2283 {
2284 	struct ena_com_create_io_ctx ctx;
2285 	struct ena_com_dev *ena_dev;
2286 	struct ena_ring *tx_ring;
2287 	u32 msix_vector;
2288 	u16 ena_qid;
2289 	int rc;
2290 
2291 	ena_dev = adapter->ena_dev;
2292 
2293 	tx_ring = &adapter->tx_ring[qid];
2294 	msix_vector = ENA_IO_IRQ_IDX(qid);
2295 	ena_qid = ENA_IO_TXQ_IDX(qid);
2296 
2297 	memset(&ctx, 0x0, sizeof(ctx));
2298 
2299 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2300 	ctx.qid = ena_qid;
2301 	ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2302 	ctx.msix_vector = msix_vector;
2303 	ctx.queue_size = tx_ring->ring_size;
2304 	ctx.numa_node = cpu_to_node(tx_ring->cpu);
2305 
2306 	rc = ena_com_create_io_queue(ena_dev, &ctx);
2307 	if (rc) {
2308 		netif_err(adapter, ifup, adapter->netdev,
2309 			  "Failed to create I/O TX queue num %d rc: %d\n",
2310 			   qid, rc);
2311 		return rc;
2312 	}
2313 
2314 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2315 				     &tx_ring->ena_com_io_sq,
2316 				     &tx_ring->ena_com_io_cq);
2317 	if (rc) {
2318 		netif_err(adapter, ifup, adapter->netdev,
2319 			  "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2320 			  qid, rc);
2321 		ena_com_destroy_io_queue(ena_dev, ena_qid);
2322 		return rc;
2323 	}
2324 
2325 	ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2326 	return rc;
2327 }
2328 
2329 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2330 					    int first_index, int count)
2331 {
2332 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2333 	int rc, i;
2334 
2335 	for (i = first_index; i < first_index + count; i++) {
2336 		rc = ena_create_io_tx_queue(adapter, i);
2337 		if (rc)
2338 			goto create_err;
2339 	}
2340 
2341 	return 0;
2342 
2343 create_err:
2344 	while (i-- > first_index)
2345 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2346 
2347 	return rc;
2348 }
2349 
2350 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2351 {
2352 	struct ena_com_dev *ena_dev;
2353 	struct ena_com_create_io_ctx ctx;
2354 	struct ena_ring *rx_ring;
2355 	u32 msix_vector;
2356 	u16 ena_qid;
2357 	int rc;
2358 
2359 	ena_dev = adapter->ena_dev;
2360 
2361 	rx_ring = &adapter->rx_ring[qid];
2362 	msix_vector = ENA_IO_IRQ_IDX(qid);
2363 	ena_qid = ENA_IO_RXQ_IDX(qid);
2364 
2365 	memset(&ctx, 0x0, sizeof(ctx));
2366 
2367 	ctx.qid = ena_qid;
2368 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2369 	ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2370 	ctx.msix_vector = msix_vector;
2371 	ctx.queue_size = rx_ring->ring_size;
2372 	ctx.numa_node = cpu_to_node(rx_ring->cpu);
2373 
2374 	rc = ena_com_create_io_queue(ena_dev, &ctx);
2375 	if (rc) {
2376 		netif_err(adapter, ifup, adapter->netdev,
2377 			  "Failed to create I/O RX queue num %d rc: %d\n",
2378 			  qid, rc);
2379 		return rc;
2380 	}
2381 
2382 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2383 				     &rx_ring->ena_com_io_sq,
2384 				     &rx_ring->ena_com_io_cq);
2385 	if (rc) {
2386 		netif_err(adapter, ifup, adapter->netdev,
2387 			  "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2388 			  qid, rc);
2389 		goto err;
2390 	}
2391 
2392 	ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2393 
2394 	return rc;
2395 err:
2396 	ena_com_destroy_io_queue(ena_dev, ena_qid);
2397 	return rc;
2398 }
2399 
2400 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2401 {
2402 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2403 	int rc, i;
2404 
2405 	for (i = 0; i < adapter->num_io_queues; i++) {
2406 		rc = ena_create_io_rx_queue(adapter, i);
2407 		if (rc)
2408 			goto create_err;
2409 		INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2410 	}
2411 
2412 	return 0;
2413 
2414 create_err:
2415 	while (i--) {
2416 		cancel_work_sync(&adapter->ena_napi[i].dim.work);
2417 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2418 	}
2419 
2420 	return rc;
2421 }
2422 
2423 static void set_io_rings_size(struct ena_adapter *adapter,
2424 			      int new_tx_size,
2425 			      int new_rx_size)
2426 {
2427 	int i;
2428 
2429 	for (i = 0; i < adapter->num_io_queues; i++) {
2430 		adapter->tx_ring[i].ring_size = new_tx_size;
2431 		adapter->rx_ring[i].ring_size = new_rx_size;
2432 	}
2433 }
2434 
2435 /* This function allows queue allocation to backoff when the system is
2436  * low on memory. If there is not enough memory to allocate io queues
2437  * the driver will try to allocate smaller queues.
2438  *
2439  * The backoff algorithm is as follows:
2440  *  1. Try to allocate TX and RX and if successful.
2441  *  1.1. return success
2442  *
2443  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2444  *
2445  *  3. If TX or RX is smaller than 256
2446  *  3.1. return failure.
2447  *  4. else
2448  *  4.1. go back to 1.
2449  */
2450 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2451 {
2452 	int rc, cur_rx_ring_size, cur_tx_ring_size;
2453 	int new_rx_ring_size, new_tx_ring_size;
2454 
2455 	/* current queue sizes might be set to smaller than the requested
2456 	 * ones due to past queue allocation failures.
2457 	 */
2458 	set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2459 			adapter->requested_rx_ring_size);
2460 
2461 	while (1) {
2462 		if (ena_xdp_present(adapter)) {
2463 			rc = ena_setup_and_create_all_xdp_queues(adapter);
2464 
2465 			if (rc)
2466 				goto err_setup_tx;
2467 		}
2468 		rc = ena_setup_tx_resources_in_range(adapter,
2469 						     0,
2470 						     adapter->num_io_queues);
2471 		if (rc)
2472 			goto err_setup_tx;
2473 
2474 		rc = ena_create_io_tx_queues_in_range(adapter,
2475 						      0,
2476 						      adapter->num_io_queues);
2477 		if (rc)
2478 			goto err_create_tx_queues;
2479 
2480 		rc = ena_setup_all_rx_resources(adapter);
2481 		if (rc)
2482 			goto err_setup_rx;
2483 
2484 		rc = ena_create_all_io_rx_queues(adapter);
2485 		if (rc)
2486 			goto err_create_rx_queues;
2487 
2488 		return 0;
2489 
2490 err_create_rx_queues:
2491 		ena_free_all_io_rx_resources(adapter);
2492 err_setup_rx:
2493 		ena_destroy_all_tx_queues(adapter);
2494 err_create_tx_queues:
2495 		ena_free_all_io_tx_resources(adapter);
2496 err_setup_tx:
2497 		if (rc != -ENOMEM) {
2498 			netif_err(adapter, ifup, adapter->netdev,
2499 				  "Queue creation failed with error code %d\n",
2500 				   rc);
2501 			return rc;
2502 		}
2503 
2504 		cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2505 		cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2506 
2507 		netif_err(adapter, ifup, adapter->netdev,
2508 			  "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2509 			  cur_tx_ring_size, cur_rx_ring_size);
2510 
2511 		new_tx_ring_size = cur_tx_ring_size;
2512 		new_rx_ring_size = cur_rx_ring_size;
2513 
2514 		/* Decrease the size of the larger queue, or
2515 		 * decrease both if they are the same size.
2516 		 */
2517 		if (cur_rx_ring_size <= cur_tx_ring_size)
2518 			new_tx_ring_size = cur_tx_ring_size / 2;
2519 		if (cur_rx_ring_size >= cur_tx_ring_size)
2520 			new_rx_ring_size = cur_rx_ring_size / 2;
2521 
2522 		if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2523 				new_rx_ring_size < ENA_MIN_RING_SIZE) {
2524 			netif_err(adapter, ifup, adapter->netdev,
2525 				  "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2526 				  ENA_MIN_RING_SIZE);
2527 			return rc;
2528 		}
2529 
2530 		netif_err(adapter, ifup, adapter->netdev,
2531 			  "Retrying queue creation with sizes TX=%d, RX=%d\n",
2532 			  new_tx_ring_size,
2533 			  new_rx_ring_size);
2534 
2535 		set_io_rings_size(adapter, new_tx_ring_size,
2536 				  new_rx_ring_size);
2537 	}
2538 }
2539 
2540 static int ena_up(struct ena_adapter *adapter)
2541 {
2542 	int io_queue_count, rc, i;
2543 
2544 	netdev_dbg(adapter->netdev, "%s\n", __func__);
2545 
2546 	io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2547 	ena_setup_io_intr(adapter);
2548 
2549 	/* napi poll functions should be initialized before running
2550 	 * request_irq(), to handle a rare condition where there is a pending
2551 	 * interrupt, causing the ISR to fire immediately while the poll
2552 	 * function wasn't set yet, causing a null dereference
2553 	 */
2554 	ena_init_napi_in_range(adapter, 0, io_queue_count);
2555 
2556 	rc = ena_request_io_irq(adapter);
2557 	if (rc)
2558 		goto err_req_irq;
2559 
2560 	rc = create_queues_with_size_backoff(adapter);
2561 	if (rc)
2562 		goto err_create_queues_with_backoff;
2563 
2564 	rc = ena_up_complete(adapter);
2565 	if (rc)
2566 		goto err_up;
2567 
2568 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2569 		netif_carrier_on(adapter->netdev);
2570 
2571 	u64_stats_update_begin(&adapter->syncp);
2572 	adapter->dev_stats.interface_up++;
2573 	u64_stats_update_end(&adapter->syncp);
2574 
2575 	set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2576 
2577 	/* Enable completion queues interrupt */
2578 	for (i = 0; i < adapter->num_io_queues; i++)
2579 		ena_unmask_interrupt(&adapter->tx_ring[i],
2580 				     &adapter->rx_ring[i]);
2581 
2582 	/* schedule napi in case we had pending packets
2583 	 * from the last time we disable napi
2584 	 */
2585 	for (i = 0; i < io_queue_count; i++)
2586 		napi_schedule(&adapter->ena_napi[i].napi);
2587 
2588 	return rc;
2589 
2590 err_up:
2591 	ena_destroy_all_tx_queues(adapter);
2592 	ena_free_all_io_tx_resources(adapter);
2593 	ena_destroy_all_rx_queues(adapter);
2594 	ena_free_all_io_rx_resources(adapter);
2595 err_create_queues_with_backoff:
2596 	ena_free_io_irq(adapter);
2597 err_req_irq:
2598 	ena_del_napi_in_range(adapter, 0, io_queue_count);
2599 
2600 	return rc;
2601 }
2602 
2603 static void ena_down(struct ena_adapter *adapter)
2604 {
2605 	int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2606 
2607 	netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2608 
2609 	clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2610 
2611 	u64_stats_update_begin(&adapter->syncp);
2612 	adapter->dev_stats.interface_down++;
2613 	u64_stats_update_end(&adapter->syncp);
2614 
2615 	netif_carrier_off(adapter->netdev);
2616 	netif_tx_disable(adapter->netdev);
2617 
2618 	/* After this point the napi handler won't enable the tx queue */
2619 	ena_napi_disable_in_range(adapter, 0, io_queue_count);
2620 
2621 	/* After destroy the queue there won't be any new interrupts */
2622 
2623 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2624 		int rc;
2625 
2626 		rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2627 		if (rc)
2628 			dev_err(&adapter->pdev->dev, "Device reset failed\n");
2629 		/* stop submitting admin commands on a device that was reset */
2630 		ena_com_set_admin_running_state(adapter->ena_dev, false);
2631 	}
2632 
2633 	ena_destroy_all_io_queues(adapter);
2634 
2635 	ena_disable_io_intr_sync(adapter);
2636 	ena_free_io_irq(adapter);
2637 	ena_del_napi_in_range(adapter, 0, io_queue_count);
2638 
2639 	ena_free_all_tx_bufs(adapter);
2640 	ena_free_all_rx_bufs(adapter);
2641 	ena_free_all_io_tx_resources(adapter);
2642 	ena_free_all_io_rx_resources(adapter);
2643 }
2644 
2645 /* ena_open - Called when a network interface is made active
2646  * @netdev: network interface device structure
2647  *
2648  * Returns 0 on success, negative value on failure
2649  *
2650  * The open entry point is called when a network interface is made
2651  * active by the system (IFF_UP).  At this point all resources needed
2652  * for transmit and receive operations are allocated, the interrupt
2653  * handler is registered with the OS, the watchdog timer is started,
2654  * and the stack is notified that the interface is ready.
2655  */
2656 static int ena_open(struct net_device *netdev)
2657 {
2658 	struct ena_adapter *adapter = netdev_priv(netdev);
2659 	int rc;
2660 
2661 	/* Notify the stack of the actual queue counts. */
2662 	rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2663 	if (rc) {
2664 		netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2665 		return rc;
2666 	}
2667 
2668 	rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2669 	if (rc) {
2670 		netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2671 		return rc;
2672 	}
2673 
2674 	rc = ena_up(adapter);
2675 	if (rc)
2676 		return rc;
2677 
2678 	return rc;
2679 }
2680 
2681 /* ena_close - Disables a network interface
2682  * @netdev: network interface device structure
2683  *
2684  * Returns 0, this is not allowed to fail
2685  *
2686  * The close entry point is called when an interface is de-activated
2687  * by the OS.  The hardware is still under the drivers control, but
2688  * needs to be disabled.  A global MAC reset is issued to stop the
2689  * hardware, and all transmit and receive resources are freed.
2690  */
2691 static int ena_close(struct net_device *netdev)
2692 {
2693 	struct ena_adapter *adapter = netdev_priv(netdev);
2694 
2695 	netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2696 
2697 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2698 		return 0;
2699 
2700 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2701 		ena_down(adapter);
2702 
2703 	/* Check for device status and issue reset if needed*/
2704 	check_for_admin_com_state(adapter);
2705 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2706 		netif_err(adapter, ifdown, adapter->netdev,
2707 			  "Destroy failure, restarting device\n");
2708 		ena_dump_stats_to_dmesg(adapter);
2709 		/* rtnl lock already obtained in dev_ioctl() layer */
2710 		ena_destroy_device(adapter, false);
2711 		ena_restore_device(adapter);
2712 	}
2713 
2714 	return 0;
2715 }
2716 
2717 int ena_update_queue_sizes(struct ena_adapter *adapter,
2718 			   u32 new_tx_size,
2719 			   u32 new_rx_size)
2720 {
2721 	bool dev_was_up;
2722 
2723 	dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2724 	ena_close(adapter->netdev);
2725 	adapter->requested_tx_ring_size = new_tx_size;
2726 	adapter->requested_rx_ring_size = new_rx_size;
2727 	ena_init_io_rings(adapter,
2728 			  0,
2729 			  adapter->xdp_num_queues +
2730 			  adapter->num_io_queues);
2731 	return dev_was_up ? ena_up(adapter) : 0;
2732 }
2733 
2734 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2735 {
2736 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2737 	int prev_channel_count;
2738 	bool dev_was_up;
2739 
2740 	dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2741 	ena_close(adapter->netdev);
2742 	prev_channel_count = adapter->num_io_queues;
2743 	adapter->num_io_queues = new_channel_count;
2744 	if (ena_xdp_present(adapter) &&
2745 	    ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2746 		adapter->xdp_first_ring = new_channel_count;
2747 		adapter->xdp_num_queues = new_channel_count;
2748 		if (prev_channel_count > new_channel_count)
2749 			ena_xdp_exchange_program_rx_in_range(adapter,
2750 							     NULL,
2751 							     new_channel_count,
2752 							     prev_channel_count);
2753 		else
2754 			ena_xdp_exchange_program_rx_in_range(adapter,
2755 							     adapter->xdp_bpf_prog,
2756 							     prev_channel_count,
2757 							     new_channel_count);
2758 	}
2759 
2760 	/* We need to destroy the rss table so that the indirection
2761 	 * table will be reinitialized by ena_up()
2762 	 */
2763 	ena_com_rss_destroy(ena_dev);
2764 	ena_init_io_rings(adapter,
2765 			  0,
2766 			  adapter->xdp_num_queues +
2767 			  adapter->num_io_queues);
2768 	return dev_was_up ? ena_open(adapter->netdev) : 0;
2769 }
2770 
2771 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
2772 {
2773 	u32 mss = skb_shinfo(skb)->gso_size;
2774 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2775 	u8 l4_protocol = 0;
2776 
2777 	if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2778 		ena_tx_ctx->l4_csum_enable = 1;
2779 		if (mss) {
2780 			ena_tx_ctx->tso_enable = 1;
2781 			ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2782 			ena_tx_ctx->l4_csum_partial = 0;
2783 		} else {
2784 			ena_tx_ctx->tso_enable = 0;
2785 			ena_meta->l4_hdr_len = 0;
2786 			ena_tx_ctx->l4_csum_partial = 1;
2787 		}
2788 
2789 		switch (ip_hdr(skb)->version) {
2790 		case IPVERSION:
2791 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2792 			if (ip_hdr(skb)->frag_off & htons(IP_DF))
2793 				ena_tx_ctx->df = 1;
2794 			if (mss)
2795 				ena_tx_ctx->l3_csum_enable = 1;
2796 			l4_protocol = ip_hdr(skb)->protocol;
2797 			break;
2798 		case 6:
2799 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2800 			l4_protocol = ipv6_hdr(skb)->nexthdr;
2801 			break;
2802 		default:
2803 			break;
2804 		}
2805 
2806 		if (l4_protocol == IPPROTO_TCP)
2807 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2808 		else
2809 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2810 
2811 		ena_meta->mss = mss;
2812 		ena_meta->l3_hdr_len = skb_network_header_len(skb);
2813 		ena_meta->l3_hdr_offset = skb_network_offset(skb);
2814 		ena_tx_ctx->meta_valid = 1;
2815 
2816 	} else {
2817 		ena_tx_ctx->meta_valid = 0;
2818 	}
2819 }
2820 
2821 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2822 				       struct sk_buff *skb)
2823 {
2824 	int num_frags, header_len, rc;
2825 
2826 	num_frags = skb_shinfo(skb)->nr_frags;
2827 	header_len = skb_headlen(skb);
2828 
2829 	if (num_frags < tx_ring->sgl_size)
2830 		return 0;
2831 
2832 	if ((num_frags == tx_ring->sgl_size) &&
2833 	    (header_len < tx_ring->tx_max_header_size))
2834 		return 0;
2835 
2836 	u64_stats_update_begin(&tx_ring->syncp);
2837 	tx_ring->tx_stats.linearize++;
2838 	u64_stats_update_end(&tx_ring->syncp);
2839 
2840 	rc = skb_linearize(skb);
2841 	if (unlikely(rc)) {
2842 		u64_stats_update_begin(&tx_ring->syncp);
2843 		tx_ring->tx_stats.linearize_failed++;
2844 		u64_stats_update_end(&tx_ring->syncp);
2845 	}
2846 
2847 	return rc;
2848 }
2849 
2850 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2851 			  struct ena_tx_buffer *tx_info,
2852 			  struct sk_buff *skb,
2853 			  void **push_hdr,
2854 			  u16 *header_len)
2855 {
2856 	struct ena_adapter *adapter = tx_ring->adapter;
2857 	struct ena_com_buf *ena_buf;
2858 	dma_addr_t dma;
2859 	u32 skb_head_len, frag_len, last_frag;
2860 	u16 push_len = 0;
2861 	u16 delta = 0;
2862 	int i = 0;
2863 
2864 	skb_head_len = skb_headlen(skb);
2865 	tx_info->skb = skb;
2866 	ena_buf = tx_info->bufs;
2867 
2868 	if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2869 		/* When the device is LLQ mode, the driver will copy
2870 		 * the header into the device memory space.
2871 		 * the ena_com layer assume the header is in a linear
2872 		 * memory space.
2873 		 * This assumption might be wrong since part of the header
2874 		 * can be in the fragmented buffers.
2875 		 * Use skb_header_pointer to make sure the header is in a
2876 		 * linear memory space.
2877 		 */
2878 
2879 		push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2880 		*push_hdr = skb_header_pointer(skb, 0, push_len,
2881 					       tx_ring->push_buf_intermediate_buf);
2882 		*header_len = push_len;
2883 		if (unlikely(skb->data != *push_hdr)) {
2884 			u64_stats_update_begin(&tx_ring->syncp);
2885 			tx_ring->tx_stats.llq_buffer_copy++;
2886 			u64_stats_update_end(&tx_ring->syncp);
2887 
2888 			delta = push_len - skb_head_len;
2889 		}
2890 	} else {
2891 		*push_hdr = NULL;
2892 		*header_len = min_t(u32, skb_head_len,
2893 				    tx_ring->tx_max_header_size);
2894 	}
2895 
2896 	netif_dbg(adapter, tx_queued, adapter->netdev,
2897 		  "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2898 		  *push_hdr, push_len);
2899 
2900 	if (skb_head_len > push_len) {
2901 		dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2902 				     skb_head_len - push_len, DMA_TO_DEVICE);
2903 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2904 			goto error_report_dma_error;
2905 
2906 		ena_buf->paddr = dma;
2907 		ena_buf->len = skb_head_len - push_len;
2908 
2909 		ena_buf++;
2910 		tx_info->num_of_bufs++;
2911 		tx_info->map_linear_data = 1;
2912 	} else {
2913 		tx_info->map_linear_data = 0;
2914 	}
2915 
2916 	last_frag = skb_shinfo(skb)->nr_frags;
2917 
2918 	for (i = 0; i < last_frag; i++) {
2919 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2920 
2921 		frag_len = skb_frag_size(frag);
2922 
2923 		if (unlikely(delta >= frag_len)) {
2924 			delta -= frag_len;
2925 			continue;
2926 		}
2927 
2928 		dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2929 				       frag_len - delta, DMA_TO_DEVICE);
2930 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2931 			goto error_report_dma_error;
2932 
2933 		ena_buf->paddr = dma;
2934 		ena_buf->len = frag_len - delta;
2935 		ena_buf++;
2936 		tx_info->num_of_bufs++;
2937 		delta = 0;
2938 	}
2939 
2940 	return 0;
2941 
2942 error_report_dma_error:
2943 	u64_stats_update_begin(&tx_ring->syncp);
2944 	tx_ring->tx_stats.dma_mapping_err++;
2945 	u64_stats_update_end(&tx_ring->syncp);
2946 	netdev_warn(adapter->netdev, "failed to map skb\n");
2947 
2948 	tx_info->skb = NULL;
2949 
2950 	tx_info->num_of_bufs += i;
2951 	ena_unmap_tx_buff(tx_ring, tx_info);
2952 
2953 	return -EINVAL;
2954 }
2955 
2956 /* Called with netif_tx_lock. */
2957 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
2958 {
2959 	struct ena_adapter *adapter = netdev_priv(dev);
2960 	struct ena_tx_buffer *tx_info;
2961 	struct ena_com_tx_ctx ena_tx_ctx;
2962 	struct ena_ring *tx_ring;
2963 	struct netdev_queue *txq;
2964 	void *push_hdr;
2965 	u16 next_to_use, req_id, header_len;
2966 	int qid, rc;
2967 
2968 	netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
2969 	/*  Determine which tx ring we will be placed on */
2970 	qid = skb_get_queue_mapping(skb);
2971 	tx_ring = &adapter->tx_ring[qid];
2972 	txq = netdev_get_tx_queue(dev, qid);
2973 
2974 	rc = ena_check_and_linearize_skb(tx_ring, skb);
2975 	if (unlikely(rc))
2976 		goto error_drop_packet;
2977 
2978 	skb_tx_timestamp(skb);
2979 
2980 	next_to_use = tx_ring->next_to_use;
2981 	req_id = tx_ring->free_ids[next_to_use];
2982 	tx_info = &tx_ring->tx_buffer_info[req_id];
2983 	tx_info->num_of_bufs = 0;
2984 
2985 	WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
2986 
2987 	rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
2988 	if (unlikely(rc))
2989 		goto error_drop_packet;
2990 
2991 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2992 	ena_tx_ctx.ena_bufs = tx_info->bufs;
2993 	ena_tx_ctx.push_header = push_hdr;
2994 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2995 	ena_tx_ctx.req_id = req_id;
2996 	ena_tx_ctx.header_len = header_len;
2997 
2998 	/* set flags and meta data */
2999 	ena_tx_csum(&ena_tx_ctx, skb);
3000 
3001 	rc = ena_xmit_common(dev,
3002 			     tx_ring,
3003 			     tx_info,
3004 			     &ena_tx_ctx,
3005 			     next_to_use,
3006 			     skb->len);
3007 	if (rc)
3008 		goto error_unmap_dma;
3009 
3010 	netdev_tx_sent_queue(txq, skb->len);
3011 
3012 	/* stop the queue when no more space available, the packet can have up
3013 	 * to sgl_size + 2. one for the meta descriptor and one for header
3014 	 * (if the header is larger than tx_max_header_size).
3015 	 */
3016 	if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3017 						   tx_ring->sgl_size + 2))) {
3018 		netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3019 			  __func__, qid);
3020 
3021 		netif_tx_stop_queue(txq);
3022 		u64_stats_update_begin(&tx_ring->syncp);
3023 		tx_ring->tx_stats.queue_stop++;
3024 		u64_stats_update_end(&tx_ring->syncp);
3025 
3026 		/* There is a rare condition where this function decide to
3027 		 * stop the queue but meanwhile clean_tx_irq updates
3028 		 * next_to_completion and terminates.
3029 		 * The queue will remain stopped forever.
3030 		 * To solve this issue add a mb() to make sure that
3031 		 * netif_tx_stop_queue() write is vissible before checking if
3032 		 * there is additional space in the queue.
3033 		 */
3034 		smp_mb();
3035 
3036 		if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3037 						 ENA_TX_WAKEUP_THRESH)) {
3038 			netif_tx_wake_queue(txq);
3039 			u64_stats_update_begin(&tx_ring->syncp);
3040 			tx_ring->tx_stats.queue_wakeup++;
3041 			u64_stats_update_end(&tx_ring->syncp);
3042 		}
3043 	}
3044 
3045 	if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
3046 		/* trigger the dma engine. ena_com_write_sq_doorbell()
3047 		 * has a mb
3048 		 */
3049 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
3050 		u64_stats_update_begin(&tx_ring->syncp);
3051 		tx_ring->tx_stats.doorbells++;
3052 		u64_stats_update_end(&tx_ring->syncp);
3053 	}
3054 
3055 	return NETDEV_TX_OK;
3056 
3057 error_unmap_dma:
3058 	ena_unmap_tx_buff(tx_ring, tx_info);
3059 	tx_info->skb = NULL;
3060 
3061 error_drop_packet:
3062 	dev_kfree_skb(skb);
3063 	return NETDEV_TX_OK;
3064 }
3065 
3066 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3067 			    struct net_device *sb_dev)
3068 {
3069 	u16 qid;
3070 	/* we suspect that this is good for in--kernel network services that
3071 	 * want to loop incoming skb rx to tx in normal user generated traffic,
3072 	 * most probably we will not get to this
3073 	 */
3074 	if (skb_rx_queue_recorded(skb))
3075 		qid = skb_get_rx_queue(skb);
3076 	else
3077 		qid = netdev_pick_tx(dev, skb, NULL);
3078 
3079 	return qid;
3080 }
3081 
3082 static void ena_config_host_info(struct ena_com_dev *ena_dev,
3083 				 struct pci_dev *pdev)
3084 {
3085 	struct ena_admin_host_info *host_info;
3086 	int rc;
3087 
3088 	/* Allocate only the host info */
3089 	rc = ena_com_allocate_host_info(ena_dev);
3090 	if (rc) {
3091 		pr_err("Cannot allocate host info\n");
3092 		return;
3093 	}
3094 
3095 	host_info = ena_dev->host_attr.host_info;
3096 
3097 	host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3098 	host_info->os_type = ENA_ADMIN_OS_LINUX;
3099 	host_info->kernel_ver = LINUX_VERSION_CODE;
3100 	strlcpy(host_info->kernel_ver_str, utsname()->version,
3101 		sizeof(host_info->kernel_ver_str) - 1);
3102 	host_info->os_dist = 0;
3103 	strncpy(host_info->os_dist_str, utsname()->release,
3104 		sizeof(host_info->os_dist_str) - 1);
3105 	host_info->driver_version =
3106 		(DRV_MODULE_GEN_MAJOR) |
3107 		(DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3108 		(DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3109 		("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3110 	host_info->num_cpus = num_online_cpus();
3111 
3112 	host_info->driver_supported_features =
3113 		ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK;
3114 
3115 	rc = ena_com_set_host_attributes(ena_dev);
3116 	if (rc) {
3117 		if (rc == -EOPNOTSUPP)
3118 			pr_warn("Cannot set host attributes\n");
3119 		else
3120 			pr_err("Cannot set host attributes\n");
3121 
3122 		goto err;
3123 	}
3124 
3125 	return;
3126 
3127 err:
3128 	ena_com_delete_host_info(ena_dev);
3129 }
3130 
3131 static void ena_config_debug_area(struct ena_adapter *adapter)
3132 {
3133 	u32 debug_area_size;
3134 	int rc, ss_count;
3135 
3136 	ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3137 	if (ss_count <= 0) {
3138 		netif_err(adapter, drv, adapter->netdev,
3139 			  "SS count is negative\n");
3140 		return;
3141 	}
3142 
3143 	/* allocate 32 bytes for each string and 64bit for the value */
3144 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3145 
3146 	rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3147 	if (rc) {
3148 		pr_err("Cannot allocate debug area\n");
3149 		return;
3150 	}
3151 
3152 	rc = ena_com_set_host_attributes(adapter->ena_dev);
3153 	if (rc) {
3154 		if (rc == -EOPNOTSUPP)
3155 			netif_warn(adapter, drv, adapter->netdev,
3156 				   "Cannot set host attributes\n");
3157 		else
3158 			netif_err(adapter, drv, adapter->netdev,
3159 				  "Cannot set host attributes\n");
3160 		goto err;
3161 	}
3162 
3163 	return;
3164 err:
3165 	ena_com_delete_debug_area(adapter->ena_dev);
3166 }
3167 
3168 static void ena_get_stats64(struct net_device *netdev,
3169 			    struct rtnl_link_stats64 *stats)
3170 {
3171 	struct ena_adapter *adapter = netdev_priv(netdev);
3172 	struct ena_ring *rx_ring, *tx_ring;
3173 	unsigned int start;
3174 	u64 rx_drops;
3175 	u64 tx_drops;
3176 	int i;
3177 
3178 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3179 		return;
3180 
3181 	for (i = 0; i < adapter->num_io_queues; i++) {
3182 		u64 bytes, packets;
3183 
3184 		tx_ring = &adapter->tx_ring[i];
3185 
3186 		do {
3187 			start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3188 			packets = tx_ring->tx_stats.cnt;
3189 			bytes = tx_ring->tx_stats.bytes;
3190 		} while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3191 
3192 		stats->tx_packets += packets;
3193 		stats->tx_bytes += bytes;
3194 
3195 		rx_ring = &adapter->rx_ring[i];
3196 
3197 		do {
3198 			start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3199 			packets = rx_ring->rx_stats.cnt;
3200 			bytes = rx_ring->rx_stats.bytes;
3201 		} while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3202 
3203 		stats->rx_packets += packets;
3204 		stats->rx_bytes += bytes;
3205 	}
3206 
3207 	do {
3208 		start = u64_stats_fetch_begin_irq(&adapter->syncp);
3209 		rx_drops = adapter->dev_stats.rx_drops;
3210 		tx_drops = adapter->dev_stats.tx_drops;
3211 	} while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3212 
3213 	stats->rx_dropped = rx_drops;
3214 	stats->tx_dropped = tx_drops;
3215 
3216 	stats->multicast = 0;
3217 	stats->collisions = 0;
3218 
3219 	stats->rx_length_errors = 0;
3220 	stats->rx_crc_errors = 0;
3221 	stats->rx_frame_errors = 0;
3222 	stats->rx_fifo_errors = 0;
3223 	stats->rx_missed_errors = 0;
3224 	stats->tx_window_errors = 0;
3225 
3226 	stats->rx_errors = 0;
3227 	stats->tx_errors = 0;
3228 }
3229 
3230 static const struct net_device_ops ena_netdev_ops = {
3231 	.ndo_open		= ena_open,
3232 	.ndo_stop		= ena_close,
3233 	.ndo_start_xmit		= ena_start_xmit,
3234 	.ndo_select_queue	= ena_select_queue,
3235 	.ndo_get_stats64	= ena_get_stats64,
3236 	.ndo_tx_timeout		= ena_tx_timeout,
3237 	.ndo_change_mtu		= ena_change_mtu,
3238 	.ndo_set_mac_address	= NULL,
3239 	.ndo_validate_addr	= eth_validate_addr,
3240 	.ndo_bpf		= ena_xdp,
3241 };
3242 
3243 static int ena_device_validate_params(struct ena_adapter *adapter,
3244 				      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3245 {
3246 	struct net_device *netdev = adapter->netdev;
3247 	int rc;
3248 
3249 	rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3250 			      adapter->mac_addr);
3251 	if (!rc) {
3252 		netif_err(adapter, drv, netdev,
3253 			  "Error, mac address are different\n");
3254 		return -EINVAL;
3255 	}
3256 
3257 	if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3258 		netif_err(adapter, drv, netdev,
3259 			  "Error, device max mtu is smaller than netdev MTU\n");
3260 		return -EINVAL;
3261 	}
3262 
3263 	return 0;
3264 }
3265 
3266 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3267 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
3268 			   bool *wd_state)
3269 {
3270 	struct device *dev = &pdev->dev;
3271 	bool readless_supported;
3272 	u32 aenq_groups;
3273 	int dma_width;
3274 	int rc;
3275 
3276 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
3277 	if (rc) {
3278 		dev_err(dev, "failed to init mmio read less\n");
3279 		return rc;
3280 	}
3281 
3282 	/* The PCIe configuration space revision id indicate if mmio reg
3283 	 * read is disabled
3284 	 */
3285 	readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3286 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3287 
3288 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3289 	if (rc) {
3290 		dev_err(dev, "Can not reset device\n");
3291 		goto err_mmio_read_less;
3292 	}
3293 
3294 	rc = ena_com_validate_version(ena_dev);
3295 	if (rc) {
3296 		dev_err(dev, "device version is too low\n");
3297 		goto err_mmio_read_less;
3298 	}
3299 
3300 	dma_width = ena_com_get_dma_width(ena_dev);
3301 	if (dma_width < 0) {
3302 		dev_err(dev, "Invalid dma width value %d", dma_width);
3303 		rc = dma_width;
3304 		goto err_mmio_read_less;
3305 	}
3306 
3307 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
3308 	if (rc) {
3309 		dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
3310 		goto err_mmio_read_less;
3311 	}
3312 
3313 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
3314 	if (rc) {
3315 		dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
3316 			rc);
3317 		goto err_mmio_read_less;
3318 	}
3319 
3320 	/* ENA admin level init */
3321 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3322 	if (rc) {
3323 		dev_err(dev,
3324 			"Can not initialize ena admin queue with device\n");
3325 		goto err_mmio_read_less;
3326 	}
3327 
3328 	/* To enable the msix interrupts the driver needs to know the number
3329 	 * of queues. So the driver uses polling mode to retrieve this
3330 	 * information
3331 	 */
3332 	ena_com_set_admin_polling_mode(ena_dev, true);
3333 
3334 	ena_config_host_info(ena_dev, pdev);
3335 
3336 	/* Get Device Attributes*/
3337 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3338 	if (rc) {
3339 		dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3340 		goto err_admin_init;
3341 	}
3342 
3343 	/* Try to turn all the available aenq groups */
3344 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3345 		BIT(ENA_ADMIN_FATAL_ERROR) |
3346 		BIT(ENA_ADMIN_WARNING) |
3347 		BIT(ENA_ADMIN_NOTIFICATION) |
3348 		BIT(ENA_ADMIN_KEEP_ALIVE);
3349 
3350 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
3351 
3352 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3353 	if (rc) {
3354 		dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3355 		goto err_admin_init;
3356 	}
3357 
3358 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3359 
3360 	return 0;
3361 
3362 err_admin_init:
3363 	ena_com_delete_host_info(ena_dev);
3364 	ena_com_admin_destroy(ena_dev);
3365 err_mmio_read_less:
3366 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3367 
3368 	return rc;
3369 }
3370 
3371 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3372 {
3373 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3374 	struct device *dev = &adapter->pdev->dev;
3375 	int rc;
3376 
3377 	rc = ena_enable_msix(adapter);
3378 	if (rc) {
3379 		dev_err(dev, "Can not reserve msix vectors\n");
3380 		return rc;
3381 	}
3382 
3383 	ena_setup_mgmnt_intr(adapter);
3384 
3385 	rc = ena_request_mgmnt_irq(adapter);
3386 	if (rc) {
3387 		dev_err(dev, "Can not setup management interrupts\n");
3388 		goto err_disable_msix;
3389 	}
3390 
3391 	ena_com_set_admin_polling_mode(ena_dev, false);
3392 
3393 	ena_com_admin_aenq_enable(ena_dev);
3394 
3395 	return 0;
3396 
3397 err_disable_msix:
3398 	ena_disable_msix(adapter);
3399 
3400 	return rc;
3401 }
3402 
3403 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3404 {
3405 	struct net_device *netdev = adapter->netdev;
3406 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3407 	bool dev_up;
3408 
3409 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3410 		return;
3411 
3412 	netif_carrier_off(netdev);
3413 
3414 	del_timer_sync(&adapter->timer_service);
3415 
3416 	dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3417 	adapter->dev_up_before_reset = dev_up;
3418 	if (!graceful)
3419 		ena_com_set_admin_running_state(ena_dev, false);
3420 
3421 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3422 		ena_down(adapter);
3423 
3424 	/* Stop the device from sending AENQ events (in case reset flag is set
3425 	 *  and device is up, ena_down() already reset the device.
3426 	 */
3427 	if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3428 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3429 
3430 	ena_free_mgmnt_irq(adapter);
3431 
3432 	ena_disable_msix(adapter);
3433 
3434 	ena_com_abort_admin_commands(ena_dev);
3435 
3436 	ena_com_wait_for_abort_completion(ena_dev);
3437 
3438 	ena_com_admin_destroy(ena_dev);
3439 
3440 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3441 
3442 	/* return reset reason to default value */
3443 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3444 
3445 	clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3446 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3447 }
3448 
3449 static int ena_restore_device(struct ena_adapter *adapter)
3450 {
3451 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3452 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3453 	struct pci_dev *pdev = adapter->pdev;
3454 	bool wd_state;
3455 	int rc;
3456 
3457 	set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3458 	rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3459 	if (rc) {
3460 		dev_err(&pdev->dev, "Can not initialize device\n");
3461 		goto err;
3462 	}
3463 	adapter->wd_state = wd_state;
3464 
3465 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
3466 	if (rc) {
3467 		dev_err(&pdev->dev, "Validation of device parameters failed\n");
3468 		goto err_device_destroy;
3469 	}
3470 
3471 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3472 	if (rc) {
3473 		dev_err(&pdev->dev, "Enable MSI-X failed\n");
3474 		goto err_device_destroy;
3475 	}
3476 	/* If the interface was up before the reset bring it up */
3477 	if (adapter->dev_up_before_reset) {
3478 		rc = ena_up(adapter);
3479 		if (rc) {
3480 			dev_err(&pdev->dev, "Failed to create I/O queues\n");
3481 			goto err_disable_msix;
3482 		}
3483 	}
3484 
3485 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3486 
3487 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3488 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3489 		netif_carrier_on(adapter->netdev);
3490 
3491 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3492 	dev_err(&pdev->dev, "Device reset completed successfully\n");
3493 	adapter->last_keep_alive_jiffies = jiffies;
3494 
3495 	return rc;
3496 err_disable_msix:
3497 	ena_free_mgmnt_irq(adapter);
3498 	ena_disable_msix(adapter);
3499 err_device_destroy:
3500 	ena_com_abort_admin_commands(ena_dev);
3501 	ena_com_wait_for_abort_completion(ena_dev);
3502 	ena_com_admin_destroy(ena_dev);
3503 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3504 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3505 err:
3506 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3507 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3508 	dev_err(&pdev->dev,
3509 		"Reset attempt failed. Can not reset the device\n");
3510 
3511 	return rc;
3512 }
3513 
3514 static void ena_fw_reset_device(struct work_struct *work)
3515 {
3516 	struct ena_adapter *adapter =
3517 		container_of(work, struct ena_adapter, reset_task);
3518 	struct pci_dev *pdev = adapter->pdev;
3519 
3520 	if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3521 		dev_err(&pdev->dev,
3522 			"device reset schedule while reset bit is off\n");
3523 		return;
3524 	}
3525 	rtnl_lock();
3526 	ena_destroy_device(adapter, false);
3527 	ena_restore_device(adapter);
3528 	rtnl_unlock();
3529 }
3530 
3531 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3532 					struct ena_ring *rx_ring)
3533 {
3534 	if (likely(rx_ring->first_interrupt))
3535 		return 0;
3536 
3537 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3538 		return 0;
3539 
3540 	rx_ring->no_interrupt_event_cnt++;
3541 
3542 	if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3543 		netif_err(adapter, rx_err, adapter->netdev,
3544 			  "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3545 			  rx_ring->qid);
3546 		adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3547 		smp_mb__before_atomic();
3548 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3549 		return -EIO;
3550 	}
3551 
3552 	return 0;
3553 }
3554 
3555 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3556 					  struct ena_ring *tx_ring)
3557 {
3558 	struct ena_tx_buffer *tx_buf;
3559 	unsigned long last_jiffies;
3560 	u32 missed_tx = 0;
3561 	int i, rc = 0;
3562 
3563 	for (i = 0; i < tx_ring->ring_size; i++) {
3564 		tx_buf = &tx_ring->tx_buffer_info[i];
3565 		last_jiffies = tx_buf->last_jiffies;
3566 
3567 		if (last_jiffies == 0)
3568 			/* no pending Tx at this location */
3569 			continue;
3570 
3571 		if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
3572 			     2 * adapter->missing_tx_completion_to))) {
3573 			/* If after graceful period interrupt is still not
3574 			 * received, we schedule a reset
3575 			 */
3576 			netif_err(adapter, tx_err, adapter->netdev,
3577 				  "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3578 				  tx_ring->qid);
3579 			adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3580 			smp_mb__before_atomic();
3581 			set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3582 			return -EIO;
3583 		}
3584 
3585 		if (unlikely(time_is_before_jiffies(last_jiffies +
3586 				adapter->missing_tx_completion_to))) {
3587 			if (!tx_buf->print_once)
3588 				netif_notice(adapter, tx_err, adapter->netdev,
3589 					     "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
3590 					     tx_ring->qid, i);
3591 
3592 			tx_buf->print_once = 1;
3593 			missed_tx++;
3594 		}
3595 	}
3596 
3597 	if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3598 		netif_err(adapter, tx_err, adapter->netdev,
3599 			  "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3600 			  missed_tx,
3601 			  adapter->missing_tx_completion_threshold);
3602 		adapter->reset_reason =
3603 			ENA_REGS_RESET_MISS_TX_CMPL;
3604 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3605 		rc = -EIO;
3606 	}
3607 
3608 	u64_stats_update_begin(&tx_ring->syncp);
3609 	tx_ring->tx_stats.missed_tx = missed_tx;
3610 	u64_stats_update_end(&tx_ring->syncp);
3611 
3612 	return rc;
3613 }
3614 
3615 static void check_for_missing_completions(struct ena_adapter *adapter)
3616 {
3617 	struct ena_ring *tx_ring;
3618 	struct ena_ring *rx_ring;
3619 	int i, budget, rc;
3620 	int io_queue_count;
3621 
3622 	io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3623 	/* Make sure the driver doesn't turn the device in other process */
3624 	smp_rmb();
3625 
3626 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3627 		return;
3628 
3629 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3630 		return;
3631 
3632 	if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3633 		return;
3634 
3635 	budget = ENA_MONITORED_TX_QUEUES;
3636 
3637 	for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3638 		tx_ring = &adapter->tx_ring[i];
3639 		rx_ring = &adapter->rx_ring[i];
3640 
3641 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3642 		if (unlikely(rc))
3643 			return;
3644 
3645 		rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3646 			check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3647 		if (unlikely(rc))
3648 			return;
3649 
3650 		budget--;
3651 		if (!budget)
3652 			break;
3653 	}
3654 
3655 	adapter->last_monitored_tx_qid = i % io_queue_count;
3656 }
3657 
3658 /* trigger napi schedule after 2 consecutive detections */
3659 #define EMPTY_RX_REFILL 2
3660 /* For the rare case where the device runs out of Rx descriptors and the
3661  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3662  * for example).
3663  * This case will lead to a deadlock:
3664  * The device won't send interrupts since all the new Rx packets will be dropped
3665  * The napi handler won't allocate new Rx descriptors so the device will be
3666  * able to send new packets.
3667  *
3668  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3669  * It is recommended to have at least 512MB, with a minimum of 128MB for
3670  * constrained environment).
3671  *
3672  * When such a situation is detected - Reschedule napi
3673  */
3674 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3675 {
3676 	struct ena_ring *rx_ring;
3677 	int i, refill_required;
3678 
3679 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3680 		return;
3681 
3682 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3683 		return;
3684 
3685 	for (i = 0; i < adapter->num_io_queues; i++) {
3686 		rx_ring = &adapter->rx_ring[i];
3687 
3688 		refill_required =
3689 			ena_com_free_desc(rx_ring->ena_com_io_sq);
3690 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3691 			rx_ring->empty_rx_queue++;
3692 
3693 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3694 				u64_stats_update_begin(&rx_ring->syncp);
3695 				rx_ring->rx_stats.empty_rx_ring++;
3696 				u64_stats_update_end(&rx_ring->syncp);
3697 
3698 				netif_err(adapter, drv, adapter->netdev,
3699 					  "trigger refill for ring %d\n", i);
3700 
3701 				napi_schedule(rx_ring->napi);
3702 				rx_ring->empty_rx_queue = 0;
3703 			}
3704 		} else {
3705 			rx_ring->empty_rx_queue = 0;
3706 		}
3707 	}
3708 }
3709 
3710 /* Check for keep alive expiration */
3711 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3712 {
3713 	unsigned long keep_alive_expired;
3714 
3715 	if (!adapter->wd_state)
3716 		return;
3717 
3718 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3719 		return;
3720 
3721 	keep_alive_expired = adapter->last_keep_alive_jiffies +
3722 			     adapter->keep_alive_timeout;
3723 	if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3724 		netif_err(adapter, drv, adapter->netdev,
3725 			  "Keep alive watchdog timeout.\n");
3726 		u64_stats_update_begin(&adapter->syncp);
3727 		adapter->dev_stats.wd_expired++;
3728 		u64_stats_update_end(&adapter->syncp);
3729 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3730 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3731 	}
3732 }
3733 
3734 static void check_for_admin_com_state(struct ena_adapter *adapter)
3735 {
3736 	if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3737 		netif_err(adapter, drv, adapter->netdev,
3738 			  "ENA admin queue is not in running state!\n");
3739 		u64_stats_update_begin(&adapter->syncp);
3740 		adapter->dev_stats.admin_q_pause++;
3741 		u64_stats_update_end(&adapter->syncp);
3742 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3743 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3744 	}
3745 }
3746 
3747 static void ena_update_hints(struct ena_adapter *adapter,
3748 			     struct ena_admin_ena_hw_hints *hints)
3749 {
3750 	struct net_device *netdev = adapter->netdev;
3751 
3752 	if (hints->admin_completion_tx_timeout)
3753 		adapter->ena_dev->admin_queue.completion_timeout =
3754 			hints->admin_completion_tx_timeout * 1000;
3755 
3756 	if (hints->mmio_read_timeout)
3757 		/* convert to usec */
3758 		adapter->ena_dev->mmio_read.reg_read_to =
3759 			hints->mmio_read_timeout * 1000;
3760 
3761 	if (hints->missed_tx_completion_count_threshold_to_reset)
3762 		adapter->missing_tx_completion_threshold =
3763 			hints->missed_tx_completion_count_threshold_to_reset;
3764 
3765 	if (hints->missing_tx_completion_timeout) {
3766 		if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3767 			adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3768 		else
3769 			adapter->missing_tx_completion_to =
3770 				msecs_to_jiffies(hints->missing_tx_completion_timeout);
3771 	}
3772 
3773 	if (hints->netdev_wd_timeout)
3774 		netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3775 
3776 	if (hints->driver_watchdog_timeout) {
3777 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3778 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3779 		else
3780 			adapter->keep_alive_timeout =
3781 				msecs_to_jiffies(hints->driver_watchdog_timeout);
3782 	}
3783 }
3784 
3785 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3786 				 struct net_device *netdev)
3787 {
3788 	host_info->supported_network_features[0] =
3789 		netdev->features & GENMASK_ULL(31, 0);
3790 	host_info->supported_network_features[1] =
3791 		(netdev->features & GENMASK_ULL(63, 32)) >> 32;
3792 }
3793 
3794 static void ena_timer_service(struct timer_list *t)
3795 {
3796 	struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3797 	u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3798 	struct ena_admin_host_info *host_info =
3799 		adapter->ena_dev->host_attr.host_info;
3800 
3801 	check_for_missing_keep_alive(adapter);
3802 
3803 	check_for_admin_com_state(adapter);
3804 
3805 	check_for_missing_completions(adapter);
3806 
3807 	check_for_empty_rx_ring(adapter);
3808 
3809 	if (debug_area)
3810 		ena_dump_stats_to_buf(adapter, debug_area);
3811 
3812 	if (host_info)
3813 		ena_update_host_info(host_info, adapter->netdev);
3814 
3815 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3816 		netif_err(adapter, drv, adapter->netdev,
3817 			  "Trigger reset is on\n");
3818 		ena_dump_stats_to_dmesg(adapter);
3819 		queue_work(ena_wq, &adapter->reset_task);
3820 		return;
3821 	}
3822 
3823 	/* Reset the timer */
3824 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3825 }
3826 
3827 static int ena_calc_max_io_queue_num(struct pci_dev *pdev,
3828 				     struct ena_com_dev *ena_dev,
3829 				     struct ena_com_dev_get_features_ctx *get_feat_ctx)
3830 {
3831 	int io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3832 
3833 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3834 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3835 			&get_feat_ctx->max_queue_ext.max_queue_ext;
3836 		io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3837 				  max_queue_ext->max_rx_cq_num);
3838 
3839 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3840 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3841 	} else {
3842 		struct ena_admin_queue_feature_desc *max_queues =
3843 			&get_feat_ctx->max_queues;
3844 		io_tx_sq_num = max_queues->max_sq_num;
3845 		io_tx_cq_num = max_queues->max_cq_num;
3846 		io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
3847 	}
3848 
3849 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
3850 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3851 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3852 
3853 	max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3854 	max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
3855 	max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
3856 	max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
3857 	/* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
3858 	max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
3859 	if (unlikely(!max_num_io_queues)) {
3860 		dev_err(&pdev->dev, "The device doesn't have io queues\n");
3861 		return -EFAULT;
3862 	}
3863 
3864 	return max_num_io_queues;
3865 }
3866 
3867 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3868 					   struct ena_com_dev *ena_dev,
3869 					   struct ena_admin_feature_llq_desc *llq,
3870 					   struct ena_llq_configurations *llq_default_configurations)
3871 {
3872 	bool has_mem_bar;
3873 	int rc;
3874 	u32 llq_feature_mask;
3875 
3876 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3877 	if (!(ena_dev->supported_features & llq_feature_mask)) {
3878 		dev_err(&pdev->dev,
3879 			"LLQ is not supported Fallback to host mode policy.\n");
3880 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3881 		return 0;
3882 	}
3883 
3884 	has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
3885 
3886 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3887 	if (unlikely(rc)) {
3888 		dev_err(&pdev->dev,
3889 			"Failed to configure the device mode.  Fallback to host mode policy.\n");
3890 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3891 		return 0;
3892 	}
3893 
3894 	/* Nothing to config, exit */
3895 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3896 		return 0;
3897 
3898 	if (!has_mem_bar) {
3899 		dev_err(&pdev->dev,
3900 			"ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3901 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3902 		return 0;
3903 	}
3904 
3905 	ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3906 					   pci_resource_start(pdev, ENA_MEM_BAR),
3907 					   pci_resource_len(pdev, ENA_MEM_BAR));
3908 
3909 	if (!ena_dev->mem_bar)
3910 		return -EFAULT;
3911 
3912 	return 0;
3913 }
3914 
3915 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3916 				 struct net_device *netdev)
3917 {
3918 	netdev_features_t dev_features = 0;
3919 
3920 	/* Set offload features */
3921 	if (feat->offload.tx &
3922 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3923 		dev_features |= NETIF_F_IP_CSUM;
3924 
3925 	if (feat->offload.tx &
3926 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
3927 		dev_features |= NETIF_F_IPV6_CSUM;
3928 
3929 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
3930 		dev_features |= NETIF_F_TSO;
3931 
3932 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
3933 		dev_features |= NETIF_F_TSO6;
3934 
3935 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
3936 		dev_features |= NETIF_F_TSO_ECN;
3937 
3938 	if (feat->offload.rx_supported &
3939 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
3940 		dev_features |= NETIF_F_RXCSUM;
3941 
3942 	if (feat->offload.rx_supported &
3943 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
3944 		dev_features |= NETIF_F_RXCSUM;
3945 
3946 	netdev->features =
3947 		dev_features |
3948 		NETIF_F_SG |
3949 		NETIF_F_RXHASH |
3950 		NETIF_F_HIGHDMA;
3951 
3952 	netdev->hw_features |= netdev->features;
3953 	netdev->vlan_features |= netdev->features;
3954 }
3955 
3956 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
3957 				     struct ena_com_dev_get_features_ctx *feat)
3958 {
3959 	struct net_device *netdev = adapter->netdev;
3960 
3961 	/* Copy mac address */
3962 	if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
3963 		eth_hw_addr_random(netdev);
3964 		ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
3965 	} else {
3966 		ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
3967 		ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
3968 	}
3969 
3970 	/* Set offload features */
3971 	ena_set_dev_offloads(feat, netdev);
3972 
3973 	adapter->max_mtu = feat->dev_attr.max_mtu;
3974 	netdev->max_mtu = adapter->max_mtu;
3975 	netdev->min_mtu = ENA_MIN_MTU;
3976 }
3977 
3978 static int ena_rss_init_default(struct ena_adapter *adapter)
3979 {
3980 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3981 	struct device *dev = &adapter->pdev->dev;
3982 	int rc, i;
3983 	u32 val;
3984 
3985 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3986 	if (unlikely(rc)) {
3987 		dev_err(dev, "Cannot init indirect table\n");
3988 		goto err_rss_init;
3989 	}
3990 
3991 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3992 		val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
3993 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3994 						       ENA_IO_RXQ_IDX(val));
3995 		if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3996 			dev_err(dev, "Cannot fill indirect table\n");
3997 			goto err_fill_indir;
3998 		}
3999 	}
4000 
4001 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4002 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4003 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4004 		dev_err(dev, "Cannot fill hash function\n");
4005 		goto err_fill_indir;
4006 	}
4007 
4008 	rc = ena_com_set_default_hash_ctrl(ena_dev);
4009 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4010 		dev_err(dev, "Cannot fill hash control\n");
4011 		goto err_fill_indir;
4012 	}
4013 
4014 	return 0;
4015 
4016 err_fill_indir:
4017 	ena_com_rss_destroy(ena_dev);
4018 err_rss_init:
4019 
4020 	return rc;
4021 }
4022 
4023 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4024 {
4025 	int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4026 
4027 	pci_release_selected_regions(pdev, release_bars);
4028 }
4029 
4030 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
4031 {
4032 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
4033 	llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
4034 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
4035 	llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
4036 	llq_config->llq_ring_entry_size_value = 128;
4037 }
4038 
4039 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4040 {
4041 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4042 	struct ena_com_dev *ena_dev = ctx->ena_dev;
4043 	u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4044 	u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4045 	u32 max_tx_queue_size;
4046 	u32 max_rx_queue_size;
4047 
4048 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4049 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4050 			&ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4051 		max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4052 					  max_queue_ext->max_rx_sq_depth);
4053 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4054 
4055 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4056 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4057 						  llq->max_llq_depth);
4058 		else
4059 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4060 						  max_queue_ext->max_tx_sq_depth);
4061 
4062 		ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4063 					     max_queue_ext->max_per_packet_tx_descs);
4064 		ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4065 					     max_queue_ext->max_per_packet_rx_descs);
4066 	} else {
4067 		struct ena_admin_queue_feature_desc *max_queues =
4068 			&ctx->get_feat_ctx->max_queues;
4069 		max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4070 					  max_queues->max_sq_depth);
4071 		max_tx_queue_size = max_queues->max_cq_depth;
4072 
4073 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4074 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4075 						  llq->max_llq_depth);
4076 		else
4077 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4078 						  max_queues->max_sq_depth);
4079 
4080 		ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4081 					     max_queues->max_packet_tx_descs);
4082 		ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4083 					     max_queues->max_packet_rx_descs);
4084 	}
4085 
4086 	max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4087 	max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4088 
4089 	tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4090 				  max_tx_queue_size);
4091 	rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4092 				  max_rx_queue_size);
4093 
4094 	tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4095 	rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4096 
4097 	ctx->max_tx_queue_size = max_tx_queue_size;
4098 	ctx->max_rx_queue_size = max_rx_queue_size;
4099 	ctx->tx_queue_size = tx_queue_size;
4100 	ctx->rx_queue_size = rx_queue_size;
4101 
4102 	return 0;
4103 }
4104 
4105 /* ena_probe - Device Initialization Routine
4106  * @pdev: PCI device information struct
4107  * @ent: entry in ena_pci_tbl
4108  *
4109  * Returns 0 on success, negative on failure
4110  *
4111  * ena_probe initializes an adapter identified by a pci_dev structure.
4112  * The OS initialization, configuring of the adapter private structure,
4113  * and a hardware reset occur.
4114  */
4115 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4116 {
4117 	struct ena_com_dev_get_features_ctx get_feat_ctx;
4118 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
4119 	struct ena_llq_configurations llq_config;
4120 	struct ena_com_dev *ena_dev = NULL;
4121 	struct ena_adapter *adapter;
4122 	struct net_device *netdev;
4123 	static int adapters_found;
4124 	u32 max_num_io_queues;
4125 	char *queue_type_str;
4126 	bool wd_state;
4127 	int bars, rc;
4128 
4129 	dev_dbg(&pdev->dev, "%s\n", __func__);
4130 
4131 	rc = pci_enable_device_mem(pdev);
4132 	if (rc) {
4133 		dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4134 		return rc;
4135 	}
4136 
4137 	pci_set_master(pdev);
4138 
4139 	ena_dev = vzalloc(sizeof(*ena_dev));
4140 	if (!ena_dev) {
4141 		rc = -ENOMEM;
4142 		goto err_disable_device;
4143 	}
4144 
4145 	bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4146 	rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4147 	if (rc) {
4148 		dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4149 			rc);
4150 		goto err_free_ena_dev;
4151 	}
4152 
4153 	ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4154 					pci_resource_start(pdev, ENA_REG_BAR),
4155 					pci_resource_len(pdev, ENA_REG_BAR));
4156 	if (!ena_dev->reg_bar) {
4157 		dev_err(&pdev->dev, "failed to remap regs bar\n");
4158 		rc = -EFAULT;
4159 		goto err_free_region;
4160 	}
4161 
4162 	ena_dev->dmadev = &pdev->dev;
4163 
4164 	rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4165 	if (rc) {
4166 		dev_err(&pdev->dev, "ena device init failed\n");
4167 		if (rc == -ETIME)
4168 			rc = -EPROBE_DEFER;
4169 		goto err_free_region;
4170 	}
4171 
4172 	set_default_llq_configurations(&llq_config);
4173 
4174 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
4175 					     &llq_config);
4176 	if (rc) {
4177 		dev_err(&pdev->dev, "ena device init failed\n");
4178 		goto err_device_destroy;
4179 	}
4180 
4181 	calc_queue_ctx.ena_dev = ena_dev;
4182 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4183 	calc_queue_ctx.pdev = pdev;
4184 
4185 	/* Initial Tx and RX interrupt delay. Assumes 1 usec granularity.
4186 	 * Updated during device initialization with the real granularity
4187 	 */
4188 	ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4189 	ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4190 	ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4191 	max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4192 	rc = ena_calc_io_queue_size(&calc_queue_ctx);
4193 	if (rc || !max_num_io_queues) {
4194 		rc = -EFAULT;
4195 		goto err_device_destroy;
4196 	}
4197 
4198 	/* dev zeroed in init_etherdev */
4199 	netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), max_num_io_queues);
4200 	if (!netdev) {
4201 		dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4202 		rc = -ENOMEM;
4203 		goto err_device_destroy;
4204 	}
4205 
4206 	SET_NETDEV_DEV(netdev, &pdev->dev);
4207 
4208 	adapter = netdev_priv(netdev);
4209 	pci_set_drvdata(pdev, adapter);
4210 
4211 	adapter->ena_dev = ena_dev;
4212 	adapter->netdev = netdev;
4213 	adapter->pdev = pdev;
4214 
4215 	ena_set_conf_feat_params(adapter, &get_feat_ctx);
4216 
4217 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4218 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4219 
4220 	adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4221 	adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4222 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4223 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4224 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4225 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4226 
4227 	adapter->num_io_queues = max_num_io_queues;
4228 	adapter->max_num_io_queues = max_num_io_queues;
4229 
4230 	adapter->xdp_first_ring = 0;
4231 	adapter->xdp_num_queues = 0;
4232 
4233 	adapter->last_monitored_tx_qid = 0;
4234 
4235 	adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4236 	adapter->wd_state = wd_state;
4237 
4238 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4239 
4240 	rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4241 	if (rc) {
4242 		dev_err(&pdev->dev,
4243 			"Failed to query interrupt moderation feature\n");
4244 		goto err_netdev_destroy;
4245 	}
4246 	ena_init_io_rings(adapter,
4247 			  0,
4248 			  adapter->xdp_num_queues +
4249 			  adapter->num_io_queues);
4250 
4251 	netdev->netdev_ops = &ena_netdev_ops;
4252 	netdev->watchdog_timeo = TX_TIMEOUT;
4253 	ena_set_ethtool_ops(netdev);
4254 
4255 	netdev->priv_flags |= IFF_UNICAST_FLT;
4256 
4257 	u64_stats_init(&adapter->syncp);
4258 
4259 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4260 	if (rc) {
4261 		dev_err(&pdev->dev,
4262 			"Failed to enable and set the admin interrupts\n");
4263 		goto err_worker_destroy;
4264 	}
4265 	rc = ena_rss_init_default(adapter);
4266 	if (rc && (rc != -EOPNOTSUPP)) {
4267 		dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4268 		goto err_free_msix;
4269 	}
4270 
4271 	ena_config_debug_area(adapter);
4272 
4273 	memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4274 
4275 	netif_carrier_off(netdev);
4276 
4277 	rc = register_netdev(netdev);
4278 	if (rc) {
4279 		dev_err(&pdev->dev, "Cannot register net device\n");
4280 		goto err_rss;
4281 	}
4282 
4283 	INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4284 
4285 	adapter->last_keep_alive_jiffies = jiffies;
4286 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4287 	adapter->missing_tx_completion_to = TX_TIMEOUT;
4288 	adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4289 
4290 	ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4291 
4292 	timer_setup(&adapter->timer_service, ena_timer_service, 0);
4293 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4294 
4295 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
4296 		queue_type_str = "Regular";
4297 	else
4298 		queue_type_str = "Low Latency";
4299 
4300 	dev_info(&pdev->dev,
4301 		 "%s found at mem %lx, mac addr %pM, Placement policy: %s\n",
4302 		 DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4303 		 netdev->dev_addr, queue_type_str);
4304 
4305 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4306 
4307 	adapters_found++;
4308 
4309 	return 0;
4310 
4311 err_rss:
4312 	ena_com_delete_debug_area(ena_dev);
4313 	ena_com_rss_destroy(ena_dev);
4314 err_free_msix:
4315 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4316 	/* stop submitting admin commands on a device that was reset */
4317 	ena_com_set_admin_running_state(ena_dev, false);
4318 	ena_free_mgmnt_irq(adapter);
4319 	ena_disable_msix(adapter);
4320 err_worker_destroy:
4321 	del_timer(&adapter->timer_service);
4322 err_netdev_destroy:
4323 	free_netdev(netdev);
4324 err_device_destroy:
4325 	ena_com_delete_host_info(ena_dev);
4326 	ena_com_admin_destroy(ena_dev);
4327 err_free_region:
4328 	ena_release_bars(ena_dev, pdev);
4329 err_free_ena_dev:
4330 	vfree(ena_dev);
4331 err_disable_device:
4332 	pci_disable_device(pdev);
4333 	return rc;
4334 }
4335 
4336 /*****************************************************************************/
4337 
4338 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4339  * @pdev: PCI device information struct
4340  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4341  *
4342  * __ena_shutoff is a helper routine that does the real work on shutdown and
4343  * removal paths; the difference between those paths is with regards to whether
4344  * dettach or unregister the netdevice.
4345  */
4346 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4347 {
4348 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4349 	struct ena_com_dev *ena_dev;
4350 	struct net_device *netdev;
4351 
4352 	ena_dev = adapter->ena_dev;
4353 	netdev = adapter->netdev;
4354 
4355 #ifdef CONFIG_RFS_ACCEL
4356 	if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4357 		free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4358 		netdev->rx_cpu_rmap = NULL;
4359 	}
4360 #endif /* CONFIG_RFS_ACCEL */
4361 	del_timer_sync(&adapter->timer_service);
4362 
4363 	cancel_work_sync(&adapter->reset_task);
4364 
4365 	rtnl_lock(); /* lock released inside the below if-else block */
4366 	adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4367 	ena_destroy_device(adapter, true);
4368 	if (shutdown) {
4369 		netif_device_detach(netdev);
4370 		dev_close(netdev);
4371 		rtnl_unlock();
4372 	} else {
4373 		rtnl_unlock();
4374 		unregister_netdev(netdev);
4375 		free_netdev(netdev);
4376 	}
4377 
4378 	ena_com_rss_destroy(ena_dev);
4379 
4380 	ena_com_delete_debug_area(ena_dev);
4381 
4382 	ena_com_delete_host_info(ena_dev);
4383 
4384 	ena_release_bars(ena_dev, pdev);
4385 
4386 	pci_disable_device(pdev);
4387 
4388 	vfree(ena_dev);
4389 }
4390 
4391 /* ena_remove - Device Removal Routine
4392  * @pdev: PCI device information struct
4393  *
4394  * ena_remove is called by the PCI subsystem to alert the driver
4395  * that it should release a PCI device.
4396  */
4397 
4398 static void ena_remove(struct pci_dev *pdev)
4399 {
4400 	__ena_shutoff(pdev, false);
4401 }
4402 
4403 /* ena_shutdown - Device Shutdown Routine
4404  * @pdev: PCI device information struct
4405  *
4406  * ena_shutdown is called by the PCI subsystem to alert the driver that
4407  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4408  */
4409 
4410 static void ena_shutdown(struct pci_dev *pdev)
4411 {
4412 	__ena_shutoff(pdev, true);
4413 }
4414 
4415 #ifdef CONFIG_PM
4416 /* ena_suspend - PM suspend callback
4417  * @pdev: PCI device information struct
4418  * @state:power state
4419  */
4420 static int ena_suspend(struct pci_dev *pdev,  pm_message_t state)
4421 {
4422 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4423 
4424 	u64_stats_update_begin(&adapter->syncp);
4425 	adapter->dev_stats.suspend++;
4426 	u64_stats_update_end(&adapter->syncp);
4427 
4428 	rtnl_lock();
4429 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4430 		dev_err(&pdev->dev,
4431 			"ignoring device reset request as the device is being suspended\n");
4432 		clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4433 	}
4434 	ena_destroy_device(adapter, true);
4435 	rtnl_unlock();
4436 	return 0;
4437 }
4438 
4439 /* ena_resume - PM resume callback
4440  * @pdev: PCI device information struct
4441  *
4442  */
4443 static int ena_resume(struct pci_dev *pdev)
4444 {
4445 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4446 	int rc;
4447 
4448 	u64_stats_update_begin(&adapter->syncp);
4449 	adapter->dev_stats.resume++;
4450 	u64_stats_update_end(&adapter->syncp);
4451 
4452 	rtnl_lock();
4453 	rc = ena_restore_device(adapter);
4454 	rtnl_unlock();
4455 	return rc;
4456 }
4457 #endif
4458 
4459 static struct pci_driver ena_pci_driver = {
4460 	.name		= DRV_MODULE_NAME,
4461 	.id_table	= ena_pci_tbl,
4462 	.probe		= ena_probe,
4463 	.remove		= ena_remove,
4464 	.shutdown	= ena_shutdown,
4465 #ifdef CONFIG_PM
4466 	.suspend    = ena_suspend,
4467 	.resume     = ena_resume,
4468 #endif
4469 	.sriov_configure = pci_sriov_configure_simple,
4470 };
4471 
4472 static int __init ena_init(void)
4473 {
4474 	ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4475 	if (!ena_wq) {
4476 		pr_err("Failed to create workqueue\n");
4477 		return -ENOMEM;
4478 	}
4479 
4480 	return pci_register_driver(&ena_pci_driver);
4481 }
4482 
4483 static void __exit ena_cleanup(void)
4484 {
4485 	pci_unregister_driver(&ena_pci_driver);
4486 
4487 	if (ena_wq) {
4488 		destroy_workqueue(ena_wq);
4489 		ena_wq = NULL;
4490 	}
4491 }
4492 
4493 /******************************************************************************
4494  ******************************** AENQ Handlers *******************************
4495  *****************************************************************************/
4496 /* ena_update_on_link_change:
4497  * Notify the network interface about the change in link status
4498  */
4499 static void ena_update_on_link_change(void *adapter_data,
4500 				      struct ena_admin_aenq_entry *aenq_e)
4501 {
4502 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4503 	struct ena_admin_aenq_link_change_desc *aenq_desc =
4504 		(struct ena_admin_aenq_link_change_desc *)aenq_e;
4505 	int status = aenq_desc->flags &
4506 		ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4507 
4508 	if (status) {
4509 		netdev_dbg(adapter->netdev, "%s\n", __func__);
4510 		set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4511 		if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4512 			netif_carrier_on(adapter->netdev);
4513 	} else {
4514 		clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4515 		netif_carrier_off(adapter->netdev);
4516 	}
4517 }
4518 
4519 static void ena_keep_alive_wd(void *adapter_data,
4520 			      struct ena_admin_aenq_entry *aenq_e)
4521 {
4522 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4523 	struct ena_admin_aenq_keep_alive_desc *desc;
4524 	u64 rx_drops;
4525 	u64 tx_drops;
4526 
4527 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4528 	adapter->last_keep_alive_jiffies = jiffies;
4529 
4530 	rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4531 	tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4532 
4533 	u64_stats_update_begin(&adapter->syncp);
4534 	adapter->dev_stats.rx_drops = rx_drops;
4535 	adapter->dev_stats.tx_drops = tx_drops;
4536 	u64_stats_update_end(&adapter->syncp);
4537 }
4538 
4539 static void ena_notification(void *adapter_data,
4540 			     struct ena_admin_aenq_entry *aenq_e)
4541 {
4542 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4543 	struct ena_admin_ena_hw_hints *hints;
4544 
4545 	WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4546 	     "Invalid group(%x) expected %x\n",
4547 	     aenq_e->aenq_common_desc.group,
4548 	     ENA_ADMIN_NOTIFICATION);
4549 
4550 	switch (aenq_e->aenq_common_desc.syndrom) {
4551 	case ENA_ADMIN_UPDATE_HINTS:
4552 		hints = (struct ena_admin_ena_hw_hints *)
4553 			(&aenq_e->inline_data_w4);
4554 		ena_update_hints(adapter, hints);
4555 		break;
4556 	default:
4557 		netif_err(adapter, drv, adapter->netdev,
4558 			  "Invalid aenq notification link state %d\n",
4559 			  aenq_e->aenq_common_desc.syndrom);
4560 	}
4561 }
4562 
4563 /* This handler will called for unknown event group or unimplemented handlers*/
4564 static void unimplemented_aenq_handler(void *data,
4565 				       struct ena_admin_aenq_entry *aenq_e)
4566 {
4567 	struct ena_adapter *adapter = (struct ena_adapter *)data;
4568 
4569 	netif_err(adapter, drv, adapter->netdev,
4570 		  "Unknown event was received or event with unimplemented handler\n");
4571 }
4572 
4573 static struct ena_aenq_handlers aenq_handlers = {
4574 	.handlers = {
4575 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4576 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
4577 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4578 	},
4579 	.unimplemented_handler = unimplemented_aenq_handler
4580 };
4581 
4582 module_init(ena_init);
4583 module_exit(ena_cleanup);
4584