xref: /linux/drivers/net/ethernet/amazon/ena/ena_netdev.c (revision 05ee19c18c2bb3dea69e29219017367c4a77e65a)
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 	xdp.frame_sz = ENA_PAGE_SIZE;
1610 
1611 	do {
1612 		xdp_verdict = XDP_PASS;
1613 		skb = NULL;
1614 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1615 		ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1616 		ena_rx_ctx.descs = 0;
1617 		rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1618 				    rx_ring->ena_com_io_sq,
1619 				    &ena_rx_ctx);
1620 		if (unlikely(rc))
1621 			goto error;
1622 
1623 		if (unlikely(ena_rx_ctx.descs == 0))
1624 			break;
1625 
1626 		netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1627 			  "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1628 			  rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1629 			  ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1630 
1631 		if (ena_xdp_present_ring(rx_ring))
1632 			xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1633 
1634 		/* allocate skb and fill it */
1635 		if (xdp_verdict == XDP_PASS)
1636 			skb = ena_rx_skb(rx_ring,
1637 					 rx_ring->ena_bufs,
1638 					 ena_rx_ctx.descs,
1639 					 &next_to_clean);
1640 
1641 		if (unlikely(!skb)) {
1642 			if (xdp_verdict == XDP_TX) {
1643 				ena_free_rx_page(rx_ring,
1644 						 &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id]);
1645 				res_budget--;
1646 			}
1647 			for (i = 0; i < ena_rx_ctx.descs; i++) {
1648 				rx_ring->free_ids[next_to_clean] =
1649 					rx_ring->ena_bufs[i].req_id;
1650 				next_to_clean =
1651 					ENA_RX_RING_IDX_NEXT(next_to_clean,
1652 							     rx_ring->ring_size);
1653 			}
1654 			if (xdp_verdict == XDP_TX || xdp_verdict == XDP_DROP)
1655 				continue;
1656 			break;
1657 		}
1658 
1659 		ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1660 
1661 		ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1662 
1663 		skb_record_rx_queue(skb, rx_ring->qid);
1664 
1665 		if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1666 			total_len += rx_ring->ena_bufs[0].len;
1667 			rx_copybreak_pkt++;
1668 			napi_gro_receive(napi, skb);
1669 		} else {
1670 			total_len += skb->len;
1671 			napi_gro_frags(napi);
1672 		}
1673 
1674 		res_budget--;
1675 	} while (likely(res_budget));
1676 
1677 	work_done = budget - res_budget;
1678 	rx_ring->per_napi_packets += work_done;
1679 	u64_stats_update_begin(&rx_ring->syncp);
1680 	rx_ring->rx_stats.bytes += total_len;
1681 	rx_ring->rx_stats.cnt += work_done;
1682 	rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1683 	u64_stats_update_end(&rx_ring->syncp);
1684 
1685 	rx_ring->next_to_clean = next_to_clean;
1686 
1687 	refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
1688 	refill_threshold =
1689 		min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1690 		      ENA_RX_REFILL_THRESH_PACKET);
1691 
1692 	/* Optimization, try to batch new rx buffers */
1693 	if (refill_required > refill_threshold) {
1694 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1695 		ena_refill_rx_bufs(rx_ring, refill_required);
1696 	}
1697 
1698 	return work_done;
1699 
1700 error:
1701 	adapter = netdev_priv(rx_ring->netdev);
1702 
1703 	u64_stats_update_begin(&rx_ring->syncp);
1704 	rx_ring->rx_stats.bad_desc_num++;
1705 	u64_stats_update_end(&rx_ring->syncp);
1706 
1707 	/* Too many desc from the device. Trigger reset */
1708 	adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1709 	set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1710 
1711 	return 0;
1712 }
1713 
1714 static void ena_dim_work(struct work_struct *w)
1715 {
1716 	struct dim *dim = container_of(w, struct dim, work);
1717 	struct dim_cq_moder cur_moder =
1718 		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1719 	struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1720 
1721 	ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1722 	dim->state = DIM_START_MEASURE;
1723 }
1724 
1725 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1726 {
1727 	struct dim_sample dim_sample;
1728 	struct ena_ring *rx_ring = ena_napi->rx_ring;
1729 
1730 	if (!rx_ring->per_napi_packets)
1731 		return;
1732 
1733 	rx_ring->non_empty_napi_events++;
1734 
1735 	dim_update_sample(rx_ring->non_empty_napi_events,
1736 			  rx_ring->rx_stats.cnt,
1737 			  rx_ring->rx_stats.bytes,
1738 			  &dim_sample);
1739 
1740 	net_dim(&ena_napi->dim, dim_sample);
1741 
1742 	rx_ring->per_napi_packets = 0;
1743 }
1744 
1745 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1746 					struct ena_ring *rx_ring)
1747 {
1748 	struct ena_eth_io_intr_reg intr_reg;
1749 	u32 rx_interval = 0;
1750 	/* Rx ring can be NULL when for XDP tx queues which don't have an
1751 	 * accompanying rx_ring pair.
1752 	 */
1753 	if (rx_ring)
1754 		rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1755 			rx_ring->smoothed_interval :
1756 			ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1757 
1758 	/* Update intr register: rx intr delay,
1759 	 * tx intr delay and interrupt unmask
1760 	 */
1761 	ena_com_update_intr_reg(&intr_reg,
1762 				rx_interval,
1763 				tx_ring->smoothed_interval,
1764 				true);
1765 
1766 	u64_stats_update_begin(&tx_ring->syncp);
1767 	tx_ring->tx_stats.unmask_interrupt++;
1768 	u64_stats_update_end(&tx_ring->syncp);
1769 	/* It is a shared MSI-X.
1770 	 * Tx and Rx CQ have pointer to it.
1771 	 * So we use one of them to reach the intr reg
1772 	 * The Tx ring is used because the rx_ring is NULL for XDP queues
1773 	 */
1774 	ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1775 }
1776 
1777 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1778 					     struct ena_ring *rx_ring)
1779 {
1780 	int cpu = get_cpu();
1781 	int numa_node;
1782 
1783 	/* Check only one ring since the 2 rings are running on the same cpu */
1784 	if (likely(tx_ring->cpu == cpu))
1785 		goto out;
1786 
1787 	numa_node = cpu_to_node(cpu);
1788 	put_cpu();
1789 
1790 	if (numa_node != NUMA_NO_NODE) {
1791 		ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1792 		if (rx_ring)
1793 			ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1794 						 numa_node);
1795 	}
1796 
1797 	tx_ring->cpu = cpu;
1798 	if (rx_ring)
1799 		rx_ring->cpu = cpu;
1800 
1801 	return;
1802 out:
1803 	put_cpu();
1804 }
1805 
1806 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1807 {
1808 	u32 total_done = 0;
1809 	u16 next_to_clean;
1810 	u32 tx_bytes = 0;
1811 	int tx_pkts = 0;
1812 	u16 req_id;
1813 	int rc;
1814 
1815 	if (unlikely(!xdp_ring))
1816 		return 0;
1817 	next_to_clean = xdp_ring->next_to_clean;
1818 
1819 	while (tx_pkts < budget) {
1820 		struct ena_tx_buffer *tx_info;
1821 		struct xdp_frame *xdpf;
1822 
1823 		rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1824 						&req_id);
1825 		if (rc)
1826 			break;
1827 
1828 		rc = validate_xdp_req_id(xdp_ring, req_id);
1829 		if (rc)
1830 			break;
1831 
1832 		tx_info = &xdp_ring->tx_buffer_info[req_id];
1833 		xdpf = tx_info->xdpf;
1834 
1835 		tx_info->xdpf = NULL;
1836 		tx_info->last_jiffies = 0;
1837 		ena_unmap_tx_buff(xdp_ring, tx_info);
1838 
1839 		netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1840 			  "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1841 			  xdpf);
1842 
1843 		tx_bytes += xdpf->len;
1844 		tx_pkts++;
1845 		total_done += tx_info->tx_descs;
1846 
1847 		__free_page(tx_info->xdp_rx_page);
1848 		xdp_ring->free_ids[next_to_clean] = req_id;
1849 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1850 						     xdp_ring->ring_size);
1851 	}
1852 
1853 	xdp_ring->next_to_clean = next_to_clean;
1854 	ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1855 	ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1856 
1857 	netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1858 		  "tx_poll: q %d done. total pkts: %d\n",
1859 		  xdp_ring->qid, tx_pkts);
1860 
1861 	return tx_pkts;
1862 }
1863 
1864 static int ena_io_poll(struct napi_struct *napi, int budget)
1865 {
1866 	struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1867 	struct ena_ring *tx_ring, *rx_ring;
1868 	int tx_work_done;
1869 	int rx_work_done = 0;
1870 	int tx_budget;
1871 	int napi_comp_call = 0;
1872 	int ret;
1873 
1874 	tx_ring = ena_napi->tx_ring;
1875 	rx_ring = ena_napi->rx_ring;
1876 
1877 	tx_ring->first_interrupt = ena_napi->first_interrupt;
1878 	rx_ring->first_interrupt = ena_napi->first_interrupt;
1879 
1880 	tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1881 
1882 	if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1883 	    test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1884 		napi_complete_done(napi, 0);
1885 		return 0;
1886 	}
1887 
1888 	tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1889 	/* On netpoll the budget is zero and the handler should only clean the
1890 	 * tx completions.
1891 	 */
1892 	if (likely(budget))
1893 		rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1894 
1895 	/* If the device is about to reset or down, avoid unmask
1896 	 * the interrupt and return 0 so NAPI won't reschedule
1897 	 */
1898 	if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1899 		     test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1900 		napi_complete_done(napi, 0);
1901 		ret = 0;
1902 
1903 	} else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1904 		napi_comp_call = 1;
1905 
1906 		/* Update numa and unmask the interrupt only when schedule
1907 		 * from the interrupt context (vs from sk_busy_loop)
1908 		 */
1909 		if (napi_complete_done(napi, rx_work_done)) {
1910 			/* We apply adaptive moderation on Rx path only.
1911 			 * Tx uses static interrupt moderation.
1912 			 */
1913 			if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1914 				ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1915 
1916 			ena_unmask_interrupt(tx_ring, rx_ring);
1917 		}
1918 
1919 		ena_update_ring_numa_node(tx_ring, rx_ring);
1920 
1921 		ret = rx_work_done;
1922 	} else {
1923 		ret = budget;
1924 	}
1925 
1926 	u64_stats_update_begin(&tx_ring->syncp);
1927 	tx_ring->tx_stats.napi_comp += napi_comp_call;
1928 	tx_ring->tx_stats.tx_poll++;
1929 	u64_stats_update_end(&tx_ring->syncp);
1930 
1931 	return ret;
1932 }
1933 
1934 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1935 {
1936 	struct ena_adapter *adapter = (struct ena_adapter *)data;
1937 
1938 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1939 
1940 	/* Don't call the aenq handler before probe is done */
1941 	if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1942 		ena_com_aenq_intr_handler(adapter->ena_dev, data);
1943 
1944 	return IRQ_HANDLED;
1945 }
1946 
1947 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1948  * @irq: interrupt number
1949  * @data: pointer to a network interface private napi device structure
1950  */
1951 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1952 {
1953 	struct ena_napi *ena_napi = data;
1954 
1955 	ena_napi->first_interrupt = true;
1956 
1957 	napi_schedule_irqoff(&ena_napi->napi);
1958 
1959 	return IRQ_HANDLED;
1960 }
1961 
1962 /* Reserve a single MSI-X vector for management (admin + aenq).
1963  * plus reserve one vector for each potential io queue.
1964  * the number of potential io queues is the minimum of what the device
1965  * supports and the number of vCPUs.
1966  */
1967 static int ena_enable_msix(struct ena_adapter *adapter)
1968 {
1969 	int msix_vecs, irq_cnt;
1970 
1971 	if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1972 		netif_err(adapter, probe, adapter->netdev,
1973 			  "Error, MSI-X is already enabled\n");
1974 		return -EPERM;
1975 	}
1976 
1977 	/* Reserved the max msix vectors we might need */
1978 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
1979 	netif_dbg(adapter, probe, adapter->netdev,
1980 		  "trying to enable MSI-X, vectors %d\n", msix_vecs);
1981 
1982 	irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
1983 					msix_vecs, PCI_IRQ_MSIX);
1984 
1985 	if (irq_cnt < 0) {
1986 		netif_err(adapter, probe, adapter->netdev,
1987 			  "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
1988 		return -ENOSPC;
1989 	}
1990 
1991 	if (irq_cnt != msix_vecs) {
1992 		netif_notice(adapter, probe, adapter->netdev,
1993 			     "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1994 			     irq_cnt, msix_vecs);
1995 		adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
1996 	}
1997 
1998 	if (ena_init_rx_cpu_rmap(adapter))
1999 		netif_warn(adapter, probe, adapter->netdev,
2000 			   "Failed to map IRQs to CPUs\n");
2001 
2002 	adapter->msix_vecs = irq_cnt;
2003 	set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2004 
2005 	return 0;
2006 }
2007 
2008 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2009 {
2010 	u32 cpu;
2011 
2012 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2013 		 ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2014 		 pci_name(adapter->pdev));
2015 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2016 		ena_intr_msix_mgmnt;
2017 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2018 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2019 		pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2020 	cpu = cpumask_first(cpu_online_mask);
2021 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2022 	cpumask_set_cpu(cpu,
2023 			&adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2024 }
2025 
2026 static void ena_setup_io_intr(struct ena_adapter *adapter)
2027 {
2028 	struct net_device *netdev;
2029 	int irq_idx, i, cpu;
2030 	int io_queue_count;
2031 
2032 	netdev = adapter->netdev;
2033 	io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2034 
2035 	for (i = 0; i < io_queue_count; i++) {
2036 		irq_idx = ENA_IO_IRQ_IDX(i);
2037 		cpu = i % num_online_cpus();
2038 
2039 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2040 			 "%s-Tx-Rx-%d", netdev->name, i);
2041 		adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2042 		adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2043 		adapter->irq_tbl[irq_idx].vector =
2044 			pci_irq_vector(adapter->pdev, irq_idx);
2045 		adapter->irq_tbl[irq_idx].cpu = cpu;
2046 
2047 		cpumask_set_cpu(cpu,
2048 				&adapter->irq_tbl[irq_idx].affinity_hint_mask);
2049 	}
2050 }
2051 
2052 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2053 {
2054 	unsigned long flags = 0;
2055 	struct ena_irq *irq;
2056 	int rc;
2057 
2058 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2059 	rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2060 			 irq->data);
2061 	if (rc) {
2062 		netif_err(adapter, probe, adapter->netdev,
2063 			  "failed to request admin irq\n");
2064 		return rc;
2065 	}
2066 
2067 	netif_dbg(adapter, probe, adapter->netdev,
2068 		  "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2069 		  irq->affinity_hint_mask.bits[0], irq->vector);
2070 
2071 	irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2072 
2073 	return rc;
2074 }
2075 
2076 static int ena_request_io_irq(struct ena_adapter *adapter)
2077 {
2078 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2079 	unsigned long flags = 0;
2080 	struct ena_irq *irq;
2081 	int rc = 0, i, k;
2082 
2083 	if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2084 		netif_err(adapter, ifup, adapter->netdev,
2085 			  "Failed to request I/O IRQ: MSI-X is not enabled\n");
2086 		return -EINVAL;
2087 	}
2088 
2089 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2090 		irq = &adapter->irq_tbl[i];
2091 		rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2092 				 irq->data);
2093 		if (rc) {
2094 			netif_err(adapter, ifup, adapter->netdev,
2095 				  "Failed to request I/O IRQ. index %d rc %d\n",
2096 				   i, rc);
2097 			goto err;
2098 		}
2099 
2100 		netif_dbg(adapter, ifup, adapter->netdev,
2101 			  "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2102 			  i, irq->affinity_hint_mask.bits[0], irq->vector);
2103 
2104 		irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2105 	}
2106 
2107 	return rc;
2108 
2109 err:
2110 	for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2111 		irq = &adapter->irq_tbl[k];
2112 		free_irq(irq->vector, irq->data);
2113 	}
2114 
2115 	return rc;
2116 }
2117 
2118 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2119 {
2120 	struct ena_irq *irq;
2121 
2122 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2123 	synchronize_irq(irq->vector);
2124 	irq_set_affinity_hint(irq->vector, NULL);
2125 	free_irq(irq->vector, irq->data);
2126 }
2127 
2128 static void ena_free_io_irq(struct ena_adapter *adapter)
2129 {
2130 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2131 	struct ena_irq *irq;
2132 	int i;
2133 
2134 #ifdef CONFIG_RFS_ACCEL
2135 	if (adapter->msix_vecs >= 1) {
2136 		free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2137 		adapter->netdev->rx_cpu_rmap = NULL;
2138 	}
2139 #endif /* CONFIG_RFS_ACCEL */
2140 
2141 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2142 		irq = &adapter->irq_tbl[i];
2143 		irq_set_affinity_hint(irq->vector, NULL);
2144 		free_irq(irq->vector, irq->data);
2145 	}
2146 }
2147 
2148 static void ena_disable_msix(struct ena_adapter *adapter)
2149 {
2150 	if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2151 		pci_free_irq_vectors(adapter->pdev);
2152 }
2153 
2154 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2155 {
2156 	u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2157 	int i;
2158 
2159 	if (!netif_running(adapter->netdev))
2160 		return;
2161 
2162 	for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2163 		synchronize_irq(adapter->irq_tbl[i].vector);
2164 }
2165 
2166 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2167 				  int first_index,
2168 				  int count)
2169 {
2170 	int i;
2171 
2172 	for (i = first_index; i < first_index + count; i++) {
2173 		/* Check if napi was initialized before */
2174 		if (!ENA_IS_XDP_INDEX(adapter, i) ||
2175 		    adapter->ena_napi[i].xdp_ring)
2176 			netif_napi_del(&adapter->ena_napi[i].napi);
2177 		else
2178 			WARN_ON(ENA_IS_XDP_INDEX(adapter, i) &&
2179 				adapter->ena_napi[i].xdp_ring);
2180 	}
2181 }
2182 
2183 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2184 				   int first_index, int count)
2185 {
2186 	struct ena_napi *napi = {0};
2187 	int i;
2188 
2189 	for (i = first_index; i < first_index + count; i++) {
2190 		napi = &adapter->ena_napi[i];
2191 
2192 		netif_napi_add(adapter->netdev,
2193 			       &adapter->ena_napi[i].napi,
2194 			       ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2195 			       ENA_NAPI_BUDGET);
2196 
2197 		if (!ENA_IS_XDP_INDEX(adapter, i)) {
2198 			napi->rx_ring = &adapter->rx_ring[i];
2199 			napi->tx_ring = &adapter->tx_ring[i];
2200 		} else {
2201 			napi->xdp_ring = &adapter->tx_ring[i];
2202 		}
2203 		napi->qid = i;
2204 	}
2205 }
2206 
2207 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2208 				      int first_index,
2209 				      int count)
2210 {
2211 	int i;
2212 
2213 	for (i = first_index; i < first_index + count; i++)
2214 		napi_disable(&adapter->ena_napi[i].napi);
2215 }
2216 
2217 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2218 				     int first_index,
2219 				     int count)
2220 {
2221 	int i;
2222 
2223 	for (i = first_index; i < first_index + count; i++)
2224 		napi_enable(&adapter->ena_napi[i].napi);
2225 }
2226 
2227 /* Configure the Rx forwarding */
2228 static int ena_rss_configure(struct ena_adapter *adapter)
2229 {
2230 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2231 	int rc;
2232 
2233 	/* In case the RSS table wasn't initialized by probe */
2234 	if (!ena_dev->rss.tbl_log_size) {
2235 		rc = ena_rss_init_default(adapter);
2236 		if (rc && (rc != -EOPNOTSUPP)) {
2237 			netif_err(adapter, ifup, adapter->netdev,
2238 					"Failed to init RSS rc: %d\n", rc);
2239 			return rc;
2240 		}
2241 	}
2242 
2243 	/* Set indirect table */
2244 	rc = ena_com_indirect_table_set(ena_dev);
2245 	if (unlikely(rc && rc != -EOPNOTSUPP))
2246 		return rc;
2247 
2248 	/* Configure hash function (if supported) */
2249 	rc = ena_com_set_hash_function(ena_dev);
2250 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
2251 		return rc;
2252 
2253 	/* Configure hash inputs (if supported) */
2254 	rc = ena_com_set_hash_ctrl(ena_dev);
2255 	if (unlikely(rc && (rc != -EOPNOTSUPP)))
2256 		return rc;
2257 
2258 	return 0;
2259 }
2260 
2261 static int ena_up_complete(struct ena_adapter *adapter)
2262 {
2263 	int rc;
2264 
2265 	rc = ena_rss_configure(adapter);
2266 	if (rc)
2267 		return rc;
2268 
2269 	ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2270 
2271 	ena_refill_all_rx_bufs(adapter);
2272 
2273 	/* enable transmits */
2274 	netif_tx_start_all_queues(adapter->netdev);
2275 
2276 	ena_napi_enable_in_range(adapter,
2277 				 0,
2278 				 adapter->xdp_num_queues + adapter->num_io_queues);
2279 
2280 	return 0;
2281 }
2282 
2283 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2284 {
2285 	struct ena_com_create_io_ctx ctx;
2286 	struct ena_com_dev *ena_dev;
2287 	struct ena_ring *tx_ring;
2288 	u32 msix_vector;
2289 	u16 ena_qid;
2290 	int rc;
2291 
2292 	ena_dev = adapter->ena_dev;
2293 
2294 	tx_ring = &adapter->tx_ring[qid];
2295 	msix_vector = ENA_IO_IRQ_IDX(qid);
2296 	ena_qid = ENA_IO_TXQ_IDX(qid);
2297 
2298 	memset(&ctx, 0x0, sizeof(ctx));
2299 
2300 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2301 	ctx.qid = ena_qid;
2302 	ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2303 	ctx.msix_vector = msix_vector;
2304 	ctx.queue_size = tx_ring->ring_size;
2305 	ctx.numa_node = cpu_to_node(tx_ring->cpu);
2306 
2307 	rc = ena_com_create_io_queue(ena_dev, &ctx);
2308 	if (rc) {
2309 		netif_err(adapter, ifup, adapter->netdev,
2310 			  "Failed to create I/O TX queue num %d rc: %d\n",
2311 			   qid, rc);
2312 		return rc;
2313 	}
2314 
2315 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2316 				     &tx_ring->ena_com_io_sq,
2317 				     &tx_ring->ena_com_io_cq);
2318 	if (rc) {
2319 		netif_err(adapter, ifup, adapter->netdev,
2320 			  "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2321 			  qid, rc);
2322 		ena_com_destroy_io_queue(ena_dev, ena_qid);
2323 		return rc;
2324 	}
2325 
2326 	ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2327 	return rc;
2328 }
2329 
2330 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2331 					    int first_index, int count)
2332 {
2333 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2334 	int rc, i;
2335 
2336 	for (i = first_index; i < first_index + count; i++) {
2337 		rc = ena_create_io_tx_queue(adapter, i);
2338 		if (rc)
2339 			goto create_err;
2340 	}
2341 
2342 	return 0;
2343 
2344 create_err:
2345 	while (i-- > first_index)
2346 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2347 
2348 	return rc;
2349 }
2350 
2351 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2352 {
2353 	struct ena_com_dev *ena_dev;
2354 	struct ena_com_create_io_ctx ctx;
2355 	struct ena_ring *rx_ring;
2356 	u32 msix_vector;
2357 	u16 ena_qid;
2358 	int rc;
2359 
2360 	ena_dev = adapter->ena_dev;
2361 
2362 	rx_ring = &adapter->rx_ring[qid];
2363 	msix_vector = ENA_IO_IRQ_IDX(qid);
2364 	ena_qid = ENA_IO_RXQ_IDX(qid);
2365 
2366 	memset(&ctx, 0x0, sizeof(ctx));
2367 
2368 	ctx.qid = ena_qid;
2369 	ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2370 	ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2371 	ctx.msix_vector = msix_vector;
2372 	ctx.queue_size = rx_ring->ring_size;
2373 	ctx.numa_node = cpu_to_node(rx_ring->cpu);
2374 
2375 	rc = ena_com_create_io_queue(ena_dev, &ctx);
2376 	if (rc) {
2377 		netif_err(adapter, ifup, adapter->netdev,
2378 			  "Failed to create I/O RX queue num %d rc: %d\n",
2379 			  qid, rc);
2380 		return rc;
2381 	}
2382 
2383 	rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2384 				     &rx_ring->ena_com_io_sq,
2385 				     &rx_ring->ena_com_io_cq);
2386 	if (rc) {
2387 		netif_err(adapter, ifup, adapter->netdev,
2388 			  "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2389 			  qid, rc);
2390 		goto err;
2391 	}
2392 
2393 	ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2394 
2395 	return rc;
2396 err:
2397 	ena_com_destroy_io_queue(ena_dev, ena_qid);
2398 	return rc;
2399 }
2400 
2401 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2402 {
2403 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2404 	int rc, i;
2405 
2406 	for (i = 0; i < adapter->num_io_queues; i++) {
2407 		rc = ena_create_io_rx_queue(adapter, i);
2408 		if (rc)
2409 			goto create_err;
2410 		INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2411 	}
2412 
2413 	return 0;
2414 
2415 create_err:
2416 	while (i--) {
2417 		cancel_work_sync(&adapter->ena_napi[i].dim.work);
2418 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2419 	}
2420 
2421 	return rc;
2422 }
2423 
2424 static void set_io_rings_size(struct ena_adapter *adapter,
2425 			      int new_tx_size,
2426 			      int new_rx_size)
2427 {
2428 	int i;
2429 
2430 	for (i = 0; i < adapter->num_io_queues; i++) {
2431 		adapter->tx_ring[i].ring_size = new_tx_size;
2432 		adapter->rx_ring[i].ring_size = new_rx_size;
2433 	}
2434 }
2435 
2436 /* This function allows queue allocation to backoff when the system is
2437  * low on memory. If there is not enough memory to allocate io queues
2438  * the driver will try to allocate smaller queues.
2439  *
2440  * The backoff algorithm is as follows:
2441  *  1. Try to allocate TX and RX and if successful.
2442  *  1.1. return success
2443  *
2444  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2445  *
2446  *  3. If TX or RX is smaller than 256
2447  *  3.1. return failure.
2448  *  4. else
2449  *  4.1. go back to 1.
2450  */
2451 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2452 {
2453 	int rc, cur_rx_ring_size, cur_tx_ring_size;
2454 	int new_rx_ring_size, new_tx_ring_size;
2455 
2456 	/* current queue sizes might be set to smaller than the requested
2457 	 * ones due to past queue allocation failures.
2458 	 */
2459 	set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2460 			adapter->requested_rx_ring_size);
2461 
2462 	while (1) {
2463 		if (ena_xdp_present(adapter)) {
2464 			rc = ena_setup_and_create_all_xdp_queues(adapter);
2465 
2466 			if (rc)
2467 				goto err_setup_tx;
2468 		}
2469 		rc = ena_setup_tx_resources_in_range(adapter,
2470 						     0,
2471 						     adapter->num_io_queues);
2472 		if (rc)
2473 			goto err_setup_tx;
2474 
2475 		rc = ena_create_io_tx_queues_in_range(adapter,
2476 						      0,
2477 						      adapter->num_io_queues);
2478 		if (rc)
2479 			goto err_create_tx_queues;
2480 
2481 		rc = ena_setup_all_rx_resources(adapter);
2482 		if (rc)
2483 			goto err_setup_rx;
2484 
2485 		rc = ena_create_all_io_rx_queues(adapter);
2486 		if (rc)
2487 			goto err_create_rx_queues;
2488 
2489 		return 0;
2490 
2491 err_create_rx_queues:
2492 		ena_free_all_io_rx_resources(adapter);
2493 err_setup_rx:
2494 		ena_destroy_all_tx_queues(adapter);
2495 err_create_tx_queues:
2496 		ena_free_all_io_tx_resources(adapter);
2497 err_setup_tx:
2498 		if (rc != -ENOMEM) {
2499 			netif_err(adapter, ifup, adapter->netdev,
2500 				  "Queue creation failed with error code %d\n",
2501 				   rc);
2502 			return rc;
2503 		}
2504 
2505 		cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2506 		cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2507 
2508 		netif_err(adapter, ifup, adapter->netdev,
2509 			  "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2510 			  cur_tx_ring_size, cur_rx_ring_size);
2511 
2512 		new_tx_ring_size = cur_tx_ring_size;
2513 		new_rx_ring_size = cur_rx_ring_size;
2514 
2515 		/* Decrease the size of the larger queue, or
2516 		 * decrease both if they are the same size.
2517 		 */
2518 		if (cur_rx_ring_size <= cur_tx_ring_size)
2519 			new_tx_ring_size = cur_tx_ring_size / 2;
2520 		if (cur_rx_ring_size >= cur_tx_ring_size)
2521 			new_rx_ring_size = cur_rx_ring_size / 2;
2522 
2523 		if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2524 				new_rx_ring_size < ENA_MIN_RING_SIZE) {
2525 			netif_err(adapter, ifup, adapter->netdev,
2526 				  "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2527 				  ENA_MIN_RING_SIZE);
2528 			return rc;
2529 		}
2530 
2531 		netif_err(adapter, ifup, adapter->netdev,
2532 			  "Retrying queue creation with sizes TX=%d, RX=%d\n",
2533 			  new_tx_ring_size,
2534 			  new_rx_ring_size);
2535 
2536 		set_io_rings_size(adapter, new_tx_ring_size,
2537 				  new_rx_ring_size);
2538 	}
2539 }
2540 
2541 static int ena_up(struct ena_adapter *adapter)
2542 {
2543 	int io_queue_count, rc, i;
2544 
2545 	netdev_dbg(adapter->netdev, "%s\n", __func__);
2546 
2547 	io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2548 	ena_setup_io_intr(adapter);
2549 
2550 	/* napi poll functions should be initialized before running
2551 	 * request_irq(), to handle a rare condition where there is a pending
2552 	 * interrupt, causing the ISR to fire immediately while the poll
2553 	 * function wasn't set yet, causing a null dereference
2554 	 */
2555 	ena_init_napi_in_range(adapter, 0, io_queue_count);
2556 
2557 	rc = ena_request_io_irq(adapter);
2558 	if (rc)
2559 		goto err_req_irq;
2560 
2561 	rc = create_queues_with_size_backoff(adapter);
2562 	if (rc)
2563 		goto err_create_queues_with_backoff;
2564 
2565 	rc = ena_up_complete(adapter);
2566 	if (rc)
2567 		goto err_up;
2568 
2569 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2570 		netif_carrier_on(adapter->netdev);
2571 
2572 	u64_stats_update_begin(&adapter->syncp);
2573 	adapter->dev_stats.interface_up++;
2574 	u64_stats_update_end(&adapter->syncp);
2575 
2576 	set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2577 
2578 	/* Enable completion queues interrupt */
2579 	for (i = 0; i < adapter->num_io_queues; i++)
2580 		ena_unmask_interrupt(&adapter->tx_ring[i],
2581 				     &adapter->rx_ring[i]);
2582 
2583 	/* schedule napi in case we had pending packets
2584 	 * from the last time we disable napi
2585 	 */
2586 	for (i = 0; i < io_queue_count; i++)
2587 		napi_schedule(&adapter->ena_napi[i].napi);
2588 
2589 	return rc;
2590 
2591 err_up:
2592 	ena_destroy_all_tx_queues(adapter);
2593 	ena_free_all_io_tx_resources(adapter);
2594 	ena_destroy_all_rx_queues(adapter);
2595 	ena_free_all_io_rx_resources(adapter);
2596 err_create_queues_with_backoff:
2597 	ena_free_io_irq(adapter);
2598 err_req_irq:
2599 	ena_del_napi_in_range(adapter, 0, io_queue_count);
2600 
2601 	return rc;
2602 }
2603 
2604 static void ena_down(struct ena_adapter *adapter)
2605 {
2606 	int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2607 
2608 	netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2609 
2610 	clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2611 
2612 	u64_stats_update_begin(&adapter->syncp);
2613 	adapter->dev_stats.interface_down++;
2614 	u64_stats_update_end(&adapter->syncp);
2615 
2616 	netif_carrier_off(adapter->netdev);
2617 	netif_tx_disable(adapter->netdev);
2618 
2619 	/* After this point the napi handler won't enable the tx queue */
2620 	ena_napi_disable_in_range(adapter, 0, io_queue_count);
2621 
2622 	/* After destroy the queue there won't be any new interrupts */
2623 
2624 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2625 		int rc;
2626 
2627 		rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2628 		if (rc)
2629 			dev_err(&adapter->pdev->dev, "Device reset failed\n");
2630 		/* stop submitting admin commands on a device that was reset */
2631 		ena_com_set_admin_running_state(adapter->ena_dev, false);
2632 	}
2633 
2634 	ena_destroy_all_io_queues(adapter);
2635 
2636 	ena_disable_io_intr_sync(adapter);
2637 	ena_free_io_irq(adapter);
2638 	ena_del_napi_in_range(adapter, 0, io_queue_count);
2639 
2640 	ena_free_all_tx_bufs(adapter);
2641 	ena_free_all_rx_bufs(adapter);
2642 	ena_free_all_io_tx_resources(adapter);
2643 	ena_free_all_io_rx_resources(adapter);
2644 }
2645 
2646 /* ena_open - Called when a network interface is made active
2647  * @netdev: network interface device structure
2648  *
2649  * Returns 0 on success, negative value on failure
2650  *
2651  * The open entry point is called when a network interface is made
2652  * active by the system (IFF_UP).  At this point all resources needed
2653  * for transmit and receive operations are allocated, the interrupt
2654  * handler is registered with the OS, the watchdog timer is started,
2655  * and the stack is notified that the interface is ready.
2656  */
2657 static int ena_open(struct net_device *netdev)
2658 {
2659 	struct ena_adapter *adapter = netdev_priv(netdev);
2660 	int rc;
2661 
2662 	/* Notify the stack of the actual queue counts. */
2663 	rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2664 	if (rc) {
2665 		netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2666 		return rc;
2667 	}
2668 
2669 	rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2670 	if (rc) {
2671 		netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2672 		return rc;
2673 	}
2674 
2675 	rc = ena_up(adapter);
2676 	if (rc)
2677 		return rc;
2678 
2679 	return rc;
2680 }
2681 
2682 /* ena_close - Disables a network interface
2683  * @netdev: network interface device structure
2684  *
2685  * Returns 0, this is not allowed to fail
2686  *
2687  * The close entry point is called when an interface is de-activated
2688  * by the OS.  The hardware is still under the drivers control, but
2689  * needs to be disabled.  A global MAC reset is issued to stop the
2690  * hardware, and all transmit and receive resources are freed.
2691  */
2692 static int ena_close(struct net_device *netdev)
2693 {
2694 	struct ena_adapter *adapter = netdev_priv(netdev);
2695 
2696 	netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2697 
2698 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2699 		return 0;
2700 
2701 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2702 		ena_down(adapter);
2703 
2704 	/* Check for device status and issue reset if needed*/
2705 	check_for_admin_com_state(adapter);
2706 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2707 		netif_err(adapter, ifdown, adapter->netdev,
2708 			  "Destroy failure, restarting device\n");
2709 		ena_dump_stats_to_dmesg(adapter);
2710 		/* rtnl lock already obtained in dev_ioctl() layer */
2711 		ena_destroy_device(adapter, false);
2712 		ena_restore_device(adapter);
2713 	}
2714 
2715 	return 0;
2716 }
2717 
2718 int ena_update_queue_sizes(struct ena_adapter *adapter,
2719 			   u32 new_tx_size,
2720 			   u32 new_rx_size)
2721 {
2722 	bool dev_was_up;
2723 
2724 	dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2725 	ena_close(adapter->netdev);
2726 	adapter->requested_tx_ring_size = new_tx_size;
2727 	adapter->requested_rx_ring_size = new_rx_size;
2728 	ena_init_io_rings(adapter,
2729 			  0,
2730 			  adapter->xdp_num_queues +
2731 			  adapter->num_io_queues);
2732 	return dev_was_up ? ena_up(adapter) : 0;
2733 }
2734 
2735 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2736 {
2737 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2738 	int prev_channel_count;
2739 	bool dev_was_up;
2740 
2741 	dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2742 	ena_close(adapter->netdev);
2743 	prev_channel_count = adapter->num_io_queues;
2744 	adapter->num_io_queues = new_channel_count;
2745 	if (ena_xdp_present(adapter) &&
2746 	    ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2747 		adapter->xdp_first_ring = new_channel_count;
2748 		adapter->xdp_num_queues = new_channel_count;
2749 		if (prev_channel_count > new_channel_count)
2750 			ena_xdp_exchange_program_rx_in_range(adapter,
2751 							     NULL,
2752 							     new_channel_count,
2753 							     prev_channel_count);
2754 		else
2755 			ena_xdp_exchange_program_rx_in_range(adapter,
2756 							     adapter->xdp_bpf_prog,
2757 							     prev_channel_count,
2758 							     new_channel_count);
2759 	}
2760 
2761 	/* We need to destroy the rss table so that the indirection
2762 	 * table will be reinitialized by ena_up()
2763 	 */
2764 	ena_com_rss_destroy(ena_dev);
2765 	ena_init_io_rings(adapter,
2766 			  0,
2767 			  adapter->xdp_num_queues +
2768 			  adapter->num_io_queues);
2769 	return dev_was_up ? ena_open(adapter->netdev) : 0;
2770 }
2771 
2772 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
2773 {
2774 	u32 mss = skb_shinfo(skb)->gso_size;
2775 	struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2776 	u8 l4_protocol = 0;
2777 
2778 	if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2779 		ena_tx_ctx->l4_csum_enable = 1;
2780 		if (mss) {
2781 			ena_tx_ctx->tso_enable = 1;
2782 			ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2783 			ena_tx_ctx->l4_csum_partial = 0;
2784 		} else {
2785 			ena_tx_ctx->tso_enable = 0;
2786 			ena_meta->l4_hdr_len = 0;
2787 			ena_tx_ctx->l4_csum_partial = 1;
2788 		}
2789 
2790 		switch (ip_hdr(skb)->version) {
2791 		case IPVERSION:
2792 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2793 			if (ip_hdr(skb)->frag_off & htons(IP_DF))
2794 				ena_tx_ctx->df = 1;
2795 			if (mss)
2796 				ena_tx_ctx->l3_csum_enable = 1;
2797 			l4_protocol = ip_hdr(skb)->protocol;
2798 			break;
2799 		case 6:
2800 			ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2801 			l4_protocol = ipv6_hdr(skb)->nexthdr;
2802 			break;
2803 		default:
2804 			break;
2805 		}
2806 
2807 		if (l4_protocol == IPPROTO_TCP)
2808 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2809 		else
2810 			ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2811 
2812 		ena_meta->mss = mss;
2813 		ena_meta->l3_hdr_len = skb_network_header_len(skb);
2814 		ena_meta->l3_hdr_offset = skb_network_offset(skb);
2815 		ena_tx_ctx->meta_valid = 1;
2816 
2817 	} else {
2818 		ena_tx_ctx->meta_valid = 0;
2819 	}
2820 }
2821 
2822 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2823 				       struct sk_buff *skb)
2824 {
2825 	int num_frags, header_len, rc;
2826 
2827 	num_frags = skb_shinfo(skb)->nr_frags;
2828 	header_len = skb_headlen(skb);
2829 
2830 	if (num_frags < tx_ring->sgl_size)
2831 		return 0;
2832 
2833 	if ((num_frags == tx_ring->sgl_size) &&
2834 	    (header_len < tx_ring->tx_max_header_size))
2835 		return 0;
2836 
2837 	u64_stats_update_begin(&tx_ring->syncp);
2838 	tx_ring->tx_stats.linearize++;
2839 	u64_stats_update_end(&tx_ring->syncp);
2840 
2841 	rc = skb_linearize(skb);
2842 	if (unlikely(rc)) {
2843 		u64_stats_update_begin(&tx_ring->syncp);
2844 		tx_ring->tx_stats.linearize_failed++;
2845 		u64_stats_update_end(&tx_ring->syncp);
2846 	}
2847 
2848 	return rc;
2849 }
2850 
2851 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2852 			  struct ena_tx_buffer *tx_info,
2853 			  struct sk_buff *skb,
2854 			  void **push_hdr,
2855 			  u16 *header_len)
2856 {
2857 	struct ena_adapter *adapter = tx_ring->adapter;
2858 	struct ena_com_buf *ena_buf;
2859 	dma_addr_t dma;
2860 	u32 skb_head_len, frag_len, last_frag;
2861 	u16 push_len = 0;
2862 	u16 delta = 0;
2863 	int i = 0;
2864 
2865 	skb_head_len = skb_headlen(skb);
2866 	tx_info->skb = skb;
2867 	ena_buf = tx_info->bufs;
2868 
2869 	if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2870 		/* When the device is LLQ mode, the driver will copy
2871 		 * the header into the device memory space.
2872 		 * the ena_com layer assume the header is in a linear
2873 		 * memory space.
2874 		 * This assumption might be wrong since part of the header
2875 		 * can be in the fragmented buffers.
2876 		 * Use skb_header_pointer to make sure the header is in a
2877 		 * linear memory space.
2878 		 */
2879 
2880 		push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2881 		*push_hdr = skb_header_pointer(skb, 0, push_len,
2882 					       tx_ring->push_buf_intermediate_buf);
2883 		*header_len = push_len;
2884 		if (unlikely(skb->data != *push_hdr)) {
2885 			u64_stats_update_begin(&tx_ring->syncp);
2886 			tx_ring->tx_stats.llq_buffer_copy++;
2887 			u64_stats_update_end(&tx_ring->syncp);
2888 
2889 			delta = push_len - skb_head_len;
2890 		}
2891 	} else {
2892 		*push_hdr = NULL;
2893 		*header_len = min_t(u32, skb_head_len,
2894 				    tx_ring->tx_max_header_size);
2895 	}
2896 
2897 	netif_dbg(adapter, tx_queued, adapter->netdev,
2898 		  "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2899 		  *push_hdr, push_len);
2900 
2901 	if (skb_head_len > push_len) {
2902 		dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2903 				     skb_head_len - push_len, DMA_TO_DEVICE);
2904 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2905 			goto error_report_dma_error;
2906 
2907 		ena_buf->paddr = dma;
2908 		ena_buf->len = skb_head_len - push_len;
2909 
2910 		ena_buf++;
2911 		tx_info->num_of_bufs++;
2912 		tx_info->map_linear_data = 1;
2913 	} else {
2914 		tx_info->map_linear_data = 0;
2915 	}
2916 
2917 	last_frag = skb_shinfo(skb)->nr_frags;
2918 
2919 	for (i = 0; i < last_frag; i++) {
2920 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2921 
2922 		frag_len = skb_frag_size(frag);
2923 
2924 		if (unlikely(delta >= frag_len)) {
2925 			delta -= frag_len;
2926 			continue;
2927 		}
2928 
2929 		dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2930 				       frag_len - delta, DMA_TO_DEVICE);
2931 		if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2932 			goto error_report_dma_error;
2933 
2934 		ena_buf->paddr = dma;
2935 		ena_buf->len = frag_len - delta;
2936 		ena_buf++;
2937 		tx_info->num_of_bufs++;
2938 		delta = 0;
2939 	}
2940 
2941 	return 0;
2942 
2943 error_report_dma_error:
2944 	u64_stats_update_begin(&tx_ring->syncp);
2945 	tx_ring->tx_stats.dma_mapping_err++;
2946 	u64_stats_update_end(&tx_ring->syncp);
2947 	netdev_warn(adapter->netdev, "failed to map skb\n");
2948 
2949 	tx_info->skb = NULL;
2950 
2951 	tx_info->num_of_bufs += i;
2952 	ena_unmap_tx_buff(tx_ring, tx_info);
2953 
2954 	return -EINVAL;
2955 }
2956 
2957 /* Called with netif_tx_lock. */
2958 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
2959 {
2960 	struct ena_adapter *adapter = netdev_priv(dev);
2961 	struct ena_tx_buffer *tx_info;
2962 	struct ena_com_tx_ctx ena_tx_ctx;
2963 	struct ena_ring *tx_ring;
2964 	struct netdev_queue *txq;
2965 	void *push_hdr;
2966 	u16 next_to_use, req_id, header_len;
2967 	int qid, rc;
2968 
2969 	netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
2970 	/*  Determine which tx ring we will be placed on */
2971 	qid = skb_get_queue_mapping(skb);
2972 	tx_ring = &adapter->tx_ring[qid];
2973 	txq = netdev_get_tx_queue(dev, qid);
2974 
2975 	rc = ena_check_and_linearize_skb(tx_ring, skb);
2976 	if (unlikely(rc))
2977 		goto error_drop_packet;
2978 
2979 	skb_tx_timestamp(skb);
2980 
2981 	next_to_use = tx_ring->next_to_use;
2982 	req_id = tx_ring->free_ids[next_to_use];
2983 	tx_info = &tx_ring->tx_buffer_info[req_id];
2984 	tx_info->num_of_bufs = 0;
2985 
2986 	WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
2987 
2988 	rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
2989 	if (unlikely(rc))
2990 		goto error_drop_packet;
2991 
2992 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2993 	ena_tx_ctx.ena_bufs = tx_info->bufs;
2994 	ena_tx_ctx.push_header = push_hdr;
2995 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2996 	ena_tx_ctx.req_id = req_id;
2997 	ena_tx_ctx.header_len = header_len;
2998 
2999 	/* set flags and meta data */
3000 	ena_tx_csum(&ena_tx_ctx, skb);
3001 
3002 	rc = ena_xmit_common(dev,
3003 			     tx_ring,
3004 			     tx_info,
3005 			     &ena_tx_ctx,
3006 			     next_to_use,
3007 			     skb->len);
3008 	if (rc)
3009 		goto error_unmap_dma;
3010 
3011 	netdev_tx_sent_queue(txq, skb->len);
3012 
3013 	/* stop the queue when no more space available, the packet can have up
3014 	 * to sgl_size + 2. one for the meta descriptor and one for header
3015 	 * (if the header is larger than tx_max_header_size).
3016 	 */
3017 	if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3018 						   tx_ring->sgl_size + 2))) {
3019 		netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3020 			  __func__, qid);
3021 
3022 		netif_tx_stop_queue(txq);
3023 		u64_stats_update_begin(&tx_ring->syncp);
3024 		tx_ring->tx_stats.queue_stop++;
3025 		u64_stats_update_end(&tx_ring->syncp);
3026 
3027 		/* There is a rare condition where this function decide to
3028 		 * stop the queue but meanwhile clean_tx_irq updates
3029 		 * next_to_completion and terminates.
3030 		 * The queue will remain stopped forever.
3031 		 * To solve this issue add a mb() to make sure that
3032 		 * netif_tx_stop_queue() write is vissible before checking if
3033 		 * there is additional space in the queue.
3034 		 */
3035 		smp_mb();
3036 
3037 		if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3038 						 ENA_TX_WAKEUP_THRESH)) {
3039 			netif_tx_wake_queue(txq);
3040 			u64_stats_update_begin(&tx_ring->syncp);
3041 			tx_ring->tx_stats.queue_wakeup++;
3042 			u64_stats_update_end(&tx_ring->syncp);
3043 		}
3044 	}
3045 
3046 	if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
3047 		/* trigger the dma engine. ena_com_write_sq_doorbell()
3048 		 * has a mb
3049 		 */
3050 		ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
3051 		u64_stats_update_begin(&tx_ring->syncp);
3052 		tx_ring->tx_stats.doorbells++;
3053 		u64_stats_update_end(&tx_ring->syncp);
3054 	}
3055 
3056 	return NETDEV_TX_OK;
3057 
3058 error_unmap_dma:
3059 	ena_unmap_tx_buff(tx_ring, tx_info);
3060 	tx_info->skb = NULL;
3061 
3062 error_drop_packet:
3063 	dev_kfree_skb(skb);
3064 	return NETDEV_TX_OK;
3065 }
3066 
3067 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3068 			    struct net_device *sb_dev)
3069 {
3070 	u16 qid;
3071 	/* we suspect that this is good for in--kernel network services that
3072 	 * want to loop incoming skb rx to tx in normal user generated traffic,
3073 	 * most probably we will not get to this
3074 	 */
3075 	if (skb_rx_queue_recorded(skb))
3076 		qid = skb_get_rx_queue(skb);
3077 	else
3078 		qid = netdev_pick_tx(dev, skb, NULL);
3079 
3080 	return qid;
3081 }
3082 
3083 static void ena_config_host_info(struct ena_com_dev *ena_dev,
3084 				 struct pci_dev *pdev)
3085 {
3086 	struct ena_admin_host_info *host_info;
3087 	int rc;
3088 
3089 	/* Allocate only the host info */
3090 	rc = ena_com_allocate_host_info(ena_dev);
3091 	if (rc) {
3092 		pr_err("Cannot allocate host info\n");
3093 		return;
3094 	}
3095 
3096 	host_info = ena_dev->host_attr.host_info;
3097 
3098 	host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3099 	host_info->os_type = ENA_ADMIN_OS_LINUX;
3100 	host_info->kernel_ver = LINUX_VERSION_CODE;
3101 	strlcpy(host_info->kernel_ver_str, utsname()->version,
3102 		sizeof(host_info->kernel_ver_str) - 1);
3103 	host_info->os_dist = 0;
3104 	strncpy(host_info->os_dist_str, utsname()->release,
3105 		sizeof(host_info->os_dist_str) - 1);
3106 	host_info->driver_version =
3107 		(DRV_MODULE_GEN_MAJOR) |
3108 		(DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3109 		(DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3110 		("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3111 	host_info->num_cpus = num_online_cpus();
3112 
3113 	host_info->driver_supported_features =
3114 		ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK;
3115 
3116 	rc = ena_com_set_host_attributes(ena_dev);
3117 	if (rc) {
3118 		if (rc == -EOPNOTSUPP)
3119 			pr_warn("Cannot set host attributes\n");
3120 		else
3121 			pr_err("Cannot set host attributes\n");
3122 
3123 		goto err;
3124 	}
3125 
3126 	return;
3127 
3128 err:
3129 	ena_com_delete_host_info(ena_dev);
3130 }
3131 
3132 static void ena_config_debug_area(struct ena_adapter *adapter)
3133 {
3134 	u32 debug_area_size;
3135 	int rc, ss_count;
3136 
3137 	ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3138 	if (ss_count <= 0) {
3139 		netif_err(adapter, drv, adapter->netdev,
3140 			  "SS count is negative\n");
3141 		return;
3142 	}
3143 
3144 	/* allocate 32 bytes for each string and 64bit for the value */
3145 	debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3146 
3147 	rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3148 	if (rc) {
3149 		pr_err("Cannot allocate debug area\n");
3150 		return;
3151 	}
3152 
3153 	rc = ena_com_set_host_attributes(adapter->ena_dev);
3154 	if (rc) {
3155 		if (rc == -EOPNOTSUPP)
3156 			netif_warn(adapter, drv, adapter->netdev,
3157 				   "Cannot set host attributes\n");
3158 		else
3159 			netif_err(adapter, drv, adapter->netdev,
3160 				  "Cannot set host attributes\n");
3161 		goto err;
3162 	}
3163 
3164 	return;
3165 err:
3166 	ena_com_delete_debug_area(adapter->ena_dev);
3167 }
3168 
3169 static void ena_get_stats64(struct net_device *netdev,
3170 			    struct rtnl_link_stats64 *stats)
3171 {
3172 	struct ena_adapter *adapter = netdev_priv(netdev);
3173 	struct ena_ring *rx_ring, *tx_ring;
3174 	unsigned int start;
3175 	u64 rx_drops;
3176 	u64 tx_drops;
3177 	int i;
3178 
3179 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3180 		return;
3181 
3182 	for (i = 0; i < adapter->num_io_queues; i++) {
3183 		u64 bytes, packets;
3184 
3185 		tx_ring = &adapter->tx_ring[i];
3186 
3187 		do {
3188 			start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3189 			packets = tx_ring->tx_stats.cnt;
3190 			bytes = tx_ring->tx_stats.bytes;
3191 		} while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3192 
3193 		stats->tx_packets += packets;
3194 		stats->tx_bytes += bytes;
3195 
3196 		rx_ring = &adapter->rx_ring[i];
3197 
3198 		do {
3199 			start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3200 			packets = rx_ring->rx_stats.cnt;
3201 			bytes = rx_ring->rx_stats.bytes;
3202 		} while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3203 
3204 		stats->rx_packets += packets;
3205 		stats->rx_bytes += bytes;
3206 	}
3207 
3208 	do {
3209 		start = u64_stats_fetch_begin_irq(&adapter->syncp);
3210 		rx_drops = adapter->dev_stats.rx_drops;
3211 		tx_drops = adapter->dev_stats.tx_drops;
3212 	} while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3213 
3214 	stats->rx_dropped = rx_drops;
3215 	stats->tx_dropped = tx_drops;
3216 
3217 	stats->multicast = 0;
3218 	stats->collisions = 0;
3219 
3220 	stats->rx_length_errors = 0;
3221 	stats->rx_crc_errors = 0;
3222 	stats->rx_frame_errors = 0;
3223 	stats->rx_fifo_errors = 0;
3224 	stats->rx_missed_errors = 0;
3225 	stats->tx_window_errors = 0;
3226 
3227 	stats->rx_errors = 0;
3228 	stats->tx_errors = 0;
3229 }
3230 
3231 static const struct net_device_ops ena_netdev_ops = {
3232 	.ndo_open		= ena_open,
3233 	.ndo_stop		= ena_close,
3234 	.ndo_start_xmit		= ena_start_xmit,
3235 	.ndo_select_queue	= ena_select_queue,
3236 	.ndo_get_stats64	= ena_get_stats64,
3237 	.ndo_tx_timeout		= ena_tx_timeout,
3238 	.ndo_change_mtu		= ena_change_mtu,
3239 	.ndo_set_mac_address	= NULL,
3240 	.ndo_validate_addr	= eth_validate_addr,
3241 	.ndo_bpf		= ena_xdp,
3242 };
3243 
3244 static int ena_device_validate_params(struct ena_adapter *adapter,
3245 				      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3246 {
3247 	struct net_device *netdev = adapter->netdev;
3248 	int rc;
3249 
3250 	rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3251 			      adapter->mac_addr);
3252 	if (!rc) {
3253 		netif_err(adapter, drv, netdev,
3254 			  "Error, mac address are different\n");
3255 		return -EINVAL;
3256 	}
3257 
3258 	if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3259 		netif_err(adapter, drv, netdev,
3260 			  "Error, device max mtu is smaller than netdev MTU\n");
3261 		return -EINVAL;
3262 	}
3263 
3264 	return 0;
3265 }
3266 
3267 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3268 			   struct ena_com_dev_get_features_ctx *get_feat_ctx,
3269 			   bool *wd_state)
3270 {
3271 	struct device *dev = &pdev->dev;
3272 	bool readless_supported;
3273 	u32 aenq_groups;
3274 	int dma_width;
3275 	int rc;
3276 
3277 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
3278 	if (rc) {
3279 		dev_err(dev, "failed to init mmio read less\n");
3280 		return rc;
3281 	}
3282 
3283 	/* The PCIe configuration space revision id indicate if mmio reg
3284 	 * read is disabled
3285 	 */
3286 	readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3287 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3288 
3289 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3290 	if (rc) {
3291 		dev_err(dev, "Can not reset device\n");
3292 		goto err_mmio_read_less;
3293 	}
3294 
3295 	rc = ena_com_validate_version(ena_dev);
3296 	if (rc) {
3297 		dev_err(dev, "device version is too low\n");
3298 		goto err_mmio_read_less;
3299 	}
3300 
3301 	dma_width = ena_com_get_dma_width(ena_dev);
3302 	if (dma_width < 0) {
3303 		dev_err(dev, "Invalid dma width value %d", dma_width);
3304 		rc = dma_width;
3305 		goto err_mmio_read_less;
3306 	}
3307 
3308 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
3309 	if (rc) {
3310 		dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
3311 		goto err_mmio_read_less;
3312 	}
3313 
3314 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
3315 	if (rc) {
3316 		dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
3317 			rc);
3318 		goto err_mmio_read_less;
3319 	}
3320 
3321 	/* ENA admin level init */
3322 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3323 	if (rc) {
3324 		dev_err(dev,
3325 			"Can not initialize ena admin queue with device\n");
3326 		goto err_mmio_read_less;
3327 	}
3328 
3329 	/* To enable the msix interrupts the driver needs to know the number
3330 	 * of queues. So the driver uses polling mode to retrieve this
3331 	 * information
3332 	 */
3333 	ena_com_set_admin_polling_mode(ena_dev, true);
3334 
3335 	ena_config_host_info(ena_dev, pdev);
3336 
3337 	/* Get Device Attributes*/
3338 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3339 	if (rc) {
3340 		dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3341 		goto err_admin_init;
3342 	}
3343 
3344 	/* Try to turn all the available aenq groups */
3345 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3346 		BIT(ENA_ADMIN_FATAL_ERROR) |
3347 		BIT(ENA_ADMIN_WARNING) |
3348 		BIT(ENA_ADMIN_NOTIFICATION) |
3349 		BIT(ENA_ADMIN_KEEP_ALIVE);
3350 
3351 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
3352 
3353 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3354 	if (rc) {
3355 		dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3356 		goto err_admin_init;
3357 	}
3358 
3359 	*wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3360 
3361 	return 0;
3362 
3363 err_admin_init:
3364 	ena_com_delete_host_info(ena_dev);
3365 	ena_com_admin_destroy(ena_dev);
3366 err_mmio_read_less:
3367 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3368 
3369 	return rc;
3370 }
3371 
3372 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3373 {
3374 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3375 	struct device *dev = &adapter->pdev->dev;
3376 	int rc;
3377 
3378 	rc = ena_enable_msix(adapter);
3379 	if (rc) {
3380 		dev_err(dev, "Can not reserve msix vectors\n");
3381 		return rc;
3382 	}
3383 
3384 	ena_setup_mgmnt_intr(adapter);
3385 
3386 	rc = ena_request_mgmnt_irq(adapter);
3387 	if (rc) {
3388 		dev_err(dev, "Can not setup management interrupts\n");
3389 		goto err_disable_msix;
3390 	}
3391 
3392 	ena_com_set_admin_polling_mode(ena_dev, false);
3393 
3394 	ena_com_admin_aenq_enable(ena_dev);
3395 
3396 	return 0;
3397 
3398 err_disable_msix:
3399 	ena_disable_msix(adapter);
3400 
3401 	return rc;
3402 }
3403 
3404 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3405 {
3406 	struct net_device *netdev = adapter->netdev;
3407 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3408 	bool dev_up;
3409 
3410 	if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3411 		return;
3412 
3413 	netif_carrier_off(netdev);
3414 
3415 	del_timer_sync(&adapter->timer_service);
3416 
3417 	dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3418 	adapter->dev_up_before_reset = dev_up;
3419 	if (!graceful)
3420 		ena_com_set_admin_running_state(ena_dev, false);
3421 
3422 	if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3423 		ena_down(adapter);
3424 
3425 	/* Stop the device from sending AENQ events (in case reset flag is set
3426 	 *  and device is up, ena_down() already reset the device.
3427 	 */
3428 	if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3429 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3430 
3431 	ena_free_mgmnt_irq(adapter);
3432 
3433 	ena_disable_msix(adapter);
3434 
3435 	ena_com_abort_admin_commands(ena_dev);
3436 
3437 	ena_com_wait_for_abort_completion(ena_dev);
3438 
3439 	ena_com_admin_destroy(ena_dev);
3440 
3441 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3442 
3443 	/* return reset reason to default value */
3444 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3445 
3446 	clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3447 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3448 }
3449 
3450 static int ena_restore_device(struct ena_adapter *adapter)
3451 {
3452 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3453 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3454 	struct pci_dev *pdev = adapter->pdev;
3455 	bool wd_state;
3456 	int rc;
3457 
3458 	set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3459 	rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3460 	if (rc) {
3461 		dev_err(&pdev->dev, "Can not initialize device\n");
3462 		goto err;
3463 	}
3464 	adapter->wd_state = wd_state;
3465 
3466 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
3467 	if (rc) {
3468 		dev_err(&pdev->dev, "Validation of device parameters failed\n");
3469 		goto err_device_destroy;
3470 	}
3471 
3472 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3473 	if (rc) {
3474 		dev_err(&pdev->dev, "Enable MSI-X failed\n");
3475 		goto err_device_destroy;
3476 	}
3477 	/* If the interface was up before the reset bring it up */
3478 	if (adapter->dev_up_before_reset) {
3479 		rc = ena_up(adapter);
3480 		if (rc) {
3481 			dev_err(&pdev->dev, "Failed to create I/O queues\n");
3482 			goto err_disable_msix;
3483 		}
3484 	}
3485 
3486 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3487 
3488 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3489 	if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3490 		netif_carrier_on(adapter->netdev);
3491 
3492 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3493 	dev_err(&pdev->dev, "Device reset completed successfully\n");
3494 	adapter->last_keep_alive_jiffies = jiffies;
3495 
3496 	return rc;
3497 err_disable_msix:
3498 	ena_free_mgmnt_irq(adapter);
3499 	ena_disable_msix(adapter);
3500 err_device_destroy:
3501 	ena_com_abort_admin_commands(ena_dev);
3502 	ena_com_wait_for_abort_completion(ena_dev);
3503 	ena_com_admin_destroy(ena_dev);
3504 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3505 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3506 err:
3507 	clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3508 	clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3509 	dev_err(&pdev->dev,
3510 		"Reset attempt failed. Can not reset the device\n");
3511 
3512 	return rc;
3513 }
3514 
3515 static void ena_fw_reset_device(struct work_struct *work)
3516 {
3517 	struct ena_adapter *adapter =
3518 		container_of(work, struct ena_adapter, reset_task);
3519 	struct pci_dev *pdev = adapter->pdev;
3520 
3521 	if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3522 		dev_err(&pdev->dev,
3523 			"device reset schedule while reset bit is off\n");
3524 		return;
3525 	}
3526 	rtnl_lock();
3527 	ena_destroy_device(adapter, false);
3528 	ena_restore_device(adapter);
3529 	rtnl_unlock();
3530 }
3531 
3532 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3533 					struct ena_ring *rx_ring)
3534 {
3535 	if (likely(rx_ring->first_interrupt))
3536 		return 0;
3537 
3538 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3539 		return 0;
3540 
3541 	rx_ring->no_interrupt_event_cnt++;
3542 
3543 	if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3544 		netif_err(adapter, rx_err, adapter->netdev,
3545 			  "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3546 			  rx_ring->qid);
3547 		adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3548 		smp_mb__before_atomic();
3549 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3550 		return -EIO;
3551 	}
3552 
3553 	return 0;
3554 }
3555 
3556 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3557 					  struct ena_ring *tx_ring)
3558 {
3559 	struct ena_tx_buffer *tx_buf;
3560 	unsigned long last_jiffies;
3561 	u32 missed_tx = 0;
3562 	int i, rc = 0;
3563 
3564 	for (i = 0; i < tx_ring->ring_size; i++) {
3565 		tx_buf = &tx_ring->tx_buffer_info[i];
3566 		last_jiffies = tx_buf->last_jiffies;
3567 
3568 		if (last_jiffies == 0)
3569 			/* no pending Tx at this location */
3570 			continue;
3571 
3572 		if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
3573 			     2 * adapter->missing_tx_completion_to))) {
3574 			/* If after graceful period interrupt is still not
3575 			 * received, we schedule a reset
3576 			 */
3577 			netif_err(adapter, tx_err, adapter->netdev,
3578 				  "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3579 				  tx_ring->qid);
3580 			adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3581 			smp_mb__before_atomic();
3582 			set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3583 			return -EIO;
3584 		}
3585 
3586 		if (unlikely(time_is_before_jiffies(last_jiffies +
3587 				adapter->missing_tx_completion_to))) {
3588 			if (!tx_buf->print_once)
3589 				netif_notice(adapter, tx_err, adapter->netdev,
3590 					     "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
3591 					     tx_ring->qid, i);
3592 
3593 			tx_buf->print_once = 1;
3594 			missed_tx++;
3595 		}
3596 	}
3597 
3598 	if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3599 		netif_err(adapter, tx_err, adapter->netdev,
3600 			  "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3601 			  missed_tx,
3602 			  adapter->missing_tx_completion_threshold);
3603 		adapter->reset_reason =
3604 			ENA_REGS_RESET_MISS_TX_CMPL;
3605 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3606 		rc = -EIO;
3607 	}
3608 
3609 	u64_stats_update_begin(&tx_ring->syncp);
3610 	tx_ring->tx_stats.missed_tx = missed_tx;
3611 	u64_stats_update_end(&tx_ring->syncp);
3612 
3613 	return rc;
3614 }
3615 
3616 static void check_for_missing_completions(struct ena_adapter *adapter)
3617 {
3618 	struct ena_ring *tx_ring;
3619 	struct ena_ring *rx_ring;
3620 	int i, budget, rc;
3621 	int io_queue_count;
3622 
3623 	io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3624 	/* Make sure the driver doesn't turn the device in other process */
3625 	smp_rmb();
3626 
3627 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3628 		return;
3629 
3630 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3631 		return;
3632 
3633 	if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3634 		return;
3635 
3636 	budget = ENA_MONITORED_TX_QUEUES;
3637 
3638 	for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3639 		tx_ring = &adapter->tx_ring[i];
3640 		rx_ring = &adapter->rx_ring[i];
3641 
3642 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3643 		if (unlikely(rc))
3644 			return;
3645 
3646 		rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3647 			check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3648 		if (unlikely(rc))
3649 			return;
3650 
3651 		budget--;
3652 		if (!budget)
3653 			break;
3654 	}
3655 
3656 	adapter->last_monitored_tx_qid = i % io_queue_count;
3657 }
3658 
3659 /* trigger napi schedule after 2 consecutive detections */
3660 #define EMPTY_RX_REFILL 2
3661 /* For the rare case where the device runs out of Rx descriptors and the
3662  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3663  * for example).
3664  * This case will lead to a deadlock:
3665  * The device won't send interrupts since all the new Rx packets will be dropped
3666  * The napi handler won't allocate new Rx descriptors so the device will be
3667  * able to send new packets.
3668  *
3669  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3670  * It is recommended to have at least 512MB, with a minimum of 128MB for
3671  * constrained environment).
3672  *
3673  * When such a situation is detected - Reschedule napi
3674  */
3675 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3676 {
3677 	struct ena_ring *rx_ring;
3678 	int i, refill_required;
3679 
3680 	if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3681 		return;
3682 
3683 	if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3684 		return;
3685 
3686 	for (i = 0; i < adapter->num_io_queues; i++) {
3687 		rx_ring = &adapter->rx_ring[i];
3688 
3689 		refill_required =
3690 			ena_com_free_desc(rx_ring->ena_com_io_sq);
3691 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3692 			rx_ring->empty_rx_queue++;
3693 
3694 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3695 				u64_stats_update_begin(&rx_ring->syncp);
3696 				rx_ring->rx_stats.empty_rx_ring++;
3697 				u64_stats_update_end(&rx_ring->syncp);
3698 
3699 				netif_err(adapter, drv, adapter->netdev,
3700 					  "trigger refill for ring %d\n", i);
3701 
3702 				napi_schedule(rx_ring->napi);
3703 				rx_ring->empty_rx_queue = 0;
3704 			}
3705 		} else {
3706 			rx_ring->empty_rx_queue = 0;
3707 		}
3708 	}
3709 }
3710 
3711 /* Check for keep alive expiration */
3712 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3713 {
3714 	unsigned long keep_alive_expired;
3715 
3716 	if (!adapter->wd_state)
3717 		return;
3718 
3719 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3720 		return;
3721 
3722 	keep_alive_expired = adapter->last_keep_alive_jiffies +
3723 			     adapter->keep_alive_timeout;
3724 	if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3725 		netif_err(adapter, drv, adapter->netdev,
3726 			  "Keep alive watchdog timeout.\n");
3727 		u64_stats_update_begin(&adapter->syncp);
3728 		adapter->dev_stats.wd_expired++;
3729 		u64_stats_update_end(&adapter->syncp);
3730 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3731 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3732 	}
3733 }
3734 
3735 static void check_for_admin_com_state(struct ena_adapter *adapter)
3736 {
3737 	if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3738 		netif_err(adapter, drv, adapter->netdev,
3739 			  "ENA admin queue is not in running state!\n");
3740 		u64_stats_update_begin(&adapter->syncp);
3741 		adapter->dev_stats.admin_q_pause++;
3742 		u64_stats_update_end(&adapter->syncp);
3743 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3744 		set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3745 	}
3746 }
3747 
3748 static void ena_update_hints(struct ena_adapter *adapter,
3749 			     struct ena_admin_ena_hw_hints *hints)
3750 {
3751 	struct net_device *netdev = adapter->netdev;
3752 
3753 	if (hints->admin_completion_tx_timeout)
3754 		adapter->ena_dev->admin_queue.completion_timeout =
3755 			hints->admin_completion_tx_timeout * 1000;
3756 
3757 	if (hints->mmio_read_timeout)
3758 		/* convert to usec */
3759 		adapter->ena_dev->mmio_read.reg_read_to =
3760 			hints->mmio_read_timeout * 1000;
3761 
3762 	if (hints->missed_tx_completion_count_threshold_to_reset)
3763 		adapter->missing_tx_completion_threshold =
3764 			hints->missed_tx_completion_count_threshold_to_reset;
3765 
3766 	if (hints->missing_tx_completion_timeout) {
3767 		if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3768 			adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3769 		else
3770 			adapter->missing_tx_completion_to =
3771 				msecs_to_jiffies(hints->missing_tx_completion_timeout);
3772 	}
3773 
3774 	if (hints->netdev_wd_timeout)
3775 		netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3776 
3777 	if (hints->driver_watchdog_timeout) {
3778 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3779 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3780 		else
3781 			adapter->keep_alive_timeout =
3782 				msecs_to_jiffies(hints->driver_watchdog_timeout);
3783 	}
3784 }
3785 
3786 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3787 				 struct net_device *netdev)
3788 {
3789 	host_info->supported_network_features[0] =
3790 		netdev->features & GENMASK_ULL(31, 0);
3791 	host_info->supported_network_features[1] =
3792 		(netdev->features & GENMASK_ULL(63, 32)) >> 32;
3793 }
3794 
3795 static void ena_timer_service(struct timer_list *t)
3796 {
3797 	struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3798 	u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3799 	struct ena_admin_host_info *host_info =
3800 		adapter->ena_dev->host_attr.host_info;
3801 
3802 	check_for_missing_keep_alive(adapter);
3803 
3804 	check_for_admin_com_state(adapter);
3805 
3806 	check_for_missing_completions(adapter);
3807 
3808 	check_for_empty_rx_ring(adapter);
3809 
3810 	if (debug_area)
3811 		ena_dump_stats_to_buf(adapter, debug_area);
3812 
3813 	if (host_info)
3814 		ena_update_host_info(host_info, adapter->netdev);
3815 
3816 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3817 		netif_err(adapter, drv, adapter->netdev,
3818 			  "Trigger reset is on\n");
3819 		ena_dump_stats_to_dmesg(adapter);
3820 		queue_work(ena_wq, &adapter->reset_task);
3821 		return;
3822 	}
3823 
3824 	/* Reset the timer */
3825 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3826 }
3827 
3828 static int ena_calc_max_io_queue_num(struct pci_dev *pdev,
3829 				     struct ena_com_dev *ena_dev,
3830 				     struct ena_com_dev_get_features_ctx *get_feat_ctx)
3831 {
3832 	int io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3833 
3834 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3835 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3836 			&get_feat_ctx->max_queue_ext.max_queue_ext;
3837 		io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3838 				  max_queue_ext->max_rx_cq_num);
3839 
3840 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3841 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3842 	} else {
3843 		struct ena_admin_queue_feature_desc *max_queues =
3844 			&get_feat_ctx->max_queues;
3845 		io_tx_sq_num = max_queues->max_sq_num;
3846 		io_tx_cq_num = max_queues->max_cq_num;
3847 		io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
3848 	}
3849 
3850 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
3851 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3852 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3853 
3854 	max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3855 	max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
3856 	max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
3857 	max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
3858 	/* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
3859 	max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
3860 	if (unlikely(!max_num_io_queues)) {
3861 		dev_err(&pdev->dev, "The device doesn't have io queues\n");
3862 		return -EFAULT;
3863 	}
3864 
3865 	return max_num_io_queues;
3866 }
3867 
3868 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3869 					   struct ena_com_dev *ena_dev,
3870 					   struct ena_admin_feature_llq_desc *llq,
3871 					   struct ena_llq_configurations *llq_default_configurations)
3872 {
3873 	bool has_mem_bar;
3874 	int rc;
3875 	u32 llq_feature_mask;
3876 
3877 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3878 	if (!(ena_dev->supported_features & llq_feature_mask)) {
3879 		dev_err(&pdev->dev,
3880 			"LLQ is not supported Fallback to host mode policy.\n");
3881 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3882 		return 0;
3883 	}
3884 
3885 	has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
3886 
3887 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3888 	if (unlikely(rc)) {
3889 		dev_err(&pdev->dev,
3890 			"Failed to configure the device mode.  Fallback to host mode policy.\n");
3891 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3892 		return 0;
3893 	}
3894 
3895 	/* Nothing to config, exit */
3896 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3897 		return 0;
3898 
3899 	if (!has_mem_bar) {
3900 		dev_err(&pdev->dev,
3901 			"ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3902 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3903 		return 0;
3904 	}
3905 
3906 	ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3907 					   pci_resource_start(pdev, ENA_MEM_BAR),
3908 					   pci_resource_len(pdev, ENA_MEM_BAR));
3909 
3910 	if (!ena_dev->mem_bar)
3911 		return -EFAULT;
3912 
3913 	return 0;
3914 }
3915 
3916 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3917 				 struct net_device *netdev)
3918 {
3919 	netdev_features_t dev_features = 0;
3920 
3921 	/* Set offload features */
3922 	if (feat->offload.tx &
3923 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3924 		dev_features |= NETIF_F_IP_CSUM;
3925 
3926 	if (feat->offload.tx &
3927 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
3928 		dev_features |= NETIF_F_IPV6_CSUM;
3929 
3930 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
3931 		dev_features |= NETIF_F_TSO;
3932 
3933 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
3934 		dev_features |= NETIF_F_TSO6;
3935 
3936 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
3937 		dev_features |= NETIF_F_TSO_ECN;
3938 
3939 	if (feat->offload.rx_supported &
3940 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
3941 		dev_features |= NETIF_F_RXCSUM;
3942 
3943 	if (feat->offload.rx_supported &
3944 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
3945 		dev_features |= NETIF_F_RXCSUM;
3946 
3947 	netdev->features =
3948 		dev_features |
3949 		NETIF_F_SG |
3950 		NETIF_F_RXHASH |
3951 		NETIF_F_HIGHDMA;
3952 
3953 	netdev->hw_features |= netdev->features;
3954 	netdev->vlan_features |= netdev->features;
3955 }
3956 
3957 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
3958 				     struct ena_com_dev_get_features_ctx *feat)
3959 {
3960 	struct net_device *netdev = adapter->netdev;
3961 
3962 	/* Copy mac address */
3963 	if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
3964 		eth_hw_addr_random(netdev);
3965 		ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
3966 	} else {
3967 		ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
3968 		ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
3969 	}
3970 
3971 	/* Set offload features */
3972 	ena_set_dev_offloads(feat, netdev);
3973 
3974 	adapter->max_mtu = feat->dev_attr.max_mtu;
3975 	netdev->max_mtu = adapter->max_mtu;
3976 	netdev->min_mtu = ENA_MIN_MTU;
3977 }
3978 
3979 static int ena_rss_init_default(struct ena_adapter *adapter)
3980 {
3981 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3982 	struct device *dev = &adapter->pdev->dev;
3983 	int rc, i;
3984 	u32 val;
3985 
3986 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3987 	if (unlikely(rc)) {
3988 		dev_err(dev, "Cannot init indirect table\n");
3989 		goto err_rss_init;
3990 	}
3991 
3992 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3993 		val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
3994 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3995 						       ENA_IO_RXQ_IDX(val));
3996 		if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3997 			dev_err(dev, "Cannot fill indirect table\n");
3998 			goto err_fill_indir;
3999 		}
4000 	}
4001 
4002 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4003 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4004 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4005 		dev_err(dev, "Cannot fill hash function\n");
4006 		goto err_fill_indir;
4007 	}
4008 
4009 	rc = ena_com_set_default_hash_ctrl(ena_dev);
4010 	if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4011 		dev_err(dev, "Cannot fill hash control\n");
4012 		goto err_fill_indir;
4013 	}
4014 
4015 	return 0;
4016 
4017 err_fill_indir:
4018 	ena_com_rss_destroy(ena_dev);
4019 err_rss_init:
4020 
4021 	return rc;
4022 }
4023 
4024 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4025 {
4026 	int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4027 
4028 	pci_release_selected_regions(pdev, release_bars);
4029 }
4030 
4031 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
4032 {
4033 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
4034 	llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
4035 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
4036 	llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
4037 	llq_config->llq_ring_entry_size_value = 128;
4038 }
4039 
4040 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4041 {
4042 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4043 	struct ena_com_dev *ena_dev = ctx->ena_dev;
4044 	u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4045 	u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4046 	u32 max_tx_queue_size;
4047 	u32 max_rx_queue_size;
4048 
4049 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4050 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4051 			&ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4052 		max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4053 					  max_queue_ext->max_rx_sq_depth);
4054 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4055 
4056 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4057 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4058 						  llq->max_llq_depth);
4059 		else
4060 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4061 						  max_queue_ext->max_tx_sq_depth);
4062 
4063 		ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4064 					     max_queue_ext->max_per_packet_tx_descs);
4065 		ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4066 					     max_queue_ext->max_per_packet_rx_descs);
4067 	} else {
4068 		struct ena_admin_queue_feature_desc *max_queues =
4069 			&ctx->get_feat_ctx->max_queues;
4070 		max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4071 					  max_queues->max_sq_depth);
4072 		max_tx_queue_size = max_queues->max_cq_depth;
4073 
4074 		if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4075 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4076 						  llq->max_llq_depth);
4077 		else
4078 			max_tx_queue_size = min_t(u32, max_tx_queue_size,
4079 						  max_queues->max_sq_depth);
4080 
4081 		ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4082 					     max_queues->max_packet_tx_descs);
4083 		ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4084 					     max_queues->max_packet_rx_descs);
4085 	}
4086 
4087 	max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4088 	max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4089 
4090 	tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4091 				  max_tx_queue_size);
4092 	rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4093 				  max_rx_queue_size);
4094 
4095 	tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4096 	rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4097 
4098 	ctx->max_tx_queue_size = max_tx_queue_size;
4099 	ctx->max_rx_queue_size = max_rx_queue_size;
4100 	ctx->tx_queue_size = tx_queue_size;
4101 	ctx->rx_queue_size = rx_queue_size;
4102 
4103 	return 0;
4104 }
4105 
4106 /* ena_probe - Device Initialization Routine
4107  * @pdev: PCI device information struct
4108  * @ent: entry in ena_pci_tbl
4109  *
4110  * Returns 0 on success, negative on failure
4111  *
4112  * ena_probe initializes an adapter identified by a pci_dev structure.
4113  * The OS initialization, configuring of the adapter private structure,
4114  * and a hardware reset occur.
4115  */
4116 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4117 {
4118 	struct ena_com_dev_get_features_ctx get_feat_ctx;
4119 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
4120 	struct ena_llq_configurations llq_config;
4121 	struct ena_com_dev *ena_dev = NULL;
4122 	struct ena_adapter *adapter;
4123 	struct net_device *netdev;
4124 	static int adapters_found;
4125 	u32 max_num_io_queues;
4126 	char *queue_type_str;
4127 	bool wd_state;
4128 	int bars, rc;
4129 
4130 	dev_dbg(&pdev->dev, "%s\n", __func__);
4131 
4132 	rc = pci_enable_device_mem(pdev);
4133 	if (rc) {
4134 		dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4135 		return rc;
4136 	}
4137 
4138 	pci_set_master(pdev);
4139 
4140 	ena_dev = vzalloc(sizeof(*ena_dev));
4141 	if (!ena_dev) {
4142 		rc = -ENOMEM;
4143 		goto err_disable_device;
4144 	}
4145 
4146 	bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4147 	rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4148 	if (rc) {
4149 		dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4150 			rc);
4151 		goto err_free_ena_dev;
4152 	}
4153 
4154 	ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4155 					pci_resource_start(pdev, ENA_REG_BAR),
4156 					pci_resource_len(pdev, ENA_REG_BAR));
4157 	if (!ena_dev->reg_bar) {
4158 		dev_err(&pdev->dev, "failed to remap regs bar\n");
4159 		rc = -EFAULT;
4160 		goto err_free_region;
4161 	}
4162 
4163 	ena_dev->dmadev = &pdev->dev;
4164 
4165 	rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4166 	if (rc) {
4167 		dev_err(&pdev->dev, "ena device init failed\n");
4168 		if (rc == -ETIME)
4169 			rc = -EPROBE_DEFER;
4170 		goto err_free_region;
4171 	}
4172 
4173 	set_default_llq_configurations(&llq_config);
4174 
4175 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
4176 					     &llq_config);
4177 	if (rc) {
4178 		dev_err(&pdev->dev, "ena device init failed\n");
4179 		goto err_device_destroy;
4180 	}
4181 
4182 	calc_queue_ctx.ena_dev = ena_dev;
4183 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4184 	calc_queue_ctx.pdev = pdev;
4185 
4186 	/* Initial Tx and RX interrupt delay. Assumes 1 usec granularity.
4187 	 * Updated during device initialization with the real granularity
4188 	 */
4189 	ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4190 	ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4191 	ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4192 	max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4193 	rc = ena_calc_io_queue_size(&calc_queue_ctx);
4194 	if (rc || !max_num_io_queues) {
4195 		rc = -EFAULT;
4196 		goto err_device_destroy;
4197 	}
4198 
4199 	/* dev zeroed in init_etherdev */
4200 	netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), max_num_io_queues);
4201 	if (!netdev) {
4202 		dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4203 		rc = -ENOMEM;
4204 		goto err_device_destroy;
4205 	}
4206 
4207 	SET_NETDEV_DEV(netdev, &pdev->dev);
4208 
4209 	adapter = netdev_priv(netdev);
4210 	pci_set_drvdata(pdev, adapter);
4211 
4212 	adapter->ena_dev = ena_dev;
4213 	adapter->netdev = netdev;
4214 	adapter->pdev = pdev;
4215 
4216 	ena_set_conf_feat_params(adapter, &get_feat_ctx);
4217 
4218 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4219 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4220 
4221 	adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4222 	adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4223 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4224 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4225 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4226 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4227 
4228 	adapter->num_io_queues = max_num_io_queues;
4229 	adapter->max_num_io_queues = max_num_io_queues;
4230 
4231 	adapter->xdp_first_ring = 0;
4232 	adapter->xdp_num_queues = 0;
4233 
4234 	adapter->last_monitored_tx_qid = 0;
4235 
4236 	adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4237 	adapter->wd_state = wd_state;
4238 
4239 	snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4240 
4241 	rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4242 	if (rc) {
4243 		dev_err(&pdev->dev,
4244 			"Failed to query interrupt moderation feature\n");
4245 		goto err_netdev_destroy;
4246 	}
4247 	ena_init_io_rings(adapter,
4248 			  0,
4249 			  adapter->xdp_num_queues +
4250 			  adapter->num_io_queues);
4251 
4252 	netdev->netdev_ops = &ena_netdev_ops;
4253 	netdev->watchdog_timeo = TX_TIMEOUT;
4254 	ena_set_ethtool_ops(netdev);
4255 
4256 	netdev->priv_flags |= IFF_UNICAST_FLT;
4257 
4258 	u64_stats_init(&adapter->syncp);
4259 
4260 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4261 	if (rc) {
4262 		dev_err(&pdev->dev,
4263 			"Failed to enable and set the admin interrupts\n");
4264 		goto err_worker_destroy;
4265 	}
4266 	rc = ena_rss_init_default(adapter);
4267 	if (rc && (rc != -EOPNOTSUPP)) {
4268 		dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4269 		goto err_free_msix;
4270 	}
4271 
4272 	ena_config_debug_area(adapter);
4273 
4274 	memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4275 
4276 	netif_carrier_off(netdev);
4277 
4278 	rc = register_netdev(netdev);
4279 	if (rc) {
4280 		dev_err(&pdev->dev, "Cannot register net device\n");
4281 		goto err_rss;
4282 	}
4283 
4284 	INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4285 
4286 	adapter->last_keep_alive_jiffies = jiffies;
4287 	adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4288 	adapter->missing_tx_completion_to = TX_TIMEOUT;
4289 	adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4290 
4291 	ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4292 
4293 	timer_setup(&adapter->timer_service, ena_timer_service, 0);
4294 	mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4295 
4296 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
4297 		queue_type_str = "Regular";
4298 	else
4299 		queue_type_str = "Low Latency";
4300 
4301 	dev_info(&pdev->dev,
4302 		 "%s found at mem %lx, mac addr %pM, Placement policy: %s\n",
4303 		 DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4304 		 netdev->dev_addr, queue_type_str);
4305 
4306 	set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4307 
4308 	adapters_found++;
4309 
4310 	return 0;
4311 
4312 err_rss:
4313 	ena_com_delete_debug_area(ena_dev);
4314 	ena_com_rss_destroy(ena_dev);
4315 err_free_msix:
4316 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4317 	/* stop submitting admin commands on a device that was reset */
4318 	ena_com_set_admin_running_state(ena_dev, false);
4319 	ena_free_mgmnt_irq(adapter);
4320 	ena_disable_msix(adapter);
4321 err_worker_destroy:
4322 	del_timer(&adapter->timer_service);
4323 err_netdev_destroy:
4324 	free_netdev(netdev);
4325 err_device_destroy:
4326 	ena_com_delete_host_info(ena_dev);
4327 	ena_com_admin_destroy(ena_dev);
4328 err_free_region:
4329 	ena_release_bars(ena_dev, pdev);
4330 err_free_ena_dev:
4331 	vfree(ena_dev);
4332 err_disable_device:
4333 	pci_disable_device(pdev);
4334 	return rc;
4335 }
4336 
4337 /*****************************************************************************/
4338 
4339 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4340  * @pdev: PCI device information struct
4341  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4342  *
4343  * __ena_shutoff is a helper routine that does the real work on shutdown and
4344  * removal paths; the difference between those paths is with regards to whether
4345  * dettach or unregister the netdevice.
4346  */
4347 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4348 {
4349 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4350 	struct ena_com_dev *ena_dev;
4351 	struct net_device *netdev;
4352 
4353 	ena_dev = adapter->ena_dev;
4354 	netdev = adapter->netdev;
4355 
4356 #ifdef CONFIG_RFS_ACCEL
4357 	if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4358 		free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4359 		netdev->rx_cpu_rmap = NULL;
4360 	}
4361 #endif /* CONFIG_RFS_ACCEL */
4362 	del_timer_sync(&adapter->timer_service);
4363 
4364 	cancel_work_sync(&adapter->reset_task);
4365 
4366 	rtnl_lock(); /* lock released inside the below if-else block */
4367 	adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4368 	ena_destroy_device(adapter, true);
4369 	if (shutdown) {
4370 		netif_device_detach(netdev);
4371 		dev_close(netdev);
4372 		rtnl_unlock();
4373 	} else {
4374 		rtnl_unlock();
4375 		unregister_netdev(netdev);
4376 		free_netdev(netdev);
4377 	}
4378 
4379 	ena_com_rss_destroy(ena_dev);
4380 
4381 	ena_com_delete_debug_area(ena_dev);
4382 
4383 	ena_com_delete_host_info(ena_dev);
4384 
4385 	ena_release_bars(ena_dev, pdev);
4386 
4387 	pci_disable_device(pdev);
4388 
4389 	vfree(ena_dev);
4390 }
4391 
4392 /* ena_remove - Device Removal Routine
4393  * @pdev: PCI device information struct
4394  *
4395  * ena_remove is called by the PCI subsystem to alert the driver
4396  * that it should release a PCI device.
4397  */
4398 
4399 static void ena_remove(struct pci_dev *pdev)
4400 {
4401 	__ena_shutoff(pdev, false);
4402 }
4403 
4404 /* ena_shutdown - Device Shutdown Routine
4405  * @pdev: PCI device information struct
4406  *
4407  * ena_shutdown is called by the PCI subsystem to alert the driver that
4408  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4409  */
4410 
4411 static void ena_shutdown(struct pci_dev *pdev)
4412 {
4413 	__ena_shutoff(pdev, true);
4414 }
4415 
4416 #ifdef CONFIG_PM
4417 /* ena_suspend - PM suspend callback
4418  * @pdev: PCI device information struct
4419  * @state:power state
4420  */
4421 static int ena_suspend(struct pci_dev *pdev,  pm_message_t state)
4422 {
4423 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4424 
4425 	u64_stats_update_begin(&adapter->syncp);
4426 	adapter->dev_stats.suspend++;
4427 	u64_stats_update_end(&adapter->syncp);
4428 
4429 	rtnl_lock();
4430 	if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4431 		dev_err(&pdev->dev,
4432 			"ignoring device reset request as the device is being suspended\n");
4433 		clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4434 	}
4435 	ena_destroy_device(adapter, true);
4436 	rtnl_unlock();
4437 	return 0;
4438 }
4439 
4440 /* ena_resume - PM resume callback
4441  * @pdev: PCI device information struct
4442  *
4443  */
4444 static int ena_resume(struct pci_dev *pdev)
4445 {
4446 	struct ena_adapter *adapter = pci_get_drvdata(pdev);
4447 	int rc;
4448 
4449 	u64_stats_update_begin(&adapter->syncp);
4450 	adapter->dev_stats.resume++;
4451 	u64_stats_update_end(&adapter->syncp);
4452 
4453 	rtnl_lock();
4454 	rc = ena_restore_device(adapter);
4455 	rtnl_unlock();
4456 	return rc;
4457 }
4458 #endif
4459 
4460 static struct pci_driver ena_pci_driver = {
4461 	.name		= DRV_MODULE_NAME,
4462 	.id_table	= ena_pci_tbl,
4463 	.probe		= ena_probe,
4464 	.remove		= ena_remove,
4465 	.shutdown	= ena_shutdown,
4466 #ifdef CONFIG_PM
4467 	.suspend    = ena_suspend,
4468 	.resume     = ena_resume,
4469 #endif
4470 	.sriov_configure = pci_sriov_configure_simple,
4471 };
4472 
4473 static int __init ena_init(void)
4474 {
4475 	ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4476 	if (!ena_wq) {
4477 		pr_err("Failed to create workqueue\n");
4478 		return -ENOMEM;
4479 	}
4480 
4481 	return pci_register_driver(&ena_pci_driver);
4482 }
4483 
4484 static void __exit ena_cleanup(void)
4485 {
4486 	pci_unregister_driver(&ena_pci_driver);
4487 
4488 	if (ena_wq) {
4489 		destroy_workqueue(ena_wq);
4490 		ena_wq = NULL;
4491 	}
4492 }
4493 
4494 /******************************************************************************
4495  ******************************** AENQ Handlers *******************************
4496  *****************************************************************************/
4497 /* ena_update_on_link_change:
4498  * Notify the network interface about the change in link status
4499  */
4500 static void ena_update_on_link_change(void *adapter_data,
4501 				      struct ena_admin_aenq_entry *aenq_e)
4502 {
4503 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4504 	struct ena_admin_aenq_link_change_desc *aenq_desc =
4505 		(struct ena_admin_aenq_link_change_desc *)aenq_e;
4506 	int status = aenq_desc->flags &
4507 		ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4508 
4509 	if (status) {
4510 		netdev_dbg(adapter->netdev, "%s\n", __func__);
4511 		set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4512 		if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4513 			netif_carrier_on(adapter->netdev);
4514 	} else {
4515 		clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4516 		netif_carrier_off(adapter->netdev);
4517 	}
4518 }
4519 
4520 static void ena_keep_alive_wd(void *adapter_data,
4521 			      struct ena_admin_aenq_entry *aenq_e)
4522 {
4523 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4524 	struct ena_admin_aenq_keep_alive_desc *desc;
4525 	u64 rx_drops;
4526 	u64 tx_drops;
4527 
4528 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4529 	adapter->last_keep_alive_jiffies = jiffies;
4530 
4531 	rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4532 	tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4533 
4534 	u64_stats_update_begin(&adapter->syncp);
4535 	adapter->dev_stats.rx_drops = rx_drops;
4536 	adapter->dev_stats.tx_drops = tx_drops;
4537 	u64_stats_update_end(&adapter->syncp);
4538 }
4539 
4540 static void ena_notification(void *adapter_data,
4541 			     struct ena_admin_aenq_entry *aenq_e)
4542 {
4543 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4544 	struct ena_admin_ena_hw_hints *hints;
4545 
4546 	WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4547 	     "Invalid group(%x) expected %x\n",
4548 	     aenq_e->aenq_common_desc.group,
4549 	     ENA_ADMIN_NOTIFICATION);
4550 
4551 	switch (aenq_e->aenq_common_desc.syndrom) {
4552 	case ENA_ADMIN_UPDATE_HINTS:
4553 		hints = (struct ena_admin_ena_hw_hints *)
4554 			(&aenq_e->inline_data_w4);
4555 		ena_update_hints(adapter, hints);
4556 		break;
4557 	default:
4558 		netif_err(adapter, drv, adapter->netdev,
4559 			  "Invalid aenq notification link state %d\n",
4560 			  aenq_e->aenq_common_desc.syndrom);
4561 	}
4562 }
4563 
4564 /* This handler will called for unknown event group or unimplemented handlers*/
4565 static void unimplemented_aenq_handler(void *data,
4566 				       struct ena_admin_aenq_entry *aenq_e)
4567 {
4568 	struct ena_adapter *adapter = (struct ena_adapter *)data;
4569 
4570 	netif_err(adapter, drv, adapter->netdev,
4571 		  "Unknown event was received or event with unimplemented handler\n");
4572 }
4573 
4574 static struct ena_aenq_handlers aenq_handlers = {
4575 	.handlers = {
4576 		[ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4577 		[ENA_ADMIN_NOTIFICATION] = ena_notification,
4578 		[ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4579 	},
4580 	.unimplemented_handler = unimplemented_aenq_handler
4581 };
4582 
4583 module_init(ena_init);
4584 module_exit(ena_cleanup);
4585