xref: /freebsd/sys/dev/ena/ena.c (revision b9e5884ef786a6c0fd203064baa07ef308025707)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2015-2021 Amazon.com, Inc. or its affiliates.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  *
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
23  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
24  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
28  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_rss.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/bus.h>
38 #include <sys/endian.h>
39 #include <sys/kernel.h>
40 #include <sys/kthread.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/module.h>
44 #include <sys/rman.h>
45 #include <sys/smp.h>
46 #include <sys/socket.h>
47 #include <sys/sockio.h>
48 #include <sys/sysctl.h>
49 #include <sys/taskqueue.h>
50 #include <sys/time.h>
51 #include <sys/eventhandler.h>
52 
53 #include <machine/bus.h>
54 #include <machine/resource.h>
55 #include <machine/in_cksum.h>
56 
57 #include <net/bpf.h>
58 #include <net/ethernet.h>
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_arp.h>
62 #include <net/if_dl.h>
63 #include <net/if_media.h>
64 #include <net/if_types.h>
65 #include <net/if_vlan_var.h>
66 
67 #include <netinet/in_systm.h>
68 #include <netinet/in.h>
69 #include <netinet/if_ether.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet/tcp.h>
73 #include <netinet/udp.h>
74 
75 #include <dev/pci/pcivar.h>
76 #include <dev/pci/pcireg.h>
77 
78 #include <vm/vm.h>
79 #include <vm/pmap.h>
80 
81 #include "ena_datapath.h"
82 #include "ena.h"
83 #include "ena_sysctl.h"
84 #include "ena_rss.h"
85 
86 #ifdef DEV_NETMAP
87 #include "ena_netmap.h"
88 #endif /* DEV_NETMAP */
89 
90 /*********************************************************
91  *  Function prototypes
92  *********************************************************/
93 static int	ena_probe(device_t);
94 static void	ena_intr_msix_mgmnt(void *);
95 static void	ena_free_pci_resources(struct ena_adapter *);
96 static int	ena_change_mtu(if_t, int);
97 static inline void ena_alloc_counters(counter_u64_t *, int);
98 static inline void ena_free_counters(counter_u64_t *, int);
99 static inline void ena_reset_counters(counter_u64_t *, int);
100 static void	ena_init_io_rings_common(struct ena_adapter *,
101     struct ena_ring *, uint16_t);
102 static void	ena_init_io_rings_basic(struct ena_adapter *);
103 static void	ena_init_io_rings_advanced(struct ena_adapter *);
104 static void	ena_init_io_rings(struct ena_adapter *);
105 static void	ena_free_io_ring_resources(struct ena_adapter *, unsigned int);
106 static void	ena_free_all_io_rings_resources(struct ena_adapter *);
107 static int	ena_setup_tx_dma_tag(struct ena_adapter *);
108 static int	ena_free_tx_dma_tag(struct ena_adapter *);
109 static int	ena_setup_rx_dma_tag(struct ena_adapter *);
110 static int	ena_free_rx_dma_tag(struct ena_adapter *);
111 static void	ena_release_all_tx_dmamap(struct ena_ring *);
112 static int	ena_setup_tx_resources(struct ena_adapter *, int);
113 static void	ena_free_tx_resources(struct ena_adapter *, int);
114 static int	ena_setup_all_tx_resources(struct ena_adapter *);
115 static void	ena_free_all_tx_resources(struct ena_adapter *);
116 static int	ena_setup_rx_resources(struct ena_adapter *, unsigned int);
117 static void	ena_free_rx_resources(struct ena_adapter *, unsigned int);
118 static int	ena_setup_all_rx_resources(struct ena_adapter *);
119 static void	ena_free_all_rx_resources(struct ena_adapter *);
120 static inline int ena_alloc_rx_mbuf(struct ena_adapter *, struct ena_ring *,
121     struct ena_rx_buffer *);
122 static void	ena_free_rx_mbuf(struct ena_adapter *, struct ena_ring *,
123     struct ena_rx_buffer *);
124 static void	ena_free_rx_bufs(struct ena_adapter *, unsigned int);
125 static void	ena_refill_all_rx_bufs(struct ena_adapter *);
126 static void	ena_free_all_rx_bufs(struct ena_adapter *);
127 static void	ena_free_tx_bufs(struct ena_adapter *, unsigned int);
128 static void	ena_free_all_tx_bufs(struct ena_adapter *);
129 static void	ena_destroy_all_tx_queues(struct ena_adapter *);
130 static void	ena_destroy_all_rx_queues(struct ena_adapter *);
131 static void	ena_destroy_all_io_queues(struct ena_adapter *);
132 static int	ena_create_io_queues(struct ena_adapter *);
133 static int	ena_handle_msix(void *);
134 static int	ena_enable_msix(struct ena_adapter *);
135 static void	ena_setup_mgmnt_intr(struct ena_adapter *);
136 static int	ena_setup_io_intr(struct ena_adapter *);
137 static int	ena_request_mgmnt_irq(struct ena_adapter *);
138 static int	ena_request_io_irq(struct ena_adapter *);
139 static void	ena_free_mgmnt_irq(struct ena_adapter *);
140 static void	ena_free_io_irq(struct ena_adapter *);
141 static void	ena_free_irqs(struct ena_adapter*);
142 static void	ena_disable_msix(struct ena_adapter *);
143 static void	ena_unmask_all_io_irqs(struct ena_adapter *);
144 static int	ena_up_complete(struct ena_adapter *);
145 static uint64_t	ena_get_counter(if_t, ift_counter);
146 static int	ena_media_change(if_t);
147 static void	ena_media_status(if_t, struct ifmediareq *);
148 static void	ena_init(void *);
149 static int	ena_ioctl(if_t, u_long, caddr_t);
150 static int	ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *);
151 static void	ena_update_host_info(struct ena_admin_host_info *, if_t);
152 static void	ena_update_hwassist(struct ena_adapter *);
153 static int	ena_setup_ifnet(device_t, struct ena_adapter *,
154     struct ena_com_dev_get_features_ctx *);
155 static int	ena_enable_wc(device_t, struct resource *);
156 static int	ena_set_queues_placement_policy(device_t, struct ena_com_dev *,
157     struct ena_admin_feature_llq_desc *, struct ena_llq_configurations *);
158 static uint32_t	ena_calc_max_io_queue_num(device_t, struct ena_com_dev *,
159     struct ena_com_dev_get_features_ctx *);
160 static int	ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *);
161 static void	ena_config_host_info(struct ena_com_dev *, device_t);
162 static int	ena_attach(device_t);
163 static int	ena_detach(device_t);
164 static int	ena_device_init(struct ena_adapter *, device_t,
165     struct ena_com_dev_get_features_ctx *, int *);
166 static int	ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *);
167 static void ena_update_on_link_change(void *, struct ena_admin_aenq_entry *);
168 static void	unimplemented_aenq_handler(void *,
169     struct ena_admin_aenq_entry *);
170 static int	ena_copy_eni_metrics(struct ena_adapter *);
171 static void	ena_timer_service(void *);
172 
173 static char ena_version[] = DEVICE_NAME DRV_MODULE_NAME " v" DRV_MODULE_VERSION;
174 
175 static ena_vendor_info_t ena_vendor_info_array[] = {
176     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0},
177     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF_RSERV0, 0},
178     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0},
179     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF_RSERV0, 0},
180     /* Last entry */
181     { 0, 0, 0 }
182 };
183 
184 struct sx ena_global_lock;
185 
186 /*
187  * Contains pointers to event handlers, e.g. link state chage.
188  */
189 static struct ena_aenq_handlers aenq_handlers;
190 
191 void
192 ena_dmamap_callback(void *arg, bus_dma_segment_t *segs, int nseg, int error)
193 {
194 	if (error != 0)
195 		return;
196 	*(bus_addr_t *) arg = segs[0].ds_addr;
197 }
198 
199 int
200 ena_dma_alloc(device_t dmadev, bus_size_t size,
201     ena_mem_handle_t *dma, int mapflags, bus_size_t alignment, int domain)
202 {
203 	struct ena_adapter* adapter = device_get_softc(dmadev);
204 	device_t pdev = adapter->pdev;
205 	uint32_t maxsize;
206 	uint64_t dma_space_addr;
207 	int error;
208 
209 	maxsize = ((size - 1) / PAGE_SIZE + 1) * PAGE_SIZE;
210 
211 	dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
212 	if (unlikely(dma_space_addr == 0))
213 		dma_space_addr = BUS_SPACE_MAXADDR;
214 
215 	error = bus_dma_tag_create(bus_get_dma_tag(dmadev), /* parent */
216 	    alignment, 0,     /* alignment, bounds 		*/
217 	    dma_space_addr,   /* lowaddr of exclusion window	*/
218 	    BUS_SPACE_MAXADDR,/* highaddr of exclusion window	*/
219 	    NULL, NULL,	      /* filter, filterarg 		*/
220 	    maxsize,	      /* maxsize 			*/
221 	    1,		      /* nsegments 			*/
222 	    maxsize,	      /* maxsegsize 			*/
223 	    BUS_DMA_ALLOCNOW, /* flags 				*/
224 	    NULL,	      /* lockfunc 			*/
225 	    NULL,	      /* lockarg 			*/
226 	    &dma->tag);
227 	if (unlikely(error != 0)) {
228 		ena_log(pdev, ERR, "bus_dma_tag_create failed: %d\n", error);
229 		goto fail_tag;
230 	}
231 
232 	error = bus_dma_tag_set_domain(dma->tag, domain);
233 	if (unlikely(error != 0)) {
234 		ena_log(pdev, ERR, "bus_dma_tag_set_domain failed: %d\n",
235 		    error);
236 		goto fail_map_create;
237 	}
238 
239 	error = bus_dmamem_alloc(dma->tag, (void**) &dma->vaddr,
240 	    BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->map);
241 	if (unlikely(error != 0)) {
242 		ena_log(pdev, ERR, "bus_dmamem_alloc(%ju) failed: %d\n",
243 		    (uintmax_t)size, error);
244 		goto fail_map_create;
245 	}
246 
247 	dma->paddr = 0;
248 	error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr,
249 	    size, ena_dmamap_callback, &dma->paddr, mapflags);
250 	if (unlikely((error != 0) || (dma->paddr == 0))) {
251 		ena_log(pdev, ERR, "bus_dmamap_load failed: %d\n", error);
252 		goto fail_map_load;
253 	}
254 
255 	bus_dmamap_sync(dma->tag, dma->map,
256 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
257 
258 	return (0);
259 
260 fail_map_load:
261 	bus_dmamem_free(dma->tag, dma->vaddr, dma->map);
262 fail_map_create:
263 	bus_dma_tag_destroy(dma->tag);
264 fail_tag:
265 	dma->tag = NULL;
266 	dma->vaddr = NULL;
267 	dma->paddr = 0;
268 
269 	return (error);
270 }
271 
272 static void
273 ena_free_pci_resources(struct ena_adapter *adapter)
274 {
275 	device_t pdev = adapter->pdev;
276 
277 	if (adapter->memory != NULL) {
278 		bus_release_resource(pdev, SYS_RES_MEMORY,
279 		    PCIR_BAR(ENA_MEM_BAR), adapter->memory);
280 	}
281 
282 	if (adapter->registers != NULL) {
283 		bus_release_resource(pdev, SYS_RES_MEMORY,
284 		    PCIR_BAR(ENA_REG_BAR), adapter->registers);
285 	}
286 
287 	if (adapter->msix != NULL) {
288 		bus_release_resource(pdev, SYS_RES_MEMORY,
289 		    adapter->msix_rid, adapter->msix);
290 	}
291 }
292 
293 static int
294 ena_probe(device_t dev)
295 {
296 	ena_vendor_info_t *ent;
297 	char		adapter_name[60];
298 	uint16_t	pci_vendor_id = 0;
299 	uint16_t	pci_device_id = 0;
300 
301 	pci_vendor_id = pci_get_vendor(dev);
302 	pci_device_id = pci_get_device(dev);
303 
304 	ent = ena_vendor_info_array;
305 	while (ent->vendor_id != 0) {
306 		if ((pci_vendor_id == ent->vendor_id) &&
307 		    (pci_device_id == ent->device_id)) {
308 			ena_log_raw(DBG, "vendor=%x device=%x\n",
309 			    pci_vendor_id, pci_device_id);
310 
311 			sprintf(adapter_name, DEVICE_DESC);
312 			device_set_desc_copy(dev, adapter_name);
313 			return (BUS_PROBE_DEFAULT);
314 		}
315 
316 		ent++;
317 
318 	}
319 
320 	return (ENXIO);
321 }
322 
323 static int
324 ena_change_mtu(if_t ifp, int new_mtu)
325 {
326 	struct ena_adapter *adapter = if_getsoftc(ifp);
327 	device_t pdev = adapter->pdev;
328 	int rc;
329 
330 	if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
331 		ena_log(pdev, ERR, "Invalid MTU setting. "
332 		    "new_mtu: %d max mtu: %d min mtu: %d\n",
333 		    new_mtu, adapter->max_mtu, ENA_MIN_MTU);
334 		return (EINVAL);
335 	}
336 
337 	rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
338 	if (likely(rc == 0)) {
339 		ena_log(pdev, DBG, "set MTU to %d\n", new_mtu);
340 		if_setmtu(ifp, new_mtu);
341 	} else {
342 		ena_log(pdev, ERR, "Failed to set MTU to %d\n", new_mtu);
343 	}
344 
345 	return (rc);
346 }
347 
348 static inline void
349 ena_alloc_counters(counter_u64_t *begin, int size)
350 {
351 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
352 
353 	for (; begin < end; ++begin)
354 		*begin = counter_u64_alloc(M_WAITOK);
355 }
356 
357 static inline void
358 ena_free_counters(counter_u64_t *begin, int size)
359 {
360 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
361 
362 	for (; begin < end; ++begin)
363 		counter_u64_free(*begin);
364 }
365 
366 static inline void
367 ena_reset_counters(counter_u64_t *begin, int size)
368 {
369 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
370 
371 	for (; begin < end; ++begin)
372 		counter_u64_zero(*begin);
373 }
374 
375 static void
376 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
377     uint16_t qid)
378 {
379 
380 	ring->qid = qid;
381 	ring->adapter = adapter;
382 	ring->ena_dev = adapter->ena_dev;
383 	ring->first_interrupt = false;
384 	ring->no_interrupt_event_cnt = 0;
385 }
386 
387 static void
388 ena_init_io_rings_basic(struct ena_adapter *adapter)
389 {
390 	struct ena_com_dev *ena_dev;
391 	struct ena_ring *txr, *rxr;
392 	struct ena_que *que;
393 	int i;
394 
395 	ena_dev = adapter->ena_dev;
396 
397 	for (i = 0; i < adapter->num_io_queues; i++) {
398 		txr = &adapter->tx_ring[i];
399 		rxr = &adapter->rx_ring[i];
400 
401 		/* TX/RX common ring state */
402 		ena_init_io_rings_common(adapter, txr, i);
403 		ena_init_io_rings_common(adapter, rxr, i);
404 
405 		/* TX specific ring state */
406 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
407 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
408 
409 		que = &adapter->que[i];
410 		que->adapter = adapter;
411 		que->id = i;
412 		que->tx_ring = txr;
413 		que->rx_ring = rxr;
414 
415 		txr->que = que;
416 		rxr->que = que;
417 
418 		rxr->empty_rx_queue = 0;
419 		rxr->rx_mbuf_sz = ena_mbuf_sz;
420 	}
421 }
422 
423 static void
424 ena_init_io_rings_advanced(struct ena_adapter *adapter)
425 {
426 	struct ena_ring *txr, *rxr;
427 	int i;
428 
429 	for (i = 0; i < adapter->num_io_queues; i++) {
430 		txr = &adapter->tx_ring[i];
431 		rxr = &adapter->rx_ring[i];
432 
433 		/* Allocate a buf ring */
434 		txr->buf_ring_size = adapter->buf_ring_size;
435 		txr->br = buf_ring_alloc(txr->buf_ring_size, M_DEVBUF,
436 		    M_WAITOK, &txr->ring_mtx);
437 
438 		/* Allocate Tx statistics. */
439 		ena_alloc_counters((counter_u64_t *)&txr->tx_stats,
440 		    sizeof(txr->tx_stats));
441 
442 		/* Allocate Rx statistics. */
443 		ena_alloc_counters((counter_u64_t *)&rxr->rx_stats,
444 		    sizeof(rxr->rx_stats));
445 
446 		/* Initialize locks */
447 		snprintf(txr->mtx_name, nitems(txr->mtx_name), "%s:tx(%d)",
448 		    device_get_nameunit(adapter->pdev), i);
449 		snprintf(rxr->mtx_name, nitems(rxr->mtx_name), "%s:rx(%d)",
450 		    device_get_nameunit(adapter->pdev), i);
451 
452 		mtx_init(&txr->ring_mtx, txr->mtx_name, NULL, MTX_DEF);
453 	}
454 }
455 
456 static void
457 ena_init_io_rings(struct ena_adapter *adapter)
458 {
459 	/*
460 	 * IO rings initialization can be divided into the 2 steps:
461 	 *   1. Initialize variables and fields with initial values and copy
462 	 *      them from adapter/ena_dev (basic)
463 	 *   2. Allocate mutex, counters and buf_ring (advanced)
464 	 */
465 	ena_init_io_rings_basic(adapter);
466 	ena_init_io_rings_advanced(adapter);
467 }
468 
469 static void
470 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
471 {
472 	struct ena_ring *txr = &adapter->tx_ring[qid];
473 	struct ena_ring *rxr = &adapter->rx_ring[qid];
474 
475 	ena_free_counters((counter_u64_t *)&txr->tx_stats,
476 	    sizeof(txr->tx_stats));
477 	ena_free_counters((counter_u64_t *)&rxr->rx_stats,
478 	    sizeof(rxr->rx_stats));
479 
480 	ENA_RING_MTX_LOCK(txr);
481 	drbr_free(txr->br, M_DEVBUF);
482 	ENA_RING_MTX_UNLOCK(txr);
483 
484 	mtx_destroy(&txr->ring_mtx);
485 }
486 
487 static void
488 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
489 {
490 	int i;
491 
492 	for (i = 0; i < adapter->num_io_queues; i++)
493 		ena_free_io_ring_resources(adapter, i);
494 
495 }
496 
497 static int
498 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
499 {
500 	int ret;
501 
502 	/* Create DMA tag for Tx buffers */
503 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
504 	    1, 0,				  /* alignment, bounds 	     */
505 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
506 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
507 	    NULL, NULL,				  /* filter, filterarg 	     */
508 	    ENA_TSO_MAXSIZE,			  /* maxsize 		     */
509 	    adapter->max_tx_sgl_size - 1,	  /* nsegments 		     */
510 	    ENA_TSO_MAXSIZE,			  /* maxsegsize 	     */
511 	    0,					  /* flags 		     */
512 	    NULL,				  /* lockfunc 		     */
513 	    NULL,				  /* lockfuncarg 	     */
514 	    &adapter->tx_buf_tag);
515 
516 	return (ret);
517 }
518 
519 static int
520 ena_free_tx_dma_tag(struct ena_adapter *adapter)
521 {
522 	int ret;
523 
524 	ret = bus_dma_tag_destroy(adapter->tx_buf_tag);
525 
526 	if (likely(ret == 0))
527 		adapter->tx_buf_tag = NULL;
528 
529 	return (ret);
530 }
531 
532 static int
533 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
534 {
535 	int ret;
536 
537 	/* Create DMA tag for Rx buffers*/
538 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent   */
539 	    1, 0,				  /* alignment, bounds 	     */
540 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
541 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
542 	    NULL, NULL,				  /* filter, filterarg 	     */
543 	    ena_mbuf_sz,			  /* maxsize 		     */
544 	    adapter->max_rx_sgl_size,		  /* nsegments 		     */
545 	    ena_mbuf_sz,			  /* maxsegsize 	     */
546 	    0,					  /* flags 		     */
547 	    NULL,				  /* lockfunc 		     */
548 	    NULL,				  /* lockarg 		     */
549 	    &adapter->rx_buf_tag);
550 
551 	return (ret);
552 }
553 
554 static int
555 ena_free_rx_dma_tag(struct ena_adapter *adapter)
556 {
557 	int ret;
558 
559 	ret = bus_dma_tag_destroy(adapter->rx_buf_tag);
560 
561 	if (likely(ret == 0))
562 		adapter->rx_buf_tag = NULL;
563 
564 	return (ret);
565 }
566 
567 static void
568 ena_release_all_tx_dmamap(struct ena_ring *tx_ring)
569 {
570 	struct ena_adapter *adapter = tx_ring->adapter;
571 	struct ena_tx_buffer *tx_info;
572 	bus_dma_tag_t tx_tag = adapter->tx_buf_tag;;
573 	int i;
574 #ifdef DEV_NETMAP
575 	struct ena_netmap_tx_info *nm_info;
576 	int j;
577 #endif /* DEV_NETMAP */
578 
579 	for (i = 0; i < tx_ring->ring_size; ++i) {
580 		tx_info = &tx_ring->tx_buffer_info[i];
581 #ifdef DEV_NETMAP
582 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
583 			nm_info = &tx_info->nm_info;
584 			for (j = 0; j < ENA_PKT_MAX_BUFS; ++j) {
585 				if (nm_info->map_seg[j] != NULL) {
586 					bus_dmamap_destroy(tx_tag,
587 					    nm_info->map_seg[j]);
588 					nm_info->map_seg[j] = NULL;
589 				}
590 			}
591 		}
592 #endif /* DEV_NETMAP */
593 		if (tx_info->dmamap != NULL) {
594 			bus_dmamap_destroy(tx_tag, tx_info->dmamap);
595 			tx_info->dmamap = NULL;
596 		}
597 	}
598 }
599 
600 /**
601  * ena_setup_tx_resources - allocate Tx resources (Descriptors)
602  * @adapter: network interface device structure
603  * @qid: queue index
604  *
605  * Returns 0 on success, otherwise on failure.
606  **/
607 static int
608 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
609 {
610 	device_t pdev = adapter->pdev;
611 	char thread_name[MAXCOMLEN + 1];
612 	struct ena_que *que = &adapter->que[qid];
613 	struct ena_ring *tx_ring = que->tx_ring;
614 	cpuset_t *cpu_mask = NULL;
615 	int size, i, err;
616 #ifdef DEV_NETMAP
617 	bus_dmamap_t *map;
618 	int j;
619 
620 	ena_netmap_reset_tx_ring(adapter, qid);
621 #endif /* DEV_NETMAP */
622 
623 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
624 
625 	tx_ring->tx_buffer_info = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
626 	if (unlikely(tx_ring->tx_buffer_info == NULL))
627 		return (ENOMEM);
628 
629 	size = sizeof(uint16_t) * tx_ring->ring_size;
630 	tx_ring->free_tx_ids = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
631 	if (unlikely(tx_ring->free_tx_ids == NULL))
632 		goto err_buf_info_free;
633 
634 	size = tx_ring->tx_max_header_size;
635 	tx_ring->push_buf_intermediate_buf = malloc(size, M_DEVBUF,
636 	    M_NOWAIT | M_ZERO);
637 	if (unlikely(tx_ring->push_buf_intermediate_buf == NULL))
638 		goto err_tx_ids_free;
639 
640 	/* Req id stack for TX OOO completions */
641 	for (i = 0; i < tx_ring->ring_size; i++)
642 		tx_ring->free_tx_ids[i] = i;
643 
644 	/* Reset TX statistics. */
645 	ena_reset_counters((counter_u64_t *)&tx_ring->tx_stats,
646 	    sizeof(tx_ring->tx_stats));
647 
648 	tx_ring->next_to_use = 0;
649 	tx_ring->next_to_clean = 0;
650 	tx_ring->acum_pkts = 0;
651 
652 	/* Make sure that drbr is empty */
653 	ENA_RING_MTX_LOCK(tx_ring);
654 	drbr_flush(adapter->ifp, tx_ring->br);
655 	ENA_RING_MTX_UNLOCK(tx_ring);
656 
657 	/* ... and create the buffer DMA maps */
658 	for (i = 0; i < tx_ring->ring_size; i++) {
659 		err = bus_dmamap_create(adapter->tx_buf_tag, 0,
660 		    &tx_ring->tx_buffer_info[i].dmamap);
661 		if (unlikely(err != 0)) {
662 			ena_log(pdev, ERR,
663 			    "Unable to create Tx DMA map for buffer %d\n",
664 			    i);
665 			goto err_map_release;
666 		}
667 
668 #ifdef DEV_NETMAP
669 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
670 			map = tx_ring->tx_buffer_info[i].nm_info.map_seg;
671 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
672 				err = bus_dmamap_create(adapter->tx_buf_tag, 0,
673 				    &map[j]);
674 				if (unlikely(err != 0)) {
675 					ena_log(pdev, ERR,
676 					    "Unable to create "
677 					    "Tx DMA for buffer %d %d\n", i, j);
678 					goto err_map_release;
679 				}
680 			}
681 		}
682 #endif /* DEV_NETMAP */
683 	}
684 
685 	/* Allocate taskqueues */
686 	TASK_INIT(&tx_ring->enqueue_task, 0, ena_deferred_mq_start, tx_ring);
687 	tx_ring->enqueue_tq = taskqueue_create_fast("ena_tx_enque", M_NOWAIT,
688 	    taskqueue_thread_enqueue, &tx_ring->enqueue_tq);
689 	if (unlikely(tx_ring->enqueue_tq == NULL)) {
690 		ena_log(pdev, ERR,
691 		    "Unable to create taskqueue for enqueue task\n");
692 		i = tx_ring->ring_size;
693 		goto err_map_release;
694 	}
695 
696 	tx_ring->running = true;
697 
698 #ifdef RSS
699 	cpu_mask = &que->cpu_mask;
700 	snprintf(thread_name, sizeof(thread_name), "%s txeq %d",
701 	    device_get_nameunit(adapter->pdev), que->cpu);
702 #else
703 	snprintf(thread_name, sizeof(thread_name), "%s txeq %d",
704 	    device_get_nameunit(adapter->pdev), que->id);
705 #endif
706 	taskqueue_start_threads_cpuset(&tx_ring->enqueue_tq, 1, PI_NET,
707 	    cpu_mask, "%s", thread_name);
708 
709 	return (0);
710 
711 err_map_release:
712 	ena_release_all_tx_dmamap(tx_ring);
713 err_tx_ids_free:
714 	free(tx_ring->free_tx_ids, M_DEVBUF);
715 	tx_ring->free_tx_ids = NULL;
716 err_buf_info_free:
717 	free(tx_ring->tx_buffer_info, M_DEVBUF);
718 	tx_ring->tx_buffer_info = NULL;
719 
720 	return (ENOMEM);
721 }
722 
723 /**
724  * ena_free_tx_resources - Free Tx Resources per Queue
725  * @adapter: network interface device structure
726  * @qid: queue index
727  *
728  * Free all transmit software resources
729  **/
730 static void
731 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
732 {
733 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
734 #ifdef DEV_NETMAP
735 	struct ena_netmap_tx_info *nm_info;
736 	int j;
737 #endif /* DEV_NETMAP */
738 
739 	while (taskqueue_cancel(tx_ring->enqueue_tq, &tx_ring->enqueue_task,
740 	    NULL))
741 		taskqueue_drain(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
742 
743 	taskqueue_free(tx_ring->enqueue_tq);
744 
745 	ENA_RING_MTX_LOCK(tx_ring);
746 	/* Flush buffer ring, */
747 	drbr_flush(adapter->ifp, tx_ring->br);
748 
749 	/* Free buffer DMA maps, */
750 	for (int i = 0; i < tx_ring->ring_size; i++) {
751 		bus_dmamap_sync(adapter->tx_buf_tag,
752 		    tx_ring->tx_buffer_info[i].dmamap, BUS_DMASYNC_POSTWRITE);
753 		bus_dmamap_unload(adapter->tx_buf_tag,
754 		    tx_ring->tx_buffer_info[i].dmamap);
755 		bus_dmamap_destroy(adapter->tx_buf_tag,
756 		    tx_ring->tx_buffer_info[i].dmamap);
757 
758 #ifdef DEV_NETMAP
759 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
760 			nm_info = &tx_ring->tx_buffer_info[i].nm_info;
761 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
762 				if (nm_info->socket_buf_idx[j] != 0) {
763 					bus_dmamap_sync(adapter->tx_buf_tag,
764 					    nm_info->map_seg[j],
765 					    BUS_DMASYNC_POSTWRITE);
766 					ena_netmap_unload(adapter,
767 					    nm_info->map_seg[j]);
768 				}
769 				bus_dmamap_destroy(adapter->tx_buf_tag,
770 				    nm_info->map_seg[j]);
771 				nm_info->socket_buf_idx[j] = 0;
772 			}
773 		}
774 #endif /* DEV_NETMAP */
775 
776 		m_freem(tx_ring->tx_buffer_info[i].mbuf);
777 		tx_ring->tx_buffer_info[i].mbuf = NULL;
778 	}
779 	ENA_RING_MTX_UNLOCK(tx_ring);
780 
781 	/* And free allocated memory. */
782 	free(tx_ring->tx_buffer_info, M_DEVBUF);
783 	tx_ring->tx_buffer_info = NULL;
784 
785 	free(tx_ring->free_tx_ids, M_DEVBUF);
786 	tx_ring->free_tx_ids = NULL;
787 
788 	free(tx_ring->push_buf_intermediate_buf, M_DEVBUF);
789 	tx_ring->push_buf_intermediate_buf = NULL;
790 }
791 
792 /**
793  * ena_setup_all_tx_resources - allocate all queues Tx resources
794  * @adapter: network interface device structure
795  *
796  * Returns 0 on success, otherwise on failure.
797  **/
798 static int
799 ena_setup_all_tx_resources(struct ena_adapter *adapter)
800 {
801 	int i, rc;
802 
803 	for (i = 0; i < adapter->num_io_queues; i++) {
804 		rc = ena_setup_tx_resources(adapter, i);
805 		if (rc != 0) {
806 			ena_log(adapter->pdev, ERR,
807 			    "Allocation for Tx Queue %u failed\n", i);
808 			goto err_setup_tx;
809 		}
810 	}
811 
812 	return (0);
813 
814 err_setup_tx:
815 	/* Rewind the index freeing the rings as we go */
816 	while (i--)
817 		ena_free_tx_resources(adapter, i);
818 	return (rc);
819 }
820 
821 /**
822  * ena_free_all_tx_resources - Free Tx Resources for All Queues
823  * @adapter: network interface device structure
824  *
825  * Free all transmit software resources
826  **/
827 static void
828 ena_free_all_tx_resources(struct ena_adapter *adapter)
829 {
830 	int i;
831 
832 	for (i = 0; i < adapter->num_io_queues; i++)
833 		ena_free_tx_resources(adapter, i);
834 }
835 
836 /**
837  * ena_setup_rx_resources - allocate Rx resources (Descriptors)
838  * @adapter: network interface device structure
839  * @qid: queue index
840  *
841  * Returns 0 on success, otherwise on failure.
842  **/
843 static int
844 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
845 {
846 	device_t pdev = adapter->pdev;
847 	struct ena_que *que = &adapter->que[qid];
848 	struct ena_ring *rx_ring = que->rx_ring;
849 	int size, err, i;
850 
851 	size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
852 
853 #ifdef DEV_NETMAP
854 	ena_netmap_reset_rx_ring(adapter, qid);
855 	rx_ring->initialized = false;
856 #endif /* DEV_NETMAP */
857 
858 	/*
859 	 * Alloc extra element so in rx path
860 	 * we can always prefetch rx_info + 1
861 	 */
862 	size += sizeof(struct ena_rx_buffer);
863 
864 	rx_ring->rx_buffer_info = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
865 
866 	size = sizeof(uint16_t) * rx_ring->ring_size;
867 	rx_ring->free_rx_ids = malloc(size, M_DEVBUF, M_WAITOK);
868 
869 	for (i = 0; i < rx_ring->ring_size; i++)
870 		rx_ring->free_rx_ids[i] = i;
871 
872 	/* Reset RX statistics. */
873 	ena_reset_counters((counter_u64_t *)&rx_ring->rx_stats,
874 	    sizeof(rx_ring->rx_stats));
875 
876 	rx_ring->next_to_clean = 0;
877 	rx_ring->next_to_use = 0;
878 
879 	/* ... and create the buffer DMA maps */
880 	for (i = 0; i < rx_ring->ring_size; i++) {
881 		err = bus_dmamap_create(adapter->rx_buf_tag, 0,
882 		    &(rx_ring->rx_buffer_info[i].map));
883 		if (err != 0) {
884 			ena_log(pdev, ERR,
885 			    "Unable to create Rx DMA map for buffer %d\n", i);
886 			goto err_buf_info_unmap;
887 		}
888 	}
889 
890 	/* Create LRO for the ring */
891 	if ((adapter->ifp->if_capenable & IFCAP_LRO) != 0) {
892 		int err = tcp_lro_init(&rx_ring->lro);
893 		if (err != 0) {
894 			ena_log(pdev, ERR, "LRO[%d] Initialization failed!\n",
895 			    qid);
896 		} else {
897 			ena_log(pdev, DBG, "RX Soft LRO[%d] Initialized\n",
898 			    qid);
899 			rx_ring->lro.ifp = adapter->ifp;
900 		}
901 	}
902 
903 	return (0);
904 
905 err_buf_info_unmap:
906 	while (i--) {
907 		bus_dmamap_destroy(adapter->rx_buf_tag,
908 		    rx_ring->rx_buffer_info[i].map);
909 	}
910 
911 	free(rx_ring->free_rx_ids, M_DEVBUF);
912 	rx_ring->free_rx_ids = NULL;
913 	free(rx_ring->rx_buffer_info, M_DEVBUF);
914 	rx_ring->rx_buffer_info = NULL;
915 	return (ENOMEM);
916 }
917 
918 /**
919  * ena_free_rx_resources - Free Rx Resources
920  * @adapter: network interface device structure
921  * @qid: queue index
922  *
923  * Free all receive software resources
924  **/
925 static void
926 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
927 {
928 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
929 
930 	/* Free buffer DMA maps, */
931 	for (int i = 0; i < rx_ring->ring_size; i++) {
932 		bus_dmamap_sync(adapter->rx_buf_tag,
933 		    rx_ring->rx_buffer_info[i].map, BUS_DMASYNC_POSTREAD);
934 		m_freem(rx_ring->rx_buffer_info[i].mbuf);
935 		rx_ring->rx_buffer_info[i].mbuf = NULL;
936 		bus_dmamap_unload(adapter->rx_buf_tag,
937 		    rx_ring->rx_buffer_info[i].map);
938 		bus_dmamap_destroy(adapter->rx_buf_tag,
939 		    rx_ring->rx_buffer_info[i].map);
940 	}
941 
942 	/* free LRO resources, */
943 	tcp_lro_free(&rx_ring->lro);
944 
945 	/* free allocated memory */
946 	free(rx_ring->rx_buffer_info, M_DEVBUF);
947 	rx_ring->rx_buffer_info = NULL;
948 
949 	free(rx_ring->free_rx_ids, M_DEVBUF);
950 	rx_ring->free_rx_ids = NULL;
951 }
952 
953 /**
954  * ena_setup_all_rx_resources - allocate all queues Rx resources
955  * @adapter: network interface device structure
956  *
957  * Returns 0 on success, otherwise on failure.
958  **/
959 static int
960 ena_setup_all_rx_resources(struct ena_adapter *adapter)
961 {
962 	int i, rc = 0;
963 
964 	for (i = 0; i < adapter->num_io_queues; i++) {
965 		rc = ena_setup_rx_resources(adapter, i);
966 		if (rc != 0) {
967 			ena_log(adapter->pdev, ERR,
968 			    "Allocation for Rx Queue %u failed\n", i);
969 			goto err_setup_rx;
970 		}
971 	}
972 	return (0);
973 
974 err_setup_rx:
975 	/* rewind the index freeing the rings as we go */
976 	while (i--)
977 		ena_free_rx_resources(adapter, i);
978 	return (rc);
979 }
980 
981 /**
982  * ena_free_all_rx_resources - Free Rx resources for all queues
983  * @adapter: network interface device structure
984  *
985  * Free all receive software resources
986  **/
987 static void
988 ena_free_all_rx_resources(struct ena_adapter *adapter)
989 {
990 	int i;
991 
992 	for (i = 0; i < adapter->num_io_queues; i++)
993 		ena_free_rx_resources(adapter, i);
994 }
995 
996 static inline int
997 ena_alloc_rx_mbuf(struct ena_adapter *adapter,
998     struct ena_ring *rx_ring, struct ena_rx_buffer *rx_info)
999 {
1000 	device_t pdev = adapter->pdev;
1001 	struct ena_com_buf *ena_buf;
1002 	bus_dma_segment_t segs[1];
1003 	int nsegs, error;
1004 	int mlen;
1005 
1006 	/* if previous allocated frag is not used */
1007 	if (unlikely(rx_info->mbuf != NULL))
1008 		return (0);
1009 
1010 	/* Get mbuf using UMA allocator */
1011 	rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR,
1012 	    rx_ring->rx_mbuf_sz);
1013 
1014 	if (unlikely(rx_info->mbuf == NULL)) {
1015 		counter_u64_add(rx_ring->rx_stats.mjum_alloc_fail, 1);
1016 		rx_info->mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1017 		if (unlikely(rx_info->mbuf == NULL)) {
1018 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
1019 			return (ENOMEM);
1020 		}
1021 		mlen = MCLBYTES;
1022 	} else {
1023 		mlen = rx_ring->rx_mbuf_sz;
1024 	}
1025 	/* Set mbuf length*/
1026 	rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = mlen;
1027 
1028 	/* Map packets for DMA */
1029 	ena_log(pdev, DBG, "Using tag %p for buffers' DMA mapping, mbuf %p len: %d\n",
1030 	    adapter->rx_buf_tag,rx_info->mbuf, rx_info->mbuf->m_len);
1031 	error = bus_dmamap_load_mbuf_sg(adapter->rx_buf_tag, rx_info->map,
1032 	    rx_info->mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
1033 	if (unlikely((error != 0) || (nsegs != 1))) {
1034 		ena_log(pdev, WARN,
1035 		    "failed to map mbuf, error: %d, nsegs: %d\n", error, nsegs);
1036 		counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
1037 		goto exit;
1038 
1039 	}
1040 
1041 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map, BUS_DMASYNC_PREREAD);
1042 
1043 	ena_buf = &rx_info->ena_buf;
1044 	ena_buf->paddr = segs[0].ds_addr;
1045 	ena_buf->len = mlen;
1046 
1047 	ena_log(pdev, DBG, "ALLOC RX BUF: mbuf %p, rx_info %p, len %d, paddr %#jx\n",
1048 	    rx_info->mbuf, rx_info,ena_buf->len, (uintmax_t)ena_buf->paddr);
1049 
1050 	return (0);
1051 
1052 exit:
1053 	m_freem(rx_info->mbuf);
1054 	rx_info->mbuf = NULL;
1055 	return (EFAULT);
1056 }
1057 
1058 static void
1059 ena_free_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
1060     struct ena_rx_buffer *rx_info)
1061 {
1062 
1063 	if (rx_info->mbuf == NULL) {
1064 		ena_log(adapter->pdev, WARN,
1065 		    "Trying to free unallocated buffer\n");
1066 		return;
1067 	}
1068 
1069 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map,
1070 	    BUS_DMASYNC_POSTREAD);
1071 	bus_dmamap_unload(adapter->rx_buf_tag, rx_info->map);
1072 	m_freem(rx_info->mbuf);
1073 	rx_info->mbuf = NULL;
1074 }
1075 
1076 /**
1077  * ena_refill_rx_bufs - Refills ring with descriptors
1078  * @rx_ring: the ring which we want to feed with free descriptors
1079  * @num: number of descriptors to refill
1080  * Refills the ring with newly allocated DMA-mapped mbufs for receiving
1081  **/
1082 int
1083 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
1084 {
1085 	struct ena_adapter *adapter = rx_ring->adapter;
1086 	device_t pdev = adapter->pdev;
1087 	uint16_t next_to_use, req_id;
1088 	uint32_t i;
1089 	int rc;
1090 
1091 	ena_log_io(adapter->pdev, DBG, "refill qid: %d\n", rx_ring->qid);
1092 
1093 	next_to_use = rx_ring->next_to_use;
1094 
1095 	for (i = 0; i < num; i++) {
1096 		struct ena_rx_buffer *rx_info;
1097 
1098 		ena_log_io(pdev, DBG, "RX buffer - next to use: %d\n",
1099 		    next_to_use);
1100 
1101 		req_id = rx_ring->free_rx_ids[next_to_use];
1102 		rx_info = &rx_ring->rx_buffer_info[req_id];
1103 #ifdef DEV_NETMAP
1104 		if (ena_rx_ring_in_netmap(adapter, rx_ring->qid))
1105 			rc = ena_netmap_alloc_rx_slot(adapter, rx_ring, rx_info);
1106 		else
1107 #endif /* DEV_NETMAP */
1108 			rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1109 		if (unlikely(rc != 0)) {
1110 			ena_log_io(pdev, WARN,
1111 			    "failed to alloc buffer for rx queue %d\n",
1112 			    rx_ring->qid);
1113 			break;
1114 		}
1115 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1116 		    &rx_info->ena_buf, req_id);
1117 		if (unlikely(rc != 0)) {
1118 			ena_log_io(pdev, WARN,
1119 			    "failed to add buffer for rx queue %d\n",
1120 			    rx_ring->qid);
1121 			break;
1122 		}
1123 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1124 		    rx_ring->ring_size);
1125 	}
1126 
1127 	if (unlikely(i < num)) {
1128 		counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1129 		ena_log_io(pdev, WARN,
1130 		     "refilled rx qid %d with only %d mbufs (from %d)\n",
1131 		     rx_ring->qid, i, num);
1132 	}
1133 
1134 	if (likely(i != 0))
1135 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1136 
1137 	rx_ring->next_to_use = next_to_use;
1138 	return (i);
1139 }
1140 
1141 int
1142 ena_update_buf_ring_size(struct ena_adapter *adapter,
1143     uint32_t new_buf_ring_size)
1144 {
1145 	uint32_t old_buf_ring_size;
1146 	int rc = 0;
1147 	bool dev_was_up;
1148 
1149 	old_buf_ring_size = adapter->buf_ring_size;
1150 	adapter->buf_ring_size = new_buf_ring_size;
1151 
1152 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1153 	ena_down(adapter);
1154 
1155 	/* Reconfigure buf ring for all Tx rings. */
1156 	ena_free_all_io_rings_resources(adapter);
1157 	ena_init_io_rings_advanced(adapter);
1158 	if (dev_was_up) {
1159 		/*
1160 		 * If ena_up() fails, it's not because of recent buf_ring size
1161 		 * changes. Because of that, we just want to revert old drbr
1162 		 * value and trigger the reset because something else had to
1163 		 * go wrong.
1164 		 */
1165 		rc = ena_up(adapter);
1166 		if (unlikely(rc != 0)) {
1167 			ena_log(adapter->pdev, ERR,
1168 			    "Failed to configure device after setting new drbr size: %u. Reverting old value: %u and triggering the reset\n",
1169 			    new_buf_ring_size, old_buf_ring_size);
1170 
1171 			/* Revert old size and trigger the reset */
1172 			adapter->buf_ring_size = old_buf_ring_size;
1173 			ena_free_all_io_rings_resources(adapter);
1174 			ena_init_io_rings_advanced(adapter);
1175 
1176 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET,
1177 			    adapter);
1178 			ena_trigger_reset(adapter, ENA_REGS_RESET_OS_TRIGGER);
1179 
1180 		}
1181 	}
1182 
1183 	return (rc);
1184 }
1185 
1186 int
1187 ena_update_queue_size(struct ena_adapter *adapter, uint32_t new_tx_size,
1188     uint32_t new_rx_size)
1189 {
1190 	uint32_t old_tx_size, old_rx_size;
1191 	int rc = 0;
1192 	bool dev_was_up;
1193 
1194 	old_tx_size = adapter->requested_tx_ring_size;
1195 	old_rx_size = adapter->requested_rx_ring_size;
1196 	adapter->requested_tx_ring_size = new_tx_size;
1197 	adapter->requested_rx_ring_size = new_rx_size;
1198 
1199 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1200 	ena_down(adapter);
1201 
1202 	/* Configure queues with new size. */
1203 	ena_init_io_rings_basic(adapter);
1204 	if (dev_was_up) {
1205 		rc = ena_up(adapter);
1206 		if (unlikely(rc != 0)) {
1207 			ena_log(adapter->pdev, ERR,
1208 			    "Failed to configure device with the new sizes - Tx: %u Rx: %u. Reverting old values - Tx: %u Rx: %u\n",
1209 			    new_tx_size, new_rx_size, old_tx_size, old_rx_size);
1210 
1211 			/* Revert old size. */
1212 			adapter->requested_tx_ring_size = old_tx_size;
1213 			adapter->requested_rx_ring_size = old_rx_size;
1214 			ena_init_io_rings_basic(adapter);
1215 
1216 			/* And try again. */
1217 			rc = ena_up(adapter);
1218 			if (unlikely(rc != 0)) {
1219 				ena_log(adapter->pdev, ERR,
1220 				    "Failed to revert old queue sizes. Triggering device reset.\n");
1221 				/*
1222 				 * If we've failed again, something had to go
1223 				 * wrong. After reset, the device should try to
1224 				 * go up
1225 				 */
1226 				ENA_FLAG_SET_ATOMIC(
1227 				    ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
1228 				ena_trigger_reset(adapter,
1229 				    ENA_REGS_RESET_OS_TRIGGER);
1230 			}
1231 		}
1232 	}
1233 
1234 	return (rc);
1235 }
1236 
1237 static void
1238 ena_update_io_rings(struct ena_adapter *adapter, uint32_t num)
1239 {
1240 	ena_free_all_io_rings_resources(adapter);
1241 	/* Force indirection table to be reinitialized */
1242 	ena_com_rss_destroy(adapter->ena_dev);
1243 
1244 	adapter->num_io_queues = num;
1245 	ena_init_io_rings(adapter);
1246 }
1247 
1248 /* Caller should sanitize new_num */
1249 int
1250 ena_update_io_queue_nb(struct ena_adapter *adapter, uint32_t new_num)
1251 {
1252 	uint32_t old_num;
1253 	int rc = 0;
1254 	bool dev_was_up;
1255 
1256 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1257 	old_num = adapter->num_io_queues;
1258 	ena_down(adapter);
1259 
1260 	ena_update_io_rings(adapter, new_num);
1261 
1262 	if (dev_was_up) {
1263 		rc = ena_up(adapter);
1264 		if (unlikely(rc != 0)) {
1265 			ena_log(adapter->pdev, ERR,
1266 			    "Failed to configure device with %u IO queues. "
1267 			    "Reverting to previous value: %u\n",
1268 			    new_num, old_num);
1269 
1270 			ena_update_io_rings(adapter, old_num);
1271 
1272 			rc = ena_up(adapter);
1273 			if (unlikely(rc != 0)) {
1274 				ena_log(adapter->pdev, ERR,
1275 				    "Failed to revert to previous setup IO "
1276 				    "queues. Triggering device reset.\n");
1277 				ENA_FLAG_SET_ATOMIC(
1278 				    ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
1279 				ena_trigger_reset(adapter,
1280 				    ENA_REGS_RESET_OS_TRIGGER);
1281 			}
1282 		}
1283 	}
1284 
1285 	return (rc);
1286 }
1287 
1288 static void
1289 ena_free_rx_bufs(struct ena_adapter *adapter, unsigned int qid)
1290 {
1291 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1292 	unsigned int i;
1293 
1294 	for (i = 0; i < rx_ring->ring_size; i++) {
1295 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1296 
1297 		if (rx_info->mbuf != NULL)
1298 			ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1299 #ifdef DEV_NETMAP
1300 		if (((if_getflags(adapter->ifp) & IFF_DYING) == 0) &&
1301 		    (adapter->ifp->if_capenable & IFCAP_NETMAP)) {
1302 			if (rx_info->netmap_buf_idx != 0)
1303 				ena_netmap_free_rx_slot(adapter, rx_ring,
1304 				    rx_info);
1305 		}
1306 #endif /* DEV_NETMAP */
1307 	}
1308 }
1309 
1310 /**
1311  * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1312  * @adapter: network interface device structure
1313  *
1314  */
1315 static void
1316 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1317 {
1318 	struct ena_ring *rx_ring;
1319 	int i, rc, bufs_num;
1320 
1321 	for (i = 0; i < adapter->num_io_queues; i++) {
1322 		rx_ring = &adapter->rx_ring[i];
1323 		bufs_num = rx_ring->ring_size - 1;
1324 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1325 		if (unlikely(rc != bufs_num))
1326 			ena_log_io(adapter->pdev, WARN,
1327 			    "refilling Queue %d failed. "
1328 			    "Allocated %d buffers from: %d\n", i, rc, bufs_num);
1329 #ifdef DEV_NETMAP
1330 		rx_ring->initialized = true;
1331 #endif /* DEV_NETMAP */
1332 	}
1333 }
1334 
1335 static void
1336 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1337 {
1338 	int i;
1339 
1340 	for (i = 0; i < adapter->num_io_queues; i++)
1341 		ena_free_rx_bufs(adapter, i);
1342 }
1343 
1344 /**
1345  * ena_free_tx_bufs - Free Tx Buffers per Queue
1346  * @adapter: network interface device structure
1347  * @qid: queue index
1348  **/
1349 static void
1350 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1351 {
1352 	bool print_once = true;
1353 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1354 
1355 	ENA_RING_MTX_LOCK(tx_ring);
1356 	for (int i = 0; i < tx_ring->ring_size; i++) {
1357 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1358 
1359 		if (tx_info->mbuf == NULL)
1360 			continue;
1361 
1362 		if (print_once) {
1363 			ena_log(adapter->pdev, WARN,
1364 			    "free uncompleted tx mbuf qid %d idx 0x%x\n",
1365 			    qid, i);
1366 			print_once = false;
1367 		} else {
1368 			ena_log(adapter->pdev, DBG,
1369 			    "free uncompleted tx mbuf qid %d idx 0x%x\n",
1370 			     qid, i);
1371 		}
1372 
1373 		bus_dmamap_sync(adapter->tx_buf_tag, tx_info->dmamap,
1374 		    BUS_DMASYNC_POSTWRITE);
1375 		bus_dmamap_unload(adapter->tx_buf_tag, tx_info->dmamap);
1376 
1377 		m_free(tx_info->mbuf);
1378 		tx_info->mbuf = NULL;
1379 	}
1380 	ENA_RING_MTX_UNLOCK(tx_ring);
1381 }
1382 
1383 static void
1384 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1385 {
1386 
1387 	for (int i = 0; i < adapter->num_io_queues; i++)
1388 		ena_free_tx_bufs(adapter, i);
1389 }
1390 
1391 static void
1392 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1393 {
1394 	uint16_t ena_qid;
1395 	int i;
1396 
1397 	for (i = 0; i < adapter->num_io_queues; i++) {
1398 		ena_qid = ENA_IO_TXQ_IDX(i);
1399 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1400 	}
1401 }
1402 
1403 static void
1404 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1405 {
1406 	uint16_t ena_qid;
1407 	int i;
1408 
1409 	for (i = 0; i < adapter->num_io_queues; i++) {
1410 		ena_qid = ENA_IO_RXQ_IDX(i);
1411 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1412 	}
1413 }
1414 
1415 static void
1416 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1417 {
1418 	struct ena_que *queue;
1419 	int i;
1420 
1421 	for (i = 0; i < adapter->num_io_queues; i++) {
1422 		queue = &adapter->que[i];
1423 		while (taskqueue_cancel(queue->cleanup_tq,
1424 		    &queue->cleanup_task, NULL))
1425 			taskqueue_drain(queue->cleanup_tq,
1426 			    &queue->cleanup_task);
1427 		taskqueue_free(queue->cleanup_tq);
1428 	}
1429 
1430 	ena_destroy_all_tx_queues(adapter);
1431 	ena_destroy_all_rx_queues(adapter);
1432 }
1433 
1434 static int
1435 ena_create_io_queues(struct ena_adapter *adapter)
1436 {
1437 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1438 	struct ena_com_create_io_ctx ctx;
1439 	struct ena_ring *ring;
1440 	struct ena_que *queue;
1441 	uint16_t ena_qid;
1442 	uint32_t msix_vector;
1443 	cpuset_t *cpu_mask = NULL;
1444 	int rc, i;
1445 
1446 	/* Create TX queues */
1447 	for (i = 0; i < adapter->num_io_queues; i++) {
1448 		msix_vector = ENA_IO_IRQ_IDX(i);
1449 		ena_qid = ENA_IO_TXQ_IDX(i);
1450 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1451 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1452 		ctx.queue_size = adapter->requested_tx_ring_size;
1453 		ctx.msix_vector = msix_vector;
1454 		ctx.qid = ena_qid;
1455 		ctx.numa_node = adapter->que[i].domain;
1456 
1457 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1458 		if (rc != 0) {
1459 			ena_log(adapter->pdev, ERR,
1460 			    "Failed to create io TX queue #%d rc: %d\n", i, rc);
1461 			goto err_tx;
1462 		}
1463 		ring = &adapter->tx_ring[i];
1464 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1465 		    &ring->ena_com_io_sq,
1466 		    &ring->ena_com_io_cq);
1467 		if (rc != 0) {
1468 			ena_log(adapter->pdev, ERR,
1469 			    "Failed to get TX queue handlers. TX queue num"
1470 			    " %d rc: %d\n", i, rc);
1471 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1472 			goto err_tx;
1473 		}
1474 
1475 		if (ctx.numa_node >= 0) {
1476 			ena_com_update_numa_node(ring->ena_com_io_cq,
1477 			    ctx.numa_node);
1478 		}
1479 	}
1480 
1481 	/* Create RX queues */
1482 	for (i = 0; i < adapter->num_io_queues; i++) {
1483 		msix_vector = ENA_IO_IRQ_IDX(i);
1484 		ena_qid = ENA_IO_RXQ_IDX(i);
1485 		ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1486 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1487 		ctx.queue_size = adapter->requested_rx_ring_size;
1488 		ctx.msix_vector = msix_vector;
1489 		ctx.qid = ena_qid;
1490 		ctx.numa_node = adapter->que[i].domain;
1491 
1492 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1493 		if (unlikely(rc != 0)) {
1494 			ena_log(adapter->pdev, ERR,
1495 			    "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1496 			goto err_rx;
1497 		}
1498 
1499 		ring = &adapter->rx_ring[i];
1500 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1501 		    &ring->ena_com_io_sq,
1502 		    &ring->ena_com_io_cq);
1503 		if (unlikely(rc != 0)) {
1504 			ena_log(adapter->pdev, ERR,
1505 			    "Failed to get RX queue handlers. RX queue num"
1506 			    " %d rc: %d\n", i, rc);
1507 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1508 			goto err_rx;
1509 		}
1510 
1511 		if (ctx.numa_node >= 0) {
1512 			ena_com_update_numa_node(ring->ena_com_io_cq,
1513 			    ctx.numa_node);
1514 		}
1515 	}
1516 
1517 	for (i = 0; i < adapter->num_io_queues; i++) {
1518 		queue = &adapter->que[i];
1519 
1520 		NET_TASK_INIT(&queue->cleanup_task, 0, ena_cleanup, queue);
1521 		queue->cleanup_tq = taskqueue_create_fast("ena cleanup",
1522 		    M_WAITOK, taskqueue_thread_enqueue, &queue->cleanup_tq);
1523 
1524 #ifdef RSS
1525 		cpu_mask = &queue->cpu_mask;
1526 #endif
1527 		taskqueue_start_threads_cpuset(&queue->cleanup_tq, 1, PI_NET,
1528 		    cpu_mask,
1529 		    "%s queue %d cleanup",
1530 		    device_get_nameunit(adapter->pdev), i);
1531 	}
1532 
1533 	return (0);
1534 
1535 err_rx:
1536 	while (i--)
1537 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1538 	i = adapter->num_io_queues;
1539 err_tx:
1540 	while (i--)
1541 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1542 
1543 	return (ENXIO);
1544 }
1545 
1546 /*********************************************************************
1547  *
1548  *  MSIX & Interrupt Service routine
1549  *
1550  **********************************************************************/
1551 
1552 /**
1553  * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1554  * @arg: interrupt number
1555  **/
1556 static void
1557 ena_intr_msix_mgmnt(void *arg)
1558 {
1559 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1560 
1561 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1562 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter)))
1563 		ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1564 }
1565 
1566 /**
1567  * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1568  * @arg: queue
1569  **/
1570 static int
1571 ena_handle_msix(void *arg)
1572 {
1573 	struct ena_que *queue = arg;
1574 	struct ena_adapter *adapter = queue->adapter;
1575 	if_t ifp = adapter->ifp;
1576 
1577 	if (unlikely((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0))
1578 		return (FILTER_STRAY);
1579 
1580 	taskqueue_enqueue(queue->cleanup_tq, &queue->cleanup_task);
1581 
1582 	return (FILTER_HANDLED);
1583 }
1584 
1585 static int
1586 ena_enable_msix(struct ena_adapter *adapter)
1587 {
1588 	device_t dev = adapter->pdev;
1589 	int msix_vecs, msix_req;
1590 	int i, rc = 0;
1591 
1592 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1593 		ena_log(dev, ERR, "Error, MSI-X is already enabled\n");
1594 		return (EINVAL);
1595 	}
1596 
1597 	/* Reserved the max msix vectors we might need */
1598 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
1599 
1600 	adapter->msix_entries = malloc(msix_vecs * sizeof(struct msix_entry),
1601 	    M_DEVBUF, M_WAITOK | M_ZERO);
1602 
1603 	ena_log(dev, DBG, "trying to enable MSI-X, vectors: %d\n",
1604 	    msix_vecs);
1605 
1606 	for (i = 0; i < msix_vecs; i++) {
1607 		adapter->msix_entries[i].entry = i;
1608 		/* Vectors must start from 1 */
1609 		adapter->msix_entries[i].vector = i + 1;
1610 	}
1611 
1612 	msix_req = msix_vecs;
1613 	rc = pci_alloc_msix(dev, &msix_vecs);
1614 	if (unlikely(rc != 0)) {
1615 		ena_log(dev, ERR,
1616 		    "Failed to enable MSIX, vectors %d rc %d\n", msix_vecs, rc);
1617 
1618 		rc = ENOSPC;
1619 		goto err_msix_free;
1620 	}
1621 
1622 	if (msix_vecs != msix_req) {
1623 		if (msix_vecs == ENA_ADMIN_MSIX_VEC) {
1624 			ena_log(dev, ERR,
1625 			    "Not enough number of MSI-x allocated: %d\n",
1626 			    msix_vecs);
1627 			pci_release_msi(dev);
1628 			rc = ENOSPC;
1629 			goto err_msix_free;
1630 		}
1631 		ena_log(dev, ERR, "Enable only %d MSI-x (out of %d), reduce "
1632 		    "the number of queues\n", msix_vecs, msix_req);
1633 	}
1634 
1635 	adapter->msix_vecs = msix_vecs;
1636 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1637 
1638 	return (0);
1639 
1640 err_msix_free:
1641 	free(adapter->msix_entries, M_DEVBUF);
1642 	adapter->msix_entries = NULL;
1643 
1644 	return (rc);
1645 }
1646 
1647 static void
1648 ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1649 {
1650 
1651 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1652 	    ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1653 	    device_get_nameunit(adapter->pdev));
1654 	/*
1655 	 * Handler is NULL on purpose, it will be set
1656 	 * when mgmnt interrupt is acquired
1657 	 */
1658 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler = NULL;
1659 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1660 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1661 	    adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
1662 }
1663 
1664 static int
1665 ena_setup_io_intr(struct ena_adapter *adapter)
1666 {
1667 #ifdef RSS
1668 	int num_buckets = rss_getnumbuckets();
1669 	static int last_bind = 0;
1670 	int cur_bind;
1671 	int idx;
1672 #endif
1673 	int irq_idx;
1674 
1675 	if (adapter->msix_entries == NULL)
1676 		return (EINVAL);
1677 
1678 #ifdef RSS
1679 	if (adapter->first_bind < 0) {
1680 		adapter->first_bind = last_bind;
1681 		last_bind = (last_bind + adapter->num_io_queues) % num_buckets;
1682 	}
1683 	cur_bind = adapter->first_bind;
1684 #endif
1685 
1686 	for (int i = 0; i < adapter->num_io_queues; i++) {
1687 		irq_idx = ENA_IO_IRQ_IDX(i);
1688 
1689 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1690 		    "%s-TxRx-%d", device_get_nameunit(adapter->pdev), i);
1691 		adapter->irq_tbl[irq_idx].handler = ena_handle_msix;
1692 		adapter->irq_tbl[irq_idx].data = &adapter->que[i];
1693 		adapter->irq_tbl[irq_idx].vector =
1694 		    adapter->msix_entries[irq_idx].vector;
1695 		ena_log(adapter->pdev, DBG, "ena_setup_io_intr vector: %d\n",
1696 		    adapter->msix_entries[irq_idx].vector);
1697 
1698 #ifdef RSS
1699 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1700 		    rss_getcpu(cur_bind);
1701 		cur_bind = (cur_bind + 1) % num_buckets;
1702 		CPU_SETOF(adapter->que[i].cpu, &adapter->que[i].cpu_mask);
1703 
1704 		for (idx = 0; idx < MAXMEMDOM; ++idx) {
1705 			if (CPU_ISSET(adapter->que[i].cpu, &cpuset_domain[idx]))
1706 				break;
1707 		}
1708 		adapter->que[i].domain = idx;
1709 #else
1710 		adapter->que[i].domain = -1;
1711 #endif
1712 	}
1713 
1714 	return (0);
1715 }
1716 
1717 static int
1718 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1719 {
1720 	device_t pdev = adapter->pdev;
1721 	struct ena_irq *irq;
1722 	unsigned long flags;
1723 	int rc, rcc;
1724 
1725 	flags = RF_ACTIVE | RF_SHAREABLE;
1726 
1727 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1728 	irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1729 	    &irq->vector, flags);
1730 
1731 	if (unlikely(irq->res == NULL)) {
1732 		ena_log(pdev, ERR, "could not allocate irq vector: %d\n",
1733 		    irq->vector);
1734 		return (ENXIO);
1735 	}
1736 
1737 	rc = bus_setup_intr(adapter->pdev, irq->res,
1738 	    INTR_TYPE_NET | INTR_MPSAFE, NULL, ena_intr_msix_mgmnt,
1739 	    irq->data, &irq->cookie);
1740 	if (unlikely(rc != 0)) {
1741 		ena_log(pdev, ERR, "failed to register "
1742 		    "interrupt handler for irq %ju: %d\n",
1743 		    rman_get_start(irq->res), rc);
1744 		goto err_res_free;
1745 	}
1746 	irq->requested = true;
1747 
1748 	return (rc);
1749 
1750 err_res_free:
1751 	ena_log(pdev, INFO, "releasing resource for irq %d\n", irq->vector);
1752 	rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1753 	    irq->vector, irq->res);
1754 	if (unlikely(rcc != 0))
1755 		ena_log(pdev, ERR, "dev has no parent while "
1756 		    "releasing res for irq: %d\n", irq->vector);
1757 	irq->res = NULL;
1758 
1759 	return (rc);
1760 }
1761 
1762 static int
1763 ena_request_io_irq(struct ena_adapter *adapter)
1764 {
1765 	device_t pdev = adapter->pdev;
1766 	struct ena_irq *irq;
1767 	unsigned long flags = 0;
1768 	int rc = 0, i, rcc;
1769 
1770 	if (unlikely(!ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter))) {
1771 		ena_log(pdev, ERR,
1772 		    "failed to request I/O IRQ: MSI-X is not enabled\n");
1773 		return (EINVAL);
1774 	} else {
1775 		flags = RF_ACTIVE | RF_SHAREABLE;
1776 	}
1777 
1778 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1779 		irq = &adapter->irq_tbl[i];
1780 
1781 		if (unlikely(irq->requested))
1782 			continue;
1783 
1784 		irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1785 		    &irq->vector, flags);
1786 		if (unlikely(irq->res == NULL)) {
1787 			rc = ENOMEM;
1788 			ena_log(pdev, ERR, "could not allocate irq vector: %d\n",
1789 			    irq->vector);
1790 			goto err;
1791 		}
1792 
1793 		rc = bus_setup_intr(adapter->pdev, irq->res,
1794 		    INTR_TYPE_NET | INTR_MPSAFE, irq->handler, NULL,
1795 		    irq->data, &irq->cookie);
1796 		 if (unlikely(rc != 0)) {
1797 			ena_log(pdev, ERR, "failed to register "
1798 			    "interrupt handler for irq %ju: %d\n",
1799 			    rman_get_start(irq->res), rc);
1800 			goto err;
1801 		}
1802 		irq->requested = true;
1803 
1804 #ifdef RSS
1805 		rc = bus_bind_intr(adapter->pdev, irq->res, irq->cpu);
1806 		if (unlikely(rc != 0)) {
1807 			ena_log(pdev, ERR, "failed to bind "
1808 			    "interrupt handler for irq %ju to cpu %d: %d\n",
1809 			    rman_get_start(irq->res), irq->cpu, rc);
1810 			goto err;
1811 		}
1812 
1813 		ena_log(pdev, INFO, "queue %d - cpu %d\n",
1814 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
1815 #endif
1816 	}
1817 
1818 	return (rc);
1819 
1820 err:
1821 
1822 	for (; i >= ENA_IO_IRQ_FIRST_IDX; i--) {
1823 		irq = &adapter->irq_tbl[i];
1824 		rcc = 0;
1825 
1826 		/* Once we entered err: section and irq->requested is true we
1827 		   free both intr and resources */
1828 		if (irq->requested)
1829 			rcc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
1830 		if (unlikely(rcc != 0))
1831 			ena_log(pdev, ERR, "could not release irq: %d, error: %d\n",
1832 			    irq->vector, rcc);
1833 
1834 		/* If we entred err: section without irq->requested set we know
1835 		   it was bus_alloc_resource_any() that needs cleanup, provided
1836 		   res is not NULL. In case res is NULL no work in needed in
1837 		   this iteration */
1838 		rcc = 0;
1839 		if (irq->res != NULL) {
1840 			rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1841 			    irq->vector, irq->res);
1842 		}
1843 		if (unlikely(rcc != 0))
1844 			ena_log(pdev, ERR, "dev has no parent while "
1845 			    "releasing res for irq: %d\n", irq->vector);
1846 		irq->requested = false;
1847 		irq->res = NULL;
1848 	}
1849 
1850 	return (rc);
1851 }
1852 
1853 static void
1854 ena_free_mgmnt_irq(struct ena_adapter *adapter)
1855 {
1856 	device_t pdev = adapter->pdev;
1857 	struct ena_irq *irq;
1858 	int rc;
1859 
1860 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1861 	if (irq->requested) {
1862 		ena_log(pdev, DBG, "tear down irq: %d\n", irq->vector);
1863 		rc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
1864 		if (unlikely(rc != 0))
1865 			ena_log(pdev, ERR, "failed to tear down irq: %d\n",
1866 			    irq->vector);
1867 		irq->requested = 0;
1868 	}
1869 
1870 	if (irq->res != NULL) {
1871 		ena_log(pdev, DBG, "release resource irq: %d\n", irq->vector);
1872 		rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1873 		    irq->vector, irq->res);
1874 		irq->res = NULL;
1875 		if (unlikely(rc != 0))
1876 			ena_log(pdev, ERR, "dev has no parent while "
1877 			    "releasing res for irq: %d\n", irq->vector);
1878 	}
1879 }
1880 
1881 static void
1882 ena_free_io_irq(struct ena_adapter *adapter)
1883 {
1884 	device_t pdev = adapter->pdev;
1885 	struct ena_irq *irq;
1886 	int rc;
1887 
1888 	for (int i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1889 		irq = &adapter->irq_tbl[i];
1890 		if (irq->requested) {
1891 			ena_log(pdev, DBG, "tear down irq: %d\n", irq->vector);
1892 			rc = bus_teardown_intr(adapter->pdev, irq->res,
1893 			    irq->cookie);
1894 			if (unlikely(rc != 0)) {
1895 				ena_log(pdev, ERR, "failed to tear down irq: %d\n",
1896 				    irq->vector);
1897 			}
1898 			irq->requested = 0;
1899 		}
1900 
1901 		if (irq->res != NULL) {
1902 			ena_log(pdev, DBG, "release resource irq: %d\n",
1903 			    irq->vector);
1904 			rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1905 			    irq->vector, irq->res);
1906 			irq->res = NULL;
1907 			if (unlikely(rc != 0)) {
1908 				ena_log(pdev, ERR, "dev has no parent"
1909 				    " while releasing res for irq: %d\n",
1910 				    irq->vector);
1911 			}
1912 		}
1913 	}
1914 }
1915 
1916 static void
1917 ena_free_irqs(struct ena_adapter* adapter)
1918 {
1919 
1920 	ena_free_io_irq(adapter);
1921 	ena_free_mgmnt_irq(adapter);
1922 	ena_disable_msix(adapter);
1923 }
1924 
1925 static void
1926 ena_disable_msix(struct ena_adapter *adapter)
1927 {
1928 
1929 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1930 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1931 		pci_release_msi(adapter->pdev);
1932 	}
1933 
1934 	adapter->msix_vecs = 0;
1935 	free(adapter->msix_entries, M_DEVBUF);
1936 	adapter->msix_entries = NULL;
1937 }
1938 
1939 static void
1940 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
1941 {
1942 	struct ena_com_io_cq* io_cq;
1943 	struct ena_eth_io_intr_reg intr_reg;
1944 	struct ena_ring *tx_ring;
1945 	uint16_t ena_qid;
1946 	int i;
1947 
1948 	/* Unmask interrupts for all queues */
1949 	for (i = 0; i < adapter->num_io_queues; i++) {
1950 		ena_qid = ENA_IO_TXQ_IDX(i);
1951 		io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1952 		ena_com_update_intr_reg(&intr_reg, 0, 0, true);
1953 		tx_ring = &adapter->tx_ring[i];
1954 		counter_u64_add(tx_ring->tx_stats.unmask_interrupt_num, 1);
1955 		ena_com_unmask_intr(io_cq, &intr_reg);
1956 	}
1957 }
1958 
1959 static int
1960 ena_up_complete(struct ena_adapter *adapter)
1961 {
1962 	int rc;
1963 
1964 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter))) {
1965 		rc = ena_rss_configure(adapter);
1966 		if (rc != 0) {
1967 			ena_log(adapter->pdev, ERR,
1968 			    "Failed to configure RSS\n");
1969 			return (rc);
1970 		}
1971 	}
1972 
1973 	rc = ena_change_mtu(adapter->ifp, adapter->ifp->if_mtu);
1974 	if (unlikely(rc != 0))
1975 		return (rc);
1976 
1977 	ena_refill_all_rx_bufs(adapter);
1978 	ena_reset_counters((counter_u64_t *)&adapter->hw_stats,
1979 	    sizeof(adapter->hw_stats));
1980 
1981 	return (0);
1982 }
1983 
1984 static void
1985 set_io_rings_size(struct ena_adapter *adapter, int new_tx_size,
1986     int new_rx_size)
1987 {
1988 	int i;
1989 
1990 	for (i = 0; i < adapter->num_io_queues; i++) {
1991 		adapter->tx_ring[i].ring_size = new_tx_size;
1992 		adapter->rx_ring[i].ring_size = new_rx_size;
1993 	}
1994 }
1995 
1996 static int
1997 create_queues_with_size_backoff(struct ena_adapter *adapter)
1998 {
1999 	device_t pdev = adapter->pdev;
2000 	int rc;
2001 	uint32_t cur_rx_ring_size, cur_tx_ring_size;
2002 	uint32_t new_rx_ring_size, new_tx_ring_size;
2003 
2004 	/*
2005 	 * Current queue sizes might be set to smaller than the requested
2006 	 * ones due to past queue allocation failures.
2007 	 */
2008 	set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2009 	    adapter->requested_rx_ring_size);
2010 
2011 	while (1) {
2012 		/* Allocate transmit descriptors */
2013 		rc = ena_setup_all_tx_resources(adapter);
2014 		if (unlikely(rc != 0)) {
2015 			ena_log(pdev, ERR, "err_setup_tx\n");
2016 			goto err_setup_tx;
2017 		}
2018 
2019 		/* Allocate receive descriptors */
2020 		rc = ena_setup_all_rx_resources(adapter);
2021 		if (unlikely(rc != 0)) {
2022 			ena_log(pdev, ERR, "err_setup_rx\n");
2023 			goto err_setup_rx;
2024 		}
2025 
2026 		/* Create IO queues for Rx & Tx */
2027 		rc = ena_create_io_queues(adapter);
2028 		if (unlikely(rc != 0)) {
2029 			ena_log(pdev, ERR,
2030 			    "create IO queues failed\n");
2031 			goto err_io_que;
2032 		}
2033 
2034 		return (0);
2035 
2036 err_io_que:
2037 		ena_free_all_rx_resources(adapter);
2038 err_setup_rx:
2039 		ena_free_all_tx_resources(adapter);
2040 err_setup_tx:
2041 		/*
2042 		 * Lower the ring size if ENOMEM. Otherwise, return the
2043 		 * error straightaway.
2044 		 */
2045 		if (unlikely(rc != ENOMEM)) {
2046 			ena_log(pdev, ERR,
2047 			    "Queue creation failed with error code: %d\n", rc);
2048 			return (rc);
2049 		}
2050 
2051 		cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2052 		cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2053 
2054 		ena_log(pdev, ERR,
2055 		    "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2056 		    cur_tx_ring_size, cur_rx_ring_size);
2057 
2058 		new_tx_ring_size = cur_tx_ring_size;
2059 		new_rx_ring_size = cur_rx_ring_size;
2060 
2061 		/*
2062 		 * Decrease the size of a larger queue, or decrease both if they are
2063 		 * the same size.
2064 		 */
2065 		if (cur_rx_ring_size <= cur_tx_ring_size)
2066 			new_tx_ring_size = cur_tx_ring_size / 2;
2067 		if (cur_rx_ring_size >= cur_tx_ring_size)
2068 			new_rx_ring_size = cur_rx_ring_size / 2;
2069 
2070 		if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2071 		    new_rx_ring_size < ENA_MIN_RING_SIZE) {
2072 			ena_log(pdev, ERR,
2073 			    "Queue creation failed with the smallest possible queue size"
2074 			    "of %d for both queues. Not retrying with smaller queues\n",
2075 			    ENA_MIN_RING_SIZE);
2076 			return (rc);
2077 		}
2078 
2079 		ena_log(pdev, INFO,
2080 		    "Retrying queue creation with sizes TX=%d, RX=%d\n",
2081 		    new_tx_ring_size, new_rx_ring_size);
2082 
2083 		set_io_rings_size(adapter, new_tx_ring_size, new_rx_ring_size);
2084 	}
2085 }
2086 
2087 int
2088 ena_up(struct ena_adapter *adapter)
2089 {
2090 	int rc = 0;
2091 
2092 	ENA_LOCK_ASSERT();
2093 
2094 	if (unlikely(device_is_attached(adapter->pdev) == 0)) {
2095 		ena_log(adapter->pdev, ERR, "device is not attached!\n");
2096 		return (ENXIO);
2097 	}
2098 
2099 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2100 		return (0);
2101 
2102 	ena_log(adapter->pdev, INFO, "device is going UP\n");
2103 
2104 	/*
2105 	 * ena_timer_service can use functions, which write to the admin queue.
2106 	 * Those calls are not protected by ENA_LOCK, and because of that, the
2107 	 * timer should be stopped when bringing the device up or down.
2108 	 */
2109 	ENA_TIMER_DRAIN(adapter);
2110 
2111 	/* setup interrupts for IO queues */
2112 	rc = ena_setup_io_intr(adapter);
2113 	if (unlikely(rc != 0)) {
2114 		ena_log(adapter->pdev, ERR, "error setting up IO interrupt\n");
2115 		goto error;
2116 	}
2117 	rc = ena_request_io_irq(adapter);
2118 	if (unlikely(rc != 0)) {
2119 		ena_log(adapter->pdev, ERR, "err_req_irq\n");
2120 		goto error;
2121 	}
2122 
2123 	ena_log(adapter->pdev, INFO,
2124 	    "Creating %u IO queues. Rx queue size: %d, Tx queue size: %d, "
2125 	    "LLQ is %s\n",
2126 	    adapter->num_io_queues,
2127 	    adapter->requested_rx_ring_size,
2128 	    adapter->requested_tx_ring_size,
2129 	    (adapter->ena_dev->tx_mem_queue_type ==
2130 	        ENA_ADMIN_PLACEMENT_POLICY_DEV) ?  "ENABLED" : "DISABLED");
2131 
2132 	rc = create_queues_with_size_backoff(adapter);
2133 	if (unlikely(rc != 0)) {
2134 		ena_log(adapter->pdev, ERR,
2135 		    "error creating queues with size backoff\n");
2136 		goto err_create_queues_with_backoff;
2137 	}
2138 
2139 	if (ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter))
2140 		if_link_state_change(adapter->ifp, LINK_STATE_UP);
2141 
2142 	rc = ena_up_complete(adapter);
2143 	if (unlikely(rc != 0))
2144 		goto err_up_complete;
2145 
2146 	counter_u64_add(adapter->dev_stats.interface_up, 1);
2147 
2148 	ena_update_hwassist(adapter);
2149 
2150 	if_setdrvflagbits(adapter->ifp, IFF_DRV_RUNNING,
2151 		IFF_DRV_OACTIVE);
2152 
2153 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP, adapter);
2154 
2155 	ena_unmask_all_io_irqs(adapter);
2156 
2157 	ENA_TIMER_RESET(adapter);
2158 
2159 	return (0);
2160 
2161 err_up_complete:
2162 	ena_destroy_all_io_queues(adapter);
2163 	ena_free_all_rx_resources(adapter);
2164 	ena_free_all_tx_resources(adapter);
2165 err_create_queues_with_backoff:
2166 	ena_free_io_irq(adapter);
2167 error:
2168 	ENA_TIMER_RESET(adapter);
2169 
2170 	return (rc);
2171 }
2172 
2173 static uint64_t
2174 ena_get_counter(if_t ifp, ift_counter cnt)
2175 {
2176 	struct ena_adapter *adapter;
2177 	struct ena_hw_stats *stats;
2178 
2179 	adapter = if_getsoftc(ifp);
2180 	stats = &adapter->hw_stats;
2181 
2182 	switch (cnt) {
2183 	case IFCOUNTER_IPACKETS:
2184 		return (counter_u64_fetch(stats->rx_packets));
2185 	case IFCOUNTER_OPACKETS:
2186 		return (counter_u64_fetch(stats->tx_packets));
2187 	case IFCOUNTER_IBYTES:
2188 		return (counter_u64_fetch(stats->rx_bytes));
2189 	case IFCOUNTER_OBYTES:
2190 		return (counter_u64_fetch(stats->tx_bytes));
2191 	case IFCOUNTER_IQDROPS:
2192 		return (counter_u64_fetch(stats->rx_drops));
2193 	case IFCOUNTER_OQDROPS:
2194 		return (counter_u64_fetch(stats->tx_drops));
2195 	default:
2196 		return (if_get_counter_default(ifp, cnt));
2197 	}
2198 }
2199 
2200 static int
2201 ena_media_change(if_t ifp)
2202 {
2203 	/* Media Change is not supported by firmware */
2204 	return (0);
2205 }
2206 
2207 static void
2208 ena_media_status(if_t ifp, struct ifmediareq *ifmr)
2209 {
2210 	struct ena_adapter *adapter = if_getsoftc(ifp);
2211 	ena_log(adapter->pdev, DBG, "Media status update\n");
2212 
2213 	ENA_LOCK_LOCK();
2214 
2215 	ifmr->ifm_status = IFM_AVALID;
2216 	ifmr->ifm_active = IFM_ETHER;
2217 
2218 	if (!ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter)) {
2219 		ENA_LOCK_UNLOCK();
2220 		ena_log(adapter->pdev, INFO, "Link is down\n");
2221 		return;
2222 	}
2223 
2224 	ifmr->ifm_status |= IFM_ACTIVE;
2225 	ifmr->ifm_active |= IFM_UNKNOWN | IFM_FDX;
2226 
2227 	ENA_LOCK_UNLOCK();
2228 }
2229 
2230 static void
2231 ena_init(void *arg)
2232 {
2233 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
2234 
2235 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter)) {
2236 		ENA_LOCK_LOCK();
2237 		ena_up(adapter);
2238 		ENA_LOCK_UNLOCK();
2239 	}
2240 }
2241 
2242 static int
2243 ena_ioctl(if_t ifp, u_long command, caddr_t data)
2244 {
2245 	struct ena_adapter *adapter;
2246 	struct ifreq *ifr;
2247 	int rc;
2248 
2249 	adapter = ifp->if_softc;
2250 	ifr = (struct ifreq *)data;
2251 
2252 	/*
2253 	 * Acquiring lock to prevent from running up and down routines parallel.
2254 	 */
2255 	rc = 0;
2256 	switch (command) {
2257 	case SIOCSIFMTU:
2258 		if (ifp->if_mtu == ifr->ifr_mtu)
2259 			break;
2260 		ENA_LOCK_LOCK();
2261 		ena_down(adapter);
2262 
2263 		ena_change_mtu(ifp, ifr->ifr_mtu);
2264 
2265 		rc = ena_up(adapter);
2266 		ENA_LOCK_UNLOCK();
2267 		break;
2268 
2269 	case SIOCSIFFLAGS:
2270 		if ((ifp->if_flags & IFF_UP) != 0) {
2271 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2272 				if ((ifp->if_flags & (IFF_PROMISC |
2273 				    IFF_ALLMULTI)) != 0) {
2274 					ena_log(adapter->pdev, INFO,
2275 					    "ioctl promisc/allmulti\n");
2276 				}
2277 			} else {
2278 				ENA_LOCK_LOCK();
2279 				rc = ena_up(adapter);
2280 				ENA_LOCK_UNLOCK();
2281 			}
2282 		} else {
2283 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2284 				ENA_LOCK_LOCK();
2285 				ena_down(adapter);
2286 				ENA_LOCK_UNLOCK();
2287 			}
2288 		}
2289 		break;
2290 
2291 	case SIOCADDMULTI:
2292 	case SIOCDELMULTI:
2293 		break;
2294 
2295 	case SIOCSIFMEDIA:
2296 	case SIOCGIFMEDIA:
2297 		rc = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
2298 		break;
2299 
2300 	case SIOCSIFCAP:
2301 		{
2302 			int reinit = 0;
2303 
2304 			if (ifr->ifr_reqcap != ifp->if_capenable) {
2305 				ifp->if_capenable = ifr->ifr_reqcap;
2306 				reinit = 1;
2307 			}
2308 
2309 			if ((reinit != 0) &&
2310 			    ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)) {
2311 				ENA_LOCK_LOCK();
2312 				ena_down(adapter);
2313 				rc = ena_up(adapter);
2314 				ENA_LOCK_UNLOCK();
2315 			}
2316 		}
2317 
2318 		break;
2319 	default:
2320 		rc = ether_ioctl(ifp, command, data);
2321 		break;
2322 	}
2323 
2324 	return (rc);
2325 }
2326 
2327 static int
2328 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2329 {
2330 	int caps = 0;
2331 
2332 	if ((feat->offload.tx &
2333 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2334 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2335 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)) != 0)
2336 		caps |= IFCAP_TXCSUM;
2337 
2338 	if ((feat->offload.tx &
2339 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2340 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)) != 0)
2341 		caps |= IFCAP_TXCSUM_IPV6;
2342 
2343 	if ((feat->offload.tx &
2344 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0)
2345 		caps |= IFCAP_TSO4;
2346 
2347 	if ((feat->offload.tx &
2348 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK) != 0)
2349 		caps |= IFCAP_TSO6;
2350 
2351 	if ((feat->offload.rx_supported &
2352 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2353 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK)) != 0)
2354 		caps |= IFCAP_RXCSUM;
2355 
2356 	if ((feat->offload.rx_supported &
2357 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK) != 0)
2358 		caps |= IFCAP_RXCSUM_IPV6;
2359 
2360 	caps |= IFCAP_LRO | IFCAP_JUMBO_MTU;
2361 
2362 	return (caps);
2363 }
2364 
2365 static void
2366 ena_update_host_info(struct ena_admin_host_info *host_info, if_t ifp)
2367 {
2368 
2369 	host_info->supported_network_features[0] =
2370 	    (uint32_t)if_getcapabilities(ifp);
2371 }
2372 
2373 static void
2374 ena_update_hwassist(struct ena_adapter *adapter)
2375 {
2376 	if_t ifp = adapter->ifp;
2377 	uint32_t feat = adapter->tx_offload_cap;
2378 	int cap = if_getcapenable(ifp);
2379 	int flags = 0;
2380 
2381 	if_clearhwassist(ifp);
2382 
2383 	if ((cap & IFCAP_TXCSUM) != 0) {
2384 		if ((feat &
2385 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK) != 0)
2386 			flags |= CSUM_IP;
2387 		if ((feat &
2388 		    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2389 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)) != 0)
2390 			flags |= CSUM_IP_UDP | CSUM_IP_TCP;
2391 	}
2392 
2393 	if ((cap & IFCAP_TXCSUM_IPV6) != 0)
2394 		flags |= CSUM_IP6_UDP | CSUM_IP6_TCP;
2395 
2396 	if ((cap & IFCAP_TSO4) != 0)
2397 		flags |= CSUM_IP_TSO;
2398 
2399 	if ((cap & IFCAP_TSO6) != 0)
2400 		flags |= CSUM_IP6_TSO;
2401 
2402 	if_sethwassistbits(ifp, flags, 0);
2403 }
2404 
2405 static int
2406 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2407     struct ena_com_dev_get_features_ctx *feat)
2408 {
2409 	if_t ifp;
2410 	int caps = 0;
2411 
2412 	ifp = adapter->ifp = if_gethandle(IFT_ETHER);
2413 	if (unlikely(ifp == NULL)) {
2414 		ena_log(pdev, ERR, "can not allocate ifnet structure\n");
2415 		return (ENXIO);
2416 	}
2417 	if_initname(ifp, device_get_name(pdev), device_get_unit(pdev));
2418 	if_setdev(ifp, pdev);
2419 	if_setsoftc(ifp, adapter);
2420 
2421 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST |
2422 	    IFF_KNOWSEPOCH);
2423 	if_setinitfn(ifp, ena_init);
2424 	if_settransmitfn(ifp, ena_mq_start);
2425 	if_setqflushfn(ifp, ena_qflush);
2426 	if_setioctlfn(ifp, ena_ioctl);
2427 	if_setgetcounterfn(ifp, ena_get_counter);
2428 
2429 	if_setsendqlen(ifp, adapter->requested_tx_ring_size);
2430 	if_setsendqready(ifp);
2431 	if_setmtu(ifp, ETHERMTU);
2432 	if_setbaudrate(ifp, 0);
2433 	/* Zeroize capabilities... */
2434 	if_setcapabilities(ifp, 0);
2435 	if_setcapenable(ifp, 0);
2436 	/* check hardware support */
2437 	caps = ena_get_dev_offloads(feat);
2438 	/* ... and set them */
2439 	if_setcapabilitiesbit(ifp, caps, 0);
2440 
2441 	/* TSO parameters */
2442 	ifp->if_hw_tsomax = ENA_TSO_MAXSIZE -
2443 	    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2444 	ifp->if_hw_tsomaxsegcount = adapter->max_tx_sgl_size - 1;
2445 	ifp->if_hw_tsomaxsegsize = ENA_TSO_MAXSIZE;
2446 
2447 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
2448 	if_setcapenable(ifp, if_getcapabilities(ifp));
2449 
2450 	/*
2451 	 * Specify the media types supported by this adapter and register
2452 	 * callbacks to update media and link information
2453 	 */
2454 	ifmedia_init(&adapter->media, IFM_IMASK,
2455 	    ena_media_change, ena_media_status);
2456 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2457 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2458 
2459 	ether_ifattach(ifp, adapter->mac_addr);
2460 
2461 	return (0);
2462 }
2463 
2464 void
2465 ena_down(struct ena_adapter *adapter)
2466 {
2467 	int rc;
2468 
2469 	ENA_LOCK_ASSERT();
2470 
2471 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2472 		return;
2473 
2474 	/* Drain timer service to avoid admin queue race condition. */
2475 	ENA_TIMER_DRAIN(adapter);
2476 
2477 	ena_log(adapter->pdev, INFO, "device is going DOWN\n");
2478 
2479 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP, adapter);
2480 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE,
2481 		IFF_DRV_RUNNING);
2482 
2483 	ena_free_io_irq(adapter);
2484 
2485 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter)) {
2486 		rc = ena_com_dev_reset(adapter->ena_dev,
2487 			adapter->reset_reason);
2488 		if (unlikely(rc != 0))
2489 			ena_log(adapter->pdev, ERR,
2490 				"Device reset failed\n");
2491 	}
2492 
2493 	ena_destroy_all_io_queues(adapter);
2494 
2495 	ena_free_all_tx_bufs(adapter);
2496 	ena_free_all_rx_bufs(adapter);
2497 	ena_free_all_tx_resources(adapter);
2498 	ena_free_all_rx_resources(adapter);
2499 
2500 	counter_u64_add(adapter->dev_stats.interface_down, 1);
2501 
2502 	ENA_TIMER_RESET(adapter);
2503 }
2504 
2505 static uint32_t
2506 ena_calc_max_io_queue_num(device_t pdev, struct ena_com_dev *ena_dev,
2507     struct ena_com_dev_get_features_ctx *get_feat_ctx)
2508 {
2509 	uint32_t io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
2510 
2511 	/* Regular queues capabilities */
2512 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2513 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2514 		    &get_feat_ctx->max_queue_ext.max_queue_ext;
2515 		io_rx_num = min_t(int, max_queue_ext->max_rx_sq_num,
2516 			max_queue_ext->max_rx_cq_num);
2517 
2518 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
2519 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
2520 	} else {
2521 		struct ena_admin_queue_feature_desc *max_queues =
2522 		    &get_feat_ctx->max_queues;
2523 		io_tx_sq_num = max_queues->max_sq_num;
2524 		io_tx_cq_num = max_queues->max_cq_num;
2525 		io_rx_num = min_t(int, io_tx_sq_num, io_tx_cq_num);
2526 	}
2527 
2528 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
2529 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
2530 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
2531 
2532 	max_num_io_queues = min_t(uint32_t, mp_ncpus, ENA_MAX_NUM_IO_QUEUES);
2533 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_rx_num);
2534 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_tx_sq_num);
2535 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_tx_cq_num);
2536 	/* 1 IRQ for mgmnt and 1 IRQ for each TX/RX pair */
2537 	max_num_io_queues = min_t(uint32_t, max_num_io_queues,
2538 	    pci_msix_count(pdev) - 1);
2539 #ifdef RSS
2540 	max_num_io_queues = min_t(uint32_t, max_num_io_queues,
2541 	    rss_getnumbuckets());
2542 #endif
2543 
2544 	return (max_num_io_queues);
2545 }
2546 
2547 static int
2548 ena_enable_wc(device_t pdev, struct resource *res)
2549 {
2550 #if defined(__i386) || defined(__amd64) || defined(__aarch64__)
2551 	vm_offset_t va;
2552 	vm_size_t len;
2553 	int rc;
2554 
2555 	va = (vm_offset_t)rman_get_virtual(res);
2556 	len = rman_get_size(res);
2557 	/* Enable write combining */
2558 	rc = pmap_change_attr(va, len, VM_MEMATTR_WRITE_COMBINING);
2559 	if (unlikely(rc != 0)) {
2560 		ena_log(pdev, ERR, "pmap_change_attr failed, %d\n", rc);
2561 		return (rc);
2562 	}
2563 
2564 	return (0);
2565 #endif
2566 	return (EOPNOTSUPP);
2567 }
2568 
2569 static int
2570 ena_set_queues_placement_policy(device_t pdev, struct ena_com_dev *ena_dev,
2571     struct ena_admin_feature_llq_desc *llq,
2572     struct ena_llq_configurations *llq_default_configurations)
2573 {
2574 	struct ena_adapter *adapter = device_get_softc(pdev);
2575 	int rc, rid;
2576 	uint32_t llq_feature_mask;
2577 
2578 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
2579 	if (!(ena_dev->supported_features & llq_feature_mask)) {
2580 		ena_log(pdev, WARN,
2581 		    "LLQ is not supported. Fallback to host mode policy.\n");
2582 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2583 		return (0);
2584 	}
2585 
2586 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
2587 	if (unlikely(rc != 0)) {
2588 		ena_log(pdev, WARN, "Failed to configure the device mode. "
2589 		    "Fallback to host mode policy.\n");
2590 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2591 		return (0);
2592 	}
2593 
2594 	/* Nothing to config, exit */
2595 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
2596 		return (0);
2597 
2598 	/* Try to allocate resources for LLQ bar */
2599 	rid = PCIR_BAR(ENA_MEM_BAR);
2600 	adapter->memory = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
2601 	    &rid, RF_ACTIVE);
2602 	if (unlikely(adapter->memory == NULL)) {
2603 		ena_log(pdev, WARN, "unable to allocate LLQ bar resource. "
2604 		    "Fallback to host mode policy.\n");
2605 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2606 		return (0);
2607 	}
2608 
2609 	/* Enable write combining for better LLQ performance */
2610 	rc = ena_enable_wc(adapter->pdev, adapter->memory);
2611 	if (unlikely(rc != 0)) {
2612 		ena_log(pdev, ERR, "failed to enable write combining.\n");
2613 		return (rc);
2614 	}
2615 
2616 	/*
2617 	 * Save virtual address of the device's memory region
2618 	 * for the ena_com layer.
2619 	 */
2620 	ena_dev->mem_bar = rman_get_virtual(adapter->memory);
2621 
2622 	return (0);
2623 }
2624 
2625 static inline
2626 void set_default_llq_configurations(struct ena_llq_configurations *llq_config,
2627 	struct ena_admin_feature_llq_desc *llq)
2628 {
2629 
2630 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
2631 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
2632 	llq_config->llq_num_decs_before_header =
2633 	    ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
2634 	if ((llq->entry_size_ctrl_supported &
2635 	     ENA_ADMIN_LIST_ENTRY_SIZE_256B) != 0 &&
2636 	    ena_force_large_llq_header) {
2637 		llq_config->llq_ring_entry_size =
2638 		    ENA_ADMIN_LIST_ENTRY_SIZE_256B;
2639 		llq_config->llq_ring_entry_size_value = 256;
2640 	} else {
2641 		llq_config->llq_ring_entry_size =
2642 		    ENA_ADMIN_LIST_ENTRY_SIZE_128B;
2643 		llq_config->llq_ring_entry_size_value = 128;
2644 	}
2645 }
2646 
2647 static int
2648 ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
2649 {
2650 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
2651 	struct ena_com_dev *ena_dev = ctx->ena_dev;
2652 	uint32_t tx_queue_size = ENA_DEFAULT_RING_SIZE;
2653 	uint32_t rx_queue_size = ENA_DEFAULT_RING_SIZE;
2654 	uint32_t max_tx_queue_size;
2655 	uint32_t max_rx_queue_size;
2656 
2657 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2658 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2659 		    &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
2660 		max_rx_queue_size = min_t(uint32_t,
2661 		    max_queue_ext->max_rx_cq_depth,
2662 		    max_queue_ext->max_rx_sq_depth);
2663 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
2664 
2665 		if (ena_dev->tx_mem_queue_type ==
2666 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2667 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2668 			    llq->max_llq_depth);
2669 		else
2670 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2671 			    max_queue_ext->max_tx_sq_depth);
2672 
2673 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2674 		    max_queue_ext->max_per_packet_tx_descs);
2675 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2676 		    max_queue_ext->max_per_packet_rx_descs);
2677 	} else {
2678 		struct ena_admin_queue_feature_desc *max_queues =
2679 		    &ctx->get_feat_ctx->max_queues;
2680 		max_rx_queue_size = min_t(uint32_t,
2681 		    max_queues->max_cq_depth,
2682 		    max_queues->max_sq_depth);
2683 		max_tx_queue_size = max_queues->max_cq_depth;
2684 
2685 		if (ena_dev->tx_mem_queue_type ==
2686 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2687 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2688 			    llq->max_llq_depth);
2689 		else
2690 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2691 			    max_queues->max_sq_depth);
2692 
2693 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2694 		    max_queues->max_packet_tx_descs);
2695 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2696 		    max_queues->max_packet_rx_descs);
2697 	}
2698 
2699 	/* round down to the nearest power of 2 */
2700 	max_tx_queue_size = 1 << (flsl(max_tx_queue_size) - 1);
2701 	max_rx_queue_size = 1 << (flsl(max_rx_queue_size) - 1);
2702 
2703 	/*
2704 	 * When forcing large headers, we multiply the entry size by 2,
2705 	 * and therefore divide the queue size by 2, leaving the amount
2706 	 * of memory used by the queues unchanged.
2707 	 */
2708 	if (ena_force_large_llq_header) {
2709 		if ((llq->entry_size_ctrl_supported &
2710 		     ENA_ADMIN_LIST_ENTRY_SIZE_256B) != 0 &&
2711 		    ena_dev->tx_mem_queue_type ==
2712 		     ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2713 			max_tx_queue_size /= 2;
2714 			ena_log(ctx->pdev, INFO,
2715 			    "Forcing large headers and decreasing maximum Tx queue size to %d\n",
2716 			    max_tx_queue_size);
2717 		} else {
2718 			ena_log(ctx->pdev, WARN,
2719 			    "Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
2720 		}
2721 	}
2722 
2723 	tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
2724 	    max_tx_queue_size);
2725 	rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
2726 	    max_rx_queue_size);
2727 
2728 	tx_queue_size = 1 << (flsl(tx_queue_size) - 1);
2729 	rx_queue_size = 1 << (flsl(rx_queue_size) - 1);
2730 
2731 	ctx->max_tx_queue_size = max_tx_queue_size;
2732 	ctx->max_rx_queue_size = max_rx_queue_size;
2733 	ctx->tx_queue_size = tx_queue_size;
2734 	ctx->rx_queue_size = rx_queue_size;
2735 
2736 	return (0);
2737 }
2738 
2739 static void
2740 ena_config_host_info(struct ena_com_dev *ena_dev, device_t dev)
2741 {
2742 	struct ena_admin_host_info *host_info;
2743 	uintptr_t rid;
2744 	int rc;
2745 
2746 	/* Allocate only the host info */
2747 	rc = ena_com_allocate_host_info(ena_dev);
2748 	if (unlikely(rc != 0)) {
2749 		ena_log(dev, ERR, "Cannot allocate host info\n");
2750 		return;
2751 	}
2752 
2753 	host_info = ena_dev->host_attr.host_info;
2754 
2755 	if (pci_get_id(dev, PCI_ID_RID, &rid) == 0)
2756 		host_info->bdf = rid;
2757 	host_info->os_type = ENA_ADMIN_OS_FREEBSD;
2758 	host_info->kernel_ver = osreldate;
2759 
2760 	sprintf(host_info->kernel_ver_str, "%d", osreldate);
2761 	host_info->os_dist = 0;
2762 	strncpy(host_info->os_dist_str, osrelease,
2763 	    sizeof(host_info->os_dist_str) - 1);
2764 
2765 	host_info->driver_version =
2766 		(DRV_MODULE_VER_MAJOR) |
2767 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2768 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2769 	host_info->num_cpus = mp_ncpus;
2770 	host_info->driver_supported_features =
2771 	    ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
2772 	    ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
2773 
2774 	rc = ena_com_set_host_attributes(ena_dev);
2775 	if (unlikely(rc != 0)) {
2776 		if (rc == EOPNOTSUPP)
2777 			ena_log(dev, WARN, "Cannot set host attributes\n");
2778 		else
2779 			ena_log(dev, ERR, "Cannot set host attributes\n");
2780 
2781 		goto err;
2782 	}
2783 
2784 	return;
2785 
2786 err:
2787 	ena_com_delete_host_info(ena_dev);
2788 }
2789 
2790 static int
2791 ena_device_init(struct ena_adapter *adapter, device_t pdev,
2792     struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
2793 {
2794 	struct ena_com_dev* ena_dev = adapter->ena_dev;
2795 	bool readless_supported;
2796 	uint32_t aenq_groups;
2797 	int dma_width;
2798 	int rc;
2799 
2800 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
2801 	if (unlikely(rc != 0)) {
2802 		ena_log(pdev, ERR, "failed to init mmio read less\n");
2803 		return (rc);
2804 	}
2805 
2806 	/*
2807 	 * The PCIe configuration space revision id indicate if mmio reg
2808 	 * read is disabled
2809 	 */
2810 	readless_supported = !(pci_get_revid(pdev) & ENA_MMIO_DISABLE_REG_READ);
2811 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2812 
2813 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2814 	if (unlikely(rc != 0)) {
2815 		ena_log(pdev, ERR, "Can not reset device\n");
2816 		goto err_mmio_read_less;
2817 	}
2818 
2819 	rc = ena_com_validate_version(ena_dev);
2820 	if (unlikely(rc != 0)) {
2821 		ena_log(pdev, ERR, "device version is too low\n");
2822 		goto err_mmio_read_less;
2823 	}
2824 
2825 	dma_width = ena_com_get_dma_width(ena_dev);
2826 	if (unlikely(dma_width < 0)) {
2827 		ena_log(pdev, ERR, "Invalid dma width value %d", dma_width);
2828 		rc = dma_width;
2829 		goto err_mmio_read_less;
2830 	}
2831 	adapter->dma_width = dma_width;
2832 
2833 	/* ENA admin level init */
2834 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
2835 	if (unlikely(rc != 0)) {
2836 		ena_log(pdev, ERR,
2837 		    "Can not initialize ena admin queue with device\n");
2838 		goto err_mmio_read_less;
2839 	}
2840 
2841 	/*
2842 	 * To enable the msix interrupts the driver needs to know the number
2843 	 * of queues. So the driver uses polling mode to retrieve this
2844 	 * information
2845 	 */
2846 	ena_com_set_admin_polling_mode(ena_dev, true);
2847 
2848 	ena_config_host_info(ena_dev, pdev);
2849 
2850 	/* Get Device Attributes */
2851 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2852 	if (unlikely(rc != 0)) {
2853 		ena_log(pdev, ERR,
2854 		    "Cannot get attribute for ena device rc: %d\n", rc);
2855 		goto err_admin_init;
2856 	}
2857 
2858 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2859 	    BIT(ENA_ADMIN_FATAL_ERROR) |
2860 	    BIT(ENA_ADMIN_WARNING) |
2861 	    BIT(ENA_ADMIN_NOTIFICATION) |
2862 	    BIT(ENA_ADMIN_KEEP_ALIVE);
2863 
2864 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
2865 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2866 	if (unlikely(rc != 0)) {
2867 		ena_log(pdev, ERR, "Cannot configure aenq groups rc: %d\n", rc);
2868 		goto err_admin_init;
2869 	}
2870 
2871 	*wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2872 
2873 	return (0);
2874 
2875 err_admin_init:
2876 	ena_com_delete_host_info(ena_dev);
2877 	ena_com_admin_destroy(ena_dev);
2878 err_mmio_read_less:
2879 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2880 
2881 	return (rc);
2882 }
2883 
2884 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
2885 {
2886 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2887 	int rc;
2888 
2889 	rc = ena_enable_msix(adapter);
2890 	if (unlikely(rc != 0)) {
2891 		ena_log(adapter->pdev, ERR, "Error with MSI-X enablement\n");
2892 		return (rc);
2893 	}
2894 
2895 	ena_setup_mgmnt_intr(adapter);
2896 
2897 	rc = ena_request_mgmnt_irq(adapter);
2898 	if (unlikely(rc != 0)) {
2899 		ena_log(adapter->pdev, ERR, "Cannot setup mgmnt queue intr\n");
2900 		goto err_disable_msix;
2901 	}
2902 
2903 	ena_com_set_admin_polling_mode(ena_dev, false);
2904 
2905 	ena_com_admin_aenq_enable(ena_dev);
2906 
2907 	return (0);
2908 
2909 err_disable_msix:
2910 	ena_disable_msix(adapter);
2911 
2912 	return (rc);
2913 }
2914 
2915 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
2916 static void ena_keep_alive_wd(void *adapter_data,
2917     struct ena_admin_aenq_entry *aenq_e)
2918 {
2919 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
2920 	struct ena_admin_aenq_keep_alive_desc *desc;
2921 	sbintime_t stime;
2922 	uint64_t rx_drops;
2923 	uint64_t tx_drops;
2924 
2925 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
2926 
2927 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
2928 	tx_drops = ((uint64_t)desc->tx_drops_high << 32) | desc->tx_drops_low;
2929 	counter_u64_zero(adapter->hw_stats.rx_drops);
2930 	counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
2931 	counter_u64_zero(adapter->hw_stats.tx_drops);
2932 	counter_u64_add(adapter->hw_stats.tx_drops, tx_drops);
2933 
2934 	stime = getsbinuptime();
2935 	atomic_store_rel_64(&adapter->keep_alive_timestamp, stime);
2936 }
2937 
2938 /* Check for keep alive expiration */
2939 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
2940 {
2941 	sbintime_t timestamp, time;
2942 
2943 	if (adapter->wd_active == 0)
2944 		return;
2945 
2946 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2947 		return;
2948 
2949 	timestamp = atomic_load_acq_64(&adapter->keep_alive_timestamp);
2950 	time = getsbinuptime() - timestamp;
2951 	if (unlikely(time > adapter->keep_alive_timeout)) {
2952 		ena_log(adapter->pdev, ERR, "Keep alive watchdog timeout.\n");
2953 		counter_u64_add(adapter->dev_stats.wd_expired, 1);
2954 		ena_trigger_reset(adapter, ENA_REGS_RESET_KEEP_ALIVE_TO);
2955 	}
2956 }
2957 
2958 /* Check if admin queue is enabled */
2959 static void check_for_admin_com_state(struct ena_adapter *adapter)
2960 {
2961 	if (unlikely(ena_com_get_admin_running_state(adapter->ena_dev) ==
2962 	    false)) {
2963 		ena_log(adapter->pdev, ERR,
2964 		    "ENA admin queue is not in running state!\n");
2965 		counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
2966 		ena_trigger_reset(adapter, ENA_REGS_RESET_ADMIN_TO);
2967 	}
2968 }
2969 
2970 static int
2971 check_for_rx_interrupt_queue(struct ena_adapter *adapter,
2972     struct ena_ring *rx_ring)
2973 {
2974 	if (likely(rx_ring->first_interrupt))
2975 		return (0);
2976 
2977 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
2978 		return (0);
2979 
2980 	rx_ring->no_interrupt_event_cnt++;
2981 
2982 	if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
2983 		ena_log(adapter->pdev, ERR, "Potential MSIX issue on Rx side "
2984 		    "Queue = %d. Reset the device\n", rx_ring->qid);
2985 		ena_trigger_reset(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
2986 		return (EIO);
2987 	}
2988 
2989 	return (0);
2990 }
2991 
2992 static int
2993 check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
2994     struct ena_ring *tx_ring)
2995 {
2996 	device_t pdev = adapter->pdev;
2997 	struct bintime curtime, time;
2998 	struct ena_tx_buffer *tx_buf;
2999 	sbintime_t time_offset;
3000 	uint32_t missed_tx = 0;
3001 	int i, rc = 0;
3002 
3003 	getbinuptime(&curtime);
3004 
3005 	for (i = 0; i < tx_ring->ring_size; i++) {
3006 		tx_buf = &tx_ring->tx_buffer_info[i];
3007 
3008 		if (bintime_isset(&tx_buf->timestamp) == 0)
3009 			continue;
3010 
3011 		time = curtime;
3012 		bintime_sub(&time, &tx_buf->timestamp);
3013 		time_offset = bttosbt(time);
3014 
3015 		if (unlikely(!tx_ring->first_interrupt &&
3016 		    time_offset > 2 * adapter->missing_tx_timeout)) {
3017 			/*
3018 			 * If after graceful period interrupt is still not
3019 			 * received, we schedule a reset.
3020 			 */
3021 			ena_log(pdev, ERR,
3022 			    "Potential MSIX issue on Tx side Queue = %d. "
3023 			    "Reset the device\n", tx_ring->qid);
3024 			ena_trigger_reset(adapter,
3025 			    ENA_REGS_RESET_MISS_INTERRUPT);
3026 			return (EIO);
3027 		}
3028 
3029 		/* Check again if packet is still waiting */
3030 		if (unlikely(time_offset > adapter->missing_tx_timeout)) {
3031 
3032 			if (!tx_buf->print_once)
3033 				ena_log(pdev, WARN, "Found a Tx that wasn't "
3034 				    "completed on time, qid %d, index %d.\n",
3035 				    tx_ring->qid, i);
3036 
3037 			tx_buf->print_once = true;
3038 			missed_tx++;
3039 		}
3040 	}
3041 
3042 	if (unlikely(missed_tx > adapter->missing_tx_threshold)) {
3043 		ena_log(pdev, ERR,
3044 		    "The number of lost tx completion is above the threshold "
3045 		    "(%d > %d). Reset the device\n",
3046 		    missed_tx, adapter->missing_tx_threshold);
3047 		ena_trigger_reset(adapter, ENA_REGS_RESET_MISS_TX_CMPL);
3048 		rc = EIO;
3049 	}
3050 
3051 	counter_u64_add(tx_ring->tx_stats.missing_tx_comp, missed_tx);
3052 
3053 	return (rc);
3054 }
3055 
3056 /*
3057  * Check for TX which were not completed on time.
3058  * Timeout is defined by "missing_tx_timeout".
3059  * Reset will be performed if number of incompleted
3060  * transactions exceeds "missing_tx_threshold".
3061  */
3062 static void
3063 check_for_missing_completions(struct ena_adapter *adapter)
3064 {
3065 	struct ena_ring *tx_ring;
3066 	struct ena_ring *rx_ring;
3067 	int i, budget, rc;
3068 
3069 	/* Make sure the driver doesn't turn the device in other process */
3070 	rmb();
3071 
3072 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3073 		return;
3074 
3075 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
3076 		return;
3077 
3078 	if (adapter->missing_tx_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3079 		return;
3080 
3081 	budget = adapter->missing_tx_max_queues;
3082 
3083 	for (i = adapter->next_monitored_tx_qid; i < adapter->num_io_queues; i++) {
3084 		tx_ring = &adapter->tx_ring[i];
3085 		rx_ring = &adapter->rx_ring[i];
3086 
3087 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3088 		if (unlikely(rc != 0))
3089 			return;
3090 
3091 		rc = check_for_rx_interrupt_queue(adapter, rx_ring);
3092 		if (unlikely(rc != 0))
3093 			return;
3094 
3095 		budget--;
3096 		if (budget == 0) {
3097 			i++;
3098 			break;
3099 		}
3100 	}
3101 
3102 	adapter->next_monitored_tx_qid = i % adapter->num_io_queues;
3103 }
3104 
3105 /* trigger rx cleanup after 2 consecutive detections */
3106 #define EMPTY_RX_REFILL 2
3107 /* For the rare case where the device runs out of Rx descriptors and the
3108  * msix handler failed to refill new Rx descriptors (due to a lack of memory
3109  * for example).
3110  * This case will lead to a deadlock:
3111  * The device won't send interrupts since all the new Rx packets will be dropped
3112  * The msix handler won't allocate new Rx descriptors so the device won't be
3113  * able to send new packets.
3114  *
3115  * When such a situation is detected - execute rx cleanup task in another thread
3116  */
3117 static void
3118 check_for_empty_rx_ring(struct ena_adapter *adapter)
3119 {
3120 	struct ena_ring *rx_ring;
3121 	int i, refill_required;
3122 
3123 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3124 		return;
3125 
3126 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
3127 		return;
3128 
3129 	for (i = 0; i < adapter->num_io_queues; i++) {
3130 		rx_ring = &adapter->rx_ring[i];
3131 
3132 		refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3133 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3134 			rx_ring->empty_rx_queue++;
3135 
3136 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL)	{
3137 				counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
3138 				    1);
3139 
3140 				ena_log(adapter->pdev, WARN,
3141 				    "Rx ring %d is stalled. Triggering the refill function\n",
3142 				    i);
3143 
3144 				taskqueue_enqueue(rx_ring->que->cleanup_tq,
3145 				    &rx_ring->que->cleanup_task);
3146 				rx_ring->empty_rx_queue = 0;
3147 			}
3148 		} else {
3149 			rx_ring->empty_rx_queue = 0;
3150 		}
3151 	}
3152 }
3153 
3154 static void ena_update_hints(struct ena_adapter *adapter,
3155 			     struct ena_admin_ena_hw_hints *hints)
3156 {
3157 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3158 
3159 	if (hints->admin_completion_tx_timeout)
3160 		ena_dev->admin_queue.completion_timeout =
3161 		    hints->admin_completion_tx_timeout * 1000;
3162 
3163 	if (hints->mmio_read_timeout)
3164 		/* convert to usec */
3165 		ena_dev->mmio_read.reg_read_to =
3166 		    hints->mmio_read_timeout * 1000;
3167 
3168 	if (hints->missed_tx_completion_count_threshold_to_reset)
3169 		adapter->missing_tx_threshold =
3170 		    hints->missed_tx_completion_count_threshold_to_reset;
3171 
3172 	if (hints->missing_tx_completion_timeout) {
3173 		if (hints->missing_tx_completion_timeout ==
3174 		     ENA_HW_HINTS_NO_TIMEOUT)
3175 			adapter->missing_tx_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3176 		else
3177 			adapter->missing_tx_timeout =
3178 			    SBT_1MS * hints->missing_tx_completion_timeout;
3179 	}
3180 
3181 	if (hints->driver_watchdog_timeout) {
3182 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3183 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3184 		else
3185 			adapter->keep_alive_timeout =
3186 			    SBT_1MS * hints->driver_watchdog_timeout;
3187 	}
3188 }
3189 
3190 /**
3191  * ena_copy_eni_metrics - Get and copy ENI metrics from the HW.
3192  * @adapter: ENA device adapter
3193  *
3194  * Returns 0 on success, EOPNOTSUPP if current HW doesn't support those metrics
3195  * and other error codes on failure.
3196  *
3197  * This function can possibly cause a race with other calls to the admin queue.
3198  * Because of that, the caller should either lock this function or make sure
3199  * that there is no race in the current context.
3200  */
3201 static int
3202 ena_copy_eni_metrics(struct ena_adapter *adapter)
3203 {
3204 	static bool print_once = true;
3205 	int rc;
3206 
3207 	rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_metrics);
3208 
3209 	if (rc != 0) {
3210 		if (rc == ENA_COM_UNSUPPORTED) {
3211 			if (print_once) {
3212 				ena_log(adapter->pdev, WARN,
3213 				    "Retrieving ENI metrics is not supported.\n");
3214 				print_once = false;
3215 			} else {
3216 				ena_log(adapter->pdev, DBG,
3217 				    "Retrieving ENI metrics is not supported.\n");
3218 			}
3219 		} else {
3220 			ena_log(adapter->pdev, ERR,
3221 			    "Failed to get ENI metrics: %d\n", rc);
3222 		}
3223 	}
3224 
3225 	return (rc);
3226 }
3227 
3228 static void
3229 ena_timer_service(void *data)
3230 {
3231 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3232 	struct ena_admin_host_info *host_info =
3233 	    adapter->ena_dev->host_attr.host_info;
3234 
3235 	check_for_missing_keep_alive(adapter);
3236 
3237 	check_for_admin_com_state(adapter);
3238 
3239 	check_for_missing_completions(adapter);
3240 
3241 	check_for_empty_rx_ring(adapter);
3242 
3243 	/*
3244 	 * User controller update of the ENI metrics.
3245 	 * If the delay was set to 0, then the stats shouldn't be updated at
3246 	 * all.
3247 	 * Otherwise, wait 'eni_metrics_sample_interval' seconds, before
3248 	 * updating stats.
3249 	 * As timer service is executed every second, it's enough to increment
3250 	 * appropriate counter each time the timer service is executed.
3251 	 */
3252 	if ((adapter->eni_metrics_sample_interval != 0) &&
3253 	    (++adapter->eni_metrics_sample_interval_cnt >=
3254 	     adapter->eni_metrics_sample_interval)) {
3255 		/*
3256 		 * There is no race with other admin queue calls, as:
3257 		 *   - Timer service runs after attach function ends, so all
3258 		 *     configuration calls to the admin queue are finished.
3259 		 *   - Timer service is temporarily stopped when bringing
3260 		 *     the interface up or down.
3261 		 *   - After interface is up, the driver doesn't use (at least
3262 		 *     for now) other functions writing to the admin queue.
3263 		 *
3264 		 * It may change in the future, so in that situation, the lock
3265 		 * will be needed. ENA_LOCK_*() cannot be used for that purpose,
3266 		 * as callout ena_timer_service is protected by them. It could
3267 		 * lead to the deadlock if callout_drain() would hold the lock
3268 		 * before ena_copy_eni_metrics() was executed. It's advised to
3269 		 * use separate lock in that situation which will be used only
3270 		 * for the admin queue.
3271 		 */
3272 		(void)ena_copy_eni_metrics(adapter);
3273 		adapter->eni_metrics_sample_interval_cnt = 0;
3274 	}
3275 
3276 
3277 	if (host_info != NULL)
3278 		ena_update_host_info(host_info, adapter->ifp);
3279 
3280 	if (unlikely(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
3281 		/*
3282 		 * Timeout when validating version indicates that the device
3283 		 * became unresponsive. If that happens skip the reset and
3284 		 * reschedule timer service, so the reset can be retried later.
3285 		 */
3286 		if (ena_com_validate_version(adapter->ena_dev) ==
3287 		    ENA_COM_TIMER_EXPIRED) {
3288 			ena_log(adapter->pdev, WARN,
3289 			    "FW unresponsive, skipping reset\n");
3290 			ENA_TIMER_RESET(adapter);
3291 			return;
3292 		}
3293 		ena_log(adapter->pdev, WARN, "Trigger reset is on\n");
3294 		taskqueue_enqueue(adapter->reset_tq, &adapter->reset_task);
3295 		return;
3296 	}
3297 
3298 	/*
3299 	 * Schedule another timeout one second from now.
3300 	 */
3301 	ENA_TIMER_RESET(adapter);
3302 }
3303 
3304 void
3305 ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3306 {
3307 	if_t ifp = adapter->ifp;
3308 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3309 	bool dev_up;
3310 
3311 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter))
3312 		return;
3313 
3314 	if_link_state_change(ifp, LINK_STATE_DOWN);
3315 
3316 	ENA_TIMER_DRAIN(adapter);
3317 
3318 	dev_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
3319 	if (dev_up)
3320 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3321 
3322 	if (!graceful)
3323 		ena_com_set_admin_running_state(ena_dev, false);
3324 
3325 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3326 		ena_down(adapter);
3327 
3328 	/*
3329 	 * Stop the device from sending AENQ events (if the device was up, and
3330 	 * the trigger reset was on, ena_down already performs device reset)
3331 	 */
3332 	if (!(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter) && dev_up))
3333 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3334 
3335 	ena_free_mgmnt_irq(adapter);
3336 
3337 	ena_disable_msix(adapter);
3338 
3339 	/*
3340 	 * IO rings resources should be freed because `ena_restore_device()`
3341 	 * calls (not directly) `ena_enable_msix()`, which re-allocates MSIX
3342 	 * vectors. The amount of MSIX vectors after destroy-restore may be
3343 	 * different than before. Therefore, IO rings resources should be
3344 	 * established from scratch each time.
3345 	 */
3346 	ena_free_all_io_rings_resources(adapter);
3347 
3348 	ena_com_abort_admin_commands(ena_dev);
3349 
3350 	ena_com_wait_for_abort_completion(ena_dev);
3351 
3352 	ena_com_admin_destroy(ena_dev);
3353 
3354 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3355 
3356 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3357 
3358 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
3359 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3360 }
3361 
3362 static int
3363 ena_device_validate_params(struct ena_adapter *adapter,
3364     struct ena_com_dev_get_features_ctx *get_feat_ctx)
3365 {
3366 
3367 	if (memcmp(get_feat_ctx->dev_attr.mac_addr, adapter->mac_addr,
3368 	    ETHER_ADDR_LEN) != 0) {
3369 		ena_log(adapter->pdev, ERR, "Error, mac addresses differ\n");
3370 		return (EINVAL);
3371 	}
3372 
3373 	if (get_feat_ctx->dev_attr.max_mtu < if_getmtu(adapter->ifp)) {
3374 		ena_log(adapter->pdev, ERR,
3375 		    "Error, device max mtu is smaller than ifp MTU\n");
3376 		return (EINVAL);
3377 	}
3378 
3379 	return 0;
3380 }
3381 
3382 int
3383 ena_restore_device(struct ena_adapter *adapter)
3384 {
3385 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3386 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3387 	if_t ifp = adapter->ifp;
3388 	device_t dev = adapter->pdev;
3389 	int wd_active;
3390 	int rc;
3391 
3392 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3393 
3394 	rc = ena_device_init(adapter, dev, &get_feat_ctx, &wd_active);
3395 	if (rc != 0) {
3396 		ena_log(dev, ERR, "Cannot initialize device\n");
3397 		goto err;
3398 	}
3399 	/*
3400 	 * Only enable WD if it was enabled before reset, so it won't override
3401 	 * value set by the user by the sysctl.
3402 	 */
3403 	if (adapter->wd_active != 0)
3404 		adapter->wd_active = wd_active;
3405 
3406 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
3407 	if (rc != 0) {
3408 		ena_log(dev, ERR, "Validation of device parameters failed\n");
3409 		goto err_device_destroy;
3410 	}
3411 
3412 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3413 	/* Make sure we don't have a race with AENQ Links state handler */
3414 	if (ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter))
3415 		if_link_state_change(ifp, LINK_STATE_UP);
3416 
3417 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3418 	if (rc != 0) {
3419 		ena_log(dev, ERR, "Enable MSI-X failed\n");
3420 		goto err_device_destroy;
3421 	}
3422 
3423 	/*
3424 	 * Effective value of used MSIX vectors should be the same as before
3425 	 * `ena_destroy_device()`, if possible, or closest to it if less vectors
3426 	 * are available.
3427 	 */
3428 	if ((adapter->msix_vecs - ENA_ADMIN_MSIX_VEC) < adapter->num_io_queues)
3429 		adapter->num_io_queues =
3430 		    adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
3431 
3432 	/* Re-initialize rings basic information */
3433 	ena_init_io_rings(adapter);
3434 
3435 	/* If the interface was up before the reset bring it up */
3436 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter)) {
3437 		rc = ena_up(adapter);
3438 		if (rc != 0) {
3439 			ena_log(dev, ERR, "Failed to create I/O queues\n");
3440 			goto err_disable_msix;
3441 		}
3442 	}
3443 
3444 	/* Indicate that device is running again and ready to work */
3445 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3446 
3447 	/*
3448 	 * As the AENQ handlers weren't executed during reset because
3449 	 * the flag ENA_FLAG_DEVICE_RUNNING was turned off, the
3450 	 * timestamp must be updated again That will prevent next reset
3451 	 * caused by missing keep alive.
3452 	 */
3453 	adapter->keep_alive_timestamp = getsbinuptime();
3454 	ENA_TIMER_RESET(adapter);
3455 
3456 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3457 
3458 	ena_log(dev, INFO,
3459 	    "Device reset completed successfully, Driver info: %s\n", ena_version);
3460 
3461 	return (rc);
3462 
3463 err_disable_msix:
3464 	ena_free_mgmnt_irq(adapter);
3465 	ena_disable_msix(adapter);
3466 err_device_destroy:
3467 	ena_com_abort_admin_commands(ena_dev);
3468 	ena_com_wait_for_abort_completion(ena_dev);
3469 	ena_com_admin_destroy(ena_dev);
3470 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3471 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3472 err:
3473 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3474 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3475 	ena_log(dev, ERR, "Reset attempt failed. Can not reset the device\n");
3476 
3477 	ENA_TIMER_RESET(adapter);
3478 
3479 	return (rc);
3480 }
3481 
3482 static void
3483 ena_reset_task(void *arg, int pending)
3484 {
3485 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3486 
3487 	ENA_LOCK_LOCK();
3488 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
3489 		ena_destroy_device(adapter, false);
3490 		ena_restore_device(adapter);
3491 	}
3492 	ENA_LOCK_UNLOCK();
3493 }
3494 
3495 /**
3496  * ena_attach - Device Initialization Routine
3497  * @pdev: device information struct
3498  *
3499  * Returns 0 on success, otherwise on failure.
3500  *
3501  * ena_attach initializes an adapter identified by a device structure.
3502  * The OS initialization, configuring of the adapter private structure,
3503  * and a hardware reset occur.
3504  **/
3505 static int
3506 ena_attach(device_t pdev)
3507 {
3508 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3509 	struct ena_llq_configurations llq_config;
3510 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
3511 	static int version_printed;
3512 	struct ena_adapter *adapter;
3513 	struct ena_com_dev *ena_dev = NULL;
3514 	uint32_t max_num_io_queues;
3515 	int msix_rid;
3516 	int rid, rc;
3517 
3518 	adapter = device_get_softc(pdev);
3519 	adapter->pdev = pdev;
3520 	adapter->first_bind = -1;
3521 
3522 	/*
3523 	 * Set up the timer service - driver is responsible for avoiding
3524 	 * concurrency, as the callout won't be using any locking inside.
3525 	 */
3526 	ENA_TIMER_INIT(adapter);
3527 	adapter->keep_alive_timeout = DEFAULT_KEEP_ALIVE_TO;
3528 	adapter->missing_tx_timeout = DEFAULT_TX_CMP_TO;
3529 	adapter->missing_tx_max_queues = DEFAULT_TX_MONITORED_QUEUES;
3530 	adapter->missing_tx_threshold = DEFAULT_TX_CMP_THRESHOLD;
3531 
3532 	if (version_printed++ == 0)
3533 		ena_log(pdev, INFO, "%s\n", ena_version);
3534 
3535 	/* Allocate memory for ena_dev structure */
3536 	ena_dev = malloc(sizeof(struct ena_com_dev), M_DEVBUF,
3537 	    M_WAITOK | M_ZERO);
3538 
3539 	adapter->ena_dev = ena_dev;
3540 	ena_dev->dmadev = pdev;
3541 
3542 	rid = PCIR_BAR(ENA_REG_BAR);
3543 	adapter->memory = NULL;
3544 	adapter->registers = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
3545 	    &rid, RF_ACTIVE);
3546 	if (unlikely(adapter->registers == NULL)) {
3547 		ena_log(pdev, ERR,
3548 		    "unable to allocate bus resource: registers!\n");
3549 		rc = ENOMEM;
3550 		goto err_dev_free;
3551 	}
3552 
3553 	/* MSIx vector table may reside on BAR0 with registers or on BAR1. */
3554 	msix_rid = pci_msix_table_bar(pdev);
3555 	if (msix_rid != rid) {
3556 		adapter->msix = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
3557 		    &msix_rid, RF_ACTIVE);
3558 		if (unlikely(adapter->msix == NULL)) {
3559 			ena_log(pdev, ERR,
3560 			    "unable to allocate bus resource: msix!\n");
3561 			rc = ENOMEM;
3562 			goto err_pci_free;
3563 		}
3564 		adapter->msix_rid = msix_rid;
3565 	}
3566 
3567 	ena_dev->bus = malloc(sizeof(struct ena_bus), M_DEVBUF,
3568 	    M_WAITOK | M_ZERO);
3569 
3570 	/* Store register resources */
3571 	((struct ena_bus*)(ena_dev->bus))->reg_bar_t =
3572 	    rman_get_bustag(adapter->registers);
3573 	((struct ena_bus*)(ena_dev->bus))->reg_bar_h =
3574 	    rman_get_bushandle(adapter->registers);
3575 
3576 	if (unlikely(((struct ena_bus*)(ena_dev->bus))->reg_bar_h == 0)) {
3577 		ena_log(pdev, ERR, "failed to pmap registers bar\n");
3578 		rc = ENXIO;
3579 		goto err_bus_free;
3580 	}
3581 
3582 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3583 
3584 	/* Initially clear all the flags */
3585 	ENA_FLAG_ZERO(adapter);
3586 
3587 	/* Device initialization */
3588 	rc = ena_device_init(adapter, pdev, &get_feat_ctx, &adapter->wd_active);
3589 	if (unlikely(rc != 0)) {
3590 		ena_log(pdev, ERR, "ENA device init failed! (err: %d)\n", rc);
3591 		rc = ENXIO;
3592 		goto err_bus_free;
3593 	}
3594 
3595 	set_default_llq_configurations(&llq_config, &get_feat_ctx.llq);
3596 
3597 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
3598 	     &llq_config);
3599 	if (unlikely(rc != 0)) {
3600 		ena_log(pdev, ERR, "failed to set placement policy\n");
3601 		goto err_com_free;
3602 	}
3603 
3604 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3605 		adapter->disable_meta_caching =
3606 		    !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
3607 		    BIT(ENA_ADMIN_DISABLE_META_CACHING));
3608 
3609 	adapter->keep_alive_timestamp = getsbinuptime();
3610 
3611 	adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3612 
3613 	memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3614 	    ETHER_ADDR_LEN);
3615 
3616 	calc_queue_ctx.pdev = pdev;
3617 	calc_queue_ctx.ena_dev = ena_dev;
3618 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
3619 
3620 	/* Calculate initial and maximum IO queue number and size */
3621 	max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev,
3622 	    &get_feat_ctx);
3623 	rc = ena_calc_io_queue_size(&calc_queue_ctx);
3624 	if (unlikely((rc != 0) || (max_num_io_queues <= 0))) {
3625 		rc = EFAULT;
3626 		goto err_com_free;
3627 	}
3628 
3629 	adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
3630 	adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
3631 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
3632 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
3633 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
3634 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
3635 
3636 	adapter->max_num_io_queues = max_num_io_queues;
3637 
3638 	adapter->buf_ring_size = ENA_DEFAULT_BUF_RING_SIZE;
3639 
3640 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
3641 
3642 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3643 
3644 	/* set up dma tags for rx and tx buffers */
3645 	rc = ena_setup_tx_dma_tag(adapter);
3646 	if (unlikely(rc != 0)) {
3647 		ena_log(pdev, ERR, "Failed to create TX DMA tag\n");
3648 		goto err_com_free;
3649 	}
3650 
3651 	rc = ena_setup_rx_dma_tag(adapter);
3652 	if (unlikely(rc != 0)) {
3653 		ena_log(pdev, ERR, "Failed to create RX DMA tag\n");
3654 		goto err_tx_tag_free;
3655 	}
3656 
3657 	/*
3658 	 * The amount of requested MSIX vectors is equal to
3659 	 * adapter::max_num_io_queues (see `ena_enable_msix()`), plus a constant
3660 	 * number of admin queue interrupts. The former is initially determined
3661 	 * by HW capabilities (see `ena_calc_max_io_queue_num())` but may not be
3662 	 * achieved if there are not enough system resources. By default, the
3663 	 * number of effectively used IO queues is the same but later on it can
3664 	 * be limited by the user using sysctl interface.
3665 	 */
3666 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3667 	if (unlikely(rc != 0)) {
3668 		ena_log(pdev, ERR,
3669 		    "Failed to enable and set the admin interrupts\n");
3670 		goto err_io_free;
3671 	}
3672 	/* By default all of allocated MSIX vectors are actively used */
3673 	adapter->num_io_queues = adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
3674 
3675 	/* initialize rings basic information */
3676 	ena_init_io_rings(adapter);
3677 
3678 	/* setup network interface */
3679 	rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
3680 	if (unlikely(rc != 0)) {
3681 		ena_log(pdev, ERR, "Error with network interface setup\n");
3682 		goto err_msix_free;
3683 	}
3684 
3685 	/* Initialize reset task queue */
3686 	TASK_INIT(&adapter->reset_task, 0, ena_reset_task, adapter);
3687 	adapter->reset_tq = taskqueue_create("ena_reset_enqueue",
3688 	    M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->reset_tq);
3689 	taskqueue_start_threads(&adapter->reset_tq, 1, PI_NET,
3690 	    "%s rstq", device_get_nameunit(adapter->pdev));
3691 
3692 	/* Initialize statistics */
3693 	ena_alloc_counters((counter_u64_t *)&adapter->dev_stats,
3694 	    sizeof(struct ena_stats_dev));
3695 	ena_alloc_counters((counter_u64_t *)&adapter->hw_stats,
3696 	    sizeof(struct ena_hw_stats));
3697 	ena_sysctl_add_nodes(adapter);
3698 
3699 #ifdef DEV_NETMAP
3700 	rc = ena_netmap_attach(adapter);
3701 	if (rc != 0) {
3702 		ena_log(pdev, ERR, "netmap attach failed: %d\n", rc);
3703 		goto err_detach;
3704 	}
3705 #endif /* DEV_NETMAP */
3706 
3707 	/* Tell the stack that the interface is not active */
3708 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
3709 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3710 
3711 	/* Run the timer service */
3712 	ENA_TIMER_RESET(adapter);
3713 
3714 	return (0);
3715 
3716 #ifdef DEV_NETMAP
3717 err_detach:
3718 	ether_ifdetach(adapter->ifp);
3719 #endif /* DEV_NETMAP */
3720 err_msix_free:
3721 	ena_com_dev_reset(adapter->ena_dev, ENA_REGS_RESET_INIT_ERR);
3722 	ena_free_mgmnt_irq(adapter);
3723 	ena_disable_msix(adapter);
3724 err_io_free:
3725 	ena_free_all_io_rings_resources(adapter);
3726 	ena_free_rx_dma_tag(adapter);
3727 err_tx_tag_free:
3728 	ena_free_tx_dma_tag(adapter);
3729 err_com_free:
3730 	ena_com_admin_destroy(ena_dev);
3731 	ena_com_delete_host_info(ena_dev);
3732 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3733 err_bus_free:
3734 	free(ena_dev->bus, M_DEVBUF);
3735 err_pci_free:
3736 	ena_free_pci_resources(adapter);
3737 err_dev_free:
3738 	free(ena_dev, M_DEVBUF);
3739 
3740 	return (rc);
3741 }
3742 
3743 /**
3744  * ena_detach - Device Removal Routine
3745  * @pdev: device information struct
3746  *
3747  * ena_detach is called by the device subsystem to alert the driver
3748  * that it should release a PCI device.
3749  **/
3750 static int
3751 ena_detach(device_t pdev)
3752 {
3753 	struct ena_adapter *adapter = device_get_softc(pdev);
3754 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3755 	int rc;
3756 
3757 	/* Make sure VLANS are not using driver */
3758 	if (adapter->ifp->if_vlantrunk != NULL) {
3759 		ena_log(adapter->pdev, ERR, "VLAN is in use, detach first\n");
3760 		return (EBUSY);
3761 	}
3762 
3763 	ether_ifdetach(adapter->ifp);
3764 
3765 	/* Stop timer service */
3766 	ENA_LOCK_LOCK();
3767 	ENA_TIMER_DRAIN(adapter);
3768 	ENA_LOCK_UNLOCK();
3769 
3770 	/* Release reset task */
3771 	while (taskqueue_cancel(adapter->reset_tq, &adapter->reset_task, NULL))
3772 		taskqueue_drain(adapter->reset_tq, &adapter->reset_task);
3773 	taskqueue_free(adapter->reset_tq);
3774 
3775 	ENA_LOCK_LOCK();
3776 	ena_down(adapter);
3777 	ena_destroy_device(adapter, true);
3778 	ENA_LOCK_UNLOCK();
3779 
3780 	/* Restore unregistered sysctl queue nodes. */
3781 	ena_sysctl_update_queue_node_nb(adapter, adapter->num_io_queues,
3782 	    adapter->max_num_io_queues);
3783 
3784 #ifdef DEV_NETMAP
3785 	netmap_detach(adapter->ifp);
3786 #endif /* DEV_NETMAP */
3787 
3788 	ena_free_counters((counter_u64_t *)&adapter->hw_stats,
3789 	    sizeof(struct ena_hw_stats));
3790 	ena_free_counters((counter_u64_t *)&adapter->dev_stats,
3791 	    sizeof(struct ena_stats_dev));
3792 
3793 	rc = ena_free_rx_dma_tag(adapter);
3794 	if (unlikely(rc != 0))
3795 		ena_log(adapter->pdev, WARN,
3796 		    "Unmapped RX DMA tag associations\n");
3797 
3798 	rc = ena_free_tx_dma_tag(adapter);
3799 	if (unlikely(rc != 0))
3800 		ena_log(adapter->pdev, WARN,
3801 		    "Unmapped TX DMA tag associations\n");
3802 
3803 	ena_free_irqs(adapter);
3804 
3805 	ena_free_pci_resources(adapter);
3806 
3807 	if (adapter->rss_indir != NULL)
3808 		free(adapter->rss_indir, M_DEVBUF);
3809 
3810 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter)))
3811 		ena_com_rss_destroy(ena_dev);
3812 
3813 	ena_com_delete_host_info(ena_dev);
3814 
3815 	if_free(adapter->ifp);
3816 
3817 	free(ena_dev->bus, M_DEVBUF);
3818 
3819 	free(ena_dev, M_DEVBUF);
3820 
3821 	return (bus_generic_detach(pdev));
3822 }
3823 
3824 /******************************************************************************
3825  ******************************** AENQ Handlers *******************************
3826  *****************************************************************************/
3827 /**
3828  * ena_update_on_link_change:
3829  * Notify the network interface about the change in link status
3830  **/
3831 static void
3832 ena_update_on_link_change(void *adapter_data,
3833     struct ena_admin_aenq_entry *aenq_e)
3834 {
3835 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3836 	struct ena_admin_aenq_link_change_desc *aenq_desc;
3837 	int status;
3838 	if_t ifp;
3839 
3840 	aenq_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
3841 	ifp = adapter->ifp;
3842 	status = aenq_desc->flags &
3843 	    ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3844 
3845 	if (status != 0) {
3846 		ena_log(adapter->pdev, INFO, "link is UP\n");
3847 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_LINK_UP, adapter);
3848 		if (!ENA_FLAG_ISSET(ENA_FLAG_ONGOING_RESET, adapter))
3849 			if_link_state_change(ifp, LINK_STATE_UP);
3850 	} else {
3851 		ena_log(adapter->pdev, INFO, "link is DOWN\n");
3852 		if_link_state_change(ifp, LINK_STATE_DOWN);
3853 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_LINK_UP, adapter);
3854 	}
3855 }
3856 
3857 static void ena_notification(void *adapter_data,
3858     struct ena_admin_aenq_entry *aenq_e)
3859 {
3860 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3861 	struct ena_admin_ena_hw_hints *hints;
3862 
3863 	ENA_WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION, adapter->ena_dev,
3864 	    "Invalid group(%x) expected %x\n",	aenq_e->aenq_common_desc.group,
3865 	    ENA_ADMIN_NOTIFICATION);
3866 
3867 	switch (aenq_e->aenq_common_desc.syndrome) {
3868 	case ENA_ADMIN_UPDATE_HINTS:
3869 		hints =
3870 		    (struct ena_admin_ena_hw_hints *)(&aenq_e->inline_data_w4);
3871 		ena_update_hints(adapter, hints);
3872 		break;
3873 	default:
3874 		ena_log(adapter->pdev, ERR,
3875 		    "Invalid aenq notification link state %d\n",
3876 		    aenq_e->aenq_common_desc.syndrome);
3877 	}
3878 }
3879 
3880 static void
3881 ena_lock_init(void *arg)
3882 {
3883 	ENA_LOCK_INIT();
3884 }
3885 SYSINIT(ena_lock_init, SI_SUB_LOCK, SI_ORDER_FIRST, ena_lock_init, NULL);
3886 
3887 static void
3888 ena_lock_uninit(void *arg)
3889 {
3890 	ENA_LOCK_DESTROY();
3891 }
3892 SYSUNINIT(ena_lock_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, ena_lock_uninit, NULL);
3893 
3894 /**
3895  * This handler will called for unknown event group or unimplemented handlers
3896  **/
3897 static void
3898 unimplemented_aenq_handler(void *adapter_data,
3899     struct ena_admin_aenq_entry *aenq_e)
3900 {
3901 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3902 
3903 	ena_log(adapter->pdev, ERR,
3904 	    "Unknown event was received or event with unimplemented handler\n");
3905 }
3906 
3907 static struct ena_aenq_handlers aenq_handlers = {
3908     .handlers = {
3909 	    [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3910 	    [ENA_ADMIN_NOTIFICATION] = ena_notification,
3911 	    [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3912     },
3913     .unimplemented_handler = unimplemented_aenq_handler
3914 };
3915 
3916 /*********************************************************************
3917  *  FreeBSD Device Interface Entry Points
3918  *********************************************************************/
3919 
3920 static device_method_t ena_methods[] = {
3921     /* Device interface */
3922     DEVMETHOD(device_probe, ena_probe),
3923     DEVMETHOD(device_attach, ena_attach),
3924     DEVMETHOD(device_detach, ena_detach),
3925     DEVMETHOD_END
3926 };
3927 
3928 static driver_t ena_driver = {
3929     "ena", ena_methods, sizeof(struct ena_adapter),
3930 };
3931 
3932 devclass_t ena_devclass;
3933 DRIVER_MODULE(ena, pci, ena_driver, ena_devclass, 0, 0);
3934 MODULE_PNP_INFO("U16:vendor;U16:device", pci, ena, ena_vendor_info_array,
3935     nitems(ena_vendor_info_array) - 1);
3936 MODULE_DEPEND(ena, pci, 1, 1, 1);
3937 MODULE_DEPEND(ena, ether, 1, 1, 1);
3938 #ifdef DEV_NETMAP
3939 MODULE_DEPEND(ena, netmap, 1, 1, 1);
3940 #endif /* DEV_NETMAP */
3941 
3942 /*********************************************************************/
3943