xref: /freebsd/sys/dev/ena/ena.c (revision e2eeea75eb8b6dd50c1298067a0655880d186734)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2015-2020 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 #ifdef RSS
67 #include <net/rss_config.h>
68 #endif
69 
70 #include <netinet/in_systm.h>
71 #include <netinet/in.h>
72 #include <netinet/if_ether.h>
73 #include <netinet/ip.h>
74 #include <netinet/ip6.h>
75 #include <netinet/tcp.h>
76 #include <netinet/udp.h>
77 
78 #include <dev/pci/pcivar.h>
79 #include <dev/pci/pcireg.h>
80 
81 #include <vm/vm.h>
82 #include <vm/pmap.h>
83 
84 #include "ena_datapath.h"
85 #include "ena.h"
86 #include "ena_sysctl.h"
87 
88 #ifdef DEV_NETMAP
89 #include "ena_netmap.h"
90 #endif /* DEV_NETMAP */
91 
92 /*********************************************************
93  *  Function prototypes
94  *********************************************************/
95 static int	ena_probe(device_t);
96 static void	ena_intr_msix_mgmnt(void *);
97 static void	ena_free_pci_resources(struct ena_adapter *);
98 static int	ena_change_mtu(if_t, int);
99 static inline void ena_alloc_counters(counter_u64_t *, int);
100 static inline void ena_free_counters(counter_u64_t *, int);
101 static inline void ena_reset_counters(counter_u64_t *, int);
102 static void	ena_init_io_rings_common(struct ena_adapter *,
103     struct ena_ring *, uint16_t);
104 static void	ena_init_io_rings_basic(struct ena_adapter *);
105 static void	ena_init_io_rings_advanced(struct ena_adapter *);
106 static void	ena_init_io_rings(struct ena_adapter *);
107 static void	ena_free_io_ring_resources(struct ena_adapter *, unsigned int);
108 static void	ena_free_all_io_rings_resources(struct ena_adapter *);
109 static int	ena_setup_tx_dma_tag(struct ena_adapter *);
110 static int	ena_free_tx_dma_tag(struct ena_adapter *);
111 static int	ena_setup_rx_dma_tag(struct ena_adapter *);
112 static int	ena_free_rx_dma_tag(struct ena_adapter *);
113 static void	ena_release_all_tx_dmamap(struct ena_ring *);
114 static int	ena_setup_tx_resources(struct ena_adapter *, int);
115 static void	ena_free_tx_resources(struct ena_adapter *, int);
116 static int	ena_setup_all_tx_resources(struct ena_adapter *);
117 static void	ena_free_all_tx_resources(struct ena_adapter *);
118 static int	ena_setup_rx_resources(struct ena_adapter *, unsigned int);
119 static void	ena_free_rx_resources(struct ena_adapter *, unsigned int);
120 static int	ena_setup_all_rx_resources(struct ena_adapter *);
121 static void	ena_free_all_rx_resources(struct ena_adapter *);
122 static inline int ena_alloc_rx_mbuf(struct ena_adapter *, struct ena_ring *,
123     struct ena_rx_buffer *);
124 static void	ena_free_rx_mbuf(struct ena_adapter *, struct ena_ring *,
125     struct ena_rx_buffer *);
126 static void	ena_free_rx_bufs(struct ena_adapter *, unsigned int);
127 static void	ena_refill_all_rx_bufs(struct ena_adapter *);
128 static void	ena_free_all_rx_bufs(struct ena_adapter *);
129 static void	ena_free_tx_bufs(struct ena_adapter *, unsigned int);
130 static void	ena_free_all_tx_bufs(struct ena_adapter *);
131 static void	ena_destroy_all_tx_queues(struct ena_adapter *);
132 static void	ena_destroy_all_rx_queues(struct ena_adapter *);
133 static void	ena_destroy_all_io_queues(struct ena_adapter *);
134 static int	ena_create_io_queues(struct ena_adapter *);
135 static int	ena_handle_msix(void *);
136 static int	ena_enable_msix(struct ena_adapter *);
137 static void	ena_setup_mgmnt_intr(struct ena_adapter *);
138 static int	ena_setup_io_intr(struct ena_adapter *);
139 static int	ena_request_mgmnt_irq(struct ena_adapter *);
140 static int	ena_request_io_irq(struct ena_adapter *);
141 static void	ena_free_mgmnt_irq(struct ena_adapter *);
142 static void	ena_free_io_irq(struct ena_adapter *);
143 static void	ena_free_irqs(struct ena_adapter*);
144 static void	ena_disable_msix(struct ena_adapter *);
145 static void	ena_unmask_all_io_irqs(struct ena_adapter *);
146 static int	ena_rss_configure(struct ena_adapter *);
147 static int	ena_up_complete(struct ena_adapter *);
148 static uint64_t	ena_get_counter(if_t, ift_counter);
149 static int	ena_media_change(if_t);
150 static void	ena_media_status(if_t, struct ifmediareq *);
151 static void	ena_init(void *);
152 static int	ena_ioctl(if_t, u_long, caddr_t);
153 static int	ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *);
154 static void	ena_update_host_info(struct ena_admin_host_info *, if_t);
155 static void	ena_update_hwassist(struct ena_adapter *);
156 static int	ena_setup_ifnet(device_t, struct ena_adapter *,
157     struct ena_com_dev_get_features_ctx *);
158 static int	ena_enable_wc(struct resource *);
159 static int	ena_set_queues_placement_policy(device_t, struct ena_com_dev *,
160     struct ena_admin_feature_llq_desc *, struct ena_llq_configurations *);
161 static uint32_t	ena_calc_max_io_queue_num(device_t, struct ena_com_dev *,
162     struct ena_com_dev_get_features_ctx *);
163 static int	ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *);
164 static int	ena_rss_init_default(struct ena_adapter *);
165 static void	ena_rss_init_default_deferred(void *);
166 static void	ena_config_host_info(struct ena_com_dev *, device_t);
167 static int	ena_attach(device_t);
168 static int	ena_detach(device_t);
169 static int	ena_device_init(struct ena_adapter *, device_t,
170     struct ena_com_dev_get_features_ctx *, int *);
171 static int	ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *);
172 static void ena_update_on_link_change(void *, struct ena_admin_aenq_entry *);
173 static void	unimplemented_aenq_handler(void *,
174     struct ena_admin_aenq_entry *);
175 static int	ena_copy_eni_metrics(struct ena_adapter *);
176 static void	ena_timer_service(void *);
177 
178 static char ena_version[] = DEVICE_NAME DRV_MODULE_NAME " v" DRV_MODULE_VERSION;
179 
180 static ena_vendor_info_t ena_vendor_info_array[] = {
181     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0},
182     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF_RSERV0, 0},
183     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0},
184     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF_RSERV0, 0},
185     /* Last entry */
186     { 0, 0, 0 }
187 };
188 
189 /*
190  * Contains pointers to event handlers, e.g. link state chage.
191  */
192 static struct ena_aenq_handlers aenq_handlers;
193 
194 void
195 ena_dmamap_callback(void *arg, bus_dma_segment_t *segs, int nseg, int error)
196 {
197 	if (error != 0)
198 		return;
199 	*(bus_addr_t *) arg = segs[0].ds_addr;
200 }
201 
202 int
203 ena_dma_alloc(device_t dmadev, bus_size_t size,
204     ena_mem_handle_t *dma, int mapflags, bus_size_t alignment)
205 {
206 	struct ena_adapter* adapter = device_get_softc(dmadev);
207 	uint32_t maxsize;
208 	uint64_t dma_space_addr;
209 	int error;
210 
211 	maxsize = ((size - 1) / PAGE_SIZE + 1) * PAGE_SIZE;
212 
213 	dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
214 	if (unlikely(dma_space_addr == 0))
215 		dma_space_addr = BUS_SPACE_MAXADDR;
216 
217 	error = bus_dma_tag_create(bus_get_dma_tag(dmadev), /* parent */
218 	    alignment, 0,     /* alignment, bounds 		*/
219 	    dma_space_addr,   /* lowaddr of exclusion window	*/
220 	    BUS_SPACE_MAXADDR,/* highaddr of exclusion window	*/
221 	    NULL, NULL,	      /* filter, filterarg 		*/
222 	    maxsize,	      /* maxsize 			*/
223 	    1,		      /* nsegments 			*/
224 	    maxsize,	      /* maxsegsize 			*/
225 	    BUS_DMA_ALLOCNOW, /* flags 				*/
226 	    NULL,	      /* lockfunc 			*/
227 	    NULL,	      /* lockarg 			*/
228 	    &dma->tag);
229 	if (unlikely(error != 0)) {
230 		ena_trace(NULL, ENA_ALERT, "bus_dma_tag_create failed: %d\n", error);
231 		goto fail_tag;
232 	}
233 
234 	error = bus_dmamem_alloc(dma->tag, (void**) &dma->vaddr,
235 	    BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->map);
236 	if (unlikely(error != 0)) {
237 		ena_trace(NULL, ENA_ALERT, "bus_dmamem_alloc(%ju) failed: %d\n",
238 		    (uintmax_t)size, error);
239 		goto fail_map_create;
240 	}
241 
242 	dma->paddr = 0;
243 	error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr,
244 	    size, ena_dmamap_callback, &dma->paddr, mapflags);
245 	if (unlikely((error != 0) || (dma->paddr == 0))) {
246 		ena_trace(NULL, ENA_ALERT, ": bus_dmamap_load failed: %d\n", error);
247 		goto fail_map_load;
248 	}
249 
250 	bus_dmamap_sync(dma->tag, dma->map,
251 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
252 
253 	return (0);
254 
255 fail_map_load:
256 	bus_dmamem_free(dma->tag, dma->vaddr, dma->map);
257 fail_map_create:
258 	bus_dma_tag_destroy(dma->tag);
259 fail_tag:
260 	dma->tag = NULL;
261 	dma->vaddr = NULL;
262 	dma->paddr = 0;
263 
264 	return (error);
265 }
266 
267 /*
268  * This function should generate unique key for the whole driver.
269  * If the key was already genereated in the previous call (for example
270  * for another adapter), then it should be returned instead.
271  */
272 void
273 ena_rss_key_fill(void *key, size_t size)
274 {
275 	static bool key_generated;
276 	static uint8_t default_key[ENA_HASH_KEY_SIZE];
277 
278 	KASSERT(size <= ENA_HASH_KEY_SIZE, ("Requested more bytes than ENA RSS key can hold"));
279 
280 	if (!key_generated) {
281 		arc4random_buf(default_key, ENA_HASH_KEY_SIZE);
282 		key_generated = true;
283 	}
284 
285 	memcpy(key, default_key, size);
286 }
287 
288 static void
289 ena_free_pci_resources(struct ena_adapter *adapter)
290 {
291 	device_t pdev = adapter->pdev;
292 
293 	if (adapter->memory != NULL) {
294 		bus_release_resource(pdev, SYS_RES_MEMORY,
295 		    PCIR_BAR(ENA_MEM_BAR), adapter->memory);
296 	}
297 
298 	if (adapter->registers != NULL) {
299 		bus_release_resource(pdev, SYS_RES_MEMORY,
300 		    PCIR_BAR(ENA_REG_BAR), adapter->registers);
301 	}
302 }
303 
304 static int
305 ena_probe(device_t dev)
306 {
307 	ena_vendor_info_t *ent;
308 	char		adapter_name[60];
309 	uint16_t	pci_vendor_id = 0;
310 	uint16_t	pci_device_id = 0;
311 
312 	pci_vendor_id = pci_get_vendor(dev);
313 	pci_device_id = pci_get_device(dev);
314 
315 	ent = ena_vendor_info_array;
316 	while (ent->vendor_id != 0) {
317 		if ((pci_vendor_id == ent->vendor_id) &&
318 		    (pci_device_id == ent->device_id)) {
319 			ena_trace(NULL, ENA_DBG, "vendor=%x device=%x\n",
320 			    pci_vendor_id, pci_device_id);
321 
322 			sprintf(adapter_name, DEVICE_DESC);
323 			device_set_desc_copy(dev, adapter_name);
324 			return (BUS_PROBE_DEFAULT);
325 		}
326 
327 		ent++;
328 
329 	}
330 
331 	return (ENXIO);
332 }
333 
334 static int
335 ena_change_mtu(if_t ifp, int new_mtu)
336 {
337 	struct ena_adapter *adapter = if_getsoftc(ifp);
338 	int rc;
339 
340 	if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
341 		device_printf(adapter->pdev, "Invalid MTU setting. "
342 		    "new_mtu: %d max mtu: %d min mtu: %d\n",
343 		    new_mtu, adapter->max_mtu, ENA_MIN_MTU);
344 		return (EINVAL);
345 	}
346 
347 	rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
348 	if (likely(rc == 0)) {
349 		ena_trace(NULL, ENA_DBG, "set MTU to %d\n", new_mtu);
350 		if_setmtu(ifp, new_mtu);
351 	} else {
352 		device_printf(adapter->pdev, "Failed to set MTU to %d\n",
353 		    new_mtu);
354 	}
355 
356 	return (rc);
357 }
358 
359 static inline void
360 ena_alloc_counters(counter_u64_t *begin, int size)
361 {
362 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
363 
364 	for (; begin < end; ++begin)
365 		*begin = counter_u64_alloc(M_WAITOK);
366 }
367 
368 static inline void
369 ena_free_counters(counter_u64_t *begin, int size)
370 {
371 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
372 
373 	for (; begin < end; ++begin)
374 		counter_u64_free(*begin);
375 }
376 
377 static inline void
378 ena_reset_counters(counter_u64_t *begin, int size)
379 {
380 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
381 
382 	for (; begin < end; ++begin)
383 		counter_u64_zero(*begin);
384 }
385 
386 static void
387 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
388     uint16_t qid)
389 {
390 
391 	ring->qid = qid;
392 	ring->adapter = adapter;
393 	ring->ena_dev = adapter->ena_dev;
394 	ring->first_interrupt = false;
395 	ring->no_interrupt_event_cnt = 0;
396 }
397 
398 static void
399 ena_init_io_rings_basic(struct ena_adapter *adapter)
400 {
401 	struct ena_com_dev *ena_dev;
402 	struct ena_ring *txr, *rxr;
403 	struct ena_que *que;
404 	int i;
405 
406 	ena_dev = adapter->ena_dev;
407 
408 	for (i = 0; i < adapter->num_io_queues; i++) {
409 		txr = &adapter->tx_ring[i];
410 		rxr = &adapter->rx_ring[i];
411 
412 		/* TX/RX common ring state */
413 		ena_init_io_rings_common(adapter, txr, i);
414 		ena_init_io_rings_common(adapter, rxr, i);
415 
416 		/* TX specific ring state */
417 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
418 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
419 
420 		que = &adapter->que[i];
421 		que->adapter = adapter;
422 		que->id = i;
423 		que->tx_ring = txr;
424 		que->rx_ring = rxr;
425 
426 		txr->que = que;
427 		rxr->que = que;
428 
429 		rxr->empty_rx_queue = 0;
430 		rxr->rx_mbuf_sz = ena_mbuf_sz;
431 	}
432 }
433 
434 static void
435 ena_init_io_rings_advanced(struct ena_adapter *adapter)
436 {
437 	struct ena_ring *txr, *rxr;
438 	int i;
439 
440 	for (i = 0; i < adapter->num_io_queues; i++) {
441 		txr = &adapter->tx_ring[i];
442 		rxr = &adapter->rx_ring[i];
443 
444 		/* Allocate a buf ring */
445 		txr->buf_ring_size = adapter->buf_ring_size;
446 		txr->br = buf_ring_alloc(txr->buf_ring_size, M_DEVBUF,
447 		    M_WAITOK, &txr->ring_mtx);
448 
449 		/* Allocate Tx statistics. */
450 		ena_alloc_counters((counter_u64_t *)&txr->tx_stats,
451 		    sizeof(txr->tx_stats));
452 
453 		/* Allocate Rx statistics. */
454 		ena_alloc_counters((counter_u64_t *)&rxr->rx_stats,
455 		    sizeof(rxr->rx_stats));
456 
457 		/* Initialize locks */
458 		snprintf(txr->mtx_name, nitems(txr->mtx_name), "%s:tx(%d)",
459 		    device_get_nameunit(adapter->pdev), i);
460 		snprintf(rxr->mtx_name, nitems(rxr->mtx_name), "%s:rx(%d)",
461 		    device_get_nameunit(adapter->pdev), i);
462 
463 		mtx_init(&txr->ring_mtx, txr->mtx_name, NULL, MTX_DEF);
464 	}
465 }
466 
467 static void
468 ena_init_io_rings(struct ena_adapter *adapter)
469 {
470 	/*
471 	 * IO rings initialization can be divided into the 2 steps:
472 	 *   1. Initialize variables and fields with initial values and copy
473 	 *      them from adapter/ena_dev (basic)
474 	 *   2. Allocate mutex, counters and buf_ring (advanced)
475 	 */
476 	ena_init_io_rings_basic(adapter);
477 	ena_init_io_rings_advanced(adapter);
478 }
479 
480 static void
481 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
482 {
483 	struct ena_ring *txr = &adapter->tx_ring[qid];
484 	struct ena_ring *rxr = &adapter->rx_ring[qid];
485 
486 	ena_free_counters((counter_u64_t *)&txr->tx_stats,
487 	    sizeof(txr->tx_stats));
488 	ena_free_counters((counter_u64_t *)&rxr->rx_stats,
489 	    sizeof(rxr->rx_stats));
490 
491 	ENA_RING_MTX_LOCK(txr);
492 	drbr_free(txr->br, M_DEVBUF);
493 	ENA_RING_MTX_UNLOCK(txr);
494 
495 	mtx_destroy(&txr->ring_mtx);
496 }
497 
498 static void
499 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
500 {
501 	int i;
502 
503 	for (i = 0; i < adapter->num_io_queues; i++)
504 		ena_free_io_ring_resources(adapter, i);
505 
506 }
507 
508 static int
509 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
510 {
511 	int ret;
512 
513 	/* Create DMA tag for Tx buffers */
514 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
515 	    1, 0,				  /* alignment, bounds 	     */
516 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
517 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
518 	    NULL, NULL,				  /* filter, filterarg 	     */
519 	    ENA_TSO_MAXSIZE,			  /* maxsize 		     */
520 	    adapter->max_tx_sgl_size - 1,	  /* nsegments 		     */
521 	    ENA_TSO_MAXSIZE,			  /* maxsegsize 	     */
522 	    0,					  /* flags 		     */
523 	    NULL,				  /* lockfunc 		     */
524 	    NULL,				  /* lockfuncarg 	     */
525 	    &adapter->tx_buf_tag);
526 
527 	return (ret);
528 }
529 
530 static int
531 ena_free_tx_dma_tag(struct ena_adapter *adapter)
532 {
533 	int ret;
534 
535 	ret = bus_dma_tag_destroy(adapter->tx_buf_tag);
536 
537 	if (likely(ret == 0))
538 		adapter->tx_buf_tag = NULL;
539 
540 	return (ret);
541 }
542 
543 static int
544 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
545 {
546 	int ret;
547 
548 	/* Create DMA tag for Rx buffers*/
549 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent   */
550 	    1, 0,				  /* alignment, bounds 	     */
551 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
552 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
553 	    NULL, NULL,				  /* filter, filterarg 	     */
554 	    ena_mbuf_sz,			  /* maxsize 		     */
555 	    adapter->max_rx_sgl_size,		  /* nsegments 		     */
556 	    ena_mbuf_sz,			  /* maxsegsize 	     */
557 	    0,					  /* flags 		     */
558 	    NULL,				  /* lockfunc 		     */
559 	    NULL,				  /* lockarg 		     */
560 	    &adapter->rx_buf_tag);
561 
562 	return (ret);
563 }
564 
565 static int
566 ena_free_rx_dma_tag(struct ena_adapter *adapter)
567 {
568 	int ret;
569 
570 	ret = bus_dma_tag_destroy(adapter->rx_buf_tag);
571 
572 	if (likely(ret == 0))
573 		adapter->rx_buf_tag = NULL;
574 
575 	return (ret);
576 }
577 
578 static void
579 ena_release_all_tx_dmamap(struct ena_ring *tx_ring)
580 {
581 	struct ena_adapter *adapter = tx_ring->adapter;
582 	struct ena_tx_buffer *tx_info;
583 	bus_dma_tag_t tx_tag = adapter->tx_buf_tag;;
584 	int i;
585 #ifdef DEV_NETMAP
586 	struct ena_netmap_tx_info *nm_info;
587 	int j;
588 #endif /* DEV_NETMAP */
589 
590 	for (i = 0; i < tx_ring->ring_size; ++i) {
591 		tx_info = &tx_ring->tx_buffer_info[i];
592 #ifdef DEV_NETMAP
593 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
594 			nm_info = &tx_info->nm_info;
595 			for (j = 0; j < ENA_PKT_MAX_BUFS; ++j) {
596 				if (nm_info->map_seg[j] != NULL) {
597 					bus_dmamap_destroy(tx_tag,
598 					    nm_info->map_seg[j]);
599 					nm_info->map_seg[j] = NULL;
600 				}
601 			}
602 		}
603 #endif /* DEV_NETMAP */
604 		if (tx_info->dmamap != NULL) {
605 			bus_dmamap_destroy(tx_tag, tx_info->dmamap);
606 			tx_info->dmamap = NULL;
607 		}
608 	}
609 }
610 
611 /**
612  * ena_setup_tx_resources - allocate Tx resources (Descriptors)
613  * @adapter: network interface device structure
614  * @qid: queue index
615  *
616  * Returns 0 on success, otherwise on failure.
617  **/
618 static int
619 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
620 {
621 	struct ena_que *que = &adapter->que[qid];
622 	struct ena_ring *tx_ring = que->tx_ring;
623 	int size, i, err;
624 #ifdef DEV_NETMAP
625 	bus_dmamap_t *map;
626 	int j;
627 
628 	ena_netmap_reset_tx_ring(adapter, qid);
629 #endif /* DEV_NETMAP */
630 
631 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
632 
633 	tx_ring->tx_buffer_info = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
634 	if (unlikely(tx_ring->tx_buffer_info == NULL))
635 		return (ENOMEM);
636 
637 	size = sizeof(uint16_t) * tx_ring->ring_size;
638 	tx_ring->free_tx_ids = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
639 	if (unlikely(tx_ring->free_tx_ids == NULL))
640 		goto err_buf_info_free;
641 
642 	size = tx_ring->tx_max_header_size;
643 	tx_ring->push_buf_intermediate_buf = malloc(size, M_DEVBUF,
644 	    M_NOWAIT | M_ZERO);
645 	if (unlikely(tx_ring->push_buf_intermediate_buf == NULL))
646 		goto err_tx_ids_free;
647 
648 	/* Req id stack for TX OOO completions */
649 	for (i = 0; i < tx_ring->ring_size; i++)
650 		tx_ring->free_tx_ids[i] = i;
651 
652 	/* Reset TX statistics. */
653 	ena_reset_counters((counter_u64_t *)&tx_ring->tx_stats,
654 	    sizeof(tx_ring->tx_stats));
655 
656 	tx_ring->next_to_use = 0;
657 	tx_ring->next_to_clean = 0;
658 	tx_ring->acum_pkts = 0;
659 
660 	/* Make sure that drbr is empty */
661 	ENA_RING_MTX_LOCK(tx_ring);
662 	drbr_flush(adapter->ifp, tx_ring->br);
663 	ENA_RING_MTX_UNLOCK(tx_ring);
664 
665 	/* ... and create the buffer DMA maps */
666 	for (i = 0; i < tx_ring->ring_size; i++) {
667 		err = bus_dmamap_create(adapter->tx_buf_tag, 0,
668 		    &tx_ring->tx_buffer_info[i].dmamap);
669 		if (unlikely(err != 0)) {
670 			ena_trace(NULL, ENA_ALERT,
671 			    "Unable to create Tx DMA map for buffer %d\n",
672 			    i);
673 			goto err_map_release;
674 		}
675 
676 #ifdef DEV_NETMAP
677 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
678 			map = tx_ring->tx_buffer_info[i].nm_info.map_seg;
679 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
680 				err = bus_dmamap_create(adapter->tx_buf_tag, 0,
681 				    &map[j]);
682 				if (unlikely(err != 0)) {
683 					ena_trace(NULL, ENA_ALERT, "Unable to create "
684 					    "Tx DMA for buffer %d %d\n", i, j);
685 					goto err_map_release;
686 				}
687 			}
688 		}
689 #endif /* DEV_NETMAP */
690 	}
691 
692 	/* Allocate taskqueues */
693 	TASK_INIT(&tx_ring->enqueue_task, 0, ena_deferred_mq_start, tx_ring);
694 	tx_ring->enqueue_tq = taskqueue_create_fast("ena_tx_enque", M_NOWAIT,
695 	    taskqueue_thread_enqueue, &tx_ring->enqueue_tq);
696 	if (unlikely(tx_ring->enqueue_tq == NULL)) {
697 		ena_trace(NULL, ENA_ALERT,
698 		    "Unable to create taskqueue for enqueue task\n");
699 		i = tx_ring->ring_size;
700 		goto err_map_release;
701 	}
702 
703 	tx_ring->running = true;
704 
705 	taskqueue_start_threads(&tx_ring->enqueue_tq, 1, PI_NET,
706 	    "%s txeq %d", device_get_nameunit(adapter->pdev), que->cpu);
707 
708 	return (0);
709 
710 err_map_release:
711 	ena_release_all_tx_dmamap(tx_ring);
712 err_tx_ids_free:
713 	free(tx_ring->free_tx_ids, M_DEVBUF);
714 	tx_ring->free_tx_ids = NULL;
715 err_buf_info_free:
716 	free(tx_ring->tx_buffer_info, M_DEVBUF);
717 	tx_ring->tx_buffer_info = NULL;
718 
719 	return (ENOMEM);
720 }
721 
722 /**
723  * ena_free_tx_resources - Free Tx Resources per Queue
724  * @adapter: network interface device structure
725  * @qid: queue index
726  *
727  * Free all transmit software resources
728  **/
729 static void
730 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
731 {
732 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
733 #ifdef DEV_NETMAP
734 	struct ena_netmap_tx_info *nm_info;
735 	int j;
736 #endif /* DEV_NETMAP */
737 
738 	while (taskqueue_cancel(tx_ring->enqueue_tq, &tx_ring->enqueue_task,
739 	    NULL))
740 		taskqueue_drain(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
741 
742 	taskqueue_free(tx_ring->enqueue_tq);
743 
744 	ENA_RING_MTX_LOCK(tx_ring);
745 	/* Flush buffer ring, */
746 	drbr_flush(adapter->ifp, tx_ring->br);
747 
748 	/* Free buffer DMA maps, */
749 	for (int i = 0; i < tx_ring->ring_size; i++) {
750 		bus_dmamap_sync(adapter->tx_buf_tag,
751 		    tx_ring->tx_buffer_info[i].dmamap, BUS_DMASYNC_POSTWRITE);
752 		bus_dmamap_unload(adapter->tx_buf_tag,
753 		    tx_ring->tx_buffer_info[i].dmamap);
754 		bus_dmamap_destroy(adapter->tx_buf_tag,
755 		    tx_ring->tx_buffer_info[i].dmamap);
756 
757 #ifdef DEV_NETMAP
758 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
759 			nm_info = &tx_ring->tx_buffer_info[i].nm_info;
760 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
761 				if (nm_info->socket_buf_idx[j] != 0) {
762 					bus_dmamap_sync(adapter->tx_buf_tag,
763 					    nm_info->map_seg[j],
764 					    BUS_DMASYNC_POSTWRITE);
765 					ena_netmap_unload(adapter,
766 					    nm_info->map_seg[j]);
767 				}
768 				bus_dmamap_destroy(adapter->tx_buf_tag,
769 				    nm_info->map_seg[j]);
770 				nm_info->socket_buf_idx[j] = 0;
771 			}
772 		}
773 #endif /* DEV_NETMAP */
774 
775 		m_freem(tx_ring->tx_buffer_info[i].mbuf);
776 		tx_ring->tx_buffer_info[i].mbuf = NULL;
777 	}
778 	ENA_RING_MTX_UNLOCK(tx_ring);
779 
780 	/* And free allocated memory. */
781 	free(tx_ring->tx_buffer_info, M_DEVBUF);
782 	tx_ring->tx_buffer_info = NULL;
783 
784 	free(tx_ring->free_tx_ids, M_DEVBUF);
785 	tx_ring->free_tx_ids = NULL;
786 
787 	free(tx_ring->push_buf_intermediate_buf, M_DEVBUF);
788 	tx_ring->push_buf_intermediate_buf = NULL;
789 }
790 
791 /**
792  * ena_setup_all_tx_resources - allocate all queues Tx resources
793  * @adapter: network interface device structure
794  *
795  * Returns 0 on success, otherwise on failure.
796  **/
797 static int
798 ena_setup_all_tx_resources(struct ena_adapter *adapter)
799 {
800 	int i, rc;
801 
802 	for (i = 0; i < adapter->num_io_queues; i++) {
803 		rc = ena_setup_tx_resources(adapter, i);
804 		if (rc != 0) {
805 			device_printf(adapter->pdev,
806 			    "Allocation for Tx Queue %u failed\n", i);
807 			goto err_setup_tx;
808 		}
809 	}
810 
811 	return (0);
812 
813 err_setup_tx:
814 	/* Rewind the index freeing the rings as we go */
815 	while (i--)
816 		ena_free_tx_resources(adapter, i);
817 	return (rc);
818 }
819 
820 /**
821  * ena_free_all_tx_resources - Free Tx Resources for All Queues
822  * @adapter: network interface device structure
823  *
824  * Free all transmit software resources
825  **/
826 static void
827 ena_free_all_tx_resources(struct ena_adapter *adapter)
828 {
829 	int i;
830 
831 	for (i = 0; i < adapter->num_io_queues; i++)
832 		ena_free_tx_resources(adapter, i);
833 }
834 
835 /**
836  * ena_setup_rx_resources - allocate Rx resources (Descriptors)
837  * @adapter: network interface device structure
838  * @qid: queue index
839  *
840  * Returns 0 on success, otherwise on failure.
841  **/
842 static int
843 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
844 {
845 	struct ena_que *que = &adapter->que[qid];
846 	struct ena_ring *rx_ring = que->rx_ring;
847 	int size, err, i;
848 
849 	size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
850 
851 #ifdef DEV_NETMAP
852 	ena_netmap_reset_rx_ring(adapter, qid);
853 	rx_ring->initialized = false;
854 #endif /* DEV_NETMAP */
855 
856 	/*
857 	 * Alloc extra element so in rx path
858 	 * we can always prefetch rx_info + 1
859 	 */
860 	size += sizeof(struct ena_rx_buffer);
861 
862 	rx_ring->rx_buffer_info = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
863 
864 	size = sizeof(uint16_t) * rx_ring->ring_size;
865 	rx_ring->free_rx_ids = malloc(size, M_DEVBUF, M_WAITOK);
866 
867 	for (i = 0; i < rx_ring->ring_size; i++)
868 		rx_ring->free_rx_ids[i] = i;
869 
870 	/* Reset RX statistics. */
871 	ena_reset_counters((counter_u64_t *)&rx_ring->rx_stats,
872 	    sizeof(rx_ring->rx_stats));
873 
874 	rx_ring->next_to_clean = 0;
875 	rx_ring->next_to_use = 0;
876 
877 	/* ... and create the buffer DMA maps */
878 	for (i = 0; i < rx_ring->ring_size; i++) {
879 		err = bus_dmamap_create(adapter->rx_buf_tag, 0,
880 		    &(rx_ring->rx_buffer_info[i].map));
881 		if (err != 0) {
882 			ena_trace(NULL, ENA_ALERT,
883 			    "Unable to create Rx DMA map for buffer %d\n", i);
884 			goto err_buf_info_unmap;
885 		}
886 	}
887 
888 	/* Create LRO for the ring */
889 	if ((adapter->ifp->if_capenable & IFCAP_LRO) != 0) {
890 		int err = tcp_lro_init(&rx_ring->lro);
891 		if (err != 0) {
892 			device_printf(adapter->pdev,
893 			    "LRO[%d] Initialization failed!\n", qid);
894 		} else {
895 			ena_trace(NULL, ENA_INFO,
896 			    "RX Soft LRO[%d] Initialized\n", qid);
897 			rx_ring->lro.ifp = adapter->ifp;
898 		}
899 	}
900 
901 	return (0);
902 
903 err_buf_info_unmap:
904 	while (i--) {
905 		bus_dmamap_destroy(adapter->rx_buf_tag,
906 		    rx_ring->rx_buffer_info[i].map);
907 	}
908 
909 	free(rx_ring->free_rx_ids, M_DEVBUF);
910 	rx_ring->free_rx_ids = NULL;
911 	free(rx_ring->rx_buffer_info, M_DEVBUF);
912 	rx_ring->rx_buffer_info = NULL;
913 	return (ENOMEM);
914 }
915 
916 /**
917  * ena_free_rx_resources - Free Rx Resources
918  * @adapter: network interface device structure
919  * @qid: queue index
920  *
921  * Free all receive software resources
922  **/
923 static void
924 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
925 {
926 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
927 
928 	/* Free buffer DMA maps, */
929 	for (int i = 0; i < rx_ring->ring_size; i++) {
930 		bus_dmamap_sync(adapter->rx_buf_tag,
931 		    rx_ring->rx_buffer_info[i].map, BUS_DMASYNC_POSTREAD);
932 		m_freem(rx_ring->rx_buffer_info[i].mbuf);
933 		rx_ring->rx_buffer_info[i].mbuf = NULL;
934 		bus_dmamap_unload(adapter->rx_buf_tag,
935 		    rx_ring->rx_buffer_info[i].map);
936 		bus_dmamap_destroy(adapter->rx_buf_tag,
937 		    rx_ring->rx_buffer_info[i].map);
938 	}
939 
940 	/* free LRO resources, */
941 	tcp_lro_free(&rx_ring->lro);
942 
943 	/* free allocated memory */
944 	free(rx_ring->rx_buffer_info, M_DEVBUF);
945 	rx_ring->rx_buffer_info = NULL;
946 
947 	free(rx_ring->free_rx_ids, M_DEVBUF);
948 	rx_ring->free_rx_ids = NULL;
949 }
950 
951 /**
952  * ena_setup_all_rx_resources - allocate all queues Rx resources
953  * @adapter: network interface device structure
954  *
955  * Returns 0 on success, otherwise on failure.
956  **/
957 static int
958 ena_setup_all_rx_resources(struct ena_adapter *adapter)
959 {
960 	int i, rc = 0;
961 
962 	for (i = 0; i < adapter->num_io_queues; i++) {
963 		rc = ena_setup_rx_resources(adapter, i);
964 		if (rc != 0) {
965 			device_printf(adapter->pdev,
966 			    "Allocation for Rx Queue %u failed\n", i);
967 			goto err_setup_rx;
968 		}
969 	}
970 	return (0);
971 
972 err_setup_rx:
973 	/* rewind the index freeing the rings as we go */
974 	while (i--)
975 		ena_free_rx_resources(adapter, i);
976 	return (rc);
977 }
978 
979 /**
980  * ena_free_all_rx_resources - Free Rx resources for all queues
981  * @adapter: network interface device structure
982  *
983  * Free all receive software resources
984  **/
985 static void
986 ena_free_all_rx_resources(struct ena_adapter *adapter)
987 {
988 	int i;
989 
990 	for (i = 0; i < adapter->num_io_queues; i++)
991 		ena_free_rx_resources(adapter, i);
992 }
993 
994 static inline int
995 ena_alloc_rx_mbuf(struct ena_adapter *adapter,
996     struct ena_ring *rx_ring, struct ena_rx_buffer *rx_info)
997 {
998 	struct ena_com_buf *ena_buf;
999 	bus_dma_segment_t segs[1];
1000 	int nsegs, error;
1001 	int mlen;
1002 
1003 	/* if previous allocated frag is not used */
1004 	if (unlikely(rx_info->mbuf != NULL))
1005 		return (0);
1006 
1007 	/* Get mbuf using UMA allocator */
1008 	rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR,
1009 	    rx_ring->rx_mbuf_sz);
1010 
1011 	if (unlikely(rx_info->mbuf == NULL)) {
1012 		counter_u64_add(rx_ring->rx_stats.mjum_alloc_fail, 1);
1013 		rx_info->mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1014 		if (unlikely(rx_info->mbuf == NULL)) {
1015 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
1016 			return (ENOMEM);
1017 		}
1018 		mlen = MCLBYTES;
1019 	} else {
1020 		mlen = rx_ring->rx_mbuf_sz;
1021 	}
1022 	/* Set mbuf length*/
1023 	rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = mlen;
1024 
1025 	/* Map packets for DMA */
1026 	ena_trace(NULL, ENA_DBG | ENA_RSC | ENA_RXPTH,
1027 	    "Using tag %p for buffers' DMA mapping, mbuf %p len: %d\n",
1028 	    adapter->rx_buf_tag,rx_info->mbuf, rx_info->mbuf->m_len);
1029 	error = bus_dmamap_load_mbuf_sg(adapter->rx_buf_tag, rx_info->map,
1030 	    rx_info->mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
1031 	if (unlikely((error != 0) || (nsegs != 1))) {
1032 		ena_trace(NULL, ENA_WARNING, "failed to map mbuf, error: %d, "
1033 		    "nsegs: %d\n", error, nsegs);
1034 		counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
1035 		goto exit;
1036 
1037 	}
1038 
1039 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map, BUS_DMASYNC_PREREAD);
1040 
1041 	ena_buf = &rx_info->ena_buf;
1042 	ena_buf->paddr = segs[0].ds_addr;
1043 	ena_buf->len = mlen;
1044 
1045 	ena_trace(NULL, ENA_DBG | ENA_RSC | ENA_RXPTH,
1046 	    "ALLOC RX BUF: mbuf %p, rx_info %p, len %d, paddr %#jx\n",
1047 	    rx_info->mbuf, rx_info,ena_buf->len, (uintmax_t)ena_buf->paddr);
1048 
1049 	return (0);
1050 
1051 exit:
1052 	m_freem(rx_info->mbuf);
1053 	rx_info->mbuf = NULL;
1054 	return (EFAULT);
1055 }
1056 
1057 static void
1058 ena_free_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
1059     struct ena_rx_buffer *rx_info)
1060 {
1061 
1062 	if (rx_info->mbuf == NULL) {
1063 		ena_trace(NULL, ENA_WARNING, "Trying to free unallocated buffer\n");
1064 		return;
1065 	}
1066 
1067 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map,
1068 	    BUS_DMASYNC_POSTREAD);
1069 	bus_dmamap_unload(adapter->rx_buf_tag, rx_info->map);
1070 	m_freem(rx_info->mbuf);
1071 	rx_info->mbuf = NULL;
1072 }
1073 
1074 /**
1075  * ena_refill_rx_bufs - Refills ring with descriptors
1076  * @rx_ring: the ring which we want to feed with free descriptors
1077  * @num: number of descriptors to refill
1078  * Refills the ring with newly allocated DMA-mapped mbufs for receiving
1079  **/
1080 int
1081 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
1082 {
1083 	struct ena_adapter *adapter = rx_ring->adapter;
1084 	uint16_t next_to_use, req_id;
1085 	uint32_t i;
1086 	int rc;
1087 
1088 	ena_trace(NULL, ENA_DBG | ENA_RXPTH | ENA_RSC, "refill qid: %d\n",
1089 	    rx_ring->qid);
1090 
1091 	next_to_use = rx_ring->next_to_use;
1092 
1093 	for (i = 0; i < num; i++) {
1094 		struct ena_rx_buffer *rx_info;
1095 
1096 		ena_trace(NULL, ENA_DBG | ENA_RXPTH | ENA_RSC,
1097 		    "RX buffer - next to use: %d\n", next_to_use);
1098 
1099 		req_id = rx_ring->free_rx_ids[next_to_use];
1100 		rx_info = &rx_ring->rx_buffer_info[req_id];
1101 #ifdef DEV_NETMAP
1102 		if (ena_rx_ring_in_netmap(adapter, rx_ring->qid))
1103 			rc = ena_netmap_alloc_rx_slot(adapter, rx_ring, rx_info);
1104 		else
1105 #endif /* DEV_NETMAP */
1106 			rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1107 		if (unlikely(rc != 0)) {
1108 			ena_trace(NULL, ENA_WARNING,
1109 			    "failed to alloc buffer for rx queue %d\n",
1110 			    rx_ring->qid);
1111 			break;
1112 		}
1113 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1114 		    &rx_info->ena_buf, req_id);
1115 		if (unlikely(rc != 0)) {
1116 			ena_trace(NULL, ENA_WARNING,
1117 			    "failed to add buffer for rx queue %d\n",
1118 			    rx_ring->qid);
1119 			break;
1120 		}
1121 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1122 		    rx_ring->ring_size);
1123 	}
1124 
1125 	if (unlikely(i < num)) {
1126 		counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1127 		ena_trace(NULL, ENA_WARNING,
1128 		     "refilled rx qid %d with only %d mbufs (from %d)\n",
1129 		     rx_ring->qid, i, num);
1130 	}
1131 
1132 	if (likely(i != 0))
1133 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1134 
1135 	rx_ring->next_to_use = next_to_use;
1136 	return (i);
1137 }
1138 
1139 int
1140 ena_update_buf_ring_size(struct ena_adapter *adapter,
1141     uint32_t new_buf_ring_size)
1142 {
1143 	uint32_t old_buf_ring_size;
1144 	int rc = 0;
1145 	bool dev_was_up;
1146 
1147 	ENA_LOCK_LOCK(adapter);
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 			device_printf(adapter->pdev,
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 	ENA_LOCK_UNLOCK(adapter);
1184 
1185 	return (rc);
1186 }
1187 
1188 int
1189 ena_update_queue_size(struct ena_adapter *adapter, uint32_t new_tx_size,
1190     uint32_t new_rx_size)
1191 {
1192 	uint32_t old_tx_size, old_rx_size;
1193 	int rc = 0;
1194 	bool dev_was_up;
1195 
1196 	ENA_LOCK_LOCK(adapter);
1197 
1198 	old_tx_size = adapter->requested_tx_ring_size;
1199 	old_rx_size = adapter->requested_rx_ring_size;
1200 	adapter->requested_tx_ring_size = new_tx_size;
1201 	adapter->requested_rx_ring_size = new_rx_size;
1202 
1203 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1204 	ena_down(adapter);
1205 
1206 	/* Configure queues with new size. */
1207 	ena_init_io_rings_basic(adapter);
1208 	if (dev_was_up) {
1209 		rc = ena_up(adapter);
1210 		if (unlikely(rc != 0)) {
1211 			device_printf(adapter->pdev,
1212 			    "Failed to configure device with the new sizes - Tx: %u Rx: %u. Reverting old values - Tx: %u Rx: %u\n",
1213 			    new_tx_size, new_rx_size, old_tx_size, old_rx_size);
1214 
1215 			/* Revert old size. */
1216 			adapter->requested_tx_ring_size = old_tx_size;
1217 			adapter->requested_rx_ring_size = old_rx_size;
1218 			ena_init_io_rings_basic(adapter);
1219 
1220 			/* And try again. */
1221 			rc = ena_up(adapter);
1222 			if (unlikely(rc != 0)) {
1223 				device_printf(adapter->pdev,
1224 				    "Failed to revert old queue sizes. Triggering device reset.\n");
1225 				/*
1226 				 * If we've failed again, something had to go
1227 				 * wrong. After reset, the device should try to
1228 				 * go up
1229 				 */
1230 				ENA_FLAG_SET_ATOMIC(
1231 				    ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
1232 				ena_trigger_reset(adapter,
1233 				    ENA_REGS_RESET_OS_TRIGGER);
1234 			}
1235 		}
1236 	}
1237 
1238 	ENA_LOCK_UNLOCK(adapter);
1239 
1240 	return (rc);
1241 }
1242 
1243 static void
1244 ena_update_io_rings(struct ena_adapter *adapter, uint32_t num)
1245 {
1246 	ena_free_all_io_rings_resources(adapter);
1247 	/* Force indirection table to be reinitialized */
1248 	ena_com_rss_destroy(adapter->ena_dev);
1249 
1250 	adapter->num_io_queues = num;
1251 	ena_init_io_rings(adapter);
1252 }
1253 
1254 /* Caller should sanitize new_num */
1255 int
1256 ena_update_io_queue_nb(struct ena_adapter *adapter, uint32_t new_num)
1257 {
1258 	uint32_t old_num;
1259 	int rc = 0;
1260 	bool dev_was_up;
1261 
1262 	ENA_LOCK_LOCK(adapter);
1263 
1264 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1265 	old_num = adapter->num_io_queues;
1266 	ena_down(adapter);
1267 
1268 	ena_update_io_rings(adapter, new_num);
1269 
1270 	if (dev_was_up) {
1271 		rc = ena_up(adapter);
1272 		if (unlikely(rc != 0)) {
1273 			device_printf(adapter->pdev,
1274 			    "Failed to configure device with %u IO queues. "
1275 			    "Reverting to previous value: %u\n",
1276 			    new_num, old_num);
1277 
1278 			ena_update_io_rings(adapter, old_num);
1279 
1280 			rc = ena_up(adapter);
1281 			if (unlikely(rc != 0)) {
1282 				device_printf(adapter->pdev,
1283 				    "Failed to revert to previous setup IO "
1284 				    "queues. Triggering device reset.\n");
1285 				ENA_FLAG_SET_ATOMIC(
1286 				    ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
1287 				ena_trigger_reset(adapter,
1288 				    ENA_REGS_RESET_OS_TRIGGER);
1289 			}
1290 		}
1291 	}
1292 
1293 	ENA_LOCK_UNLOCK(adapter);
1294 
1295 	return (rc);
1296 }
1297 
1298 static void
1299 ena_free_rx_bufs(struct ena_adapter *adapter, unsigned int qid)
1300 {
1301 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1302 	unsigned int i;
1303 
1304 	for (i = 0; i < rx_ring->ring_size; i++) {
1305 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1306 
1307 		if (rx_info->mbuf != NULL)
1308 			ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1309 #ifdef DEV_NETMAP
1310 		if (((if_getflags(adapter->ifp) & IFF_DYING) == 0) &&
1311 		    (adapter->ifp->if_capenable & IFCAP_NETMAP)) {
1312 			if (rx_info->netmap_buf_idx != 0)
1313 				ena_netmap_free_rx_slot(adapter, rx_ring,
1314 				    rx_info);
1315 		}
1316 #endif /* DEV_NETMAP */
1317 	}
1318 }
1319 
1320 /**
1321  * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1322  * @adapter: network interface device structure
1323  *
1324  */
1325 static void
1326 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1327 {
1328 	struct ena_ring *rx_ring;
1329 	int i, rc, bufs_num;
1330 
1331 	for (i = 0; i < adapter->num_io_queues; i++) {
1332 		rx_ring = &adapter->rx_ring[i];
1333 		bufs_num = rx_ring->ring_size - 1;
1334 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1335 		if (unlikely(rc != bufs_num))
1336 			ena_trace(NULL, ENA_WARNING, "refilling Queue %d failed. "
1337 			    "Allocated %d buffers from: %d\n", i, rc, bufs_num);
1338 #ifdef DEV_NETMAP
1339 		rx_ring->initialized = true;
1340 #endif /* DEV_NETMAP */
1341 	}
1342 }
1343 
1344 static void
1345 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1346 {
1347 	int i;
1348 
1349 	for (i = 0; i < adapter->num_io_queues; i++)
1350 		ena_free_rx_bufs(adapter, i);
1351 }
1352 
1353 /**
1354  * ena_free_tx_bufs - Free Tx Buffers per Queue
1355  * @adapter: network interface device structure
1356  * @qid: queue index
1357  **/
1358 static void
1359 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1360 {
1361 	bool print_once = true;
1362 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1363 
1364 	ENA_RING_MTX_LOCK(tx_ring);
1365 	for (int i = 0; i < tx_ring->ring_size; i++) {
1366 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1367 
1368 		if (tx_info->mbuf == NULL)
1369 			continue;
1370 
1371 		if (print_once) {
1372 			device_printf(adapter->pdev,
1373 			    "free uncompleted tx mbuf qid %d idx 0x%x\n",
1374 			    qid, i);
1375 			print_once = false;
1376 		} else {
1377 			ena_trace(NULL, ENA_DBG,
1378 			    "free uncompleted tx mbuf qid %d idx 0x%x\n",
1379 			     qid, i);
1380 		}
1381 
1382 		bus_dmamap_sync(adapter->tx_buf_tag, tx_info->dmamap,
1383 		    BUS_DMASYNC_POSTWRITE);
1384 		bus_dmamap_unload(adapter->tx_buf_tag, tx_info->dmamap);
1385 
1386 		m_free(tx_info->mbuf);
1387 		tx_info->mbuf = NULL;
1388 	}
1389 	ENA_RING_MTX_UNLOCK(tx_ring);
1390 }
1391 
1392 static void
1393 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1394 {
1395 
1396 	for (int i = 0; i < adapter->num_io_queues; i++)
1397 		ena_free_tx_bufs(adapter, i);
1398 }
1399 
1400 static void
1401 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1402 {
1403 	uint16_t ena_qid;
1404 	int i;
1405 
1406 	for (i = 0; i < adapter->num_io_queues; i++) {
1407 		ena_qid = ENA_IO_TXQ_IDX(i);
1408 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1409 	}
1410 }
1411 
1412 static void
1413 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1414 {
1415 	uint16_t ena_qid;
1416 	int i;
1417 
1418 	for (i = 0; i < adapter->num_io_queues; i++) {
1419 		ena_qid = ENA_IO_RXQ_IDX(i);
1420 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1421 	}
1422 }
1423 
1424 static void
1425 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1426 {
1427 	struct ena_que *queue;
1428 	int i;
1429 
1430 	for (i = 0; i < adapter->num_io_queues; i++) {
1431 		queue = &adapter->que[i];
1432 		while (taskqueue_cancel(queue->cleanup_tq,
1433 		    &queue->cleanup_task, NULL))
1434 			taskqueue_drain(queue->cleanup_tq,
1435 			    &queue->cleanup_task);
1436 		taskqueue_free(queue->cleanup_tq);
1437 	}
1438 
1439 	ena_destroy_all_tx_queues(adapter);
1440 	ena_destroy_all_rx_queues(adapter);
1441 }
1442 
1443 static int
1444 ena_create_io_queues(struct ena_adapter *adapter)
1445 {
1446 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1447 	struct ena_com_create_io_ctx ctx;
1448 	struct ena_ring *ring;
1449 	struct ena_que *queue;
1450 	uint16_t ena_qid;
1451 	uint32_t msix_vector;
1452 	int rc, i;
1453 
1454 	/* Create TX queues */
1455 	for (i = 0; i < adapter->num_io_queues; i++) {
1456 		msix_vector = ENA_IO_IRQ_IDX(i);
1457 		ena_qid = ENA_IO_TXQ_IDX(i);
1458 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1459 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1460 		ctx.queue_size = adapter->requested_tx_ring_size;
1461 		ctx.msix_vector = msix_vector;
1462 		ctx.qid = ena_qid;
1463 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1464 		if (rc != 0) {
1465 			device_printf(adapter->pdev,
1466 			    "Failed to create io TX queue #%d rc: %d\n", i, rc);
1467 			goto err_tx;
1468 		}
1469 		ring = &adapter->tx_ring[i];
1470 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1471 		    &ring->ena_com_io_sq,
1472 		    &ring->ena_com_io_cq);
1473 		if (rc != 0) {
1474 			device_printf(adapter->pdev,
1475 			    "Failed to get TX queue handlers. TX queue num"
1476 			    " %d rc: %d\n", i, rc);
1477 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1478 			goto err_tx;
1479 		}
1480 	}
1481 
1482 	/* Create RX queues */
1483 	for (i = 0; i < adapter->num_io_queues; i++) {
1484 		msix_vector = ENA_IO_IRQ_IDX(i);
1485 		ena_qid = ENA_IO_RXQ_IDX(i);
1486 		ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1487 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1488 		ctx.queue_size = adapter->requested_rx_ring_size;
1489 		ctx.msix_vector = msix_vector;
1490 		ctx.qid = ena_qid;
1491 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1492 		if (unlikely(rc != 0)) {
1493 			device_printf(adapter->pdev,
1494 			    "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1495 			goto err_rx;
1496 		}
1497 
1498 		ring = &adapter->rx_ring[i];
1499 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1500 		    &ring->ena_com_io_sq,
1501 		    &ring->ena_com_io_cq);
1502 		if (unlikely(rc != 0)) {
1503 			device_printf(adapter->pdev,
1504 			    "Failed to get RX queue handlers. RX queue num"
1505 			    " %d rc: %d\n", i, rc);
1506 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1507 			goto err_rx;
1508 		}
1509 	}
1510 
1511 	for (i = 0; i < adapter->num_io_queues; i++) {
1512 		queue = &adapter->que[i];
1513 
1514 		NET_TASK_INIT(&queue->cleanup_task, 0, ena_cleanup, queue);
1515 		queue->cleanup_tq = taskqueue_create_fast("ena cleanup",
1516 		    M_WAITOK, taskqueue_thread_enqueue, &queue->cleanup_tq);
1517 
1518 		taskqueue_start_threads(&queue->cleanup_tq, 1, PI_NET,
1519 		    "%s queue %d cleanup",
1520 		    device_get_nameunit(adapter->pdev), i);
1521 	}
1522 
1523 	return (0);
1524 
1525 err_rx:
1526 	while (i--)
1527 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1528 	i = adapter->num_io_queues;
1529 err_tx:
1530 	while (i--)
1531 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1532 
1533 	return (ENXIO);
1534 }
1535 
1536 /*********************************************************************
1537  *
1538  *  MSIX & Interrupt Service routine
1539  *
1540  **********************************************************************/
1541 
1542 /**
1543  * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1544  * @arg: interrupt number
1545  **/
1546 static void
1547 ena_intr_msix_mgmnt(void *arg)
1548 {
1549 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1550 
1551 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1552 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter)))
1553 		ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1554 }
1555 
1556 /**
1557  * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1558  * @arg: queue
1559  **/
1560 static int
1561 ena_handle_msix(void *arg)
1562 {
1563 	struct ena_que *queue = arg;
1564 	struct ena_adapter *adapter = queue->adapter;
1565 	if_t ifp = adapter->ifp;
1566 
1567 	if (unlikely((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0))
1568 		return (FILTER_STRAY);
1569 
1570 	taskqueue_enqueue(queue->cleanup_tq, &queue->cleanup_task);
1571 
1572 	return (FILTER_HANDLED);
1573 }
1574 
1575 static int
1576 ena_enable_msix(struct ena_adapter *adapter)
1577 {
1578 	device_t dev = adapter->pdev;
1579 	int msix_vecs, msix_req;
1580 	int i, rc = 0;
1581 
1582 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1583 		device_printf(dev, "Error, MSI-X is already enabled\n");
1584 		return (EINVAL);
1585 	}
1586 
1587 	/* Reserved the max msix vectors we might need */
1588 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
1589 
1590 	adapter->msix_entries = malloc(msix_vecs * sizeof(struct msix_entry),
1591 	    M_DEVBUF, M_WAITOK | M_ZERO);
1592 
1593 	ena_trace(NULL, ENA_DBG, "trying to enable MSI-X, vectors: %d\n", msix_vecs);
1594 
1595 	for (i = 0; i < msix_vecs; i++) {
1596 		adapter->msix_entries[i].entry = i;
1597 		/* Vectors must start from 1 */
1598 		adapter->msix_entries[i].vector = i + 1;
1599 	}
1600 
1601 	msix_req = msix_vecs;
1602 	rc = pci_alloc_msix(dev, &msix_vecs);
1603 	if (unlikely(rc != 0)) {
1604 		device_printf(dev,
1605 		    "Failed to enable MSIX, vectors %d rc %d\n", msix_vecs, rc);
1606 
1607 		rc = ENOSPC;
1608 		goto err_msix_free;
1609 	}
1610 
1611 	if (msix_vecs != msix_req) {
1612 		if (msix_vecs == ENA_ADMIN_MSIX_VEC) {
1613 			device_printf(dev,
1614 			    "Not enough number of MSI-x allocated: %d\n",
1615 			    msix_vecs);
1616 			pci_release_msi(dev);
1617 			rc = ENOSPC;
1618 			goto err_msix_free;
1619 		}
1620 		device_printf(dev, "Enable only %d MSI-x (out of %d), reduce "
1621 		    "the number of queues\n", msix_vecs, msix_req);
1622 	}
1623 
1624 	adapter->msix_vecs = msix_vecs;
1625 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1626 
1627 	return (0);
1628 
1629 err_msix_free:
1630 	free(adapter->msix_entries, M_DEVBUF);
1631 	adapter->msix_entries = NULL;
1632 
1633 	return (rc);
1634 }
1635 
1636 static void
1637 ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1638 {
1639 
1640 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1641 	    ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1642 	    device_get_nameunit(adapter->pdev));
1643 	/*
1644 	 * Handler is NULL on purpose, it will be set
1645 	 * when mgmnt interrupt is acquired
1646 	 */
1647 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler = NULL;
1648 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1649 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1650 	    adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
1651 }
1652 
1653 static int
1654 ena_setup_io_intr(struct ena_adapter *adapter)
1655 {
1656 	static int last_bind_cpu = -1;
1657 	int irq_idx;
1658 
1659 	if (adapter->msix_entries == NULL)
1660 		return (EINVAL);
1661 
1662 	for (int i = 0; i < adapter->num_io_queues; i++) {
1663 		irq_idx = ENA_IO_IRQ_IDX(i);
1664 
1665 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1666 		    "%s-TxRx-%d", device_get_nameunit(adapter->pdev), i);
1667 		adapter->irq_tbl[irq_idx].handler = ena_handle_msix;
1668 		adapter->irq_tbl[irq_idx].data = &adapter->que[i];
1669 		adapter->irq_tbl[irq_idx].vector =
1670 		    adapter->msix_entries[irq_idx].vector;
1671 		ena_trace(NULL, ENA_INFO | ENA_IOQ, "ena_setup_io_intr vector: %d\n",
1672 		    adapter->msix_entries[irq_idx].vector);
1673 
1674 		/*
1675 		 * We want to bind rings to the corresponding cpu
1676 		 * using something similar to the RSS round-robin technique.
1677 		 */
1678 		if (unlikely(last_bind_cpu < 0))
1679 			last_bind_cpu = CPU_FIRST();
1680 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1681 		    last_bind_cpu;
1682 		last_bind_cpu = CPU_NEXT(last_bind_cpu);
1683 	}
1684 
1685 	return (0);
1686 }
1687 
1688 static int
1689 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1690 {
1691 	struct ena_irq *irq;
1692 	unsigned long flags;
1693 	int rc, rcc;
1694 
1695 	flags = RF_ACTIVE | RF_SHAREABLE;
1696 
1697 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1698 	irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1699 	    &irq->vector, flags);
1700 
1701 	if (unlikely(irq->res == NULL)) {
1702 		device_printf(adapter->pdev, "could not allocate "
1703 		    "irq vector: %d\n", irq->vector);
1704 		return (ENXIO);
1705 	}
1706 
1707 	rc = bus_setup_intr(adapter->pdev, irq->res,
1708 	    INTR_TYPE_NET | INTR_MPSAFE, NULL, ena_intr_msix_mgmnt,
1709 	    irq->data, &irq->cookie);
1710 	if (unlikely(rc != 0)) {
1711 		device_printf(adapter->pdev, "failed to register "
1712 		    "interrupt handler for irq %ju: %d\n",
1713 		    rman_get_start(irq->res), rc);
1714 		goto err_res_free;
1715 	}
1716 	irq->requested = true;
1717 
1718 	return (rc);
1719 
1720 err_res_free:
1721 	ena_trace(NULL, ENA_INFO | ENA_ADMQ, "releasing resource for irq %d\n",
1722 	    irq->vector);
1723 	rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1724 	    irq->vector, irq->res);
1725 	if (unlikely(rcc != 0))
1726 		device_printf(adapter->pdev, "dev has no parent while "
1727 		    "releasing res for irq: %d\n", irq->vector);
1728 	irq->res = NULL;
1729 
1730 	return (rc);
1731 }
1732 
1733 static int
1734 ena_request_io_irq(struct ena_adapter *adapter)
1735 {
1736 	struct ena_irq *irq;
1737 	unsigned long flags = 0;
1738 	int rc = 0, i, rcc;
1739 
1740 	if (unlikely(!ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter))) {
1741 		device_printf(adapter->pdev,
1742 		    "failed to request I/O IRQ: MSI-X is not enabled\n");
1743 		return (EINVAL);
1744 	} else {
1745 		flags = RF_ACTIVE | RF_SHAREABLE;
1746 	}
1747 
1748 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1749 		irq = &adapter->irq_tbl[i];
1750 
1751 		if (unlikely(irq->requested))
1752 			continue;
1753 
1754 		irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1755 		    &irq->vector, flags);
1756 		if (unlikely(irq->res == NULL)) {
1757 			rc = ENOMEM;
1758 			device_printf(adapter->pdev, "could not allocate "
1759 			    "irq vector: %d\n", irq->vector);
1760 			goto err;
1761 		}
1762 
1763 		rc = bus_setup_intr(adapter->pdev, irq->res,
1764 		    INTR_TYPE_NET | INTR_MPSAFE, irq->handler, NULL,
1765 		    irq->data, &irq->cookie);
1766 		 if (unlikely(rc != 0)) {
1767 			device_printf(adapter->pdev, "failed to register "
1768 			    "interrupt handler for irq %ju: %d\n",
1769 			    rman_get_start(irq->res), rc);
1770 			goto err;
1771 		}
1772 		irq->requested = true;
1773 
1774 		ena_trace(NULL, ENA_INFO, "queue %d - cpu %d\n",
1775 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
1776 	}
1777 
1778 	return (rc);
1779 
1780 err:
1781 
1782 	for (; i >= ENA_IO_IRQ_FIRST_IDX; i--) {
1783 		irq = &adapter->irq_tbl[i];
1784 		rcc = 0;
1785 
1786 		/* Once we entered err: section and irq->requested is true we
1787 		   free both intr and resources */
1788 		if (irq->requested)
1789 			rcc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
1790 		if (unlikely(rcc != 0))
1791 			device_printf(adapter->pdev, "could not release"
1792 			    " irq: %d, error: %d\n", irq->vector, rcc);
1793 
1794 		/* If we entred err: section without irq->requested set we know
1795 		   it was bus_alloc_resource_any() that needs cleanup, provided
1796 		   res is not NULL. In case res is NULL no work in needed in
1797 		   this iteration */
1798 		rcc = 0;
1799 		if (irq->res != NULL) {
1800 			rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1801 			    irq->vector, irq->res);
1802 		}
1803 		if (unlikely(rcc != 0))
1804 			device_printf(adapter->pdev, "dev has no parent while "
1805 			    "releasing res for irq: %d\n", irq->vector);
1806 		irq->requested = false;
1807 		irq->res = NULL;
1808 	}
1809 
1810 	return (rc);
1811 }
1812 
1813 static void
1814 ena_free_mgmnt_irq(struct ena_adapter *adapter)
1815 {
1816 	struct ena_irq *irq;
1817 	int rc;
1818 
1819 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1820 	if (irq->requested) {
1821 		ena_trace(NULL, ENA_INFO | ENA_ADMQ, "tear down irq: %d\n",
1822 		    irq->vector);
1823 		rc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
1824 		if (unlikely(rc != 0))
1825 			device_printf(adapter->pdev, "failed to tear "
1826 			    "down irq: %d\n", irq->vector);
1827 		irq->requested = 0;
1828 	}
1829 
1830 	if (irq->res != NULL) {
1831 		ena_trace(NULL, ENA_INFO | ENA_ADMQ, "release resource irq: %d\n",
1832 		    irq->vector);
1833 		rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1834 		    irq->vector, irq->res);
1835 		irq->res = NULL;
1836 		if (unlikely(rc != 0))
1837 			device_printf(adapter->pdev, "dev has no parent while "
1838 			    "releasing res for irq: %d\n", irq->vector);
1839 	}
1840 }
1841 
1842 static void
1843 ena_free_io_irq(struct ena_adapter *adapter)
1844 {
1845 	struct ena_irq *irq;
1846 	int rc;
1847 
1848 	for (int i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1849 		irq = &adapter->irq_tbl[i];
1850 		if (irq->requested) {
1851 			ena_trace(NULL, ENA_INFO | ENA_IOQ, "tear down irq: %d\n",
1852 			    irq->vector);
1853 			rc = bus_teardown_intr(adapter->pdev, irq->res,
1854 			    irq->cookie);
1855 			if (unlikely(rc != 0)) {
1856 				device_printf(adapter->pdev, "failed to tear "
1857 				    "down irq: %d\n", irq->vector);
1858 			}
1859 			irq->requested = 0;
1860 		}
1861 
1862 		if (irq->res != NULL) {
1863 			ena_trace(NULL, ENA_INFO | ENA_IOQ, "release resource irq: %d\n",
1864 			    irq->vector);
1865 			rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1866 			    irq->vector, irq->res);
1867 			irq->res = NULL;
1868 			if (unlikely(rc != 0)) {
1869 				device_printf(adapter->pdev, "dev has no parent"
1870 				    " while releasing res for irq: %d\n",
1871 				    irq->vector);
1872 			}
1873 		}
1874 	}
1875 }
1876 
1877 static void
1878 ena_free_irqs(struct ena_adapter* adapter)
1879 {
1880 
1881 	ena_free_io_irq(adapter);
1882 	ena_free_mgmnt_irq(adapter);
1883 	ena_disable_msix(adapter);
1884 }
1885 
1886 static void
1887 ena_disable_msix(struct ena_adapter *adapter)
1888 {
1889 
1890 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1891 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1892 		pci_release_msi(adapter->pdev);
1893 	}
1894 
1895 	adapter->msix_vecs = 0;
1896 	if (adapter->msix_entries != NULL)
1897 		free(adapter->msix_entries, M_DEVBUF);
1898 	adapter->msix_entries = NULL;
1899 }
1900 
1901 static void
1902 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
1903 {
1904 	struct ena_com_io_cq* io_cq;
1905 	struct ena_eth_io_intr_reg intr_reg;
1906 	uint16_t ena_qid;
1907 	int i;
1908 
1909 	/* Unmask interrupts for all queues */
1910 	for (i = 0; i < adapter->num_io_queues; i++) {
1911 		ena_qid = ENA_IO_TXQ_IDX(i);
1912 		io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1913 		ena_com_update_intr_reg(&intr_reg, 0, 0, true);
1914 		ena_com_unmask_intr(io_cq, &intr_reg);
1915 	}
1916 }
1917 
1918 /* Configure the Rx forwarding */
1919 static int
1920 ena_rss_configure(struct ena_adapter *adapter)
1921 {
1922 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1923 	int rc;
1924 
1925 	/* In case the RSS table was destroyed */
1926 	if (!ena_dev->rss.tbl_log_size) {
1927 		rc = ena_rss_init_default(adapter);
1928 		if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
1929 			device_printf(adapter->pdev,
1930 			    "WARNING: RSS was not properly re-initialized,"
1931 			    " it will affect bandwidth\n");
1932 			ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_RSS_ACTIVE, adapter);
1933 			return (rc);
1934 		}
1935 	}
1936 
1937 	/* Set indirect table */
1938 	rc = ena_com_indirect_table_set(ena_dev);
1939 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
1940 		return (rc);
1941 
1942 	/* Configure hash function (if supported) */
1943 	rc = ena_com_set_hash_function(ena_dev);
1944 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
1945 		return (rc);
1946 
1947 	/* Configure hash inputs (if supported) */
1948 	rc = ena_com_set_hash_ctrl(ena_dev);
1949 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
1950 		return (rc);
1951 
1952 	return (0);
1953 }
1954 
1955 static int
1956 ena_up_complete(struct ena_adapter *adapter)
1957 {
1958 	int rc;
1959 
1960 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter))) {
1961 		rc = ena_rss_configure(adapter);
1962 		if (rc != 0) {
1963 			device_printf(adapter->pdev,
1964 			    "Failed to configure RSS\n");
1965 			return (rc);
1966 		}
1967 	}
1968 
1969 	rc = ena_change_mtu(adapter->ifp, adapter->ifp->if_mtu);
1970 	if (unlikely(rc != 0))
1971 		return (rc);
1972 
1973 	ena_refill_all_rx_bufs(adapter);
1974 	ena_reset_counters((counter_u64_t *)&adapter->hw_stats,
1975 	    sizeof(adapter->hw_stats));
1976 
1977 	return (0);
1978 }
1979 
1980 static void
1981 set_io_rings_size(struct ena_adapter *adapter, int new_tx_size,
1982     int new_rx_size)
1983 {
1984 	int i;
1985 
1986 	for (i = 0; i < adapter->num_io_queues; i++) {
1987 		adapter->tx_ring[i].ring_size = new_tx_size;
1988 		adapter->rx_ring[i].ring_size = new_rx_size;
1989 	}
1990 }
1991 
1992 static int
1993 create_queues_with_size_backoff(struct ena_adapter *adapter)
1994 {
1995 	int rc;
1996 	uint32_t cur_rx_ring_size, cur_tx_ring_size;
1997 	uint32_t new_rx_ring_size, new_tx_ring_size;
1998 
1999 	/*
2000 	 * Current queue sizes might be set to smaller than the requested
2001 	 * ones due to past queue allocation failures.
2002 	 */
2003 	set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2004 	    adapter->requested_rx_ring_size);
2005 
2006 	while (1) {
2007 		/* Allocate transmit descriptors */
2008 		rc = ena_setup_all_tx_resources(adapter);
2009 		if (unlikely(rc != 0)) {
2010 			ena_trace(NULL, ENA_ALERT, "err_setup_tx\n");
2011 			goto err_setup_tx;
2012 		}
2013 
2014 		/* Allocate receive descriptors */
2015 		rc = ena_setup_all_rx_resources(adapter);
2016 		if (unlikely(rc != 0)) {
2017 			ena_trace(NULL, ENA_ALERT, "err_setup_rx\n");
2018 			goto err_setup_rx;
2019 		}
2020 
2021 		/* Create IO queues for Rx & Tx */
2022 		rc = ena_create_io_queues(adapter);
2023 		if (unlikely(rc != 0)) {
2024 			ena_trace(NULL, ENA_ALERT,
2025 			    "create IO queues failed\n");
2026 			goto err_io_que;
2027 		}
2028 
2029 		return (0);
2030 
2031 err_io_que:
2032 		ena_free_all_rx_resources(adapter);
2033 err_setup_rx:
2034 		ena_free_all_tx_resources(adapter);
2035 err_setup_tx:
2036 		/*
2037 		 * Lower the ring size if ENOMEM. Otherwise, return the
2038 		 * error straightaway.
2039 		 */
2040 		if (unlikely(rc != ENOMEM)) {
2041 			ena_trace(NULL, ENA_ALERT,
2042 			    "Queue creation failed with error code: %d\n", rc);
2043 			return (rc);
2044 		}
2045 
2046 		cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2047 		cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2048 
2049 		device_printf(adapter->pdev,
2050 		    "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2051 		    cur_tx_ring_size, cur_rx_ring_size);
2052 
2053 		new_tx_ring_size = cur_tx_ring_size;
2054 		new_rx_ring_size = cur_rx_ring_size;
2055 
2056 		/*
2057 		 * Decrease the size of a larger queue, or decrease both if they are
2058 		 * the same size.
2059 		 */
2060 		if (cur_rx_ring_size <= cur_tx_ring_size)
2061 			new_tx_ring_size = cur_tx_ring_size / 2;
2062 		if (cur_rx_ring_size >= cur_tx_ring_size)
2063 			new_rx_ring_size = cur_rx_ring_size / 2;
2064 
2065 		if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2066 		    new_rx_ring_size < ENA_MIN_RING_SIZE) {
2067 			device_printf(adapter->pdev,
2068 			    "Queue creation failed with the smallest possible queue size"
2069 			    "of %d for both queues. Not retrying with smaller queues\n",
2070 			    ENA_MIN_RING_SIZE);
2071 			return (rc);
2072 		}
2073 
2074 		set_io_rings_size(adapter, new_tx_ring_size, new_rx_ring_size);
2075 	}
2076 }
2077 
2078 int
2079 ena_up(struct ena_adapter *adapter)
2080 {
2081 	int rc = 0;
2082 
2083 	if (unlikely(device_is_attached(adapter->pdev) == 0)) {
2084 		device_printf(adapter->pdev, "device is not attached!\n");
2085 		return (ENXIO);
2086 	}
2087 
2088 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2089 		return (0);
2090 
2091 	device_printf(adapter->pdev, "device is going UP\n");
2092 
2093 	/* setup interrupts for IO queues */
2094 	rc = ena_setup_io_intr(adapter);
2095 	if (unlikely(rc != 0)) {
2096 		ena_trace(NULL, ENA_ALERT, "error setting up IO interrupt\n");
2097 		goto error;
2098 	}
2099 	rc = ena_request_io_irq(adapter);
2100 	if (unlikely(rc != 0)) {
2101 		ena_trace(NULL, ENA_ALERT, "err_req_irq\n");
2102 		goto error;
2103 	}
2104 
2105 	device_printf(adapter->pdev,
2106 	    "Creating %u IO queues. Rx queue size: %d, Tx queue size: %d, "
2107 	    "LLQ is %s\n",
2108 	    adapter->num_io_queues,
2109 	    adapter->requested_rx_ring_size,
2110 	    adapter->requested_tx_ring_size,
2111 	    (adapter->ena_dev->tx_mem_queue_type ==
2112 	        ENA_ADMIN_PLACEMENT_POLICY_DEV) ?  "ENABLED" : "DISABLED");
2113 
2114 	rc = create_queues_with_size_backoff(adapter);
2115 	if (unlikely(rc != 0)) {
2116 		ena_trace(NULL, ENA_ALERT,
2117 		    "error creating queues with size backoff\n");
2118 		goto err_create_queues_with_backoff;
2119 	}
2120 
2121 	if (ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter))
2122 		if_link_state_change(adapter->ifp, LINK_STATE_UP);
2123 
2124 	rc = ena_up_complete(adapter);
2125 	if (unlikely(rc != 0))
2126 		goto err_up_complete;
2127 
2128 	counter_u64_add(adapter->dev_stats.interface_up, 1);
2129 
2130 	ena_update_hwassist(adapter);
2131 
2132 	if_setdrvflagbits(adapter->ifp, IFF_DRV_RUNNING,
2133 		IFF_DRV_OACTIVE);
2134 
2135 	/* Activate timer service only if the device is running.
2136 		* If this flag is not set, it means that the driver is being
2137 		* reset and timer service will be activated afterwards.
2138 		*/
2139 	if (ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter)) {
2140 		callout_reset_sbt(&adapter->timer_service, SBT_1S,
2141 			SBT_1S, ena_timer_service, (void *)adapter, 0);
2142 	}
2143 
2144 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP, adapter);
2145 
2146 	ena_unmask_all_io_irqs(adapter);
2147 
2148 	return (0);
2149 
2150 err_up_complete:
2151 	ena_destroy_all_io_queues(adapter);
2152 	ena_free_all_rx_resources(adapter);
2153 	ena_free_all_tx_resources(adapter);
2154 err_create_queues_with_backoff:
2155 	ena_free_io_irq(adapter);
2156 error:
2157 	return (rc);
2158 }
2159 
2160 static uint64_t
2161 ena_get_counter(if_t ifp, ift_counter cnt)
2162 {
2163 	struct ena_adapter *adapter;
2164 	struct ena_hw_stats *stats;
2165 
2166 	adapter = if_getsoftc(ifp);
2167 	stats = &adapter->hw_stats;
2168 
2169 	switch (cnt) {
2170 	case IFCOUNTER_IPACKETS:
2171 		return (counter_u64_fetch(stats->rx_packets));
2172 	case IFCOUNTER_OPACKETS:
2173 		return (counter_u64_fetch(stats->tx_packets));
2174 	case IFCOUNTER_IBYTES:
2175 		return (counter_u64_fetch(stats->rx_bytes));
2176 	case IFCOUNTER_OBYTES:
2177 		return (counter_u64_fetch(stats->tx_bytes));
2178 	case IFCOUNTER_IQDROPS:
2179 		return (counter_u64_fetch(stats->rx_drops));
2180 	case IFCOUNTER_OQDROPS:
2181 		return (counter_u64_fetch(stats->tx_drops));
2182 	default:
2183 		return (if_get_counter_default(ifp, cnt));
2184 	}
2185 }
2186 
2187 static int
2188 ena_media_change(if_t ifp)
2189 {
2190 	/* Media Change is not supported by firmware */
2191 	return (0);
2192 }
2193 
2194 static void
2195 ena_media_status(if_t ifp, struct ifmediareq *ifmr)
2196 {
2197 	struct ena_adapter *adapter = if_getsoftc(ifp);
2198 	ena_trace(NULL, ENA_DBG, "enter\n");
2199 
2200 	ENA_LOCK_LOCK(adapter);
2201 
2202 	ifmr->ifm_status = IFM_AVALID;
2203 	ifmr->ifm_active = IFM_ETHER;
2204 
2205 	if (!ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter)) {
2206 		ENA_LOCK_UNLOCK(adapter);
2207 		ena_trace(NULL, ENA_INFO, "Link is down\n");
2208 		return;
2209 	}
2210 
2211 	ifmr->ifm_status |= IFM_ACTIVE;
2212 	ifmr->ifm_active |= IFM_UNKNOWN | IFM_FDX;
2213 
2214 	ENA_LOCK_UNLOCK(adapter);
2215 }
2216 
2217 static void
2218 ena_init(void *arg)
2219 {
2220 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
2221 
2222 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter)) {
2223 		ENA_LOCK_LOCK(adapter);
2224 		ena_up(adapter);
2225 		ENA_LOCK_UNLOCK(adapter);
2226 	}
2227 }
2228 
2229 static int
2230 ena_ioctl(if_t ifp, u_long command, caddr_t data)
2231 {
2232 	struct ena_adapter *adapter;
2233 	struct ifreq *ifr;
2234 	int rc;
2235 
2236 	adapter = ifp->if_softc;
2237 	ifr = (struct ifreq *)data;
2238 
2239 	/*
2240 	 * Acquiring lock to prevent from running up and down routines parallel.
2241 	 */
2242 	rc = 0;
2243 	switch (command) {
2244 	case SIOCSIFMTU:
2245 		if (ifp->if_mtu == ifr->ifr_mtu)
2246 			break;
2247 		ENA_LOCK_LOCK(adapter);
2248 		ena_down(adapter);
2249 
2250 		ena_change_mtu(ifp, ifr->ifr_mtu);
2251 
2252 		rc = ena_up(adapter);
2253 		ENA_LOCK_UNLOCK(adapter);
2254 		break;
2255 
2256 	case SIOCSIFFLAGS:
2257 		if ((ifp->if_flags & IFF_UP) != 0) {
2258 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2259 				if ((ifp->if_flags & (IFF_PROMISC |
2260 				    IFF_ALLMULTI)) != 0) {
2261 					device_printf(adapter->pdev,
2262 					    "ioctl promisc/allmulti\n");
2263 				}
2264 			} else {
2265 				ENA_LOCK_LOCK(adapter);
2266 				rc = ena_up(adapter);
2267 				ENA_LOCK_UNLOCK(adapter);
2268 			}
2269 		} else {
2270 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2271 				ENA_LOCK_LOCK(adapter);
2272 				ena_down(adapter);
2273 				ENA_LOCK_UNLOCK(adapter);
2274 			}
2275 		}
2276 		break;
2277 
2278 	case SIOCADDMULTI:
2279 	case SIOCDELMULTI:
2280 		break;
2281 
2282 	case SIOCSIFMEDIA:
2283 	case SIOCGIFMEDIA:
2284 		rc = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
2285 		break;
2286 
2287 	case SIOCSIFCAP:
2288 		{
2289 			int reinit = 0;
2290 
2291 			if (ifr->ifr_reqcap != ifp->if_capenable) {
2292 				ifp->if_capenable = ifr->ifr_reqcap;
2293 				reinit = 1;
2294 			}
2295 
2296 			if ((reinit != 0) &&
2297 			    ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)) {
2298 				ENA_LOCK_LOCK(adapter);
2299 				ena_down(adapter);
2300 				rc = ena_up(adapter);
2301 				ENA_LOCK_UNLOCK(adapter);
2302 			}
2303 		}
2304 
2305 		break;
2306 	default:
2307 		rc = ether_ioctl(ifp, command, data);
2308 		break;
2309 	}
2310 
2311 	return (rc);
2312 }
2313 
2314 static int
2315 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2316 {
2317 	int caps = 0;
2318 
2319 	if ((feat->offload.tx &
2320 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2321 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2322 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)) != 0)
2323 		caps |= IFCAP_TXCSUM;
2324 
2325 	if ((feat->offload.tx &
2326 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2327 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)) != 0)
2328 		caps |= IFCAP_TXCSUM_IPV6;
2329 
2330 	if ((feat->offload.tx &
2331 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0)
2332 		caps |= IFCAP_TSO4;
2333 
2334 	if ((feat->offload.tx &
2335 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK) != 0)
2336 		caps |= IFCAP_TSO6;
2337 
2338 	if ((feat->offload.rx_supported &
2339 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2340 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK)) != 0)
2341 		caps |= IFCAP_RXCSUM;
2342 
2343 	if ((feat->offload.rx_supported &
2344 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK) != 0)
2345 		caps |= IFCAP_RXCSUM_IPV6;
2346 
2347 	caps |= IFCAP_LRO | IFCAP_JUMBO_MTU;
2348 
2349 	return (caps);
2350 }
2351 
2352 static void
2353 ena_update_host_info(struct ena_admin_host_info *host_info, if_t ifp)
2354 {
2355 
2356 	host_info->supported_network_features[0] =
2357 	    (uint32_t)if_getcapabilities(ifp);
2358 }
2359 
2360 static void
2361 ena_update_hwassist(struct ena_adapter *adapter)
2362 {
2363 	if_t ifp = adapter->ifp;
2364 	uint32_t feat = adapter->tx_offload_cap;
2365 	int cap = if_getcapenable(ifp);
2366 	int flags = 0;
2367 
2368 	if_clearhwassist(ifp);
2369 
2370 	if ((cap & IFCAP_TXCSUM) != 0) {
2371 		if ((feat &
2372 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK) != 0)
2373 			flags |= CSUM_IP;
2374 		if ((feat &
2375 		    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2376 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)) != 0)
2377 			flags |= CSUM_IP_UDP | CSUM_IP_TCP;
2378 	}
2379 
2380 	if ((cap & IFCAP_TXCSUM_IPV6) != 0)
2381 		flags |= CSUM_IP6_UDP | CSUM_IP6_TCP;
2382 
2383 	if ((cap & IFCAP_TSO4) != 0)
2384 		flags |= CSUM_IP_TSO;
2385 
2386 	if ((cap & IFCAP_TSO6) != 0)
2387 		flags |= CSUM_IP6_TSO;
2388 
2389 	if_sethwassistbits(ifp, flags, 0);
2390 }
2391 
2392 static int
2393 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2394     struct ena_com_dev_get_features_ctx *feat)
2395 {
2396 	if_t ifp;
2397 	int caps = 0;
2398 
2399 	ifp = adapter->ifp = if_gethandle(IFT_ETHER);
2400 	if (unlikely(ifp == NULL)) {
2401 		ena_trace(NULL, ENA_ALERT, "can not allocate ifnet structure\n");
2402 		return (ENXIO);
2403 	}
2404 	if_initname(ifp, device_get_name(pdev), device_get_unit(pdev));
2405 	if_setdev(ifp, pdev);
2406 	if_setsoftc(ifp, adapter);
2407 
2408 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST |
2409 	    IFF_KNOWSEPOCH);
2410 	if_setinitfn(ifp, ena_init);
2411 	if_settransmitfn(ifp, ena_mq_start);
2412 	if_setqflushfn(ifp, ena_qflush);
2413 	if_setioctlfn(ifp, ena_ioctl);
2414 	if_setgetcounterfn(ifp, ena_get_counter);
2415 
2416 	if_setsendqlen(ifp, adapter->requested_tx_ring_size);
2417 	if_setsendqready(ifp);
2418 	if_setmtu(ifp, ETHERMTU);
2419 	if_setbaudrate(ifp, 0);
2420 	/* Zeroize capabilities... */
2421 	if_setcapabilities(ifp, 0);
2422 	if_setcapenable(ifp, 0);
2423 	/* check hardware support */
2424 	caps = ena_get_dev_offloads(feat);
2425 	/* ... and set them */
2426 	if_setcapabilitiesbit(ifp, caps, 0);
2427 
2428 	/* TSO parameters */
2429 	ifp->if_hw_tsomax = ENA_TSO_MAXSIZE -
2430 	    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2431 	ifp->if_hw_tsomaxsegcount = adapter->max_tx_sgl_size - 1;
2432 	ifp->if_hw_tsomaxsegsize = ENA_TSO_MAXSIZE;
2433 
2434 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
2435 	if_setcapenable(ifp, if_getcapabilities(ifp));
2436 
2437 	/*
2438 	 * Specify the media types supported by this adapter and register
2439 	 * callbacks to update media and link information
2440 	 */
2441 	ifmedia_init(&adapter->media, IFM_IMASK,
2442 	    ena_media_change, ena_media_status);
2443 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2444 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2445 
2446 	ether_ifattach(ifp, adapter->mac_addr);
2447 
2448 	return (0);
2449 }
2450 
2451 void
2452 ena_down(struct ena_adapter *adapter)
2453 {
2454 	int rc;
2455 
2456 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2457 		return;
2458 
2459 	device_printf(adapter->pdev, "device is going DOWN\n");
2460 
2461 	callout_drain(&adapter->timer_service);
2462 
2463 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP, adapter);
2464 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE,
2465 		IFF_DRV_RUNNING);
2466 
2467 	ena_free_io_irq(adapter);
2468 
2469 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter)) {
2470 		rc = ena_com_dev_reset(adapter->ena_dev,
2471 			adapter->reset_reason);
2472 		if (unlikely(rc != 0))
2473 			device_printf(adapter->pdev,
2474 				"Device reset failed\n");
2475 	}
2476 
2477 	ena_destroy_all_io_queues(adapter);
2478 
2479 	ena_free_all_tx_bufs(adapter);
2480 	ena_free_all_rx_bufs(adapter);
2481 	ena_free_all_tx_resources(adapter);
2482 	ena_free_all_rx_resources(adapter);
2483 
2484 	counter_u64_add(adapter->dev_stats.interface_down, 1);
2485 }
2486 
2487 static uint32_t
2488 ena_calc_max_io_queue_num(device_t pdev, struct ena_com_dev *ena_dev,
2489     struct ena_com_dev_get_features_ctx *get_feat_ctx)
2490 {
2491 	uint32_t io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
2492 
2493 	/* Regular queues capabilities */
2494 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2495 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2496 		    &get_feat_ctx->max_queue_ext.max_queue_ext;
2497 		io_rx_num = min_t(int, max_queue_ext->max_rx_sq_num,
2498 			max_queue_ext->max_rx_cq_num);
2499 
2500 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
2501 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
2502 	} else {
2503 		struct ena_admin_queue_feature_desc *max_queues =
2504 		    &get_feat_ctx->max_queues;
2505 		io_tx_sq_num = max_queues->max_sq_num;
2506 		io_tx_cq_num = max_queues->max_cq_num;
2507 		io_rx_num = min_t(int, io_tx_sq_num, io_tx_cq_num);
2508 	}
2509 
2510 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
2511 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
2512 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
2513 
2514 	max_num_io_queues = min_t(uint32_t, mp_ncpus, ENA_MAX_NUM_IO_QUEUES);
2515 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_rx_num);
2516 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_tx_sq_num);
2517 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_tx_cq_num);
2518 	/* 1 IRQ for for mgmnt and 1 IRQ for each TX/RX pair */
2519 	max_num_io_queues = min_t(uint32_t, max_num_io_queues,
2520 	    pci_msix_count(pdev) - 1);
2521 
2522 	return (max_num_io_queues);
2523 }
2524 
2525 static int
2526 ena_enable_wc(struct resource *res)
2527 {
2528 #if defined(__i386) || defined(__amd64) || defined(__aarch64__)
2529 	vm_offset_t va;
2530 	vm_size_t len;
2531 	int rc;
2532 
2533 	va = (vm_offset_t)rman_get_virtual(res);
2534 	len = rman_get_size(res);
2535 	/* Enable write combining */
2536 	rc = pmap_change_attr(va, len, VM_MEMATTR_WRITE_COMBINING);
2537 	if (unlikely(rc != 0)) {
2538 		ena_trace(NULL, ENA_ALERT, "pmap_change_attr failed, %d\n", rc);
2539 		return (rc);
2540 	}
2541 
2542 	return (0);
2543 #endif
2544 	return (EOPNOTSUPP);
2545 }
2546 
2547 static int
2548 ena_set_queues_placement_policy(device_t pdev, struct ena_com_dev *ena_dev,
2549     struct ena_admin_feature_llq_desc *llq,
2550     struct ena_llq_configurations *llq_default_configurations)
2551 {
2552 	struct ena_adapter *adapter = device_get_softc(pdev);
2553 	int rc, rid;
2554 	uint32_t llq_feature_mask;
2555 
2556 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
2557 	if (!(ena_dev->supported_features & llq_feature_mask)) {
2558 		device_printf(pdev,
2559 		    "LLQ is not supported. Fallback to host mode policy.\n");
2560 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2561 		return (0);
2562 	}
2563 
2564 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
2565 	if (unlikely(rc != 0)) {
2566 		device_printf(pdev, "Failed to configure the device mode. "
2567 		    "Fallback to host mode policy.\n");
2568 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2569 		return (0);
2570 	}
2571 
2572 	/* Nothing to config, exit */
2573 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
2574 		return (0);
2575 
2576 	/* Try to allocate resources for LLQ bar */
2577 	rid = PCIR_BAR(ENA_MEM_BAR);
2578 	adapter->memory = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
2579 	    &rid, RF_ACTIVE);
2580 	if (unlikely(adapter->memory == NULL)) {
2581 		device_printf(pdev, "unable to allocate LLQ bar resource. "
2582 		    "Fallback to host mode policy.\n");
2583 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2584 		return (0);
2585 	}
2586 
2587 	/* Enable write combining for better LLQ performance */
2588 	rc = ena_enable_wc(adapter->memory);
2589 	if (unlikely(rc != 0)) {
2590 		device_printf(pdev, "failed to enable write combining.\n");
2591 		return (rc);
2592 	}
2593 
2594 	/*
2595 	 * Save virtual address of the device's memory region
2596 	 * for the ena_com layer.
2597 	 */
2598 	ena_dev->mem_bar = rman_get_virtual(adapter->memory);
2599 
2600 	return (0);
2601 }
2602 
2603 static inline
2604 void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
2605 {
2606 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
2607 	llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
2608 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
2609 	llq_config->llq_num_decs_before_header =
2610 	    ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
2611 	llq_config->llq_ring_entry_size_value = 128;
2612 }
2613 
2614 static int
2615 ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
2616 {
2617 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
2618 	struct ena_com_dev *ena_dev = ctx->ena_dev;
2619 	uint32_t tx_queue_size = ENA_DEFAULT_RING_SIZE;
2620 	uint32_t rx_queue_size = ENA_DEFAULT_RING_SIZE;
2621 	uint32_t max_tx_queue_size;
2622 	uint32_t max_rx_queue_size;
2623 
2624 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2625 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2626 		    &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
2627 		max_rx_queue_size = min_t(uint32_t,
2628 		    max_queue_ext->max_rx_cq_depth,
2629 		    max_queue_ext->max_rx_sq_depth);
2630 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
2631 
2632 		if (ena_dev->tx_mem_queue_type ==
2633 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2634 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2635 			    llq->max_llq_depth);
2636 		else
2637 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2638 			    max_queue_ext->max_tx_sq_depth);
2639 
2640 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2641 		    max_queue_ext->max_per_packet_tx_descs);
2642 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2643 		    max_queue_ext->max_per_packet_rx_descs);
2644 	} else {
2645 		struct ena_admin_queue_feature_desc *max_queues =
2646 		    &ctx->get_feat_ctx->max_queues;
2647 		max_rx_queue_size = min_t(uint32_t,
2648 		    max_queues->max_cq_depth,
2649 		    max_queues->max_sq_depth);
2650 		max_tx_queue_size = max_queues->max_cq_depth;
2651 
2652 		if (ena_dev->tx_mem_queue_type ==
2653 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2654 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2655 			    llq->max_llq_depth);
2656 		else
2657 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2658 			    max_queues->max_sq_depth);
2659 
2660 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2661 		    max_queues->max_packet_tx_descs);
2662 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2663 		    max_queues->max_packet_rx_descs);
2664 	}
2665 
2666 	/* round down to the nearest power of 2 */
2667 	max_tx_queue_size = 1 << (flsl(max_tx_queue_size) - 1);
2668 	max_rx_queue_size = 1 << (flsl(max_rx_queue_size) - 1);
2669 
2670 	tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
2671 	    max_tx_queue_size);
2672 	rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
2673 	    max_rx_queue_size);
2674 
2675 	tx_queue_size = 1 << (flsl(tx_queue_size) - 1);
2676 	rx_queue_size = 1 << (flsl(rx_queue_size) - 1);
2677 
2678 	ctx->max_tx_queue_size = max_tx_queue_size;
2679 	ctx->max_rx_queue_size = max_rx_queue_size;
2680 	ctx->tx_queue_size = tx_queue_size;
2681 	ctx->rx_queue_size = rx_queue_size;
2682 
2683 	return (0);
2684 }
2685 
2686 static int
2687 ena_rss_init_default(struct ena_adapter *adapter)
2688 {
2689 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2690 	device_t dev = adapter->pdev;
2691 	int qid, rc, i;
2692 
2693 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
2694 	if (unlikely(rc != 0)) {
2695 		device_printf(dev, "Cannot init indirect table\n");
2696 		return (rc);
2697 	}
2698 
2699 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
2700 		qid = i % adapter->num_io_queues;
2701 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
2702 		    ENA_IO_RXQ_IDX(qid));
2703 		if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
2704 			device_printf(dev, "Cannot fill indirect table\n");
2705 			goto err_rss_destroy;
2706 		}
2707 	}
2708 
2709 #ifdef RSS
2710 	uint8_t rss_algo = rss_gethashalgo();
2711 	if (rss_algo == RSS_HASH_TOEPLITZ) {
2712 		uint8_t hash_key[RSS_KEYSIZE];
2713 
2714 		rss_getkey(hash_key);
2715 		rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ,
2716 		    hash_key, RSS_KEYSIZE, 0xFFFFFFFF);
2717 	} else
2718 #endif
2719 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
2720 	    ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
2721 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
2722 		device_printf(dev, "Cannot fill hash function\n");
2723 		goto err_rss_destroy;
2724 	}
2725 
2726 	rc = ena_com_set_default_hash_ctrl(ena_dev);
2727 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
2728 		device_printf(dev, "Cannot fill hash control\n");
2729 		goto err_rss_destroy;
2730 	}
2731 
2732 	return (0);
2733 
2734 err_rss_destroy:
2735 	ena_com_rss_destroy(ena_dev);
2736 	return (rc);
2737 }
2738 
2739 static void
2740 ena_rss_init_default_deferred(void *arg)
2741 {
2742 	struct ena_adapter *adapter;
2743 	devclass_t dc;
2744 	int max;
2745 	int rc;
2746 
2747 	dc = devclass_find("ena");
2748 	if (unlikely(dc == NULL)) {
2749 		ena_trace(NULL, ENA_ALERT, "No devclass ena\n");
2750 		return;
2751 	}
2752 
2753 	max = devclass_get_maxunit(dc);
2754 	while (max-- >= 0) {
2755 		adapter = devclass_get_softc(dc, max);
2756 		if (adapter != NULL) {
2757 			rc = ena_rss_init_default(adapter);
2758 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_RSS_ACTIVE, adapter);
2759 			if (unlikely(rc != 0)) {
2760 				device_printf(adapter->pdev,
2761 				    "WARNING: RSS was not properly initialized,"
2762 				    " it will affect bandwidth\n");
2763 				ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_RSS_ACTIVE, adapter);
2764 			}
2765 		}
2766 	}
2767 }
2768 SYSINIT(ena_rss_init, SI_SUB_KICK_SCHEDULER, SI_ORDER_SECOND, ena_rss_init_default_deferred, NULL);
2769 
2770 static void
2771 ena_config_host_info(struct ena_com_dev *ena_dev, device_t dev)
2772 {
2773 	struct ena_admin_host_info *host_info;
2774 	uintptr_t rid;
2775 	int rc;
2776 
2777 	/* Allocate only the host info */
2778 	rc = ena_com_allocate_host_info(ena_dev);
2779 	if (unlikely(rc != 0)) {
2780 		ena_trace(NULL, ENA_ALERT, "Cannot allocate host info\n");
2781 		return;
2782 	}
2783 
2784 	host_info = ena_dev->host_attr.host_info;
2785 
2786 	if (pci_get_id(dev, PCI_ID_RID, &rid) == 0)
2787 		host_info->bdf = rid;
2788 	host_info->os_type = ENA_ADMIN_OS_FREEBSD;
2789 	host_info->kernel_ver = osreldate;
2790 
2791 	sprintf(host_info->kernel_ver_str, "%d", osreldate);
2792 	host_info->os_dist = 0;
2793 	strncpy(host_info->os_dist_str, osrelease,
2794 	    sizeof(host_info->os_dist_str) - 1);
2795 
2796 	host_info->driver_version =
2797 		(DRV_MODULE_VER_MAJOR) |
2798 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2799 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2800 	host_info->num_cpus = mp_ncpus;
2801 	host_info->driver_supported_features =
2802 	    ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK;
2803 
2804 	rc = ena_com_set_host_attributes(ena_dev);
2805 	if (unlikely(rc != 0)) {
2806 		if (rc == EOPNOTSUPP)
2807 			ena_trace(NULL, ENA_WARNING, "Cannot set host attributes\n");
2808 		else
2809 			ena_trace(NULL, ENA_ALERT, "Cannot set host attributes\n");
2810 
2811 		goto err;
2812 	}
2813 
2814 	return;
2815 
2816 err:
2817 	ena_com_delete_host_info(ena_dev);
2818 }
2819 
2820 static int
2821 ena_device_init(struct ena_adapter *adapter, device_t pdev,
2822     struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
2823 {
2824 	struct ena_com_dev* ena_dev = adapter->ena_dev;
2825 	bool readless_supported;
2826 	uint32_t aenq_groups;
2827 	int dma_width;
2828 	int rc;
2829 
2830 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
2831 	if (unlikely(rc != 0)) {
2832 		device_printf(pdev, "failed to init mmio read less\n");
2833 		return (rc);
2834 	}
2835 
2836 	/*
2837 	 * The PCIe configuration space revision id indicate if mmio reg
2838 	 * read is disabled
2839 	 */
2840 	readless_supported = !(pci_get_revid(pdev) & ENA_MMIO_DISABLE_REG_READ);
2841 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2842 
2843 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2844 	if (unlikely(rc != 0)) {
2845 		device_printf(pdev, "Can not reset device\n");
2846 		goto err_mmio_read_less;
2847 	}
2848 
2849 	rc = ena_com_validate_version(ena_dev);
2850 	if (unlikely(rc != 0)) {
2851 		device_printf(pdev, "device version is too low\n");
2852 		goto err_mmio_read_less;
2853 	}
2854 
2855 	dma_width = ena_com_get_dma_width(ena_dev);
2856 	if (unlikely(dma_width < 0)) {
2857 		device_printf(pdev, "Invalid dma width value %d", dma_width);
2858 		rc = dma_width;
2859 		goto err_mmio_read_less;
2860 	}
2861 	adapter->dma_width = dma_width;
2862 
2863 	/* ENA admin level init */
2864 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
2865 	if (unlikely(rc != 0)) {
2866 		device_printf(pdev,
2867 		    "Can not initialize ena admin queue with device\n");
2868 		goto err_mmio_read_less;
2869 	}
2870 
2871 	/*
2872 	 * To enable the msix interrupts the driver needs to know the number
2873 	 * of queues. So the driver uses polling mode to retrieve this
2874 	 * information
2875 	 */
2876 	ena_com_set_admin_polling_mode(ena_dev, true);
2877 
2878 	ena_config_host_info(ena_dev, pdev);
2879 
2880 	/* Get Device Attributes */
2881 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2882 	if (unlikely(rc != 0)) {
2883 		device_printf(pdev,
2884 		    "Cannot get attribute for ena device rc: %d\n", rc);
2885 		goto err_admin_init;
2886 	}
2887 
2888 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2889 	    BIT(ENA_ADMIN_FATAL_ERROR) |
2890 	    BIT(ENA_ADMIN_WARNING) |
2891 	    BIT(ENA_ADMIN_NOTIFICATION) |
2892 	    BIT(ENA_ADMIN_KEEP_ALIVE);
2893 
2894 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
2895 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2896 	if (unlikely(rc != 0)) {
2897 		device_printf(pdev, "Cannot configure aenq groups rc: %d\n", rc);
2898 		goto err_admin_init;
2899 	}
2900 
2901 	*wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2902 
2903 	return (0);
2904 
2905 err_admin_init:
2906 	ena_com_delete_host_info(ena_dev);
2907 	ena_com_admin_destroy(ena_dev);
2908 err_mmio_read_less:
2909 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2910 
2911 	return (rc);
2912 }
2913 
2914 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
2915 {
2916 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2917 	int rc;
2918 
2919 	rc = ena_enable_msix(adapter);
2920 	if (unlikely(rc != 0)) {
2921 		device_printf(adapter->pdev, "Error with MSI-X enablement\n");
2922 		return (rc);
2923 	}
2924 
2925 	ena_setup_mgmnt_intr(adapter);
2926 
2927 	rc = ena_request_mgmnt_irq(adapter);
2928 	if (unlikely(rc != 0)) {
2929 		device_printf(adapter->pdev, "Cannot setup mgmnt queue intr\n");
2930 		goto err_disable_msix;
2931 	}
2932 
2933 	ena_com_set_admin_polling_mode(ena_dev, false);
2934 
2935 	ena_com_admin_aenq_enable(ena_dev);
2936 
2937 	return (0);
2938 
2939 err_disable_msix:
2940 	ena_disable_msix(adapter);
2941 
2942 	return (rc);
2943 }
2944 
2945 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
2946 static void ena_keep_alive_wd(void *adapter_data,
2947     struct ena_admin_aenq_entry *aenq_e)
2948 {
2949 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
2950 	struct ena_admin_aenq_keep_alive_desc *desc;
2951 	sbintime_t stime;
2952 	uint64_t rx_drops;
2953 	uint64_t tx_drops;
2954 
2955 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
2956 
2957 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
2958 	tx_drops = ((uint64_t)desc->tx_drops_high << 32) | desc->tx_drops_low;
2959 	counter_u64_zero(adapter->hw_stats.rx_drops);
2960 	counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
2961 	counter_u64_zero(adapter->hw_stats.tx_drops);
2962 	counter_u64_add(adapter->hw_stats.tx_drops, tx_drops);
2963 
2964 	stime = getsbinuptime();
2965 	atomic_store_rel_64(&adapter->keep_alive_timestamp, stime);
2966 }
2967 
2968 /* Check for keep alive expiration */
2969 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
2970 {
2971 	sbintime_t timestamp, time;
2972 
2973 	if (adapter->wd_active == 0)
2974 		return;
2975 
2976 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2977 		return;
2978 
2979 	timestamp = atomic_load_acq_64(&adapter->keep_alive_timestamp);
2980 	time = getsbinuptime() - timestamp;
2981 	if (unlikely(time > adapter->keep_alive_timeout)) {
2982 		device_printf(adapter->pdev,
2983 		    "Keep alive watchdog timeout.\n");
2984 		counter_u64_add(adapter->dev_stats.wd_expired, 1);
2985 		ena_trigger_reset(adapter, ENA_REGS_RESET_KEEP_ALIVE_TO);
2986 	}
2987 }
2988 
2989 /* Check if admin queue is enabled */
2990 static void check_for_admin_com_state(struct ena_adapter *adapter)
2991 {
2992 	if (unlikely(ena_com_get_admin_running_state(adapter->ena_dev) ==
2993 	    false)) {
2994 		device_printf(adapter->pdev,
2995 		    "ENA admin queue is not in running state!\n");
2996 		counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
2997 		ena_trigger_reset(adapter, ENA_REGS_RESET_ADMIN_TO);
2998 	}
2999 }
3000 
3001 static int
3002 check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3003     struct ena_ring *rx_ring)
3004 {
3005 	if (likely(rx_ring->first_interrupt))
3006 		return (0);
3007 
3008 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3009 		return (0);
3010 
3011 	rx_ring->no_interrupt_event_cnt++;
3012 
3013 	if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3014 		device_printf(adapter->pdev, "Potential MSIX issue on Rx side "
3015 		    "Queue = %d. Reset the device\n", rx_ring->qid);
3016 		ena_trigger_reset(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
3017 		return (EIO);
3018 	}
3019 
3020 	return (0);
3021 }
3022 
3023 static int
3024 check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3025     struct ena_ring *tx_ring)
3026 {
3027 	struct bintime curtime, time;
3028 	struct ena_tx_buffer *tx_buf;
3029 	sbintime_t time_offset;
3030 	uint32_t missed_tx = 0;
3031 	int i, rc = 0;
3032 
3033 	getbinuptime(&curtime);
3034 
3035 	for (i = 0; i < tx_ring->ring_size; i++) {
3036 		tx_buf = &tx_ring->tx_buffer_info[i];
3037 
3038 		if (bintime_isset(&tx_buf->timestamp) == 0)
3039 			continue;
3040 
3041 		time = curtime;
3042 		bintime_sub(&time, &tx_buf->timestamp);
3043 		time_offset = bttosbt(time);
3044 
3045 		if (unlikely(!tx_ring->first_interrupt &&
3046 		    time_offset > 2 * adapter->missing_tx_timeout)) {
3047 			/*
3048 			 * If after graceful period interrupt is still not
3049 			 * received, we schedule a reset.
3050 			 */
3051 			device_printf(adapter->pdev,
3052 			    "Potential MSIX issue on Tx side Queue = %d. "
3053 			    "Reset the device\n", tx_ring->qid);
3054 			ena_trigger_reset(adapter,
3055 			    ENA_REGS_RESET_MISS_INTERRUPT);
3056 			return (EIO);
3057 		}
3058 
3059 		/* Check again if packet is still waiting */
3060 		if (unlikely(time_offset > adapter->missing_tx_timeout)) {
3061 
3062 			if (!tx_buf->print_once)
3063 				ena_trace(NULL, ENA_WARNING, "Found a Tx that wasn't "
3064 				    "completed on time, qid %d, index %d.\n",
3065 				    tx_ring->qid, i);
3066 
3067 			tx_buf->print_once = true;
3068 			missed_tx++;
3069 		}
3070 	}
3071 
3072 	if (unlikely(missed_tx > adapter->missing_tx_threshold)) {
3073 		device_printf(adapter->pdev,
3074 		    "The number of lost tx completion is above the threshold "
3075 		    "(%d > %d). Reset the device\n",
3076 		    missed_tx, adapter->missing_tx_threshold);
3077 		ena_trigger_reset(adapter, ENA_REGS_RESET_MISS_TX_CMPL);
3078 		rc = EIO;
3079 	}
3080 
3081 	counter_u64_add(tx_ring->tx_stats.missing_tx_comp, missed_tx);
3082 
3083 	return (rc);
3084 }
3085 
3086 /*
3087  * Check for TX which were not completed on time.
3088  * Timeout is defined by "missing_tx_timeout".
3089  * Reset will be performed if number of incompleted
3090  * transactions exceeds "missing_tx_threshold".
3091  */
3092 static void
3093 check_for_missing_completions(struct ena_adapter *adapter)
3094 {
3095 	struct ena_ring *tx_ring;
3096 	struct ena_ring *rx_ring;
3097 	int i, budget, rc;
3098 
3099 	/* Make sure the driver doesn't turn the device in other process */
3100 	rmb();
3101 
3102 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3103 		return;
3104 
3105 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
3106 		return;
3107 
3108 	if (adapter->missing_tx_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3109 		return;
3110 
3111 	budget = adapter->missing_tx_max_queues;
3112 
3113 	for (i = adapter->next_monitored_tx_qid; i < adapter->num_io_queues; i++) {
3114 		tx_ring = &adapter->tx_ring[i];
3115 		rx_ring = &adapter->rx_ring[i];
3116 
3117 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3118 		if (unlikely(rc != 0))
3119 			return;
3120 
3121 		rc = check_for_rx_interrupt_queue(adapter, rx_ring);
3122 		if (unlikely(rc != 0))
3123 			return;
3124 
3125 		budget--;
3126 		if (budget == 0) {
3127 			i++;
3128 			break;
3129 		}
3130 	}
3131 
3132 	adapter->next_monitored_tx_qid = i % adapter->num_io_queues;
3133 }
3134 
3135 /* trigger rx cleanup after 2 consecutive detections */
3136 #define EMPTY_RX_REFILL 2
3137 /* For the rare case where the device runs out of Rx descriptors and the
3138  * msix handler failed to refill new Rx descriptors (due to a lack of memory
3139  * for example).
3140  * This case will lead to a deadlock:
3141  * The device won't send interrupts since all the new Rx packets will be dropped
3142  * The msix handler won't allocate new Rx descriptors so the device won't be
3143  * able to send new packets.
3144  *
3145  * When such a situation is detected - execute rx cleanup task in another thread
3146  */
3147 static void
3148 check_for_empty_rx_ring(struct ena_adapter *adapter)
3149 {
3150 	struct ena_ring *rx_ring;
3151 	int i, refill_required;
3152 
3153 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3154 		return;
3155 
3156 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
3157 		return;
3158 
3159 	for (i = 0; i < adapter->num_io_queues; i++) {
3160 		rx_ring = &adapter->rx_ring[i];
3161 
3162 		refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3163 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3164 			rx_ring->empty_rx_queue++;
3165 
3166 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL)	{
3167 				counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
3168 				    1);
3169 
3170 				device_printf(adapter->pdev,
3171 				    "trigger refill for ring %d\n", i);
3172 
3173 				taskqueue_enqueue(rx_ring->que->cleanup_tq,
3174 				    &rx_ring->que->cleanup_task);
3175 				rx_ring->empty_rx_queue = 0;
3176 			}
3177 		} else {
3178 			rx_ring->empty_rx_queue = 0;
3179 		}
3180 	}
3181 }
3182 
3183 static void ena_update_hints(struct ena_adapter *adapter,
3184 			     struct ena_admin_ena_hw_hints *hints)
3185 {
3186 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3187 
3188 	if (hints->admin_completion_tx_timeout)
3189 		ena_dev->admin_queue.completion_timeout =
3190 		    hints->admin_completion_tx_timeout * 1000;
3191 
3192 	if (hints->mmio_read_timeout)
3193 		/* convert to usec */
3194 		ena_dev->mmio_read.reg_read_to =
3195 		    hints->mmio_read_timeout * 1000;
3196 
3197 	if (hints->missed_tx_completion_count_threshold_to_reset)
3198 		adapter->missing_tx_threshold =
3199 		    hints->missed_tx_completion_count_threshold_to_reset;
3200 
3201 	if (hints->missing_tx_completion_timeout) {
3202 		if (hints->missing_tx_completion_timeout ==
3203 		     ENA_HW_HINTS_NO_TIMEOUT)
3204 			adapter->missing_tx_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3205 		else
3206 			adapter->missing_tx_timeout =
3207 			    SBT_1MS * hints->missing_tx_completion_timeout;
3208 	}
3209 
3210 	if (hints->driver_watchdog_timeout) {
3211 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3212 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3213 		else
3214 			adapter->keep_alive_timeout =
3215 			    SBT_1MS * hints->driver_watchdog_timeout;
3216 	}
3217 }
3218 
3219 /**
3220  * ena_copy_eni_metrics - Get and copy ENI metrics from the HW.
3221  * @adapter: ENA device adapter
3222  *
3223  * Returns 0 on success, EOPNOTSUPP if current HW doesn't support those metrics
3224  * and other error codes on failure.
3225  *
3226  * This function can possibly cause a race with other calls to the admin queue.
3227  * Because of that, the caller should either lock this function or make sure
3228  * that there is no race in the current context.
3229  */
3230 static int
3231 ena_copy_eni_metrics(struct ena_adapter *adapter)
3232 {
3233 	static bool print_once = true;
3234 	int rc;
3235 
3236 	rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_metrics);
3237 
3238 	if (rc != 0) {
3239 		if (rc == ENA_COM_UNSUPPORTED) {
3240 			if (print_once) {
3241 				device_printf(adapter->pdev,
3242 				    "Retrieving ENI metrics is not supported.\n");
3243 				print_once = false;
3244 			} else {
3245 				ena_trace(NULL, ENA_DBG,
3246 				    "Retrieving ENI metrics is not supported.\n");
3247 			}
3248 		} else {
3249 			device_printf(adapter->pdev,
3250 			    "Failed to get ENI metrics: %d\n", rc);
3251 		}
3252 	}
3253 
3254 	return (rc);
3255 }
3256 
3257 static void
3258 ena_timer_service(void *data)
3259 {
3260 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3261 	struct ena_admin_host_info *host_info =
3262 	    adapter->ena_dev->host_attr.host_info;
3263 
3264 	check_for_missing_keep_alive(adapter);
3265 
3266 	check_for_admin_com_state(adapter);
3267 
3268 	check_for_missing_completions(adapter);
3269 
3270 	check_for_empty_rx_ring(adapter);
3271 
3272 	/*
3273 	 * User controller update of the ENI metrics.
3274 	 * If the delay was set to 0, then the stats shouldn't be updated at
3275 	 * all.
3276 	 * Otherwise, wait 'eni_metrics_sample_interval' seconds, before
3277 	 * updating stats.
3278 	 * As timer service is executed every second, it's enough to increment
3279 	 * appropriate counter each time the timer service is executed.
3280 	 */
3281 	if ((adapter->eni_metrics_sample_interval != 0) &&
3282 	    (++adapter->eni_metrics_sample_interval_cnt >=
3283 	     adapter->eni_metrics_sample_interval)) {
3284 		/*
3285 		 * There is no race with other admin queue calls, as:
3286 		 *   - Timer service runs after interface is up, so all
3287 		 *     configuration calls to the admin queue are finished.
3288 		 *   - After interface is up, the driver doesn't use (at least
3289 		 *     for now) other functions writing to the admin queue.
3290 		 *
3291 		 * It may change in the future, so in that situation, the lock
3292 		 * will be needed. ENA_LOCK_*() cannot be used for that purpose,
3293 		 * as callout ena_timer_service is protected by them. It could
3294 		 * lead to the deadlock if callout_drain() would hold the lock
3295 		 * before ena_copy_eni_metrics() was executed. It's advised to
3296 		 * use separate lock in that situation which will be used only
3297 		 * for the admin queue.
3298 		 */
3299 		(void)ena_copy_eni_metrics(adapter);
3300 		adapter->eni_metrics_sample_interval_cnt = 0;
3301 	}
3302 
3303 
3304 	if (host_info != NULL)
3305 		ena_update_host_info(host_info, adapter->ifp);
3306 
3307 	if (unlikely(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
3308 		device_printf(adapter->pdev, "Trigger reset is on\n");
3309 		taskqueue_enqueue(adapter->reset_tq, &adapter->reset_task);
3310 		return;
3311 	}
3312 
3313 	/*
3314 	 * Schedule another timeout one second from now.
3315 	 */
3316 	callout_schedule_sbt(&adapter->timer_service, SBT_1S, SBT_1S, 0);
3317 }
3318 
3319 void
3320 ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3321 {
3322 	if_t ifp = adapter->ifp;
3323 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3324 	bool dev_up;
3325 
3326 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter))
3327 		return;
3328 
3329 	if_link_state_change(ifp, LINK_STATE_DOWN);
3330 
3331 	callout_drain(&adapter->timer_service);
3332 
3333 	dev_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
3334 	if (dev_up)
3335 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3336 
3337 	if (!graceful)
3338 		ena_com_set_admin_running_state(ena_dev, false);
3339 
3340 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3341 		ena_down(adapter);
3342 
3343 	/*
3344 	 * Stop the device from sending AENQ events (if the device was up, and
3345 	 * the trigger reset was on, ena_down already performs device reset)
3346 	 */
3347 	if (!(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter) && dev_up))
3348 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3349 
3350 	ena_free_mgmnt_irq(adapter);
3351 
3352 	ena_disable_msix(adapter);
3353 
3354 	/*
3355 	 * IO rings resources should be freed because `ena_restore_device()`
3356 	 * calls (not directly) `ena_enable_msix()`, which re-allocates MSIX
3357 	 * vectors. The amount of MSIX vectors after destroy-restore may be
3358 	 * different than before. Therefore, IO rings resources should be
3359 	 * established from scratch each time.
3360 	 */
3361 	ena_free_all_io_rings_resources(adapter);
3362 
3363 	ena_com_abort_admin_commands(ena_dev);
3364 
3365 	ena_com_wait_for_abort_completion(ena_dev);
3366 
3367 	ena_com_admin_destroy(ena_dev);
3368 
3369 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3370 
3371 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3372 
3373 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
3374 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3375 }
3376 
3377 static int
3378 ena_device_validate_params(struct ena_adapter *adapter,
3379     struct ena_com_dev_get_features_ctx *get_feat_ctx)
3380 {
3381 
3382 	if (memcmp(get_feat_ctx->dev_attr.mac_addr, adapter->mac_addr,
3383 	    ETHER_ADDR_LEN) != 0) {
3384 		device_printf(adapter->pdev,
3385 		    "Error, mac address are different\n");
3386 		return (EINVAL);
3387 	}
3388 
3389 	if (get_feat_ctx->dev_attr.max_mtu < if_getmtu(adapter->ifp)) {
3390 		device_printf(adapter->pdev,
3391 		    "Error, device max mtu is smaller than ifp MTU\n");
3392 		return (EINVAL);
3393 	}
3394 
3395 	return 0;
3396 }
3397 
3398 int
3399 ena_restore_device(struct ena_adapter *adapter)
3400 {
3401 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3402 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3403 	if_t ifp = adapter->ifp;
3404 	device_t dev = adapter->pdev;
3405 	int wd_active;
3406 	int rc;
3407 
3408 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3409 
3410 	rc = ena_device_init(adapter, dev, &get_feat_ctx, &wd_active);
3411 	if (rc != 0) {
3412 		device_printf(dev, "Cannot initialize device\n");
3413 		goto err;
3414 	}
3415 	/*
3416 	 * Only enable WD if it was enabled before reset, so it won't override
3417 	 * value set by the user by the sysctl.
3418 	 */
3419 	if (adapter->wd_active != 0)
3420 		adapter->wd_active = wd_active;
3421 
3422 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
3423 	if (rc != 0) {
3424 		device_printf(dev, "Validation of device parameters failed\n");
3425 		goto err_device_destroy;
3426 	}
3427 
3428 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3429 	/* Make sure we don't have a race with AENQ Links state handler */
3430 	if (ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter))
3431 		if_link_state_change(ifp, LINK_STATE_UP);
3432 
3433 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3434 	if (rc != 0) {
3435 		device_printf(dev, "Enable MSI-X failed\n");
3436 		goto err_device_destroy;
3437 	}
3438 
3439 	/*
3440 	 * Effective value of used MSIX vectors should be the same as before
3441 	 * `ena_destroy_device()`, if possible, or closest to it if less vectors
3442 	 * are available.
3443 	 */
3444 	if ((adapter->msix_vecs - ENA_ADMIN_MSIX_VEC) < adapter->num_io_queues)
3445 		adapter->num_io_queues =
3446 		    adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
3447 
3448 	/* Re-initialize rings basic information */
3449 	ena_init_io_rings(adapter);
3450 
3451 	/* If the interface was up before the reset bring it up */
3452 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter)) {
3453 		rc = ena_up(adapter);
3454 		if (rc != 0) {
3455 			device_printf(dev, "Failed to create I/O queues\n");
3456 			goto err_disable_msix;
3457 		}
3458 	}
3459 
3460 	/* Indicate that device is running again and ready to work */
3461 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3462 
3463 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter)) {
3464 		/*
3465 		 * As the AENQ handlers weren't executed during reset because
3466 		 * the flag ENA_FLAG_DEVICE_RUNNING was turned off, the
3467 		 * timestamp must be updated again That will prevent next reset
3468 		 * caused by missing keep alive.
3469 		 */
3470 		adapter->keep_alive_timestamp = getsbinuptime();
3471 		callout_reset_sbt(&adapter->timer_service, SBT_1S, SBT_1S,
3472 		    ena_timer_service, (void *)adapter, 0);
3473 	}
3474 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3475 
3476 	device_printf(dev,
3477 	    "Device reset completed successfully, Driver info: %s\n", ena_version);
3478 
3479 	return (rc);
3480 
3481 err_disable_msix:
3482 	ena_free_mgmnt_irq(adapter);
3483 	ena_disable_msix(adapter);
3484 err_device_destroy:
3485 	ena_com_abort_admin_commands(ena_dev);
3486 	ena_com_wait_for_abort_completion(ena_dev);
3487 	ena_com_admin_destroy(ena_dev);
3488 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3489 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3490 err:
3491 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3492 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3493 	device_printf(dev, "Reset attempt failed. Can not reset the device\n");
3494 
3495 	return (rc);
3496 }
3497 
3498 static void
3499 ena_reset_task(void *arg, int pending)
3500 {
3501 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3502 
3503 	if (unlikely(!ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
3504 		device_printf(adapter->pdev,
3505 		    "device reset scheduled but trigger_reset is off\n");
3506 		return;
3507 	}
3508 
3509 	ENA_LOCK_LOCK(adapter);
3510 	ena_destroy_device(adapter, false);
3511 	ena_restore_device(adapter);
3512 	ENA_LOCK_UNLOCK(adapter);
3513 }
3514 
3515 /**
3516  * ena_attach - Device Initialization Routine
3517  * @pdev: device information struct
3518  *
3519  * Returns 0 on success, otherwise on failure.
3520  *
3521  * ena_attach initializes an adapter identified by a device structure.
3522  * The OS initialization, configuring of the adapter private structure,
3523  * and a hardware reset occur.
3524  **/
3525 static int
3526 ena_attach(device_t pdev)
3527 {
3528 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3529 	struct ena_llq_configurations llq_config;
3530 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
3531 	static int version_printed;
3532 	struct ena_adapter *adapter;
3533 	struct ena_com_dev *ena_dev = NULL;
3534 	uint32_t max_num_io_queues;
3535 	int rid, rc;
3536 
3537 	adapter = device_get_softc(pdev);
3538 	adapter->pdev = pdev;
3539 
3540 	ENA_LOCK_INIT(adapter);
3541 
3542 	/*
3543 	 * Set up the timer service - driver is responsible for avoiding
3544 	 * concurrency, as the callout won't be using any locking inside.
3545 	 */
3546 	callout_init(&adapter->timer_service, true);
3547 	adapter->keep_alive_timeout = DEFAULT_KEEP_ALIVE_TO;
3548 	adapter->missing_tx_timeout = DEFAULT_TX_CMP_TO;
3549 	adapter->missing_tx_max_queues = DEFAULT_TX_MONITORED_QUEUES;
3550 	adapter->missing_tx_threshold = DEFAULT_TX_CMP_THRESHOLD;
3551 
3552 	if (version_printed++ == 0)
3553 		device_printf(pdev, "%s\n", ena_version);
3554 
3555 	/* Allocate memory for ena_dev structure */
3556 	ena_dev = malloc(sizeof(struct ena_com_dev), M_DEVBUF,
3557 	    M_WAITOK | M_ZERO);
3558 
3559 	adapter->ena_dev = ena_dev;
3560 	ena_dev->dmadev = pdev;
3561 
3562 	rid = PCIR_BAR(ENA_REG_BAR);
3563 	adapter->memory = NULL;
3564 	adapter->registers = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
3565 	    &rid, RF_ACTIVE);
3566 	if (unlikely(adapter->registers == NULL)) {
3567 		device_printf(pdev,
3568 		    "unable to allocate bus resource: registers!\n");
3569 		rc = ENOMEM;
3570 		goto err_dev_free;
3571 	}
3572 
3573 	ena_dev->bus = malloc(sizeof(struct ena_bus), M_DEVBUF,
3574 	    M_WAITOK | M_ZERO);
3575 
3576 	/* Store register resources */
3577 	((struct ena_bus*)(ena_dev->bus))->reg_bar_t =
3578 	    rman_get_bustag(adapter->registers);
3579 	((struct ena_bus*)(ena_dev->bus))->reg_bar_h =
3580 	    rman_get_bushandle(adapter->registers);
3581 
3582 	if (unlikely(((struct ena_bus*)(ena_dev->bus))->reg_bar_h == 0)) {
3583 		device_printf(pdev, "failed to pmap registers bar\n");
3584 		rc = ENXIO;
3585 		goto err_bus_free;
3586 	}
3587 
3588 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3589 
3590 	/* Initially clear all the flags */
3591 	ENA_FLAG_ZERO(adapter);
3592 
3593 	/* Device initialization */
3594 	rc = ena_device_init(adapter, pdev, &get_feat_ctx, &adapter->wd_active);
3595 	if (unlikely(rc != 0)) {
3596 		device_printf(pdev, "ENA device init failed! (err: %d)\n", rc);
3597 		rc = ENXIO;
3598 		goto err_bus_free;
3599 	}
3600 
3601 	set_default_llq_configurations(&llq_config);
3602 
3603 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
3604 	     &llq_config);
3605 	if (unlikely(rc != 0)) {
3606 		device_printf(pdev, "failed to set placement policy\n");
3607 		goto err_com_free;
3608 	}
3609 
3610 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3611 		adapter->disable_meta_caching =
3612 		    !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
3613 		    BIT(ENA_ADMIN_DISABLE_META_CACHING));
3614 
3615 	adapter->keep_alive_timestamp = getsbinuptime();
3616 
3617 	adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3618 
3619 	memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3620 	    ETHER_ADDR_LEN);
3621 
3622 	calc_queue_ctx.pdev = pdev;
3623 	calc_queue_ctx.ena_dev = ena_dev;
3624 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
3625 
3626 	/* Calculate initial and maximum IO queue number and size */
3627 	max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev,
3628 	    &get_feat_ctx);
3629 	rc = ena_calc_io_queue_size(&calc_queue_ctx);
3630 	if (unlikely((rc != 0) || (max_num_io_queues <= 0))) {
3631 		rc = EFAULT;
3632 		goto err_com_free;
3633 	}
3634 
3635 	adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
3636 	adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
3637 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
3638 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
3639 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
3640 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
3641 
3642 	adapter->max_num_io_queues = max_num_io_queues;
3643 
3644 	adapter->buf_ring_size = ENA_DEFAULT_BUF_RING_SIZE;
3645 
3646 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
3647 
3648 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3649 
3650 	/* set up dma tags for rx and tx buffers */
3651 	rc = ena_setup_tx_dma_tag(adapter);
3652 	if (unlikely(rc != 0)) {
3653 		device_printf(pdev, "Failed to create TX DMA tag\n");
3654 		goto err_com_free;
3655 	}
3656 
3657 	rc = ena_setup_rx_dma_tag(adapter);
3658 	if (unlikely(rc != 0)) {
3659 		device_printf(pdev, "Failed to create RX DMA tag\n");
3660 		goto err_tx_tag_free;
3661 	}
3662 
3663 	/*
3664 	 * The amount of requested MSIX vectors is equal to
3665 	 * adapter::max_num_io_queues (see `ena_enable_msix()`), plus a constant
3666 	 * number of admin queue interrupts. The former is initially determined
3667 	 * by HW capabilities (see `ena_calc_max_io_queue_num())` but may not be
3668 	 * achieved if there are not enough system resources. By default, the
3669 	 * number of effectively used IO queues is the same but later on it can
3670 	 * be limited by the user using sysctl interface.
3671 	 */
3672 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3673 	if (unlikely(rc != 0)) {
3674 		device_printf(pdev,
3675 		    "Failed to enable and set the admin interrupts\n");
3676 		goto err_io_free;
3677 	}
3678 	/* By default all of allocated MSIX vectors are actively used */
3679 	adapter->num_io_queues = adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
3680 
3681 	/* initialize rings basic information */
3682 	ena_init_io_rings(adapter);
3683 
3684 	/* setup network interface */
3685 	rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
3686 	if (unlikely(rc != 0)) {
3687 		device_printf(pdev, "Error with network interface setup\n");
3688 		goto err_msix_free;
3689 	}
3690 
3691 	/* Initialize reset task queue */
3692 	TASK_INIT(&adapter->reset_task, 0, ena_reset_task, adapter);
3693 	adapter->reset_tq = taskqueue_create("ena_reset_enqueue",
3694 	    M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->reset_tq);
3695 	taskqueue_start_threads(&adapter->reset_tq, 1, PI_NET,
3696 	    "%s rstq", device_get_nameunit(adapter->pdev));
3697 
3698 	/* Initialize statistics */
3699 	ena_alloc_counters((counter_u64_t *)&adapter->dev_stats,
3700 	    sizeof(struct ena_stats_dev));
3701 	ena_alloc_counters((counter_u64_t *)&adapter->hw_stats,
3702 	    sizeof(struct ena_hw_stats));
3703 	ena_sysctl_add_nodes(adapter);
3704 
3705 #ifdef DEV_NETMAP
3706 	rc = ena_netmap_attach(adapter);
3707 	if (rc != 0) {
3708 		device_printf(pdev, "netmap attach failed: %d\n", rc);
3709 		goto err_detach;
3710 	}
3711 #endif /* DEV_NETMAP */
3712 
3713 	/* Tell the stack that the interface is not active */
3714 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
3715 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3716 
3717 	return (0);
3718 
3719 #ifdef DEV_NETMAP
3720 err_detach:
3721 	ether_ifdetach(adapter->ifp);
3722 #endif /* DEV_NETMAP */
3723 err_msix_free:
3724 	ena_com_dev_reset(adapter->ena_dev, ENA_REGS_RESET_INIT_ERR);
3725 	ena_free_mgmnt_irq(adapter);
3726 	ena_disable_msix(adapter);
3727 err_io_free:
3728 	ena_free_all_io_rings_resources(adapter);
3729 	ena_free_rx_dma_tag(adapter);
3730 err_tx_tag_free:
3731 	ena_free_tx_dma_tag(adapter);
3732 err_com_free:
3733 	ena_com_admin_destroy(ena_dev);
3734 	ena_com_delete_host_info(ena_dev);
3735 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3736 err_bus_free:
3737 	free(ena_dev->bus, M_DEVBUF);
3738 	ena_free_pci_resources(adapter);
3739 err_dev_free:
3740 	free(ena_dev, M_DEVBUF);
3741 
3742 	return (rc);
3743 }
3744 
3745 /**
3746  * ena_detach - Device Removal Routine
3747  * @pdev: device information struct
3748  *
3749  * ena_detach is called by the device subsystem to alert the driver
3750  * that it should release a PCI device.
3751  **/
3752 static int
3753 ena_detach(device_t pdev)
3754 {
3755 	struct ena_adapter *adapter = device_get_softc(pdev);
3756 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3757 	int rc;
3758 
3759 	/* Make sure VLANS are not using driver */
3760 	if (adapter->ifp->if_vlantrunk != NULL) {
3761 		device_printf(adapter->pdev ,"VLAN is in use, detach first\n");
3762 		return (EBUSY);
3763 	}
3764 
3765 	ether_ifdetach(adapter->ifp);
3766 
3767 	/* Stop timer service */
3768 	ENA_LOCK_LOCK(adapter);
3769 	callout_drain(&adapter->timer_service);
3770 	ENA_LOCK_UNLOCK(adapter);
3771 
3772 	/* Release reset task */
3773 	while (taskqueue_cancel(adapter->reset_tq, &adapter->reset_task, NULL))
3774 		taskqueue_drain(adapter->reset_tq, &adapter->reset_task);
3775 	taskqueue_free(adapter->reset_tq);
3776 
3777 	ENA_LOCK_LOCK(adapter);
3778 	ena_down(adapter);
3779 	ena_destroy_device(adapter, true);
3780 	ENA_LOCK_UNLOCK(adapter);
3781 
3782 #ifdef DEV_NETMAP
3783 	netmap_detach(adapter->ifp);
3784 #endif /* DEV_NETMAP */
3785 
3786 	ena_free_counters((counter_u64_t *)&adapter->hw_stats,
3787 	    sizeof(struct ena_hw_stats));
3788 	ena_free_counters((counter_u64_t *)&adapter->dev_stats,
3789 	    sizeof(struct ena_stats_dev));
3790 
3791 	rc = ena_free_rx_dma_tag(adapter);
3792 	if (unlikely(rc != 0))
3793 		device_printf(adapter->pdev,
3794 		    "Unmapped RX DMA tag associations\n");
3795 
3796 	rc = ena_free_tx_dma_tag(adapter);
3797 	if (unlikely(rc != 0))
3798 		device_printf(adapter->pdev,
3799 		    "Unmapped TX DMA tag associations\n");
3800 
3801 	ena_free_irqs(adapter);
3802 
3803 	ena_free_pci_resources(adapter);
3804 
3805 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter)))
3806 		ena_com_rss_destroy(ena_dev);
3807 
3808 	ena_com_delete_host_info(ena_dev);
3809 
3810 	ENA_LOCK_DESTROY(adapter);
3811 
3812 	if_free(adapter->ifp);
3813 
3814 	if (ena_dev->bus != NULL)
3815 		free(ena_dev->bus, M_DEVBUF);
3816 
3817 	if (ena_dev != NULL)
3818 		free(ena_dev, M_DEVBUF);
3819 
3820 	return (bus_generic_detach(pdev));
3821 }
3822 
3823 /******************************************************************************
3824  ******************************** AENQ Handlers *******************************
3825  *****************************************************************************/
3826 /**
3827  * ena_update_on_link_change:
3828  * Notify the network interface about the change in link status
3829  **/
3830 static void
3831 ena_update_on_link_change(void *adapter_data,
3832     struct ena_admin_aenq_entry *aenq_e)
3833 {
3834 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3835 	struct ena_admin_aenq_link_change_desc *aenq_desc;
3836 	int status;
3837 	if_t ifp;
3838 
3839 	aenq_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
3840 	ifp = adapter->ifp;
3841 	status = aenq_desc->flags &
3842 	    ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3843 
3844 	if (status != 0) {
3845 		device_printf(adapter->pdev, "link is UP\n");
3846 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_LINK_UP, adapter);
3847 		if (!ENA_FLAG_ISSET(ENA_FLAG_ONGOING_RESET, adapter))
3848 			if_link_state_change(ifp, LINK_STATE_UP);
3849 	} else {
3850 		device_printf(adapter->pdev, "link is DOWN\n");
3851 		if_link_state_change(ifp, LINK_STATE_DOWN);
3852 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_LINK_UP, adapter);
3853 	}
3854 }
3855 
3856 static void ena_notification(void *adapter_data,
3857     struct ena_admin_aenq_entry *aenq_e)
3858 {
3859 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3860 	struct ena_admin_ena_hw_hints *hints;
3861 
3862 	ENA_WARN(NULL, aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3863 	    "Invalid group(%x) expected %x\n",	aenq_e->aenq_common_desc.group,
3864 	    ENA_ADMIN_NOTIFICATION);
3865 
3866 	switch (aenq_e->aenq_common_desc.syndrome) {
3867 	case ENA_ADMIN_UPDATE_HINTS:
3868 		hints =
3869 		    (struct ena_admin_ena_hw_hints *)(&aenq_e->inline_data_w4);
3870 		ena_update_hints(adapter, hints);
3871 		break;
3872 	default:
3873 		device_printf(adapter->pdev,
3874 		    "Invalid aenq notification link state %d\n",
3875 		    aenq_e->aenq_common_desc.syndrome);
3876 	}
3877 }
3878 
3879 /**
3880  * This handler will called for unknown event group or unimplemented handlers
3881  **/
3882 static void
3883 unimplemented_aenq_handler(void *adapter_data,
3884     struct ena_admin_aenq_entry *aenq_e)
3885 {
3886 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3887 
3888 	device_printf(adapter->pdev,
3889 	    "Unknown event was received or event with unimplemented handler\n");
3890 }
3891 
3892 static struct ena_aenq_handlers aenq_handlers = {
3893     .handlers = {
3894 	    [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3895 	    [ENA_ADMIN_NOTIFICATION] = ena_notification,
3896 	    [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3897     },
3898     .unimplemented_handler = unimplemented_aenq_handler
3899 };
3900 
3901 /*********************************************************************
3902  *  FreeBSD Device Interface Entry Points
3903  *********************************************************************/
3904 
3905 static device_method_t ena_methods[] = {
3906     /* Device interface */
3907     DEVMETHOD(device_probe, ena_probe),
3908     DEVMETHOD(device_attach, ena_attach),
3909     DEVMETHOD(device_detach, ena_detach),
3910     DEVMETHOD_END
3911 };
3912 
3913 static driver_t ena_driver = {
3914     "ena", ena_methods, sizeof(struct ena_adapter),
3915 };
3916 
3917 devclass_t ena_devclass;
3918 DRIVER_MODULE(ena, pci, ena_driver, ena_devclass, 0, 0);
3919 MODULE_PNP_INFO("U16:vendor;U16:device", pci, ena, ena_vendor_info_array,
3920     nitems(ena_vendor_info_array) - 1);
3921 MODULE_DEPEND(ena, pci, 1, 1, 1);
3922 MODULE_DEPEND(ena, ether, 1, 1, 1);
3923 #ifdef DEV_NETMAP
3924 MODULE_DEPEND(ena, netmap, 1, 1, 1);
3925 #endif /* DEV_NETMAP */
3926 
3927 /*********************************************************************/
3928