xref: /freebsd/sys/dev/ena/ena.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2015-2021 Amazon.com, Inc. or its affiliates.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  *
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
23  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
24  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
28  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 #include <sys/cdefs.h>
31 #include "opt_rss.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/bus.h>
36 #include <sys/endian.h>
37 #include <sys/eventhandler.h>
38 #include <sys/kernel.h>
39 #include <sys/kthread.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/module.h>
43 #include <sys/rman.h>
44 #include <sys/smp.h>
45 #include <sys/socket.h>
46 #include <sys/sockio.h>
47 #include <sys/sysctl.h>
48 #include <sys/taskqueue.h>
49 #include <sys/time.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 
54 #include <machine/atomic.h>
55 #include <machine/bus.h>
56 #include <machine/in_cksum.h>
57 #include <machine/resource.h>
58 
59 #include <dev/pci/pcireg.h>
60 #include <dev/pci/pcivar.h>
61 
62 #include <net/bpf.h>
63 #include <net/ethernet.h>
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_media.h>
68 #include <net/if_types.h>
69 #include <net/if_var.h>
70 #include <net/if_vlan_var.h>
71 #include <netinet/in.h>
72 #include <netinet/in_systm.h>
73 #include <netinet/if_ether.h>
74 #include <netinet/ip.h>
75 #include <netinet/ip6.h>
76 #include <netinet/tcp.h>
77 #include <netinet/udp.h>
78 
79 #include "ena.h"
80 #include "ena_datapath.h"
81 #include "ena_rss.h"
82 #include "ena_sysctl.h"
83 
84 #ifdef DEV_NETMAP
85 #include "ena_netmap.h"
86 #endif /* DEV_NETMAP */
87 
88 /*********************************************************
89  *  Function prototypes
90  *********************************************************/
91 static int ena_probe(device_t);
92 static void ena_intr_msix_mgmnt(void *);
93 static void ena_free_pci_resources(struct ena_adapter *);
94 static int ena_change_mtu(if_t, int);
95 static inline void ena_alloc_counters(counter_u64_t *, int);
96 static inline void ena_free_counters(counter_u64_t *, int);
97 static inline void ena_reset_counters(counter_u64_t *, int);
98 static void ena_init_io_rings_common(struct ena_adapter *, struct ena_ring *,
99     uint16_t);
100 static void ena_init_io_rings_basic(struct ena_adapter *);
101 static void ena_init_io_rings_advanced(struct ena_adapter *);
102 static void ena_init_io_rings(struct ena_adapter *);
103 static void ena_free_io_ring_resources(struct ena_adapter *, unsigned int);
104 static void ena_free_all_io_rings_resources(struct ena_adapter *);
105 static int ena_setup_tx_dma_tag(struct ena_adapter *);
106 static int ena_free_tx_dma_tag(struct ena_adapter *);
107 static int ena_setup_rx_dma_tag(struct ena_adapter *);
108 static int ena_free_rx_dma_tag(struct ena_adapter *);
109 static void ena_release_all_tx_dmamap(struct ena_ring *);
110 static int ena_setup_tx_resources(struct ena_adapter *, int);
111 static void ena_free_tx_resources(struct ena_adapter *, int);
112 static int ena_setup_all_tx_resources(struct ena_adapter *);
113 static void ena_free_all_tx_resources(struct ena_adapter *);
114 static int ena_setup_rx_resources(struct ena_adapter *, unsigned int);
115 static void ena_free_rx_resources(struct ena_adapter *, unsigned int);
116 static int ena_setup_all_rx_resources(struct ena_adapter *);
117 static void ena_free_all_rx_resources(struct ena_adapter *);
118 static inline int ena_alloc_rx_mbuf(struct ena_adapter *, struct ena_ring *,
119     struct ena_rx_buffer *);
120 static void ena_free_rx_mbuf(struct ena_adapter *, struct ena_ring *,
121     struct ena_rx_buffer *);
122 static void ena_free_rx_bufs(struct ena_adapter *, unsigned int);
123 static void ena_refill_all_rx_bufs(struct ena_adapter *);
124 static void ena_free_all_rx_bufs(struct ena_adapter *);
125 static void ena_free_tx_bufs(struct ena_adapter *, unsigned int);
126 static void ena_free_all_tx_bufs(struct ena_adapter *);
127 static void ena_destroy_all_tx_queues(struct ena_adapter *);
128 static void ena_destroy_all_rx_queues(struct ena_adapter *);
129 static void ena_destroy_all_io_queues(struct ena_adapter *);
130 static int ena_create_io_queues(struct ena_adapter *);
131 static int ena_handle_msix(void *);
132 static int ena_enable_msix(struct ena_adapter *);
133 static void ena_setup_mgmnt_intr(struct ena_adapter *);
134 static int ena_setup_io_intr(struct ena_adapter *);
135 static int ena_request_mgmnt_irq(struct ena_adapter *);
136 static int ena_request_io_irq(struct ena_adapter *);
137 static void ena_free_mgmnt_irq(struct ena_adapter *);
138 static void ena_free_io_irq(struct ena_adapter *);
139 static void ena_free_irqs(struct ena_adapter *);
140 static void ena_disable_msix(struct ena_adapter *);
141 static void ena_unmask_all_io_irqs(struct ena_adapter *);
142 static int ena_up_complete(struct ena_adapter *);
143 static uint64_t ena_get_counter(if_t, ift_counter);
144 static int ena_media_change(if_t);
145 static void ena_media_status(if_t, struct ifmediareq *);
146 static void ena_init(void *);
147 static int ena_ioctl(if_t, u_long, caddr_t);
148 static int ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *);
149 static void ena_update_host_info(struct ena_admin_host_info *, if_t);
150 static void ena_update_hwassist(struct ena_adapter *);
151 static int ena_setup_ifnet(device_t, struct ena_adapter *,
152     struct ena_com_dev_get_features_ctx *);
153 static int ena_enable_wc(device_t, struct resource *);
154 static int ena_set_queues_placement_policy(device_t, struct ena_com_dev *,
155     struct ena_admin_feature_llq_desc *, struct ena_llq_configurations *);
156 static int ena_map_llq_mem_bar(device_t, struct ena_com_dev *);
157 static uint32_t ena_calc_max_io_queue_num(device_t, struct ena_com_dev *,
158     struct ena_com_dev_get_features_ctx *);
159 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *);
160 static void ena_config_host_info(struct ena_com_dev *, device_t);
161 static int ena_attach(device_t);
162 static int ena_detach(device_t);
163 static int ena_device_init(struct ena_adapter *, device_t,
164     struct ena_com_dev_get_features_ctx *, int *);
165 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *);
166 static void ena_update_on_link_change(void *, struct ena_admin_aenq_entry *);
167 static void unimplemented_aenq_handler(void *, struct ena_admin_aenq_entry *);
168 static int ena_copy_eni_metrics(struct ena_adapter *);
169 static void ena_timer_service(void *);
170 
171 static char ena_version[] = ENA_DEVICE_NAME ENA_DRV_MODULE_NAME
172     " v" ENA_DRV_MODULE_VERSION;
173 
174 static ena_vendor_info_t ena_vendor_info_array[] = {
175 	{ PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0 },
176 	{ PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF_RSERV0, 0 },
177 	{ PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0 },
178 	{ PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF_RSERV0, 0 },
179 	/* Last entry */
180 	{ 0, 0, 0 }
181 };
182 
183 struct sx ena_global_lock;
184 
185 /*
186  * Contains pointers to event handlers, e.g. link state chage.
187  */
188 static struct ena_aenq_handlers aenq_handlers;
189 
190 void
191 ena_dmamap_callback(void *arg, bus_dma_segment_t *segs, int nseg, int error)
192 {
193 	if (error != 0)
194 		return;
195 	*(bus_addr_t *)arg = segs[0].ds_addr;
196 }
197 
198 int
199 ena_dma_alloc(device_t dmadev, bus_size_t size, ena_mem_handle_t *dma,
200     int mapflags, bus_size_t alignment, int domain)
201 {
202 	struct ena_adapter *adapter = device_get_softc(dmadev);
203 	device_t pdev = adapter->pdev;
204 	uint32_t maxsize;
205 	uint64_t dma_space_addr;
206 	int error;
207 
208 	maxsize = ((size - 1) / PAGE_SIZE + 1) * PAGE_SIZE;
209 
210 	dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
211 	if (unlikely(dma_space_addr == 0))
212 		dma_space_addr = BUS_SPACE_MAXADDR;
213 
214 	error = bus_dma_tag_create(bus_get_dma_tag(dmadev), /* parent */
215 	    alignment, 0,      /* alignment, bounds 		*/
216 	    dma_space_addr,    /* lowaddr of exclusion window	*/
217 	    BUS_SPACE_MAXADDR, /* highaddr of exclusion window	*/
218 	    NULL, NULL,	       /* filter, filterarg 		*/
219 	    maxsize,	       /* maxsize 			*/
220 	    1,		       /* nsegments 			*/
221 	    maxsize,	       /* maxsegsize 			*/
222 	    BUS_DMA_ALLOCNOW,  /* flags 			*/
223 	    NULL,	       /* lockfunc 			*/
224 	    NULL,	       /* lockarg 			*/
225 	    &dma->tag);
226 	if (unlikely(error != 0)) {
227 		ena_log(pdev, ERR, "bus_dma_tag_create failed: %d\n", error);
228 		goto fail_tag;
229 	}
230 
231 	error = bus_dma_tag_set_domain(dma->tag, domain);
232 	if (unlikely(error != 0)) {
233 		ena_log(pdev, ERR, "bus_dma_tag_set_domain failed: %d\n",
234 		    error);
235 		goto fail_map_create;
236 	}
237 
238 	error = bus_dmamem_alloc(dma->tag, (void **)&dma->vaddr,
239 	    BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->map);
240 	if (unlikely(error != 0)) {
241 		ena_log(pdev, ERR, "bus_dmamem_alloc(%ju) failed: %d\n",
242 		    (uintmax_t)size, error);
243 		goto fail_map_create;
244 	}
245 
246 	dma->paddr = 0;
247 	error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr, size,
248 	    ena_dmamap_callback, &dma->paddr, mapflags);
249 	if (unlikely((error != 0) || (dma->paddr == 0))) {
250 		ena_log(pdev, ERR, "bus_dmamap_load failed: %d\n", error);
251 		goto fail_map_load;
252 	}
253 
254 	bus_dmamap_sync(dma->tag, dma->map,
255 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
256 
257 	return (0);
258 
259 fail_map_load:
260 	bus_dmamem_free(dma->tag, dma->vaddr, dma->map);
261 fail_map_create:
262 	bus_dma_tag_destroy(dma->tag);
263 fail_tag:
264 	dma->tag = NULL;
265 	dma->vaddr = NULL;
266 	dma->paddr = 0;
267 
268 	return (error);
269 }
270 
271 static void
272 ena_free_pci_resources(struct ena_adapter *adapter)
273 {
274 	device_t pdev = adapter->pdev;
275 
276 	if (adapter->memory != NULL) {
277 		bus_release_resource(pdev, SYS_RES_MEMORY,
278 		    PCIR_BAR(ENA_MEM_BAR), adapter->memory);
279 	}
280 
281 	if (adapter->registers != NULL) {
282 		bus_release_resource(pdev, SYS_RES_MEMORY,
283 		    PCIR_BAR(ENA_REG_BAR), adapter->registers);
284 	}
285 
286 	if (adapter->msix != NULL) {
287 		bus_release_resource(pdev, SYS_RES_MEMORY, adapter->msix_rid,
288 		    adapter->msix);
289 	}
290 }
291 
292 static int
293 ena_probe(device_t dev)
294 {
295 	ena_vendor_info_t *ent;
296 	uint16_t pci_vendor_id = 0;
297 	uint16_t pci_device_id = 0;
298 
299 	pci_vendor_id = pci_get_vendor(dev);
300 	pci_device_id = pci_get_device(dev);
301 
302 	ent = ena_vendor_info_array;
303 	while (ent->vendor_id != 0) {
304 		if ((pci_vendor_id == ent->vendor_id) &&
305 		    (pci_device_id == ent->device_id)) {
306 			ena_log_raw(DBG, "vendor=%x device=%x\n", pci_vendor_id,
307 			    pci_device_id);
308 
309 			device_set_desc(dev, ENA_DEVICE_DESC);
310 			return (BUS_PROBE_DEFAULT);
311 		}
312 
313 		ent++;
314 	}
315 
316 	return (ENXIO);
317 }
318 
319 static int
320 ena_change_mtu(if_t ifp, int new_mtu)
321 {
322 	struct ena_adapter *adapter = if_getsoftc(ifp);
323 	device_t pdev = adapter->pdev;
324 	int rc;
325 
326 	if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
327 		ena_log(pdev, ERR, "Invalid MTU setting. new_mtu: %d max mtu: %d min mtu: %d\n",
328 		    new_mtu, adapter->max_mtu, ENA_MIN_MTU);
329 		return (EINVAL);
330 	}
331 
332 	rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
333 	if (likely(rc == 0)) {
334 		ena_log(pdev, DBG, "set MTU to %d\n", new_mtu);
335 		if_setmtu(ifp, new_mtu);
336 	} else {
337 		ena_log(pdev, ERR, "Failed to set MTU to %d\n", new_mtu);
338 	}
339 
340 	return (rc);
341 }
342 
343 static inline void
344 ena_alloc_counters(counter_u64_t *begin, int size)
345 {
346 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
347 
348 	for (; begin < end; ++begin)
349 		*begin = counter_u64_alloc(M_WAITOK);
350 }
351 
352 static inline void
353 ena_free_counters(counter_u64_t *begin, int size)
354 {
355 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
356 
357 	for (; begin < end; ++begin)
358 		counter_u64_free(*begin);
359 }
360 
361 static inline void
362 ena_reset_counters(counter_u64_t *begin, int size)
363 {
364 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
365 
366 	for (; begin < end; ++begin)
367 		counter_u64_zero(*begin);
368 }
369 
370 static void
371 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
372     uint16_t qid)
373 {
374 	ring->qid = qid;
375 	ring->adapter = adapter;
376 	ring->ena_dev = adapter->ena_dev;
377 	atomic_store_8(&ring->first_interrupt, 0);
378 	ring->no_interrupt_event_cnt = 0;
379 }
380 
381 static void
382 ena_init_io_rings_basic(struct ena_adapter *adapter)
383 {
384 	struct ena_com_dev *ena_dev;
385 	struct ena_ring *txr, *rxr;
386 	struct ena_que *que;
387 	int i;
388 
389 	ena_dev = adapter->ena_dev;
390 
391 	for (i = 0; i < adapter->num_io_queues; i++) {
392 		txr = &adapter->tx_ring[i];
393 		rxr = &adapter->rx_ring[i];
394 
395 		/* TX/RX common ring state */
396 		ena_init_io_rings_common(adapter, txr, i);
397 		ena_init_io_rings_common(adapter, rxr, i);
398 
399 		/* TX specific ring state */
400 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
401 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
402 
403 		que = &adapter->que[i];
404 		que->adapter = adapter;
405 		que->id = i;
406 		que->tx_ring = txr;
407 		que->rx_ring = rxr;
408 
409 		txr->que = que;
410 		rxr->que = que;
411 
412 		rxr->empty_rx_queue = 0;
413 		rxr->rx_mbuf_sz = ena_mbuf_sz;
414 	}
415 }
416 
417 static void
418 ena_init_io_rings_advanced(struct ena_adapter *adapter)
419 {
420 	struct ena_ring *txr, *rxr;
421 	int i;
422 
423 	for (i = 0; i < adapter->num_io_queues; i++) {
424 		txr = &adapter->tx_ring[i];
425 		rxr = &adapter->rx_ring[i];
426 
427 		/* Allocate a buf ring */
428 		txr->buf_ring_size = adapter->buf_ring_size;
429 		txr->br = buf_ring_alloc(txr->buf_ring_size, M_DEVBUF, M_WAITOK,
430 		    &txr->ring_mtx);
431 
432 		/* Allocate Tx statistics. */
433 		ena_alloc_counters((counter_u64_t *)&txr->tx_stats,
434 		    sizeof(txr->tx_stats));
435 		txr->tx_last_cleanup_ticks = ticks;
436 
437 		/* Allocate Rx statistics. */
438 		ena_alloc_counters((counter_u64_t *)&rxr->rx_stats,
439 		    sizeof(rxr->rx_stats));
440 
441 		/* Initialize locks */
442 		snprintf(txr->mtx_name, nitems(txr->mtx_name), "%s:tx(%d)",
443 		    device_get_nameunit(adapter->pdev), i);
444 		snprintf(rxr->mtx_name, nitems(rxr->mtx_name), "%s:rx(%d)",
445 		    device_get_nameunit(adapter->pdev), i);
446 
447 		mtx_init(&txr->ring_mtx, txr->mtx_name, NULL, MTX_DEF);
448 	}
449 }
450 
451 static void
452 ena_init_io_rings(struct ena_adapter *adapter)
453 {
454 	/*
455 	 * IO rings initialization can be divided into the 2 steps:
456 	 *   1. Initialize variables and fields with initial values and copy
457 	 *      them from adapter/ena_dev (basic)
458 	 *   2. Allocate mutex, counters and buf_ring (advanced)
459 	 */
460 	ena_init_io_rings_basic(adapter);
461 	ena_init_io_rings_advanced(adapter);
462 }
463 
464 static void
465 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
466 {
467 	struct ena_ring *txr = &adapter->tx_ring[qid];
468 	struct ena_ring *rxr = &adapter->rx_ring[qid];
469 
470 	ena_free_counters((counter_u64_t *)&txr->tx_stats,
471 	    sizeof(txr->tx_stats));
472 	ena_free_counters((counter_u64_t *)&rxr->rx_stats,
473 	    sizeof(rxr->rx_stats));
474 
475 	ENA_RING_MTX_LOCK(txr);
476 	drbr_free(txr->br, M_DEVBUF);
477 	ENA_RING_MTX_UNLOCK(txr);
478 
479 	mtx_destroy(&txr->ring_mtx);
480 }
481 
482 static void
483 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
484 {
485 	int i;
486 
487 	for (i = 0; i < adapter->num_io_queues; i++)
488 		ena_free_io_ring_resources(adapter, i);
489 }
490 
491 static int
492 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
493 {
494 	int ret;
495 
496 	/* Create DMA tag for Tx buffers */
497 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
498 	    1, 0,				  /* alignment, bounds 	     */
499 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
500 	    BUS_SPACE_MAXADDR,			  /* highaddr of excl window */
501 	    NULL, NULL,				  /* filter, filterarg 	     */
502 	    ENA_TSO_MAXSIZE,			  /* maxsize 		     */
503 	    adapter->max_tx_sgl_size - 1,	  /* nsegments 		     */
504 	    ENA_TSO_MAXSIZE,			  /* maxsegsize 	     */
505 	    0,					  /* flags 		     */
506 	    NULL,				  /* lockfunc 		     */
507 	    NULL,				  /* lockfuncarg 	     */
508 	    &adapter->tx_buf_tag);
509 
510 	return (ret);
511 }
512 
513 static int
514 ena_free_tx_dma_tag(struct ena_adapter *adapter)
515 {
516 	int ret;
517 
518 	ret = bus_dma_tag_destroy(adapter->tx_buf_tag);
519 
520 	if (likely(ret == 0))
521 		adapter->tx_buf_tag = NULL;
522 
523 	return (ret);
524 }
525 
526 static int
527 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
528 {
529 	int ret;
530 
531 	/* Create DMA tag for Rx buffers*/
532 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent   */
533 	    1, 0,				  /* alignment, bounds 	     */
534 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
535 	    BUS_SPACE_MAXADDR,			  /* highaddr of excl window */
536 	    NULL, NULL,				  /* filter, filterarg 	     */
537 	    ena_mbuf_sz,			  /* maxsize 		     */
538 	    adapter->max_rx_sgl_size,		  /* nsegments 		     */
539 	    ena_mbuf_sz,			  /* maxsegsize 	     */
540 	    0,					  /* flags 		     */
541 	    NULL,				  /* lockfunc 		     */
542 	    NULL,				  /* lockarg 		     */
543 	    &adapter->rx_buf_tag);
544 
545 	return (ret);
546 }
547 
548 static int
549 ena_free_rx_dma_tag(struct ena_adapter *adapter)
550 {
551 	int ret;
552 
553 	ret = bus_dma_tag_destroy(adapter->rx_buf_tag);
554 
555 	if (likely(ret == 0))
556 		adapter->rx_buf_tag = NULL;
557 
558 	return (ret);
559 }
560 
561 static void
562 ena_release_all_tx_dmamap(struct ena_ring *tx_ring)
563 {
564 	struct ena_adapter *adapter = tx_ring->adapter;
565 	struct ena_tx_buffer *tx_info;
566 	bus_dma_tag_t tx_tag = adapter->tx_buf_tag;
567 	int i;
568 #ifdef DEV_NETMAP
569 	struct ena_netmap_tx_info *nm_info;
570 	int j;
571 #endif /* DEV_NETMAP */
572 
573 	for (i = 0; i < tx_ring->ring_size; ++i) {
574 		tx_info = &tx_ring->tx_buffer_info[i];
575 #ifdef DEV_NETMAP
576 		if (if_getcapenable(adapter->ifp) & IFCAP_NETMAP) {
577 			nm_info = &tx_info->nm_info;
578 			for (j = 0; j < ENA_PKT_MAX_BUFS; ++j) {
579 				if (nm_info->map_seg[j] != NULL) {
580 					bus_dmamap_destroy(tx_tag,
581 					    nm_info->map_seg[j]);
582 					nm_info->map_seg[j] = NULL;
583 				}
584 			}
585 		}
586 #endif /* DEV_NETMAP */
587 		if (tx_info->dmamap != NULL) {
588 			bus_dmamap_destroy(tx_tag, tx_info->dmamap);
589 			tx_info->dmamap = NULL;
590 		}
591 	}
592 }
593 
594 /**
595  * ena_setup_tx_resources - allocate Tx resources (Descriptors)
596  * @adapter: network interface device structure
597  * @qid: queue index
598  *
599  * Returns 0 on success, otherwise on failure.
600  **/
601 static int
602 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
603 {
604 	device_t pdev = adapter->pdev;
605 	char thread_name[MAXCOMLEN + 1];
606 	struct ena_que *que = &adapter->que[qid];
607 	struct ena_ring *tx_ring = que->tx_ring;
608 	cpuset_t *cpu_mask = NULL;
609 	int size, i, err;
610 #ifdef DEV_NETMAP
611 	bus_dmamap_t *map;
612 	int j;
613 
614 	ena_netmap_reset_tx_ring(adapter, qid);
615 #endif /* DEV_NETMAP */
616 
617 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
618 
619 	tx_ring->tx_buffer_info = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
620 	if (unlikely(tx_ring->tx_buffer_info == NULL))
621 		return (ENOMEM);
622 
623 	size = sizeof(uint16_t) * tx_ring->ring_size;
624 	tx_ring->free_tx_ids = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
625 	if (unlikely(tx_ring->free_tx_ids == NULL))
626 		goto err_buf_info_free;
627 
628 	size = tx_ring->tx_max_header_size;
629 	tx_ring->push_buf_intermediate_buf = malloc(size, M_DEVBUF,
630 	    M_NOWAIT | M_ZERO);
631 	if (unlikely(tx_ring->push_buf_intermediate_buf == NULL))
632 		goto err_tx_ids_free;
633 
634 	/* Req id stack for TX OOO completions */
635 	for (i = 0; i < tx_ring->ring_size; i++)
636 		tx_ring->free_tx_ids[i] = i;
637 
638 	/* Reset TX statistics. */
639 	ena_reset_counters((counter_u64_t *)&tx_ring->tx_stats,
640 	    sizeof(tx_ring->tx_stats));
641 
642 	tx_ring->next_to_use = 0;
643 	tx_ring->next_to_clean = 0;
644 	tx_ring->acum_pkts = 0;
645 
646 	/* Make sure that drbr is empty */
647 	ENA_RING_MTX_LOCK(tx_ring);
648 	drbr_flush(adapter->ifp, tx_ring->br);
649 	ENA_RING_MTX_UNLOCK(tx_ring);
650 
651 	/* ... and create the buffer DMA maps */
652 	for (i = 0; i < tx_ring->ring_size; i++) {
653 		err = bus_dmamap_create(adapter->tx_buf_tag, 0,
654 		    &tx_ring->tx_buffer_info[i].dmamap);
655 		if (unlikely(err != 0)) {
656 			ena_log(pdev, ERR,
657 			    "Unable to create Tx DMA map for buffer %d\n", i);
658 			goto err_map_release;
659 		}
660 
661 #ifdef DEV_NETMAP
662 		if (if_getcapenable(adapter->ifp) & IFCAP_NETMAP) {
663 			map = tx_ring->tx_buffer_info[i].nm_info.map_seg;
664 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
665 				err = bus_dmamap_create(adapter->tx_buf_tag, 0,
666 				    &map[j]);
667 				if (unlikely(err != 0)) {
668 					ena_log(pdev, ERR,
669 					    "Unable to create Tx DMA for buffer %d %d\n",
670 					    i, j);
671 					goto err_map_release;
672 				}
673 			}
674 		}
675 #endif /* DEV_NETMAP */
676 	}
677 
678 	/* Allocate taskqueues */
679 	TASK_INIT(&tx_ring->enqueue_task, 0, ena_deferred_mq_start, tx_ring);
680 	tx_ring->enqueue_tq = taskqueue_create_fast("ena_tx_enque", M_NOWAIT,
681 	    taskqueue_thread_enqueue, &tx_ring->enqueue_tq);
682 	if (unlikely(tx_ring->enqueue_tq == NULL)) {
683 		ena_log(pdev, ERR,
684 		    "Unable to create taskqueue for enqueue task\n");
685 		i = tx_ring->ring_size;
686 		goto err_map_release;
687 	}
688 
689 	tx_ring->running = true;
690 
691 #ifdef RSS
692 	cpu_mask = &que->cpu_mask;
693 	snprintf(thread_name, sizeof(thread_name), "%s txeq %d",
694 	    device_get_nameunit(adapter->pdev), que->cpu);
695 #else
696 	snprintf(thread_name, sizeof(thread_name), "%s txeq %d",
697 	    device_get_nameunit(adapter->pdev), que->id);
698 #endif
699 	taskqueue_start_threads_cpuset(&tx_ring->enqueue_tq, 1, PI_NET,
700 	    cpu_mask, "%s", thread_name);
701 
702 	return (0);
703 
704 err_map_release:
705 	ena_release_all_tx_dmamap(tx_ring);
706 err_tx_ids_free:
707 	free(tx_ring->free_tx_ids, M_DEVBUF);
708 	tx_ring->free_tx_ids = NULL;
709 err_buf_info_free:
710 	free(tx_ring->tx_buffer_info, M_DEVBUF);
711 	tx_ring->tx_buffer_info = NULL;
712 
713 	return (ENOMEM);
714 }
715 
716 /**
717  * ena_free_tx_resources - Free Tx Resources per Queue
718  * @adapter: network interface device structure
719  * @qid: queue index
720  *
721  * Free all transmit software resources
722  **/
723 static void
724 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
725 {
726 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
727 #ifdef DEV_NETMAP
728 	struct ena_netmap_tx_info *nm_info;
729 	int j;
730 #endif /* DEV_NETMAP */
731 
732 	while (taskqueue_cancel(tx_ring->enqueue_tq, &tx_ring->enqueue_task, NULL))
733 		taskqueue_drain(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
734 
735 	taskqueue_free(tx_ring->enqueue_tq);
736 
737 	ENA_RING_MTX_LOCK(tx_ring);
738 	/* Flush buffer ring, */
739 	drbr_flush(adapter->ifp, tx_ring->br);
740 
741 	/* Free buffer DMA maps, */
742 	for (int i = 0; i < tx_ring->ring_size; i++) {
743 		bus_dmamap_sync(adapter->tx_buf_tag,
744 		    tx_ring->tx_buffer_info[i].dmamap, BUS_DMASYNC_POSTWRITE);
745 		bus_dmamap_unload(adapter->tx_buf_tag,
746 		    tx_ring->tx_buffer_info[i].dmamap);
747 		bus_dmamap_destroy(adapter->tx_buf_tag,
748 		    tx_ring->tx_buffer_info[i].dmamap);
749 
750 #ifdef DEV_NETMAP
751 		if (if_getcapenable(adapter->ifp) & IFCAP_NETMAP) {
752 			nm_info = &tx_ring->tx_buffer_info[i].nm_info;
753 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
754 				if (nm_info->socket_buf_idx[j] != 0) {
755 					bus_dmamap_sync(adapter->tx_buf_tag,
756 					    nm_info->map_seg[j],
757 					    BUS_DMASYNC_POSTWRITE);
758 					ena_netmap_unload(adapter,
759 					    nm_info->map_seg[j]);
760 				}
761 				bus_dmamap_destroy(adapter->tx_buf_tag,
762 				    nm_info->map_seg[j]);
763 				nm_info->socket_buf_idx[j] = 0;
764 			}
765 		}
766 #endif /* DEV_NETMAP */
767 
768 		m_freem(tx_ring->tx_buffer_info[i].mbuf);
769 		tx_ring->tx_buffer_info[i].mbuf = NULL;
770 	}
771 	ENA_RING_MTX_UNLOCK(tx_ring);
772 
773 	/* And free allocated memory. */
774 	free(tx_ring->tx_buffer_info, M_DEVBUF);
775 	tx_ring->tx_buffer_info = NULL;
776 
777 	free(tx_ring->free_tx_ids, M_DEVBUF);
778 	tx_ring->free_tx_ids = NULL;
779 
780 	free(tx_ring->push_buf_intermediate_buf, M_DEVBUF);
781 	tx_ring->push_buf_intermediate_buf = NULL;
782 }
783 
784 /**
785  * ena_setup_all_tx_resources - allocate all queues Tx resources
786  * @adapter: network interface device structure
787  *
788  * Returns 0 on success, otherwise on failure.
789  **/
790 static int
791 ena_setup_all_tx_resources(struct ena_adapter *adapter)
792 {
793 	int i, rc;
794 
795 	for (i = 0; i < adapter->num_io_queues; i++) {
796 		rc = ena_setup_tx_resources(adapter, i);
797 		if (rc != 0) {
798 			ena_log(adapter->pdev, ERR,
799 			    "Allocation for Tx Queue %u failed\n", i);
800 			goto err_setup_tx;
801 		}
802 	}
803 
804 	return (0);
805 
806 err_setup_tx:
807 	/* Rewind the index freeing the rings as we go */
808 	while (i--)
809 		ena_free_tx_resources(adapter, i);
810 	return (rc);
811 }
812 
813 /**
814  * ena_free_all_tx_resources - Free Tx Resources for All Queues
815  * @adapter: network interface device structure
816  *
817  * Free all transmit software resources
818  **/
819 static void
820 ena_free_all_tx_resources(struct ena_adapter *adapter)
821 {
822 	int i;
823 
824 	for (i = 0; i < adapter->num_io_queues; i++)
825 		ena_free_tx_resources(adapter, i);
826 }
827 
828 /**
829  * ena_setup_rx_resources - allocate Rx resources (Descriptors)
830  * @adapter: network interface device structure
831  * @qid: queue index
832  *
833  * Returns 0 on success, otherwise on failure.
834  **/
835 static int
836 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
837 {
838 	device_t pdev = adapter->pdev;
839 	struct ena_que *que = &adapter->que[qid];
840 	struct ena_ring *rx_ring = que->rx_ring;
841 	int size, err, i;
842 
843 	size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
844 
845 #ifdef DEV_NETMAP
846 	ena_netmap_reset_rx_ring(adapter, qid);
847 	rx_ring->initialized = false;
848 #endif /* DEV_NETMAP */
849 
850 	/*
851 	 * Alloc extra element so in rx path
852 	 * we can always prefetch rx_info + 1
853 	 */
854 	size += sizeof(struct ena_rx_buffer);
855 
856 	rx_ring->rx_buffer_info = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
857 
858 	size = sizeof(uint16_t) * rx_ring->ring_size;
859 	rx_ring->free_rx_ids = malloc(size, M_DEVBUF, M_WAITOK);
860 
861 	for (i = 0; i < rx_ring->ring_size; i++)
862 		rx_ring->free_rx_ids[i] = i;
863 
864 	/* Reset RX statistics. */
865 	ena_reset_counters((counter_u64_t *)&rx_ring->rx_stats,
866 	    sizeof(rx_ring->rx_stats));
867 
868 	rx_ring->next_to_clean = 0;
869 	rx_ring->next_to_use = 0;
870 
871 	/* ... and create the buffer DMA maps */
872 	for (i = 0; i < rx_ring->ring_size; i++) {
873 		err = bus_dmamap_create(adapter->rx_buf_tag, 0,
874 		    &(rx_ring->rx_buffer_info[i].map));
875 		if (err != 0) {
876 			ena_log(pdev, ERR,
877 			    "Unable to create Rx DMA map for buffer %d\n", i);
878 			goto err_buf_info_unmap;
879 		}
880 	}
881 
882 	/* Create LRO for the ring */
883 	if ((if_getcapenable(adapter->ifp) & IFCAP_LRO) != 0) {
884 		int err = tcp_lro_init(&rx_ring->lro);
885 		if (err != 0) {
886 			ena_log(pdev, ERR, "LRO[%d] Initialization failed!\n",
887 			    qid);
888 		} else {
889 			ena_log(pdev, DBG, "RX Soft LRO[%d] Initialized\n",
890 			    qid);
891 			rx_ring->lro.ifp = adapter->ifp;
892 		}
893 	}
894 
895 	return (0);
896 
897 err_buf_info_unmap:
898 	while (i--) {
899 		bus_dmamap_destroy(adapter->rx_buf_tag,
900 		    rx_ring->rx_buffer_info[i].map);
901 	}
902 
903 	free(rx_ring->free_rx_ids, M_DEVBUF);
904 	rx_ring->free_rx_ids = NULL;
905 	free(rx_ring->rx_buffer_info, M_DEVBUF);
906 	rx_ring->rx_buffer_info = NULL;
907 	return (ENOMEM);
908 }
909 
910 /**
911  * ena_free_rx_resources - Free Rx Resources
912  * @adapter: network interface device structure
913  * @qid: queue index
914  *
915  * Free all receive software resources
916  **/
917 static void
918 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
919 {
920 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
921 
922 	/* Free buffer DMA maps, */
923 	for (int i = 0; i < rx_ring->ring_size; i++) {
924 		bus_dmamap_sync(adapter->rx_buf_tag,
925 		    rx_ring->rx_buffer_info[i].map, BUS_DMASYNC_POSTREAD);
926 		m_freem(rx_ring->rx_buffer_info[i].mbuf);
927 		rx_ring->rx_buffer_info[i].mbuf = NULL;
928 		bus_dmamap_unload(adapter->rx_buf_tag,
929 		    rx_ring->rx_buffer_info[i].map);
930 		bus_dmamap_destroy(adapter->rx_buf_tag,
931 		    rx_ring->rx_buffer_info[i].map);
932 	}
933 
934 	/* free LRO resources, */
935 	tcp_lro_free(&rx_ring->lro);
936 
937 	/* free allocated memory */
938 	free(rx_ring->rx_buffer_info, M_DEVBUF);
939 	rx_ring->rx_buffer_info = NULL;
940 
941 	free(rx_ring->free_rx_ids, M_DEVBUF);
942 	rx_ring->free_rx_ids = NULL;
943 }
944 
945 /**
946  * ena_setup_all_rx_resources - allocate all queues Rx resources
947  * @adapter: network interface device structure
948  *
949  * Returns 0 on success, otherwise on failure.
950  **/
951 static int
952 ena_setup_all_rx_resources(struct ena_adapter *adapter)
953 {
954 	int i, rc = 0;
955 
956 	for (i = 0; i < adapter->num_io_queues; i++) {
957 		rc = ena_setup_rx_resources(adapter, i);
958 		if (rc != 0) {
959 			ena_log(adapter->pdev, ERR,
960 			    "Allocation for Rx Queue %u failed\n", i);
961 			goto err_setup_rx;
962 		}
963 	}
964 	return (0);
965 
966 err_setup_rx:
967 	/* rewind the index freeing the rings as we go */
968 	while (i--)
969 		ena_free_rx_resources(adapter, i);
970 	return (rc);
971 }
972 
973 /**
974  * ena_free_all_rx_resources - Free Rx resources for all queues
975  * @adapter: network interface device structure
976  *
977  * Free all receive software resources
978  **/
979 static void
980 ena_free_all_rx_resources(struct ena_adapter *adapter)
981 {
982 	int i;
983 
984 	for (i = 0; i < adapter->num_io_queues; i++)
985 		ena_free_rx_resources(adapter, i);
986 }
987 
988 static inline int
989 ena_alloc_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
990     struct ena_rx_buffer *rx_info)
991 {
992 	device_t pdev = adapter->pdev;
993 	struct ena_com_buf *ena_buf;
994 	bus_dma_segment_t segs[1];
995 	int nsegs, error;
996 	int mlen;
997 
998 	/* if previous allocated frag is not used */
999 	if (unlikely(rx_info->mbuf != NULL))
1000 		return (0);
1001 
1002 	/* Get mbuf using UMA allocator */
1003 	rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR,
1004 	    rx_ring->rx_mbuf_sz);
1005 
1006 	if (unlikely(rx_info->mbuf == NULL)) {
1007 		counter_u64_add(rx_ring->rx_stats.mjum_alloc_fail, 1);
1008 		rx_info->mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1009 		if (unlikely(rx_info->mbuf == NULL)) {
1010 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
1011 			return (ENOMEM);
1012 		}
1013 		mlen = MCLBYTES;
1014 	} else {
1015 		mlen = rx_ring->rx_mbuf_sz;
1016 	}
1017 	/* Set mbuf length*/
1018 	rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = mlen;
1019 
1020 	/* Map packets for DMA */
1021 	ena_log(pdev, DBG,
1022 	    "Using tag %p for buffers' DMA mapping, mbuf %p len: %d\n",
1023 	    adapter->rx_buf_tag, rx_info->mbuf, rx_info->mbuf->m_len);
1024 	error = bus_dmamap_load_mbuf_sg(adapter->rx_buf_tag, rx_info->map,
1025 	    rx_info->mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
1026 	if (unlikely((error != 0) || (nsegs != 1))) {
1027 		ena_log(pdev, WARN,
1028 		    "failed to map mbuf, error: %d, nsegs: %d\n", error, nsegs);
1029 		counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
1030 		goto exit;
1031 	}
1032 
1033 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map, BUS_DMASYNC_PREREAD);
1034 
1035 	ena_buf = &rx_info->ena_buf;
1036 	ena_buf->paddr = segs[0].ds_addr;
1037 	ena_buf->len = mlen;
1038 
1039 	ena_log(pdev, DBG,
1040 	    "ALLOC RX BUF: mbuf %p, rx_info %p, len %d, paddr %#jx\n",
1041 	    rx_info->mbuf, rx_info, ena_buf->len, (uintmax_t)ena_buf->paddr);
1042 
1043 	return (0);
1044 
1045 exit:
1046 	m_freem(rx_info->mbuf);
1047 	rx_info->mbuf = NULL;
1048 	return (EFAULT);
1049 }
1050 
1051 static void
1052 ena_free_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
1053     struct ena_rx_buffer *rx_info)
1054 {
1055 	if (rx_info->mbuf == NULL) {
1056 		ena_log(adapter->pdev, WARN,
1057 		    "Trying to free unallocated buffer\n");
1058 		return;
1059 	}
1060 
1061 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map,
1062 	    BUS_DMASYNC_POSTREAD);
1063 	bus_dmamap_unload(adapter->rx_buf_tag, rx_info->map);
1064 	m_freem(rx_info->mbuf);
1065 	rx_info->mbuf = NULL;
1066 }
1067 
1068 /**
1069  * ena_refill_rx_bufs - Refills ring with descriptors
1070  * @rx_ring: the ring which we want to feed with free descriptors
1071  * @num: number of descriptors to refill
1072  * Refills the ring with newly allocated DMA-mapped mbufs for receiving
1073  **/
1074 int
1075 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
1076 {
1077 	struct ena_adapter *adapter = rx_ring->adapter;
1078 	device_t pdev = adapter->pdev;
1079 	uint16_t next_to_use, req_id;
1080 	uint32_t i;
1081 	int rc;
1082 
1083 	ena_log_io(adapter->pdev, DBG, "refill qid: %d\n", rx_ring->qid);
1084 
1085 	next_to_use = rx_ring->next_to_use;
1086 
1087 	for (i = 0; i < num; i++) {
1088 		struct ena_rx_buffer *rx_info;
1089 
1090 		ena_log_io(pdev, DBG, "RX buffer - next to use: %d\n",
1091 		    next_to_use);
1092 
1093 		req_id = rx_ring->free_rx_ids[next_to_use];
1094 		rx_info = &rx_ring->rx_buffer_info[req_id];
1095 #ifdef DEV_NETMAP
1096 		if (ena_rx_ring_in_netmap(adapter, rx_ring->qid))
1097 			rc = ena_netmap_alloc_rx_slot(adapter, rx_ring,
1098 			    rx_info);
1099 		else
1100 #endif /* DEV_NETMAP */
1101 			rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1102 		if (unlikely(rc != 0)) {
1103 			ena_log_io(pdev, WARN,
1104 			    "failed to alloc buffer for rx queue %d\n",
1105 			    rx_ring->qid);
1106 			break;
1107 		}
1108 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1109 		    &rx_info->ena_buf, req_id);
1110 		if (unlikely(rc != 0)) {
1111 			ena_log_io(pdev, WARN,
1112 			    "failed to add buffer for rx queue %d\n",
1113 			    rx_ring->qid);
1114 			break;
1115 		}
1116 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1117 		    rx_ring->ring_size);
1118 	}
1119 
1120 	if (unlikely(i < num)) {
1121 		counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1122 		ena_log_io(pdev, WARN,
1123 		    "refilled rx qid %d with only %d mbufs (from %d)\n",
1124 		    rx_ring->qid, i, num);
1125 	}
1126 
1127 	if (likely(i != 0))
1128 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1129 
1130 	rx_ring->next_to_use = next_to_use;
1131 	return (i);
1132 }
1133 
1134 int
1135 ena_update_buf_ring_size(struct ena_adapter *adapter,
1136     uint32_t new_buf_ring_size)
1137 {
1138 	uint32_t old_buf_ring_size;
1139 	int rc = 0;
1140 	bool dev_was_up;
1141 
1142 	old_buf_ring_size = adapter->buf_ring_size;
1143 	adapter->buf_ring_size = new_buf_ring_size;
1144 
1145 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1146 	ena_down(adapter);
1147 
1148 	/* Reconfigure buf ring for all Tx rings. */
1149 	ena_free_all_io_rings_resources(adapter);
1150 	ena_init_io_rings_advanced(adapter);
1151 	if (dev_was_up) {
1152 		/*
1153 		 * If ena_up() fails, it's not because of recent buf_ring size
1154 		 * changes. Because of that, we just want to revert old drbr
1155 		 * value and trigger the reset because something else had to
1156 		 * go wrong.
1157 		 */
1158 		rc = ena_up(adapter);
1159 		if (unlikely(rc != 0)) {
1160 			ena_log(adapter->pdev, ERR,
1161 			    "Failed to configure device after setting new drbr size: %u. Reverting old value: %u and triggering the reset\n",
1162 			    new_buf_ring_size, old_buf_ring_size);
1163 
1164 			/* Revert old size and trigger the reset */
1165 			adapter->buf_ring_size = old_buf_ring_size;
1166 			ena_free_all_io_rings_resources(adapter);
1167 			ena_init_io_rings_advanced(adapter);
1168 
1169 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET,
1170 			    adapter);
1171 			ena_trigger_reset(adapter, ENA_REGS_RESET_OS_TRIGGER);
1172 		}
1173 	}
1174 
1175 	return (rc);
1176 }
1177 
1178 int
1179 ena_update_queue_size(struct ena_adapter *adapter, uint32_t new_tx_size,
1180     uint32_t new_rx_size)
1181 {
1182 	uint32_t old_tx_size, old_rx_size;
1183 	int rc = 0;
1184 	bool dev_was_up;
1185 
1186 	old_tx_size = adapter->requested_tx_ring_size;
1187 	old_rx_size = adapter->requested_rx_ring_size;
1188 	adapter->requested_tx_ring_size = new_tx_size;
1189 	adapter->requested_rx_ring_size = new_rx_size;
1190 
1191 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1192 	ena_down(adapter);
1193 
1194 	/* Configure queues with new size. */
1195 	ena_init_io_rings_basic(adapter);
1196 	if (dev_was_up) {
1197 		rc = ena_up(adapter);
1198 		if (unlikely(rc != 0)) {
1199 			ena_log(adapter->pdev, ERR,
1200 			    "Failed to configure device with the new sizes - Tx: %u Rx: %u. Reverting old values - Tx: %u Rx: %u\n",
1201 			    new_tx_size, new_rx_size, old_tx_size, old_rx_size);
1202 
1203 			/* Revert old size. */
1204 			adapter->requested_tx_ring_size = old_tx_size;
1205 			adapter->requested_rx_ring_size = old_rx_size;
1206 			ena_init_io_rings_basic(adapter);
1207 
1208 			/* And try again. */
1209 			rc = ena_up(adapter);
1210 			if (unlikely(rc != 0)) {
1211 				ena_log(adapter->pdev, ERR,
1212 				    "Failed to revert old queue sizes. Triggering device reset.\n");
1213 				/*
1214 				 * If we've failed again, something had to go
1215 				 * wrong. After reset, the device should try to
1216 				 * go up
1217 				 */
1218 				ENA_FLAG_SET_ATOMIC(
1219 				    ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
1220 				ena_trigger_reset(adapter,
1221 				    ENA_REGS_RESET_OS_TRIGGER);
1222 			}
1223 		}
1224 	}
1225 
1226 	return (rc);
1227 }
1228 
1229 static void
1230 ena_update_io_rings(struct ena_adapter *adapter, uint32_t num)
1231 {
1232 	ena_free_all_io_rings_resources(adapter);
1233 	/* Force indirection table to be reinitialized */
1234 	ena_com_rss_destroy(adapter->ena_dev);
1235 
1236 	adapter->num_io_queues = num;
1237 	ena_init_io_rings(adapter);
1238 }
1239 
1240 /* Caller should sanitize new_num */
1241 int
1242 ena_update_io_queue_nb(struct ena_adapter *adapter, uint32_t new_num)
1243 {
1244 	uint32_t old_num;
1245 	int rc = 0;
1246 	bool dev_was_up;
1247 
1248 	dev_was_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
1249 	old_num = adapter->num_io_queues;
1250 	ena_down(adapter);
1251 
1252 	ena_update_io_rings(adapter, new_num);
1253 
1254 	if (dev_was_up) {
1255 		rc = ena_up(adapter);
1256 		if (unlikely(rc != 0)) {
1257 			ena_log(adapter->pdev, ERR,
1258 			    "Failed to configure device with %u IO queues. "
1259 			    "Reverting to previous value: %u\n",
1260 			    new_num, old_num);
1261 
1262 			ena_update_io_rings(adapter, old_num);
1263 
1264 			rc = ena_up(adapter);
1265 			if (unlikely(rc != 0)) {
1266 				ena_log(adapter->pdev, ERR,
1267 				    "Failed to revert to previous setup IO "
1268 				    "queues. Triggering device reset.\n");
1269 				ENA_FLAG_SET_ATOMIC(
1270 				    ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
1271 				ena_trigger_reset(adapter,
1272 				    ENA_REGS_RESET_OS_TRIGGER);
1273 			}
1274 		}
1275 	}
1276 
1277 	return (rc);
1278 }
1279 
1280 static void
1281 ena_free_rx_bufs(struct ena_adapter *adapter, unsigned int qid)
1282 {
1283 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1284 	unsigned int i;
1285 
1286 	for (i = 0; i < rx_ring->ring_size; i++) {
1287 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1288 
1289 		if (rx_info->mbuf != NULL)
1290 			ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1291 #ifdef DEV_NETMAP
1292 		if (((if_getflags(adapter->ifp) & IFF_DYING) == 0) &&
1293 		    (if_getcapenable(adapter->ifp) & IFCAP_NETMAP)) {
1294 			if (rx_info->netmap_buf_idx != 0)
1295 				ena_netmap_free_rx_slot(adapter, rx_ring,
1296 				    rx_info);
1297 		}
1298 #endif /* DEV_NETMAP */
1299 	}
1300 }
1301 
1302 /**
1303  * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1304  * @adapter: network interface device structure
1305  *
1306  */
1307 static void
1308 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1309 {
1310 	struct ena_ring *rx_ring;
1311 	int i, rc, bufs_num;
1312 
1313 	for (i = 0; i < adapter->num_io_queues; i++) {
1314 		rx_ring = &adapter->rx_ring[i];
1315 		bufs_num = rx_ring->ring_size - 1;
1316 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1317 		if (unlikely(rc != bufs_num))
1318 			ena_log_io(adapter->pdev, WARN,
1319 			    "refilling Queue %d failed. "
1320 			    "Allocated %d buffers from: %d\n",
1321 			    i, rc, bufs_num);
1322 #ifdef DEV_NETMAP
1323 		rx_ring->initialized = true;
1324 #endif /* DEV_NETMAP */
1325 	}
1326 }
1327 
1328 static void
1329 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1330 {
1331 	int i;
1332 
1333 	for (i = 0; i < adapter->num_io_queues; i++)
1334 		ena_free_rx_bufs(adapter, i);
1335 }
1336 
1337 /**
1338  * ena_free_tx_bufs - Free Tx Buffers per Queue
1339  * @adapter: network interface device structure
1340  * @qid: queue index
1341  **/
1342 static void
1343 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1344 {
1345 	bool print_once = true;
1346 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1347 
1348 	ENA_RING_MTX_LOCK(tx_ring);
1349 	for (int i = 0; i < tx_ring->ring_size; i++) {
1350 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1351 
1352 		if (tx_info->mbuf == NULL)
1353 			continue;
1354 
1355 		if (print_once) {
1356 			ena_log(adapter->pdev, WARN,
1357 			    "free uncompleted tx mbuf qid %d idx 0x%x\n", qid,
1358 			    i);
1359 			print_once = false;
1360 		} else {
1361 			ena_log(adapter->pdev, DBG,
1362 			    "free uncompleted tx mbuf qid %d idx 0x%x\n", qid,
1363 			    i);
1364 		}
1365 
1366 		bus_dmamap_sync(adapter->tx_buf_tag, tx_info->dmamap,
1367 		    BUS_DMASYNC_POSTWRITE);
1368 		bus_dmamap_unload(adapter->tx_buf_tag, tx_info->dmamap);
1369 
1370 		m_free(tx_info->mbuf);
1371 		tx_info->mbuf = NULL;
1372 	}
1373 	ENA_RING_MTX_UNLOCK(tx_ring);
1374 }
1375 
1376 static void
1377 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1378 {
1379 	for (int i = 0; i < adapter->num_io_queues; i++)
1380 		ena_free_tx_bufs(adapter, i);
1381 }
1382 
1383 static void
1384 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1385 {
1386 	uint16_t ena_qid;
1387 	int i;
1388 
1389 	for (i = 0; i < adapter->num_io_queues; i++) {
1390 		ena_qid = ENA_IO_TXQ_IDX(i);
1391 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1392 	}
1393 }
1394 
1395 static void
1396 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1397 {
1398 	uint16_t ena_qid;
1399 	int i;
1400 
1401 	for (i = 0; i < adapter->num_io_queues; i++) {
1402 		ena_qid = ENA_IO_RXQ_IDX(i);
1403 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1404 	}
1405 }
1406 
1407 static void
1408 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1409 {
1410 	struct ena_que *queue;
1411 	int i;
1412 
1413 	for (i = 0; i < adapter->num_io_queues; i++) {
1414 		queue = &adapter->que[i];
1415 		while (taskqueue_cancel(queue->cleanup_tq, &queue->cleanup_task, NULL))
1416 			taskqueue_drain(queue->cleanup_tq, &queue->cleanup_task);
1417 		taskqueue_free(queue->cleanup_tq);
1418 	}
1419 
1420 	ena_destroy_all_tx_queues(adapter);
1421 	ena_destroy_all_rx_queues(adapter);
1422 }
1423 
1424 static int
1425 ena_create_io_queues(struct ena_adapter *adapter)
1426 {
1427 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1428 	struct ena_com_create_io_ctx ctx;
1429 	struct ena_ring *ring;
1430 	struct ena_que *queue;
1431 	uint16_t ena_qid;
1432 	uint32_t msix_vector;
1433 	cpuset_t *cpu_mask = NULL;
1434 	int rc, i;
1435 
1436 	/* Create TX queues */
1437 	for (i = 0; i < adapter->num_io_queues; i++) {
1438 		msix_vector = ENA_IO_IRQ_IDX(i);
1439 		ena_qid = ENA_IO_TXQ_IDX(i);
1440 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1441 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1442 		ctx.queue_size = adapter->requested_tx_ring_size;
1443 		ctx.msix_vector = msix_vector;
1444 		ctx.qid = ena_qid;
1445 		ctx.numa_node = adapter->que[i].domain;
1446 
1447 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1448 		if (rc != 0) {
1449 			ena_log(adapter->pdev, ERR,
1450 			    "Failed to create io TX queue #%d rc: %d\n", i, rc);
1451 			goto err_tx;
1452 		}
1453 		ring = &adapter->tx_ring[i];
1454 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1455 		    &ring->ena_com_io_sq, &ring->ena_com_io_cq);
1456 		if (rc != 0) {
1457 			ena_log(adapter->pdev, ERR,
1458 			    "Failed to get TX queue handlers. TX queue num"
1459 			    " %d rc: %d\n",
1460 			    i, rc);
1461 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1462 			goto err_tx;
1463 		}
1464 
1465 		if (ctx.numa_node >= 0) {
1466 			ena_com_update_numa_node(ring->ena_com_io_cq,
1467 			    ctx.numa_node);
1468 		}
1469 	}
1470 
1471 	/* Create RX queues */
1472 	for (i = 0; i < adapter->num_io_queues; i++) {
1473 		msix_vector = ENA_IO_IRQ_IDX(i);
1474 		ena_qid = ENA_IO_RXQ_IDX(i);
1475 		ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1476 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1477 		ctx.queue_size = adapter->requested_rx_ring_size;
1478 		ctx.msix_vector = msix_vector;
1479 		ctx.qid = ena_qid;
1480 		ctx.numa_node = adapter->que[i].domain;
1481 
1482 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1483 		if (unlikely(rc != 0)) {
1484 			ena_log(adapter->pdev, ERR,
1485 			    "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1486 			goto err_rx;
1487 		}
1488 
1489 		ring = &adapter->rx_ring[i];
1490 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1491 		    &ring->ena_com_io_sq, &ring->ena_com_io_cq);
1492 		if (unlikely(rc != 0)) {
1493 			ena_log(adapter->pdev, ERR,
1494 			    "Failed to get RX queue handlers. RX queue num"
1495 			    " %d rc: %d\n",
1496 			    i, rc);
1497 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1498 			goto err_rx;
1499 		}
1500 
1501 		if (ctx.numa_node >= 0) {
1502 			ena_com_update_numa_node(ring->ena_com_io_cq,
1503 			    ctx.numa_node);
1504 		}
1505 	}
1506 
1507 	for (i = 0; i < adapter->num_io_queues; i++) {
1508 		queue = &adapter->que[i];
1509 
1510 		NET_TASK_INIT(&queue->cleanup_task, 0, ena_cleanup, queue);
1511 		queue->cleanup_tq = taskqueue_create_fast("ena cleanup",
1512 		    M_WAITOK, taskqueue_thread_enqueue, &queue->cleanup_tq);
1513 
1514 #ifdef RSS
1515 		cpu_mask = &queue->cpu_mask;
1516 #endif
1517 		taskqueue_start_threads_cpuset(&queue->cleanup_tq, 1, PI_NET,
1518 		    cpu_mask, "%s queue %d cleanup",
1519 		    device_get_nameunit(adapter->pdev), i);
1520 	}
1521 
1522 	return (0);
1523 
1524 err_rx:
1525 	while (i--)
1526 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1527 	i = adapter->num_io_queues;
1528 err_tx:
1529 	while (i--)
1530 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1531 
1532 	return (ENXIO);
1533 }
1534 
1535 /*********************************************************************
1536  *
1537  *  MSIX & Interrupt Service routine
1538  *
1539  **********************************************************************/
1540 
1541 /**
1542  * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1543  * @arg: interrupt number
1544  **/
1545 static void
1546 ena_intr_msix_mgmnt(void *arg)
1547 {
1548 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1549 
1550 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1551 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter)))
1552 		ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1553 }
1554 
1555 /**
1556  * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1557  * @arg: queue
1558  **/
1559 static int
1560 ena_handle_msix(void *arg)
1561 {
1562 	struct ena_que *queue = arg;
1563 	struct ena_adapter *adapter = queue->adapter;
1564 	if_t ifp = adapter->ifp;
1565 
1566 	if (unlikely((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0))
1567 		return (FILTER_STRAY);
1568 
1569 	taskqueue_enqueue(queue->cleanup_tq, &queue->cleanup_task);
1570 
1571 	return (FILTER_HANDLED);
1572 }
1573 
1574 static int
1575 ena_enable_msix(struct ena_adapter *adapter)
1576 {
1577 	device_t dev = adapter->pdev;
1578 	int msix_vecs, msix_req;
1579 	int i, rc = 0;
1580 
1581 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1582 		ena_log(dev, ERR, "Error, MSI-X is already enabled\n");
1583 		return (EINVAL);
1584 	}
1585 
1586 	/* Reserved the max msix vectors we might need */
1587 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
1588 
1589 	adapter->msix_entries = malloc(msix_vecs * sizeof(struct msix_entry),
1590 	    M_DEVBUF, M_WAITOK | M_ZERO);
1591 
1592 	ena_log(dev, DBG, "trying to enable MSI-X, vectors: %d\n", msix_vecs);
1593 
1594 	for (i = 0; i < msix_vecs; i++) {
1595 		adapter->msix_entries[i].entry = i;
1596 		/* Vectors must start from 1 */
1597 		adapter->msix_entries[i].vector = i + 1;
1598 	}
1599 
1600 	msix_req = msix_vecs;
1601 	rc = pci_alloc_msix(dev, &msix_vecs);
1602 	if (unlikely(rc != 0)) {
1603 		ena_log(dev, ERR, "Failed to enable MSIX, vectors %d rc %d\n",
1604 		    msix_vecs, rc);
1605 
1606 		rc = ENOSPC;
1607 		goto err_msix_free;
1608 	}
1609 
1610 	if (msix_vecs != msix_req) {
1611 		if (msix_vecs == ENA_ADMIN_MSIX_VEC) {
1612 			ena_log(dev, ERR,
1613 			    "Not enough number of MSI-x allocated: %d\n",
1614 			    msix_vecs);
1615 			pci_release_msi(dev);
1616 			rc = ENOSPC;
1617 			goto err_msix_free;
1618 		}
1619 		ena_log(dev, ERR,
1620 		    "Enable only %d MSI-x (out of %d), reduce "
1621 		    "the number of queues\n",
1622 		    msix_vecs, msix_req);
1623 	}
1624 
1625 	adapter->msix_vecs = msix_vecs;
1626 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1627 
1628 	return (0);
1629 
1630 err_msix_free:
1631 	free(adapter->msix_entries, M_DEVBUF);
1632 	adapter->msix_entries = NULL;
1633 
1634 	return (rc);
1635 }
1636 
1637 static void
1638 ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1639 {
1640 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name, ENA_IRQNAME_SIZE,
1641 	    "ena-mgmnt@pci:%s", device_get_nameunit(adapter->pdev));
1642 	/*
1643 	 * Handler is NULL on purpose, it will be set
1644 	 * when mgmnt interrupt is acquired
1645 	 */
1646 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler = NULL;
1647 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1648 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1649 	    adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
1650 }
1651 
1652 static int
1653 ena_setup_io_intr(struct ena_adapter *adapter)
1654 {
1655 #ifdef RSS
1656 	int num_buckets = rss_getnumbuckets();
1657 	static int last_bind = 0;
1658 	int cur_bind;
1659 	int idx;
1660 #endif
1661 	int irq_idx;
1662 
1663 	if (adapter->msix_entries == NULL)
1664 		return (EINVAL);
1665 
1666 #ifdef RSS
1667 	if (adapter->first_bind < 0) {
1668 		adapter->first_bind = last_bind;
1669 		last_bind = (last_bind + adapter->num_io_queues) % num_buckets;
1670 	}
1671 	cur_bind = adapter->first_bind;
1672 #endif
1673 
1674 	for (int i = 0; i < adapter->num_io_queues; i++) {
1675 		irq_idx = ENA_IO_IRQ_IDX(i);
1676 
1677 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1678 		    "%s-TxRx-%d", device_get_nameunit(adapter->pdev), i);
1679 		adapter->irq_tbl[irq_idx].handler = ena_handle_msix;
1680 		adapter->irq_tbl[irq_idx].data = &adapter->que[i];
1681 		adapter->irq_tbl[irq_idx].vector =
1682 		    adapter->msix_entries[irq_idx].vector;
1683 		ena_log(adapter->pdev, DBG, "ena_setup_io_intr vector: %d\n",
1684 		    adapter->msix_entries[irq_idx].vector);
1685 
1686 #ifdef RSS
1687 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1688 		    rss_getcpu(cur_bind);
1689 		cur_bind = (cur_bind + 1) % num_buckets;
1690 		CPU_SETOF(adapter->que[i].cpu, &adapter->que[i].cpu_mask);
1691 
1692 		for (idx = 0; idx < MAXMEMDOM; ++idx) {
1693 			if (CPU_ISSET(adapter->que[i].cpu, &cpuset_domain[idx]))
1694 				break;
1695 		}
1696 		adapter->que[i].domain = idx;
1697 #else
1698 		adapter->que[i].domain = -1;
1699 #endif
1700 	}
1701 
1702 	return (0);
1703 }
1704 
1705 static int
1706 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1707 {
1708 	device_t pdev = adapter->pdev;
1709 	struct ena_irq *irq;
1710 	unsigned long flags;
1711 	int rc, rcc;
1712 
1713 	flags = RF_ACTIVE | RF_SHAREABLE;
1714 
1715 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1716 	irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1717 	    &irq->vector, flags);
1718 
1719 	if (unlikely(irq->res == NULL)) {
1720 		ena_log(pdev, ERR, "could not allocate irq vector: %d\n",
1721 		    irq->vector);
1722 		return (ENXIO);
1723 	}
1724 
1725 	rc = bus_setup_intr(adapter->pdev, irq->res,
1726 	    INTR_TYPE_NET | INTR_MPSAFE, NULL, ena_intr_msix_mgmnt, irq->data,
1727 	    &irq->cookie);
1728 	if (unlikely(rc != 0)) {
1729 		ena_log(pdev, ERR,
1730 		    "failed to register interrupt handler for irq %ju: %d\n",
1731 		    rman_get_start(irq->res), rc);
1732 		goto err_res_free;
1733 	}
1734 	irq->requested = true;
1735 
1736 	return (rc);
1737 
1738 err_res_free:
1739 	ena_log(pdev, INFO, "releasing resource for irq %d\n", irq->vector);
1740 	rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ, irq->vector,
1741 	    irq->res);
1742 	if (unlikely(rcc != 0))
1743 		ena_log(pdev, ERR,
1744 		    "dev has no parent while releasing res for irq: %d\n",
1745 		    irq->vector);
1746 	irq->res = NULL;
1747 
1748 	return (rc);
1749 }
1750 
1751 static int
1752 ena_request_io_irq(struct ena_adapter *adapter)
1753 {
1754 	device_t pdev = adapter->pdev;
1755 	struct ena_irq *irq;
1756 	unsigned long flags = 0;
1757 	int rc = 0, i, rcc;
1758 
1759 	if (unlikely(!ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter))) {
1760 		ena_log(pdev, ERR,
1761 		    "failed to request I/O IRQ: MSI-X is not enabled\n");
1762 		return (EINVAL);
1763 	} else {
1764 		flags = RF_ACTIVE | RF_SHAREABLE;
1765 	}
1766 
1767 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1768 		irq = &adapter->irq_tbl[i];
1769 
1770 		if (unlikely(irq->requested))
1771 			continue;
1772 
1773 		irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1774 		    &irq->vector, flags);
1775 		if (unlikely(irq->res == NULL)) {
1776 			rc = ENOMEM;
1777 			ena_log(pdev, ERR,
1778 			    "could not allocate irq vector: %d\n", irq->vector);
1779 			goto err;
1780 		}
1781 
1782 		rc = bus_setup_intr(adapter->pdev, irq->res,
1783 		    INTR_TYPE_NET | INTR_MPSAFE, irq->handler, NULL, irq->data,
1784 		    &irq->cookie);
1785 		if (unlikely(rc != 0)) {
1786 			ena_log(pdev, ERR,
1787 			    "failed to register interrupt handler for irq %ju: %d\n",
1788 			    rman_get_start(irq->res), rc);
1789 			goto err;
1790 		}
1791 		irq->requested = true;
1792 
1793 #ifdef RSS
1794 		rc = bus_bind_intr(adapter->pdev, irq->res, irq->cpu);
1795 		if (unlikely(rc != 0)) {
1796 			ena_log(pdev, ERR,
1797 			    "failed to bind interrupt handler for irq %ju to cpu %d: %d\n",
1798 			    rman_get_start(irq->res), irq->cpu, rc);
1799 			goto err;
1800 		}
1801 
1802 		ena_log(pdev, INFO, "queue %d - cpu %d\n",
1803 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
1804 #endif
1805 	}
1806 
1807 	return (rc);
1808 
1809 err:
1810 
1811 	for (; i >= ENA_IO_IRQ_FIRST_IDX; i--) {
1812 		irq = &adapter->irq_tbl[i];
1813 		rcc = 0;
1814 
1815 		/* Once we entered err: section and irq->requested is true we
1816 		   free both intr and resources */
1817 		if (irq->requested)
1818 			rcc = bus_teardown_intr(adapter->pdev, irq->res,
1819 			    irq->cookie);
1820 		if (unlikely(rcc != 0))
1821 			ena_log(pdev, ERR,
1822 			    "could not release irq: %d, error: %d\n",
1823 			    irq->vector, rcc);
1824 
1825 		/* If we entered err: section without irq->requested set we know
1826 		   it was bus_alloc_resource_any() that needs cleanup, provided
1827 		   res is not NULL. In case res is NULL no work in needed in
1828 		   this iteration */
1829 		rcc = 0;
1830 		if (irq->res != NULL) {
1831 			rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1832 			    irq->vector, irq->res);
1833 		}
1834 		if (unlikely(rcc != 0))
1835 			ena_log(pdev, ERR,
1836 			    "dev has no parent while releasing res for irq: %d\n",
1837 			    irq->vector);
1838 		irq->requested = false;
1839 		irq->res = NULL;
1840 	}
1841 
1842 	return (rc);
1843 }
1844 
1845 static void
1846 ena_free_mgmnt_irq(struct ena_adapter *adapter)
1847 {
1848 	device_t pdev = adapter->pdev;
1849 	struct ena_irq *irq;
1850 	int rc;
1851 
1852 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1853 	if (irq->requested) {
1854 		ena_log(pdev, DBG, "tear down irq: %d\n", irq->vector);
1855 		rc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
1856 		if (unlikely(rc != 0))
1857 			ena_log(pdev, ERR, "failed to tear down irq: %d\n",
1858 			    irq->vector);
1859 		irq->requested = 0;
1860 	}
1861 
1862 	if (irq->res != NULL) {
1863 		ena_log(pdev, DBG, "release resource irq: %d\n", irq->vector);
1864 		rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1865 		    irq->vector, irq->res);
1866 		irq->res = NULL;
1867 		if (unlikely(rc != 0))
1868 			ena_log(pdev, ERR,
1869 			    "dev has no parent while releasing res for irq: %d\n",
1870 			    irq->vector);
1871 	}
1872 }
1873 
1874 static void
1875 ena_free_io_irq(struct ena_adapter *adapter)
1876 {
1877 	device_t pdev = adapter->pdev;
1878 	struct ena_irq *irq;
1879 	int rc;
1880 
1881 	for (int i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1882 		irq = &adapter->irq_tbl[i];
1883 		if (irq->requested) {
1884 			ena_log(pdev, DBG, "tear down irq: %d\n", irq->vector);
1885 			rc = bus_teardown_intr(adapter->pdev, irq->res,
1886 			    irq->cookie);
1887 			if (unlikely(rc != 0)) {
1888 				ena_log(pdev, ERR,
1889 				    "failed to tear down irq: %d\n",
1890 				    irq->vector);
1891 			}
1892 			irq->requested = 0;
1893 		}
1894 
1895 		if (irq->res != NULL) {
1896 			ena_log(pdev, DBG, "release resource irq: %d\n",
1897 			    irq->vector);
1898 			rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1899 			    irq->vector, irq->res);
1900 			irq->res = NULL;
1901 			if (unlikely(rc != 0)) {
1902 				ena_log(pdev, ERR,
1903 				    "dev has no parent while releasing res for irq: %d\n",
1904 				    irq->vector);
1905 			}
1906 		}
1907 	}
1908 }
1909 
1910 static void
1911 ena_free_irqs(struct ena_adapter *adapter)
1912 {
1913 	ena_free_io_irq(adapter);
1914 	ena_free_mgmnt_irq(adapter);
1915 	ena_disable_msix(adapter);
1916 }
1917 
1918 static void
1919 ena_disable_msix(struct ena_adapter *adapter)
1920 {
1921 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1922 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1923 		pci_release_msi(adapter->pdev);
1924 	}
1925 
1926 	adapter->msix_vecs = 0;
1927 	free(adapter->msix_entries, M_DEVBUF);
1928 	adapter->msix_entries = NULL;
1929 }
1930 
1931 static void
1932 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
1933 {
1934 	struct ena_com_io_cq *io_cq;
1935 	struct ena_eth_io_intr_reg intr_reg;
1936 	struct ena_ring *tx_ring;
1937 	uint16_t ena_qid;
1938 	int i;
1939 
1940 	/* Unmask interrupts for all queues */
1941 	for (i = 0; i < adapter->num_io_queues; i++) {
1942 		ena_qid = ENA_IO_TXQ_IDX(i);
1943 		io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1944 		ena_com_update_intr_reg(&intr_reg, 0, 0, true);
1945 		tx_ring = &adapter->tx_ring[i];
1946 		counter_u64_add(tx_ring->tx_stats.unmask_interrupt_num, 1);
1947 		ena_com_unmask_intr(io_cq, &intr_reg);
1948 	}
1949 }
1950 
1951 static int
1952 ena_up_complete(struct ena_adapter *adapter)
1953 {
1954 	int rc;
1955 
1956 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter))) {
1957 		rc = ena_rss_configure(adapter);
1958 		if (rc != 0) {
1959 			ena_log(adapter->pdev, ERR,
1960 			    "Failed to configure RSS\n");
1961 			return (rc);
1962 		}
1963 	}
1964 
1965 	rc = ena_change_mtu(adapter->ifp, if_getmtu(adapter->ifp));
1966 	if (unlikely(rc != 0))
1967 		return (rc);
1968 
1969 	ena_refill_all_rx_bufs(adapter);
1970 	ena_reset_counters((counter_u64_t *)&adapter->hw_stats,
1971 	    sizeof(adapter->hw_stats));
1972 
1973 	return (0);
1974 }
1975 
1976 static void
1977 set_io_rings_size(struct ena_adapter *adapter, int new_tx_size, int new_rx_size)
1978 {
1979 	int i;
1980 
1981 	for (i = 0; i < adapter->num_io_queues; i++) {
1982 		adapter->tx_ring[i].ring_size = new_tx_size;
1983 		adapter->rx_ring[i].ring_size = new_rx_size;
1984 	}
1985 }
1986 
1987 static int
1988 create_queues_with_size_backoff(struct ena_adapter *adapter)
1989 {
1990 	device_t pdev = adapter->pdev;
1991 	int rc;
1992 	uint32_t cur_rx_ring_size, cur_tx_ring_size;
1993 	uint32_t new_rx_ring_size, new_tx_ring_size;
1994 
1995 	/*
1996 	 * Current queue sizes might be set to smaller than the requested
1997 	 * ones due to past queue allocation failures.
1998 	 */
1999 	set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2000 	    adapter->requested_rx_ring_size);
2001 
2002 	while (1) {
2003 		/* Allocate transmit descriptors */
2004 		rc = ena_setup_all_tx_resources(adapter);
2005 		if (unlikely(rc != 0)) {
2006 			ena_log(pdev, ERR, "err_setup_tx\n");
2007 			goto err_setup_tx;
2008 		}
2009 
2010 		/* Allocate receive descriptors */
2011 		rc = ena_setup_all_rx_resources(adapter);
2012 		if (unlikely(rc != 0)) {
2013 			ena_log(pdev, ERR, "err_setup_rx\n");
2014 			goto err_setup_rx;
2015 		}
2016 
2017 		/* Create IO queues for Rx & Tx */
2018 		rc = ena_create_io_queues(adapter);
2019 		if (unlikely(rc != 0)) {
2020 			ena_log(pdev, ERR, "create IO queues failed\n");
2021 			goto err_io_que;
2022 		}
2023 
2024 		return (0);
2025 
2026 err_io_que:
2027 		ena_free_all_rx_resources(adapter);
2028 err_setup_rx:
2029 		ena_free_all_tx_resources(adapter);
2030 err_setup_tx:
2031 		/*
2032 		 * Lower the ring size if ENOMEM. Otherwise, return the
2033 		 * error straightaway.
2034 		 */
2035 		if (unlikely(rc != ENOMEM)) {
2036 			ena_log(pdev, ERR,
2037 			    "Queue creation failed with error code: %d\n", rc);
2038 			return (rc);
2039 		}
2040 
2041 		cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2042 		cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2043 
2044 		ena_log(pdev, ERR,
2045 		    "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2046 		    cur_tx_ring_size, cur_rx_ring_size);
2047 
2048 		new_tx_ring_size = cur_tx_ring_size;
2049 		new_rx_ring_size = cur_rx_ring_size;
2050 
2051 		/*
2052 		 * Decrease the size of a larger queue, or decrease both if they
2053 		 * are the same size.
2054 		 */
2055 		if (cur_rx_ring_size <= cur_tx_ring_size)
2056 			new_tx_ring_size = cur_tx_ring_size / 2;
2057 		if (cur_rx_ring_size >= cur_tx_ring_size)
2058 			new_rx_ring_size = cur_rx_ring_size / 2;
2059 
2060 		if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2061 		    new_rx_ring_size < ENA_MIN_RING_SIZE) {
2062 			ena_log(pdev, ERR,
2063 			    "Queue creation failed with the smallest possible queue size"
2064 			    "of %d for both queues. Not retrying with smaller queues\n",
2065 			    ENA_MIN_RING_SIZE);
2066 			return (rc);
2067 		}
2068 
2069 		ena_log(pdev, INFO,
2070 		    "Retrying queue creation with sizes TX=%d, RX=%d\n",
2071 		    new_tx_ring_size, new_rx_ring_size);
2072 
2073 		set_io_rings_size(adapter, new_tx_ring_size, new_rx_ring_size);
2074 	}
2075 }
2076 
2077 int
2078 ena_up(struct ena_adapter *adapter)
2079 {
2080 	int rc = 0;
2081 
2082 	ENA_LOCK_ASSERT();
2083 
2084 	if (unlikely(device_is_attached(adapter->pdev) == 0)) {
2085 		ena_log(adapter->pdev, ERR, "device is not attached!\n");
2086 		return (ENXIO);
2087 	}
2088 
2089 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2090 		return (0);
2091 
2092 	ena_log(adapter->pdev, INFO, "device is going UP\n");
2093 
2094 	/* setup interrupts for IO queues */
2095 	rc = ena_setup_io_intr(adapter);
2096 	if (unlikely(rc != 0)) {
2097 		ena_log(adapter->pdev, ERR, "error setting up IO interrupt\n");
2098 		goto error;
2099 	}
2100 	rc = ena_request_io_irq(adapter);
2101 	if (unlikely(rc != 0)) {
2102 		ena_log(adapter->pdev, ERR, "err_req_irq\n");
2103 		goto error;
2104 	}
2105 
2106 	ena_log(adapter->pdev, INFO,
2107 	    "Creating %u IO queues. Rx queue size: %d, Tx queue size: %d, 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_log(adapter->pdev, ERR,
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, IFF_DRV_OACTIVE);
2133 
2134 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP, adapter);
2135 
2136 	ena_unmask_all_io_irqs(adapter);
2137 
2138 	return (0);
2139 
2140 err_up_complete:
2141 	ena_destroy_all_io_queues(adapter);
2142 	ena_free_all_rx_resources(adapter);
2143 	ena_free_all_tx_resources(adapter);
2144 err_create_queues_with_backoff:
2145 	ena_free_io_irq(adapter);
2146 error:
2147 	return (rc);
2148 }
2149 
2150 static uint64_t
2151 ena_get_counter(if_t ifp, ift_counter cnt)
2152 {
2153 	struct ena_adapter *adapter;
2154 	struct ena_hw_stats *stats;
2155 
2156 	adapter = if_getsoftc(ifp);
2157 	stats = &adapter->hw_stats;
2158 
2159 	switch (cnt) {
2160 	case IFCOUNTER_IPACKETS:
2161 		return (counter_u64_fetch(stats->rx_packets));
2162 	case IFCOUNTER_OPACKETS:
2163 		return (counter_u64_fetch(stats->tx_packets));
2164 	case IFCOUNTER_IBYTES:
2165 		return (counter_u64_fetch(stats->rx_bytes));
2166 	case IFCOUNTER_OBYTES:
2167 		return (counter_u64_fetch(stats->tx_bytes));
2168 	case IFCOUNTER_IQDROPS:
2169 		return (counter_u64_fetch(stats->rx_drops));
2170 	case IFCOUNTER_OQDROPS:
2171 		return (counter_u64_fetch(stats->tx_drops));
2172 	default:
2173 		return (if_get_counter_default(ifp, cnt));
2174 	}
2175 }
2176 
2177 static int
2178 ena_media_change(if_t ifp)
2179 {
2180 	/* Media Change is not supported by firmware */
2181 	return (0);
2182 }
2183 
2184 static void
2185 ena_media_status(if_t ifp, struct ifmediareq *ifmr)
2186 {
2187 	struct ena_adapter *adapter = if_getsoftc(ifp);
2188 	ena_log(adapter->pdev, DBG, "Media status update\n");
2189 
2190 	ENA_LOCK_LOCK();
2191 
2192 	ifmr->ifm_status = IFM_AVALID;
2193 	ifmr->ifm_active = IFM_ETHER;
2194 
2195 	if (!ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter)) {
2196 		ENA_LOCK_UNLOCK();
2197 		ena_log(adapter->pdev, INFO, "Link is down\n");
2198 		return;
2199 	}
2200 
2201 	ifmr->ifm_status |= IFM_ACTIVE;
2202 	ifmr->ifm_active |= IFM_UNKNOWN | IFM_FDX;
2203 
2204 	ENA_LOCK_UNLOCK();
2205 }
2206 
2207 static void
2208 ena_init(void *arg)
2209 {
2210 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
2211 
2212 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter)) {
2213 		ENA_LOCK_LOCK();
2214 		ena_up(adapter);
2215 		ENA_LOCK_UNLOCK();
2216 	}
2217 }
2218 
2219 static int
2220 ena_ioctl(if_t ifp, u_long command, caddr_t data)
2221 {
2222 	struct ena_adapter *adapter;
2223 	struct ifreq *ifr;
2224 	int rc;
2225 
2226 	adapter = if_getsoftc(ifp);
2227 	ifr = (struct ifreq *)data;
2228 
2229 	/*
2230 	 * Acquiring lock to prevent from running up and down routines parallel.
2231 	 */
2232 	rc = 0;
2233 	switch (command) {
2234 	case SIOCSIFMTU:
2235 		if (if_getmtu(ifp) == ifr->ifr_mtu)
2236 			break;
2237 		ENA_LOCK_LOCK();
2238 		ena_down(adapter);
2239 
2240 		ena_change_mtu(ifp, ifr->ifr_mtu);
2241 
2242 		rc = ena_up(adapter);
2243 		ENA_LOCK_UNLOCK();
2244 		break;
2245 
2246 	case SIOCSIFFLAGS:
2247 		if ((if_getflags(ifp) & IFF_UP) != 0) {
2248 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2249 				if ((if_getflags(ifp) & (IFF_PROMISC |
2250 				    IFF_ALLMULTI)) != 0) {
2251 					ena_log(adapter->pdev, INFO,
2252 					    "ioctl promisc/allmulti\n");
2253 				}
2254 			} else {
2255 				ENA_LOCK_LOCK();
2256 				rc = ena_up(adapter);
2257 				ENA_LOCK_UNLOCK();
2258 			}
2259 		} else {
2260 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2261 				ENA_LOCK_LOCK();
2262 				ena_down(adapter);
2263 				ENA_LOCK_UNLOCK();
2264 			}
2265 		}
2266 		break;
2267 
2268 	case SIOCADDMULTI:
2269 	case SIOCDELMULTI:
2270 		break;
2271 
2272 	case SIOCSIFMEDIA:
2273 	case SIOCGIFMEDIA:
2274 		rc = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
2275 		break;
2276 
2277 	case SIOCSIFCAP:
2278 		{
2279 			int reinit = 0;
2280 
2281 			if (ifr->ifr_reqcap != if_getcapenable(ifp)) {
2282 				if_setcapenable(ifp, ifr->ifr_reqcap);
2283 				reinit = 1;
2284 			}
2285 
2286 			if ((reinit != 0) &&
2287 			    ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)) {
2288 				ENA_LOCK_LOCK();
2289 				ena_down(adapter);
2290 				rc = ena_up(adapter);
2291 				ENA_LOCK_UNLOCK();
2292 			}
2293 		}
2294 
2295 		break;
2296 	default:
2297 		rc = ether_ioctl(ifp, command, data);
2298 		break;
2299 	}
2300 
2301 	return (rc);
2302 }
2303 
2304 static int
2305 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2306 {
2307 	int caps = 0;
2308 
2309 	if ((feat->offload.tx &
2310 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2311 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2312 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)) != 0)
2313 		caps |= IFCAP_TXCSUM;
2314 
2315 	if ((feat->offload.tx &
2316 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2317 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)) != 0)
2318 		caps |= IFCAP_TXCSUM_IPV6;
2319 
2320 	if ((feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0)
2321 		caps |= IFCAP_TSO4;
2322 
2323 	if ((feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK) != 0)
2324 		caps |= IFCAP_TSO6;
2325 
2326 	if ((feat->offload.rx_supported &
2327 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2328 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK)) != 0)
2329 		caps |= IFCAP_RXCSUM;
2330 
2331 	if ((feat->offload.rx_supported &
2332 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK) != 0)
2333 		caps |= IFCAP_RXCSUM_IPV6;
2334 
2335 	caps |= IFCAP_LRO | IFCAP_JUMBO_MTU;
2336 
2337 	return (caps);
2338 }
2339 
2340 static void
2341 ena_update_host_info(struct ena_admin_host_info *host_info, if_t ifp)
2342 {
2343 	host_info->supported_network_features[0] = (uint32_t)if_getcapabilities(ifp);
2344 }
2345 
2346 static void
2347 ena_update_hwassist(struct ena_adapter *adapter)
2348 {
2349 	if_t ifp = adapter->ifp;
2350 	uint32_t feat = adapter->tx_offload_cap;
2351 	int cap = if_getcapenable(ifp);
2352 	int flags = 0;
2353 
2354 	if_clearhwassist(ifp);
2355 
2356 	if ((cap & IFCAP_TXCSUM) != 0) {
2357 		if ((feat &
2358 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK) != 0)
2359 			flags |= CSUM_IP;
2360 		if ((feat &
2361 		    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2362 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)) != 0)
2363 			flags |= CSUM_IP_UDP | CSUM_IP_TCP;
2364 	}
2365 
2366 	if ((cap & IFCAP_TXCSUM_IPV6) != 0)
2367 		flags |= CSUM_IP6_UDP | CSUM_IP6_TCP;
2368 
2369 	if ((cap & IFCAP_TSO4) != 0)
2370 		flags |= CSUM_IP_TSO;
2371 
2372 	if ((cap & IFCAP_TSO6) != 0)
2373 		flags |= CSUM_IP6_TSO;
2374 
2375 	if_sethwassistbits(ifp, flags, 0);
2376 }
2377 
2378 static int
2379 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2380     struct ena_com_dev_get_features_ctx *feat)
2381 {
2382 	if_t ifp;
2383 	int caps = 0;
2384 
2385 	ifp = adapter->ifp = if_gethandle(IFT_ETHER);
2386 	if (unlikely(ifp == NULL)) {
2387 		ena_log(pdev, ERR, "can not allocate ifnet structure\n");
2388 		return (ENXIO);
2389 	}
2390 	if_initname(ifp, device_get_name(pdev), device_get_unit(pdev));
2391 	if_setdev(ifp, pdev);
2392 	if_setsoftc(ifp, adapter);
2393 
2394 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
2395 	if_setinitfn(ifp, ena_init);
2396 	if_settransmitfn(ifp, ena_mq_start);
2397 	if_setqflushfn(ifp, ena_qflush);
2398 	if_setioctlfn(ifp, ena_ioctl);
2399 	if_setgetcounterfn(ifp, ena_get_counter);
2400 
2401 	if_setsendqlen(ifp, adapter->requested_tx_ring_size);
2402 	if_setsendqready(ifp);
2403 	if_setmtu(ifp, ETHERMTU);
2404 	if_setbaudrate(ifp, 0);
2405 	/* Zeroize capabilities... */
2406 	if_setcapabilities(ifp, 0);
2407 	if_setcapenable(ifp, 0);
2408 	/* check hardware support */
2409 	caps = ena_get_dev_offloads(feat);
2410 	/* ... and set them */
2411 	if_setcapabilitiesbit(ifp, caps, 0);
2412 
2413 	/* TSO parameters */
2414 	if_sethwtsomax(ifp, ENA_TSO_MAXSIZE -
2415 	    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
2416 	if_sethwtsomaxsegcount(ifp, adapter->max_tx_sgl_size - 1);
2417 	if_sethwtsomaxsegsize(ifp, ENA_TSO_MAXSIZE);
2418 
2419 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
2420 	if_setcapenable(ifp, if_getcapabilities(ifp));
2421 
2422 	/*
2423 	 * Specify the media types supported by this adapter and register
2424 	 * callbacks to update media and link information
2425 	 */
2426 	ifmedia_init(&adapter->media, IFM_IMASK, ena_media_change,
2427 	    ena_media_status);
2428 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2429 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2430 
2431 	ether_ifattach(ifp, adapter->mac_addr);
2432 
2433 	return (0);
2434 }
2435 
2436 void
2437 ena_down(struct ena_adapter *adapter)
2438 {
2439 	int rc;
2440 
2441 	ENA_LOCK_ASSERT();
2442 
2443 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2444 		return;
2445 
2446 	ena_log(adapter->pdev, INFO, "device is going DOWN\n");
2447 
2448 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP, adapter);
2449 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2450 
2451 	ena_free_io_irq(adapter);
2452 
2453 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter)) {
2454 		rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2455 		if (unlikely(rc != 0))
2456 			ena_log(adapter->pdev, ERR, "Device reset failed\n");
2457 	}
2458 
2459 	ena_destroy_all_io_queues(adapter);
2460 
2461 	ena_free_all_tx_bufs(adapter);
2462 	ena_free_all_rx_bufs(adapter);
2463 	ena_free_all_tx_resources(adapter);
2464 	ena_free_all_rx_resources(adapter);
2465 
2466 	counter_u64_add(adapter->dev_stats.interface_down, 1);
2467 }
2468 
2469 static uint32_t
2470 ena_calc_max_io_queue_num(device_t pdev, struct ena_com_dev *ena_dev,
2471     struct ena_com_dev_get_features_ctx *get_feat_ctx)
2472 {
2473 	uint32_t io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
2474 
2475 	/* Regular queues capabilities */
2476 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2477 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2478 		    &get_feat_ctx->max_queue_ext.max_queue_ext;
2479 		io_rx_num = min_t(int, max_queue_ext->max_rx_sq_num,
2480 		    max_queue_ext->max_rx_cq_num);
2481 
2482 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
2483 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
2484 	} else {
2485 		struct ena_admin_queue_feature_desc *max_queues =
2486 		    &get_feat_ctx->max_queues;
2487 		io_tx_sq_num = max_queues->max_sq_num;
2488 		io_tx_cq_num = max_queues->max_cq_num;
2489 		io_rx_num = min_t(int, io_tx_sq_num, io_tx_cq_num);
2490 	}
2491 
2492 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
2493 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
2494 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
2495 
2496 	max_num_io_queues = min_t(uint32_t, mp_ncpus, ENA_MAX_NUM_IO_QUEUES);
2497 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_rx_num);
2498 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_tx_sq_num);
2499 	max_num_io_queues = min_t(uint32_t, max_num_io_queues, io_tx_cq_num);
2500 	/* 1 IRQ for mgmnt and 1 IRQ for each TX/RX pair */
2501 	max_num_io_queues = min_t(uint32_t, max_num_io_queues,
2502 	    pci_msix_count(pdev) - 1);
2503 #ifdef RSS
2504 	max_num_io_queues = min_t(uint32_t, max_num_io_queues,
2505 	    rss_getnumbuckets());
2506 #endif
2507 
2508 	return (max_num_io_queues);
2509 }
2510 
2511 static int
2512 ena_enable_wc(device_t pdev, struct resource *res)
2513 {
2514 #if defined(__i386) || defined(__amd64) || defined(__aarch64__)
2515 	vm_offset_t va;
2516 	vm_size_t len;
2517 	int rc;
2518 
2519 	va = (vm_offset_t)rman_get_virtual(res);
2520 	len = rman_get_size(res);
2521 	/* Enable write combining */
2522 	rc = pmap_change_attr(va, len, VM_MEMATTR_WRITE_COMBINING);
2523 	if (unlikely(rc != 0)) {
2524 		ena_log(pdev, ERR, "pmap_change_attr failed, %d\n", rc);
2525 		return (rc);
2526 	}
2527 
2528 	return (0);
2529 #endif
2530 	return (EOPNOTSUPP);
2531 }
2532 
2533 static int
2534 ena_set_queues_placement_policy(device_t pdev, struct ena_com_dev *ena_dev,
2535     struct ena_admin_feature_llq_desc *llq,
2536     struct ena_llq_configurations *llq_default_configurations)
2537 {
2538 	int rc;
2539 	uint32_t llq_feature_mask;
2540 
2541 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
2542 	if (!(ena_dev->supported_features & llq_feature_mask)) {
2543 		ena_log(pdev, WARN,
2544 		    "LLQ is not supported. Fallback to host mode policy.\n");
2545 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2546 		return (0);
2547 	}
2548 
2549 	if (ena_dev->mem_bar == NULL) {
2550 		ena_log(pdev, WARN,
2551 		    "LLQ is advertised as supported but device doesn't expose mem bar.\n");
2552 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2553 		return (0);
2554 	}
2555 
2556 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
2557 	if (unlikely(rc != 0)) {
2558 		ena_log(pdev, WARN,
2559 		    "Failed to configure the device mode. "
2560 		    "Fallback to host mode policy.\n");
2561 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2562 	}
2563 
2564 	return (0);
2565 }
2566 
2567 static int
2568 ena_map_llq_mem_bar(device_t pdev, struct ena_com_dev *ena_dev)
2569 {
2570 	struct ena_adapter *adapter = device_get_softc(pdev);
2571 	int rc, rid;
2572 
2573 	/* Try to allocate resources for LLQ bar */
2574 	rid = PCIR_BAR(ENA_MEM_BAR);
2575 	adapter->memory = bus_alloc_resource_any(pdev, SYS_RES_MEMORY, &rid,
2576 	    RF_ACTIVE);
2577 	if (unlikely(adapter->memory == NULL)) {
2578 		ena_log(pdev, WARN,
2579 		    "Unable to allocate LLQ bar resource. LLQ mode won't be used.\n");
2580 		return (0);
2581 	}
2582 
2583 	/* Enable write combining for better LLQ performance */
2584 	rc = ena_enable_wc(adapter->pdev, adapter->memory);
2585 	if (unlikely(rc != 0)) {
2586 		ena_log(pdev, ERR, "failed to enable write combining.\n");
2587 		return (rc);
2588 	}
2589 
2590 	/*
2591 	 * Save virtual address of the device's memory region
2592 	 * for the ena_com layer.
2593 	 */
2594 	ena_dev->mem_bar = rman_get_virtual(adapter->memory);
2595 
2596 	return (0);
2597 }
2598 
2599 static inline void
2600 set_default_llq_configurations(struct ena_llq_configurations *llq_config,
2601     struct ena_admin_feature_llq_desc *llq)
2602 {
2603 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
2604 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
2605 	llq_config->llq_num_decs_before_header =
2606 	    ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
2607 	if ((llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B) !=
2608 	    0 && ena_force_large_llq_header) {
2609 		llq_config->llq_ring_entry_size =
2610 		    ENA_ADMIN_LIST_ENTRY_SIZE_256B;
2611 		llq_config->llq_ring_entry_size_value = 256;
2612 	} else {
2613 		llq_config->llq_ring_entry_size =
2614 		    ENA_ADMIN_LIST_ENTRY_SIZE_128B;
2615 		llq_config->llq_ring_entry_size_value = 128;
2616 	}
2617 }
2618 
2619 static int
2620 ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
2621 {
2622 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
2623 	struct ena_com_dev *ena_dev = ctx->ena_dev;
2624 	uint32_t tx_queue_size = ENA_DEFAULT_RING_SIZE;
2625 	uint32_t rx_queue_size = ENA_DEFAULT_RING_SIZE;
2626 	uint32_t max_tx_queue_size;
2627 	uint32_t max_rx_queue_size;
2628 
2629 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2630 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2631 		    &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
2632 		max_rx_queue_size = min_t(uint32_t,
2633 		    max_queue_ext->max_rx_cq_depth,
2634 		    max_queue_ext->max_rx_sq_depth);
2635 		max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
2636 
2637 		if (ena_dev->tx_mem_queue_type ==
2638 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2639 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2640 			    llq->max_llq_depth);
2641 		else
2642 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2643 			    max_queue_ext->max_tx_sq_depth);
2644 
2645 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2646 		    max_queue_ext->max_per_packet_tx_descs);
2647 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2648 		    max_queue_ext->max_per_packet_rx_descs);
2649 	} else {
2650 		struct ena_admin_queue_feature_desc *max_queues =
2651 		    &ctx->get_feat_ctx->max_queues;
2652 		max_rx_queue_size = min_t(uint32_t, max_queues->max_cq_depth,
2653 		    max_queues->max_sq_depth);
2654 		max_tx_queue_size = max_queues->max_cq_depth;
2655 
2656 		if (ena_dev->tx_mem_queue_type ==
2657 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2658 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2659 			    llq->max_llq_depth);
2660 		else
2661 			max_tx_queue_size = min_t(uint32_t, max_tx_queue_size,
2662 			    max_queues->max_sq_depth);
2663 
2664 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2665 		    max_queues->max_packet_tx_descs);
2666 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2667 		    max_queues->max_packet_rx_descs);
2668 	}
2669 
2670 	/* round down to the nearest power of 2 */
2671 	max_tx_queue_size = 1 << (flsl(max_tx_queue_size) - 1);
2672 	max_rx_queue_size = 1 << (flsl(max_rx_queue_size) - 1);
2673 
2674 	/*
2675 	 * When forcing large headers, we multiply the entry size by 2,
2676 	 * and therefore divide the queue size by 2, leaving the amount
2677 	 * of memory used by the queues unchanged.
2678 	 */
2679 	if (ena_force_large_llq_header) {
2680 		if ((llq->entry_size_ctrl_supported &
2681 		    ENA_ADMIN_LIST_ENTRY_SIZE_256B) != 0 &&
2682 		    ena_dev->tx_mem_queue_type ==
2683 		    ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2684 			max_tx_queue_size /= 2;
2685 			ena_log(ctx->pdev, INFO,
2686 			    "Forcing large headers and decreasing maximum Tx queue size to %d\n",
2687 			    max_tx_queue_size);
2688 		} else {
2689 			ena_log(ctx->pdev, WARN,
2690 			    "Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
2691 		}
2692 	}
2693 
2694 	tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
2695 	    max_tx_queue_size);
2696 	rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
2697 	    max_rx_queue_size);
2698 
2699 	tx_queue_size = 1 << (flsl(tx_queue_size) - 1);
2700 	rx_queue_size = 1 << (flsl(rx_queue_size) - 1);
2701 
2702 	ctx->max_tx_queue_size = max_tx_queue_size;
2703 	ctx->max_rx_queue_size = max_rx_queue_size;
2704 	ctx->tx_queue_size = tx_queue_size;
2705 	ctx->rx_queue_size = rx_queue_size;
2706 
2707 	return (0);
2708 }
2709 
2710 static void
2711 ena_config_host_info(struct ena_com_dev *ena_dev, device_t dev)
2712 {
2713 	struct ena_admin_host_info *host_info;
2714 	uintptr_t rid;
2715 	int rc;
2716 
2717 	/* Allocate only the host info */
2718 	rc = ena_com_allocate_host_info(ena_dev);
2719 	if (unlikely(rc != 0)) {
2720 		ena_log(dev, ERR, "Cannot allocate host info\n");
2721 		return;
2722 	}
2723 
2724 	host_info = ena_dev->host_attr.host_info;
2725 
2726 	if (pci_get_id(dev, PCI_ID_RID, &rid) == 0)
2727 		host_info->bdf = rid;
2728 	host_info->os_type = ENA_ADMIN_OS_FREEBSD;
2729 	host_info->kernel_ver = osreldate;
2730 
2731 	sprintf(host_info->kernel_ver_str, "%d", osreldate);
2732 	host_info->os_dist = 0;
2733 	strncpy(host_info->os_dist_str, osrelease,
2734 	    sizeof(host_info->os_dist_str) - 1);
2735 
2736 	host_info->driver_version = (ENA_DRV_MODULE_VER_MAJOR) |
2737 	    (ENA_DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2738 	    (ENA_DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2739 	host_info->num_cpus = mp_ncpus;
2740 	host_info->driver_supported_features =
2741 	    ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
2742 	    ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
2743 
2744 	rc = ena_com_set_host_attributes(ena_dev);
2745 	if (unlikely(rc != 0)) {
2746 		if (rc == EOPNOTSUPP)
2747 			ena_log(dev, WARN, "Cannot set host attributes\n");
2748 		else
2749 			ena_log(dev, ERR, "Cannot set host attributes\n");
2750 
2751 		goto err;
2752 	}
2753 
2754 	return;
2755 
2756 err:
2757 	ena_com_delete_host_info(ena_dev);
2758 }
2759 
2760 static int
2761 ena_device_init(struct ena_adapter *adapter, device_t pdev,
2762     struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
2763 {
2764 	struct ena_llq_configurations llq_config;
2765 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2766 	bool readless_supported;
2767 	uint32_t aenq_groups;
2768 	int dma_width;
2769 	int rc;
2770 
2771 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
2772 	if (unlikely(rc != 0)) {
2773 		ena_log(pdev, ERR, "failed to init mmio read less\n");
2774 		return (rc);
2775 	}
2776 
2777 	/*
2778 	 * The PCIe configuration space revision id indicate if mmio reg
2779 	 * read is disabled
2780 	 */
2781 	readless_supported = !(pci_get_revid(pdev) & ENA_MMIO_DISABLE_REG_READ);
2782 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2783 
2784 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2785 	if (unlikely(rc != 0)) {
2786 		ena_log(pdev, ERR, "Can not reset device\n");
2787 		goto err_mmio_read_less;
2788 	}
2789 
2790 	rc = ena_com_validate_version(ena_dev);
2791 	if (unlikely(rc != 0)) {
2792 		ena_log(pdev, ERR, "device version is too low\n");
2793 		goto err_mmio_read_less;
2794 	}
2795 
2796 	dma_width = ena_com_get_dma_width(ena_dev);
2797 	if (unlikely(dma_width < 0)) {
2798 		ena_log(pdev, ERR, "Invalid dma width value %d", dma_width);
2799 		rc = dma_width;
2800 		goto err_mmio_read_less;
2801 	}
2802 	adapter->dma_width = dma_width;
2803 
2804 	/* ENA admin level init */
2805 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
2806 	if (unlikely(rc != 0)) {
2807 		ena_log(pdev, ERR,
2808 		    "Can not initialize ena admin queue with device\n");
2809 		goto err_mmio_read_less;
2810 	}
2811 
2812 	/*
2813 	 * To enable the msix interrupts the driver needs to know the number
2814 	 * of queues. So the driver uses polling mode to retrieve this
2815 	 * information
2816 	 */
2817 	ena_com_set_admin_polling_mode(ena_dev, true);
2818 
2819 	ena_config_host_info(ena_dev, pdev);
2820 
2821 	/* Get Device Attributes */
2822 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2823 	if (unlikely(rc != 0)) {
2824 		ena_log(pdev, ERR,
2825 		    "Cannot get attribute for ena device rc: %d\n", rc);
2826 		goto err_admin_init;
2827 	}
2828 
2829 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2830 	    BIT(ENA_ADMIN_FATAL_ERROR) |
2831 	    BIT(ENA_ADMIN_WARNING) |
2832 	    BIT(ENA_ADMIN_NOTIFICATION) |
2833 	    BIT(ENA_ADMIN_KEEP_ALIVE);
2834 
2835 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
2836 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2837 	if (unlikely(rc != 0)) {
2838 		ena_log(pdev, ERR, "Cannot configure aenq groups rc: %d\n", rc);
2839 		goto err_admin_init;
2840 	}
2841 
2842 	*wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2843 
2844 	set_default_llq_configurations(&llq_config, &get_feat_ctx->llq);
2845 
2846 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
2847 	    &llq_config);
2848 	if (unlikely(rc != 0)) {
2849 		ena_log(pdev, ERR, "Failed to set placement policy\n");
2850 		goto err_admin_init;
2851 	}
2852 
2853 	return (0);
2854 
2855 err_admin_init:
2856 	ena_com_delete_host_info(ena_dev);
2857 	ena_com_admin_destroy(ena_dev);
2858 err_mmio_read_less:
2859 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2860 
2861 	return (rc);
2862 }
2863 
2864 static int
2865 ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
2866 {
2867 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2868 	int rc;
2869 
2870 	rc = ena_enable_msix(adapter);
2871 	if (unlikely(rc != 0)) {
2872 		ena_log(adapter->pdev, ERR, "Error with MSI-X enablement\n");
2873 		return (rc);
2874 	}
2875 
2876 	ena_setup_mgmnt_intr(adapter);
2877 
2878 	rc = ena_request_mgmnt_irq(adapter);
2879 	if (unlikely(rc != 0)) {
2880 		ena_log(adapter->pdev, ERR, "Cannot setup mgmnt queue intr\n");
2881 		goto err_disable_msix;
2882 	}
2883 
2884 	ena_com_set_admin_polling_mode(ena_dev, false);
2885 
2886 	ena_com_admin_aenq_enable(ena_dev);
2887 
2888 	return (0);
2889 
2890 err_disable_msix:
2891 	ena_disable_msix(adapter);
2892 
2893 	return (rc);
2894 }
2895 
2896 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
2897 static void
2898 ena_keep_alive_wd(void *adapter_data, struct ena_admin_aenq_entry *aenq_e)
2899 {
2900 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
2901 	struct ena_admin_aenq_keep_alive_desc *desc;
2902 	sbintime_t stime;
2903 	uint64_t rx_drops;
2904 	uint64_t tx_drops;
2905 
2906 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
2907 
2908 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
2909 	tx_drops = ((uint64_t)desc->tx_drops_high << 32) | desc->tx_drops_low;
2910 	counter_u64_zero(adapter->hw_stats.rx_drops);
2911 	counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
2912 	counter_u64_zero(adapter->hw_stats.tx_drops);
2913 	counter_u64_add(adapter->hw_stats.tx_drops, tx_drops);
2914 
2915 	stime = getsbinuptime();
2916 	atomic_store_rel_64(&adapter->keep_alive_timestamp, stime);
2917 }
2918 
2919 /* Check for keep alive expiration */
2920 static void
2921 check_for_missing_keep_alive(struct ena_adapter *adapter)
2922 {
2923 	sbintime_t timestamp, time;
2924 
2925 	if (adapter->wd_active == 0)
2926 		return;
2927 
2928 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2929 		return;
2930 
2931 	timestamp = atomic_load_acq_64(&adapter->keep_alive_timestamp);
2932 	time = getsbinuptime() - timestamp;
2933 	if (unlikely(time > adapter->keep_alive_timeout)) {
2934 		ena_log(adapter->pdev, ERR, "Keep alive watchdog timeout.\n");
2935 		counter_u64_add(adapter->dev_stats.wd_expired, 1);
2936 		ena_trigger_reset(adapter, ENA_REGS_RESET_KEEP_ALIVE_TO);
2937 	}
2938 }
2939 
2940 /* Check if admin queue is enabled */
2941 static void
2942 check_for_admin_com_state(struct ena_adapter *adapter)
2943 {
2944 	if (unlikely(ena_com_get_admin_running_state(adapter->ena_dev) == false)) {
2945 		ena_log(adapter->pdev, ERR,
2946 		    "ENA admin queue is not in running state!\n");
2947 		counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
2948 		ena_trigger_reset(adapter, ENA_REGS_RESET_ADMIN_TO);
2949 	}
2950 }
2951 
2952 static int
2953 check_for_rx_interrupt_queue(struct ena_adapter *adapter,
2954     struct ena_ring *rx_ring)
2955 {
2956 	if (likely(atomic_load_8(&rx_ring->first_interrupt)))
2957 		return (0);
2958 
2959 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
2960 		return (0);
2961 
2962 	rx_ring->no_interrupt_event_cnt++;
2963 
2964 	if (rx_ring->no_interrupt_event_cnt ==
2965 	    ENA_MAX_NO_INTERRUPT_ITERATIONS) {
2966 		ena_log(adapter->pdev, ERR,
2967 		    "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
2968 		    rx_ring->qid);
2969 		ena_trigger_reset(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
2970 		return (EIO);
2971 	}
2972 
2973 	return (0);
2974 }
2975 
2976 static int
2977 check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
2978     struct ena_ring *tx_ring)
2979 {
2980 	device_t pdev = adapter->pdev;
2981 	struct bintime curtime, time;
2982 	struct ena_tx_buffer *tx_buf;
2983 	int time_since_last_cleanup;
2984 	int missing_tx_comp_to;
2985 	sbintime_t time_offset;
2986 	uint32_t missed_tx = 0;
2987 	int i, rc = 0;
2988 
2989 	getbinuptime(&curtime);
2990 
2991 	for (i = 0; i < tx_ring->ring_size; i++) {
2992 		tx_buf = &tx_ring->tx_buffer_info[i];
2993 
2994 		if (bintime_isset(&tx_buf->timestamp) == 0)
2995 			continue;
2996 
2997 		time = curtime;
2998 		bintime_sub(&time, &tx_buf->timestamp);
2999 		time_offset = bttosbt(time);
3000 
3001 		if (unlikely(!atomic_load_8(&tx_ring->first_interrupt) &&
3002 		    time_offset > 2 * adapter->missing_tx_timeout)) {
3003 			/*
3004 			 * If after graceful period interrupt is still not
3005 			 * received, we schedule a reset.
3006 			 */
3007 			ena_log(pdev, ERR,
3008 			    "Potential MSIX issue on Tx side Queue = %d. "
3009 			    "Reset the device\n",
3010 			    tx_ring->qid);
3011 			ena_trigger_reset(adapter,
3012 			    ENA_REGS_RESET_MISS_INTERRUPT);
3013 			return (EIO);
3014 		}
3015 
3016 		/* Check again if packet is still waiting */
3017 		if (unlikely(time_offset > adapter->missing_tx_timeout)) {
3018 
3019 			if (tx_buf->print_once) {
3020 				time_since_last_cleanup = TICKS_2_USEC(ticks -
3021 				    tx_ring->tx_last_cleanup_ticks);
3022 				missing_tx_comp_to = sbttoms(
3023 				    adapter->missing_tx_timeout);
3024 				ena_log(pdev, WARN,
3025 				    "Found a Tx that wasn't completed on time, qid %d, index %d. "
3026 				    "%d usecs have passed since last cleanup. Missing Tx timeout value %d msecs.\n",
3027 				    tx_ring->qid, i, time_since_last_cleanup,
3028 				    missing_tx_comp_to);
3029 			}
3030 
3031 			tx_buf->print_once = false;
3032 			missed_tx++;
3033 		}
3034 	}
3035 
3036 	if (unlikely(missed_tx > adapter->missing_tx_threshold)) {
3037 		ena_log(pdev, ERR,
3038 		    "The number of lost tx completion is above the threshold "
3039 		    "(%d > %d). Reset the device\n",
3040 		    missed_tx, adapter->missing_tx_threshold);
3041 		ena_trigger_reset(adapter, ENA_REGS_RESET_MISS_TX_CMPL);
3042 		rc = EIO;
3043 	}
3044 
3045 	counter_u64_add(tx_ring->tx_stats.missing_tx_comp, missed_tx);
3046 
3047 	return (rc);
3048 }
3049 
3050 /*
3051  * Check for TX which were not completed on time.
3052  * Timeout is defined by "missing_tx_timeout".
3053  * Reset will be performed if number of incompleted
3054  * transactions exceeds "missing_tx_threshold".
3055  */
3056 static void
3057 check_for_missing_completions(struct ena_adapter *adapter)
3058 {
3059 	struct ena_ring *tx_ring;
3060 	struct ena_ring *rx_ring;
3061 	int i, budget, rc;
3062 
3063 	/* Make sure the driver doesn't turn the device in other process */
3064 	rmb();
3065 
3066 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3067 		return;
3068 
3069 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
3070 		return;
3071 
3072 	if (adapter->missing_tx_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3073 		return;
3074 
3075 	budget = adapter->missing_tx_max_queues;
3076 
3077 	for (i = adapter->next_monitored_tx_qid; i < adapter->num_io_queues; i++) {
3078 		tx_ring = &adapter->tx_ring[i];
3079 		rx_ring = &adapter->rx_ring[i];
3080 
3081 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3082 		if (unlikely(rc != 0))
3083 			return;
3084 
3085 		rc = check_for_rx_interrupt_queue(adapter, rx_ring);
3086 		if (unlikely(rc != 0))
3087 			return;
3088 
3089 		budget--;
3090 		if (budget == 0) {
3091 			i++;
3092 			break;
3093 		}
3094 	}
3095 
3096 	adapter->next_monitored_tx_qid = i % adapter->num_io_queues;
3097 }
3098 
3099 /* trigger rx cleanup after 2 consecutive detections */
3100 #define EMPTY_RX_REFILL 2
3101 /* For the rare case where the device runs out of Rx descriptors and the
3102  * msix handler failed to refill new Rx descriptors (due to a lack of memory
3103  * for example).
3104  * This case will lead to a deadlock:
3105  * The device won't send interrupts since all the new Rx packets will be dropped
3106  * The msix handler won't allocate new Rx descriptors so the device won't be
3107  * able to send new packets.
3108  *
3109  * When such a situation is detected - execute rx cleanup task in another thread
3110  */
3111 static void
3112 check_for_empty_rx_ring(struct ena_adapter *adapter)
3113 {
3114 	struct ena_ring *rx_ring;
3115 	int i, refill_required;
3116 
3117 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3118 		return;
3119 
3120 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
3121 		return;
3122 
3123 	for (i = 0; i < adapter->num_io_queues; i++) {
3124 		rx_ring = &adapter->rx_ring[i];
3125 
3126 		refill_required = ena_com_free_q_entries(
3127 		    rx_ring->ena_com_io_sq);
3128 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3129 			rx_ring->empty_rx_queue++;
3130 
3131 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3132 				counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
3133 				    1);
3134 
3135 				ena_log(adapter->pdev, WARN,
3136 				    "Rx ring %d is stalled. Triggering the refill function\n",
3137 				    i);
3138 
3139 				taskqueue_enqueue(rx_ring->que->cleanup_tq,
3140 				    &rx_ring->que->cleanup_task);
3141 				rx_ring->empty_rx_queue = 0;
3142 			}
3143 		} else {
3144 			rx_ring->empty_rx_queue = 0;
3145 		}
3146 	}
3147 }
3148 
3149 static void
3150 ena_update_hints(struct ena_adapter *adapter,
3151     struct ena_admin_ena_hw_hints *hints)
3152 {
3153 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3154 
3155 	if (hints->admin_completion_tx_timeout)
3156 		ena_dev->admin_queue.completion_timeout =
3157 		    hints->admin_completion_tx_timeout * 1000;
3158 
3159 	if (hints->mmio_read_timeout)
3160 		/* convert to usec */
3161 		ena_dev->mmio_read.reg_read_to = hints->mmio_read_timeout * 1000;
3162 
3163 	if (hints->missed_tx_completion_count_threshold_to_reset)
3164 		adapter->missing_tx_threshold =
3165 		    hints->missed_tx_completion_count_threshold_to_reset;
3166 
3167 	if (hints->missing_tx_completion_timeout) {
3168 		if (hints->missing_tx_completion_timeout ==
3169 		    ENA_HW_HINTS_NO_TIMEOUT)
3170 			adapter->missing_tx_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3171 		else
3172 			adapter->missing_tx_timeout = SBT_1MS *
3173 			    hints->missing_tx_completion_timeout;
3174 	}
3175 
3176 	if (hints->driver_watchdog_timeout) {
3177 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3178 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3179 		else
3180 			adapter->keep_alive_timeout = SBT_1MS *
3181 			    hints->driver_watchdog_timeout;
3182 	}
3183 }
3184 
3185 /**
3186  * ena_copy_eni_metrics - Get and copy ENI metrics from the HW.
3187  * @adapter: ENA device adapter
3188  *
3189  * Returns 0 on success, EOPNOTSUPP if current HW doesn't support those metrics
3190  * and other error codes on failure.
3191  *
3192  * This function can possibly cause a race with other calls to the admin queue.
3193  * Because of that, the caller should either lock this function or make sure
3194  * that there is no race in the current context.
3195  */
3196 static int
3197 ena_copy_eni_metrics(struct ena_adapter *adapter)
3198 {
3199 	static bool print_once = true;
3200 	int rc;
3201 
3202 	rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_metrics);
3203 
3204 	if (rc != 0) {
3205 		if (rc == ENA_COM_UNSUPPORTED) {
3206 			if (print_once) {
3207 				ena_log(adapter->pdev, WARN,
3208 				    "Retrieving ENI metrics is not supported.\n");
3209 				print_once = false;
3210 			} else {
3211 				ena_log(adapter->pdev, DBG,
3212 				    "Retrieving ENI metrics is not supported.\n");
3213 			}
3214 		} else {
3215 			ena_log(adapter->pdev, ERR,
3216 			    "Failed to get ENI metrics: %d\n", rc);
3217 		}
3218 	}
3219 
3220 	return (rc);
3221 }
3222 
3223 static void
3224 ena_timer_service(void *data)
3225 {
3226 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3227 	struct ena_admin_host_info *host_info =
3228 	    adapter->ena_dev->host_attr.host_info;
3229 
3230 	check_for_missing_keep_alive(adapter);
3231 
3232 	check_for_admin_com_state(adapter);
3233 
3234 	check_for_missing_completions(adapter);
3235 
3236 	check_for_empty_rx_ring(adapter);
3237 
3238 	/*
3239 	 * User controller update of the ENI metrics.
3240 	 * If the delay was set to 0, then the stats shouldn't be updated at
3241 	 * all.
3242 	 * Otherwise, wait 'eni_metrics_sample_interval' seconds, before
3243 	 * updating stats.
3244 	 * As timer service is executed every second, it's enough to increment
3245 	 * appropriate counter each time the timer service is executed.
3246 	 */
3247 	if ((adapter->eni_metrics_sample_interval != 0) &&
3248 	    (++adapter->eni_metrics_sample_interval_cnt >=
3249 	     adapter->eni_metrics_sample_interval)) {
3250 		taskqueue_enqueue(adapter->metrics_tq, &adapter->metrics_task);
3251 		adapter->eni_metrics_sample_interval_cnt = 0;
3252 	}
3253 
3254 
3255 	if (host_info != NULL)
3256 		ena_update_host_info(host_info, adapter->ifp);
3257 
3258 	if (unlikely(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
3259 		/*
3260 		 * Timeout when validating version indicates that the device
3261 		 * became unresponsive. If that happens skip the reset and
3262 		 * reschedule timer service, so the reset can be retried later.
3263 		 */
3264 		if (ena_com_validate_version(adapter->ena_dev) ==
3265 		    ENA_COM_TIMER_EXPIRED) {
3266 			ena_log(adapter->pdev, WARN,
3267 			    "FW unresponsive, skipping reset\n");
3268 			ENA_TIMER_RESET(adapter);
3269 			return;
3270 		}
3271 		ena_log(adapter->pdev, WARN, "Trigger reset is on\n");
3272 		taskqueue_enqueue(adapter->reset_tq, &adapter->reset_task);
3273 		return;
3274 	}
3275 
3276 	/*
3277 	 * Schedule another timeout one second from now.
3278 	 */
3279 	ENA_TIMER_RESET(adapter);
3280 }
3281 
3282 void
3283 ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3284 {
3285 	if_t ifp = adapter->ifp;
3286 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3287 	bool dev_up;
3288 
3289 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter))
3290 		return;
3291 
3292 	if (!graceful)
3293 		if_link_state_change(ifp, LINK_STATE_DOWN);
3294 
3295 	ENA_TIMER_DRAIN(adapter);
3296 
3297 	dev_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
3298 	if (dev_up)
3299 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3300 
3301 	if (!graceful)
3302 		ena_com_set_admin_running_state(ena_dev, false);
3303 
3304 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3305 		ena_down(adapter);
3306 
3307 	/*
3308 	 * Stop the device from sending AENQ events (if the device was up, and
3309 	 * the trigger reset was on, ena_down already performs device reset)
3310 	 */
3311 	if (!(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter) && dev_up))
3312 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3313 
3314 	ena_free_mgmnt_irq(adapter);
3315 
3316 	ena_disable_msix(adapter);
3317 
3318 	/*
3319 	 * IO rings resources should be freed because `ena_restore_device()`
3320 	 * calls (not directly) `ena_enable_msix()`, which re-allocates MSIX
3321 	 * vectors. The amount of MSIX vectors after destroy-restore may be
3322 	 * different than before. Therefore, IO rings resources should be
3323 	 * established from scratch each time.
3324 	 */
3325 	ena_free_all_io_rings_resources(adapter);
3326 
3327 	ena_com_abort_admin_commands(ena_dev);
3328 
3329 	ena_com_wait_for_abort_completion(ena_dev);
3330 
3331 	ena_com_admin_destroy(ena_dev);
3332 
3333 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3334 
3335 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3336 
3337 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
3338 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3339 }
3340 
3341 static int
3342 ena_device_validate_params(struct ena_adapter *adapter,
3343     struct ena_com_dev_get_features_ctx *get_feat_ctx)
3344 {
3345 	if (memcmp(get_feat_ctx->dev_attr.mac_addr, adapter->mac_addr,
3346 	    ETHER_ADDR_LEN) != 0) {
3347 		ena_log(adapter->pdev, ERR, "Error, mac addresses differ\n");
3348 		return (EINVAL);
3349 	}
3350 
3351 	if (get_feat_ctx->dev_attr.max_mtu < if_getmtu(adapter->ifp)) {
3352 		ena_log(adapter->pdev, ERR,
3353 		    "Error, device max mtu is smaller than ifp MTU\n");
3354 		return (EINVAL);
3355 	}
3356 
3357 	return 0;
3358 }
3359 
3360 int
3361 ena_restore_device(struct ena_adapter *adapter)
3362 {
3363 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3364 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3365 	if_t ifp = adapter->ifp;
3366 	device_t dev = adapter->pdev;
3367 	int wd_active;
3368 	int rc;
3369 
3370 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3371 
3372 	rc = ena_device_init(adapter, dev, &get_feat_ctx, &wd_active);
3373 	if (rc != 0) {
3374 		ena_log(dev, ERR, "Cannot initialize device\n");
3375 		goto err;
3376 	}
3377 	/*
3378 	 * Only enable WD if it was enabled before reset, so it won't override
3379 	 * value set by the user by the sysctl.
3380 	 */
3381 	if (adapter->wd_active != 0)
3382 		adapter->wd_active = wd_active;
3383 
3384 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
3385 	if (rc != 0) {
3386 		ena_log(dev, ERR, "Validation of device parameters failed\n");
3387 		goto err_device_destroy;
3388 	}
3389 
3390 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3391 	/* Make sure we don't have a race with AENQ Links state handler */
3392 	if (ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter))
3393 		if_link_state_change(ifp, LINK_STATE_UP);
3394 
3395 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3396 	if (rc != 0) {
3397 		ena_log(dev, ERR, "Enable MSI-X failed\n");
3398 		goto err_device_destroy;
3399 	}
3400 
3401 	/*
3402 	 * Effective value of used MSIX vectors should be the same as before
3403 	 * `ena_destroy_device()`, if possible, or closest to it if less vectors
3404 	 * are available.
3405 	 */
3406 	if ((adapter->msix_vecs - ENA_ADMIN_MSIX_VEC) < adapter->num_io_queues)
3407 		adapter->num_io_queues = adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
3408 
3409 	/* Re-initialize rings basic information */
3410 	ena_init_io_rings(adapter);
3411 
3412 	/* If the interface was up before the reset bring it up */
3413 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter)) {
3414 		rc = ena_up(adapter);
3415 		if (rc != 0) {
3416 			ena_log(dev, ERR, "Failed to create I/O queues\n");
3417 			goto err_disable_msix;
3418 		}
3419 	}
3420 
3421 	/* Indicate that device is running again and ready to work */
3422 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3423 
3424 	/*
3425 	 * As the AENQ handlers weren't executed during reset because
3426 	 * the flag ENA_FLAG_DEVICE_RUNNING was turned off, the
3427 	 * timestamp must be updated again That will prevent next reset
3428 	 * caused by missing keep alive.
3429 	 */
3430 	adapter->keep_alive_timestamp = getsbinuptime();
3431 	ENA_TIMER_RESET(adapter);
3432 
3433 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3434 
3435 	return (rc);
3436 
3437 err_disable_msix:
3438 	ena_free_mgmnt_irq(adapter);
3439 	ena_disable_msix(adapter);
3440 err_device_destroy:
3441 	ena_com_abort_admin_commands(ena_dev);
3442 	ena_com_wait_for_abort_completion(ena_dev);
3443 	ena_com_admin_destroy(ena_dev);
3444 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3445 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3446 err:
3447 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3448 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3449 	ena_log(dev, ERR, "Reset attempt failed. Can not reset the device\n");
3450 
3451 	return (rc);
3452 }
3453 
3454 static void
3455 ena_metrics_task(void *arg, int pending)
3456 {
3457 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3458 
3459 	ENA_LOCK_LOCK();
3460 	(void)ena_copy_eni_metrics(adapter);
3461 	ENA_LOCK_UNLOCK();
3462 }
3463 
3464 static void
3465 ena_reset_task(void *arg, int pending)
3466 {
3467 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3468 
3469 	ENA_LOCK_LOCK();
3470 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
3471 		ena_destroy_device(adapter, false);
3472 		ena_restore_device(adapter);
3473 
3474 		ena_log(adapter->pdev, INFO,
3475 		    "Device reset completed successfully, Driver info: %s\n",
3476 		    ena_version);
3477 	}
3478 	ENA_LOCK_UNLOCK();
3479 }
3480 
3481 static void
3482 ena_free_stats(struct ena_adapter *adapter)
3483 {
3484 	ena_free_counters((counter_u64_t *)&adapter->hw_stats,
3485 	    sizeof(struct ena_hw_stats));
3486 	ena_free_counters((counter_u64_t *)&adapter->dev_stats,
3487 	    sizeof(struct ena_stats_dev));
3488 
3489 }
3490 /**
3491  * ena_attach - Device Initialization Routine
3492  * @pdev: device information struct
3493  *
3494  * Returns 0 on success, otherwise on failure.
3495  *
3496  * ena_attach initializes an adapter identified by a device structure.
3497  * The OS initialization, configuring of the adapter private structure,
3498  * and a hardware reset occur.
3499  **/
3500 static int
3501 ena_attach(device_t pdev)
3502 {
3503 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3504 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
3505 	static int version_printed;
3506 	struct ena_adapter *adapter;
3507 	struct ena_com_dev *ena_dev = NULL;
3508 	uint32_t max_num_io_queues;
3509 	int msix_rid;
3510 	int rid, rc;
3511 
3512 	adapter = device_get_softc(pdev);
3513 	adapter->pdev = pdev;
3514 	adapter->first_bind = -1;
3515 
3516 	/*
3517 	 * Set up the timer service - driver is responsible for avoiding
3518 	 * concurrency, as the callout won't be using any locking inside.
3519 	 */
3520 	ENA_TIMER_INIT(adapter);
3521 	adapter->keep_alive_timeout = ENA_DEFAULT_KEEP_ALIVE_TO;
3522 	adapter->missing_tx_timeout = ENA_DEFAULT_TX_CMP_TO;
3523 	adapter->missing_tx_max_queues = ENA_DEFAULT_TX_MONITORED_QUEUES;
3524 	adapter->missing_tx_threshold = ENA_DEFAULT_TX_CMP_THRESHOLD;
3525 
3526 	if (version_printed++ == 0)
3527 		ena_log(pdev, INFO, "%s\n", ena_version);
3528 
3529 	/* Allocate memory for ena_dev structure */
3530 	ena_dev = malloc(sizeof(struct ena_com_dev), M_DEVBUF,
3531 	    M_WAITOK | M_ZERO);
3532 
3533 	adapter->ena_dev = ena_dev;
3534 	ena_dev->dmadev = pdev;
3535 
3536 	rid = PCIR_BAR(ENA_REG_BAR);
3537 	adapter->memory = NULL;
3538 	adapter->registers = bus_alloc_resource_any(pdev, SYS_RES_MEMORY, &rid,
3539 	    RF_ACTIVE);
3540 	if (unlikely(adapter->registers == NULL)) {
3541 		ena_log(pdev, ERR,
3542 		    "unable to allocate bus resource: registers!\n");
3543 		rc = ENOMEM;
3544 		goto err_dev_free;
3545 	}
3546 
3547 	/* MSIx vector table may reside on BAR0 with registers or on BAR1. */
3548 	msix_rid = pci_msix_table_bar(pdev);
3549 	if (msix_rid != rid) {
3550 		adapter->msix = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
3551 		    &msix_rid, RF_ACTIVE);
3552 		if (unlikely(adapter->msix == NULL)) {
3553 			ena_log(pdev, ERR,
3554 			    "unable to allocate bus resource: msix!\n");
3555 			rc = ENOMEM;
3556 			goto err_pci_free;
3557 		}
3558 		adapter->msix_rid = msix_rid;
3559 	}
3560 
3561 	ena_dev->bus = malloc(sizeof(struct ena_bus), M_DEVBUF,
3562 	    M_WAITOK | M_ZERO);
3563 
3564 	/* Store register resources */
3565 	((struct ena_bus *)(ena_dev->bus))->reg_bar_t = rman_get_bustag(
3566 	    adapter->registers);
3567 	((struct ena_bus *)(ena_dev->bus))->reg_bar_h = rman_get_bushandle(
3568 	    adapter->registers);
3569 
3570 	if (unlikely(((struct ena_bus *)(ena_dev->bus))->reg_bar_h == 0)) {
3571 		ena_log(pdev, ERR, "failed to pmap registers bar\n");
3572 		rc = ENXIO;
3573 		goto err_bus_free;
3574 	}
3575 
3576 	rc = ena_map_llq_mem_bar(pdev, ena_dev);
3577 	if (unlikely(rc != 0)) {
3578 		ena_log(pdev, ERR, "Failed to map ENA mem bar");
3579 		goto err_bus_free;
3580 	}
3581 
3582 	/* Initially clear all the flags */
3583 	ENA_FLAG_ZERO(adapter);
3584 
3585 	/* Device initialization */
3586 	rc = ena_device_init(adapter, pdev, &get_feat_ctx, &adapter->wd_active);
3587 	if (unlikely(rc != 0)) {
3588 		ena_log(pdev, ERR, "ENA device init failed! (err: %d)\n", rc);
3589 		rc = ENXIO;
3590 		goto err_bus_free;
3591 	}
3592 
3593 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3594 		adapter->disable_meta_caching = !!(
3595 		    get_feat_ctx.llq.accel_mode.u.get.supported_flags &
3596 		    BIT(ENA_ADMIN_DISABLE_META_CACHING));
3597 
3598 	adapter->keep_alive_timestamp = getsbinuptime();
3599 
3600 	adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3601 
3602 	memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3603 	    ETHER_ADDR_LEN);
3604 
3605 	calc_queue_ctx.pdev = pdev;
3606 	calc_queue_ctx.ena_dev = ena_dev;
3607 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
3608 
3609 	/* Calculate initial and maximum IO queue number and size */
3610 	max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev,
3611 	    &get_feat_ctx);
3612 	rc = ena_calc_io_queue_size(&calc_queue_ctx);
3613 	if (unlikely((rc != 0) || (max_num_io_queues <= 0))) {
3614 		rc = EFAULT;
3615 		goto err_com_free;
3616 	}
3617 
3618 	adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
3619 	adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
3620 	adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
3621 	adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
3622 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
3623 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
3624 
3625 	adapter->max_num_io_queues = max_num_io_queues;
3626 
3627 	adapter->buf_ring_size = ENA_DEFAULT_BUF_RING_SIZE;
3628 
3629 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
3630 
3631 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3632 
3633 	/* set up dma tags for rx and tx buffers */
3634 	rc = ena_setup_tx_dma_tag(adapter);
3635 	if (unlikely(rc != 0)) {
3636 		ena_log(pdev, ERR, "Failed to create TX DMA tag\n");
3637 		goto err_com_free;
3638 	}
3639 
3640 	rc = ena_setup_rx_dma_tag(adapter);
3641 	if (unlikely(rc != 0)) {
3642 		ena_log(pdev, ERR, "Failed to create RX DMA tag\n");
3643 		goto err_tx_tag_free;
3644 	}
3645 
3646 	/*
3647 	 * The amount of requested MSIX vectors is equal to
3648 	 * adapter::max_num_io_queues (see `ena_enable_msix()`), plus a constant
3649 	 * number of admin queue interrupts. The former is initially determined
3650 	 * by HW capabilities (see `ena_calc_max_io_queue_num())` but may not be
3651 	 * achieved if there are not enough system resources. By default, the
3652 	 * number of effectively used IO queues is the same but later on it can
3653 	 * be limited by the user using sysctl interface.
3654 	 */
3655 	rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3656 	if (unlikely(rc != 0)) {
3657 		ena_log(pdev, ERR,
3658 		    "Failed to enable and set the admin interrupts\n");
3659 		goto err_io_free;
3660 	}
3661 	/* By default all of allocated MSIX vectors are actively used */
3662 	adapter->num_io_queues = adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
3663 
3664 	/* initialize rings basic information */
3665 	ena_init_io_rings(adapter);
3666 
3667 	/* Initialize statistics */
3668 	ena_alloc_counters((counter_u64_t *)&adapter->dev_stats,
3669 	    sizeof(struct ena_stats_dev));
3670 	ena_alloc_counters((counter_u64_t *)&adapter->hw_stats,
3671 	    sizeof(struct ena_hw_stats));
3672 	ena_sysctl_add_nodes(adapter);
3673 
3674 	/* setup network interface */
3675 	rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
3676 	if (unlikely(rc != 0)) {
3677 		ena_log(pdev, ERR, "Error with network interface setup\n");
3678 		goto err_msix_free;
3679 	}
3680 
3681 	/* Initialize reset task queue */
3682 	TASK_INIT(&adapter->reset_task, 0, ena_reset_task, adapter);
3683 	adapter->reset_tq = taskqueue_create("ena_reset_enqueue",
3684 	    M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->reset_tq);
3685 	taskqueue_start_threads(&adapter->reset_tq, 1, PI_NET, "%s rstq",
3686 	    device_get_nameunit(adapter->pdev));
3687 
3688 	/* Initialize metrics task queue */
3689 	TASK_INIT(&adapter->metrics_task, 0, ena_metrics_task, adapter);
3690 	adapter->metrics_tq = taskqueue_create("ena_metrics_enqueue",
3691 	    M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->metrics_tq);
3692 	taskqueue_start_threads(&adapter->metrics_tq, 1, PI_NET, "%s metricsq",
3693 	    device_get_nameunit(adapter->pdev));
3694 
3695 #ifdef DEV_NETMAP
3696 	rc = ena_netmap_attach(adapter);
3697 	if (rc != 0) {
3698 		ena_log(pdev, ERR, "netmap attach failed: %d\n", rc);
3699 		goto err_detach;
3700 	}
3701 #endif /* DEV_NETMAP */
3702 
3703 	/* Tell the stack that the interface is not active */
3704 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
3705 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3706 
3707 	/* Run the timer service */
3708 	ENA_TIMER_RESET(adapter);
3709 
3710 	return (0);
3711 
3712 #ifdef DEV_NETMAP
3713 err_detach:
3714 	ether_ifdetach(adapter->ifp);
3715 #endif /* DEV_NETMAP */
3716 err_msix_free:
3717 	ena_free_stats(adapter);
3718 	ena_com_dev_reset(adapter->ena_dev, ENA_REGS_RESET_INIT_ERR);
3719 	ena_free_mgmnt_irq(adapter);
3720 	ena_disable_msix(adapter);
3721 err_io_free:
3722 	ena_free_all_io_rings_resources(adapter);
3723 	ena_free_rx_dma_tag(adapter);
3724 err_tx_tag_free:
3725 	ena_free_tx_dma_tag(adapter);
3726 err_com_free:
3727 	ena_com_admin_destroy(ena_dev);
3728 	ena_com_delete_host_info(ena_dev);
3729 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3730 err_bus_free:
3731 	free(ena_dev->bus, M_DEVBUF);
3732 err_pci_free:
3733 	ena_free_pci_resources(adapter);
3734 err_dev_free:
3735 	free(ena_dev, M_DEVBUF);
3736 
3737 	return (rc);
3738 }
3739 
3740 /**
3741  * ena_detach - Device Removal Routine
3742  * @pdev: device information struct
3743  *
3744  * ena_detach is called by the device subsystem to alert the driver
3745  * that it should release a PCI device.
3746  **/
3747 static int
3748 ena_detach(device_t pdev)
3749 {
3750 	struct ena_adapter *adapter = device_get_softc(pdev);
3751 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3752 	int rc;
3753 
3754 	/* Make sure VLANS are not using driver */
3755 	if (if_vlantrunkinuse(adapter->ifp)) {
3756 		ena_log(adapter->pdev, ERR, "VLAN is in use, detach first\n");
3757 		return (EBUSY);
3758 	}
3759 
3760 	ether_ifdetach(adapter->ifp);
3761 
3762 	/* Stop timer service */
3763 	ENA_LOCK_LOCK();
3764 	ENA_TIMER_DRAIN(adapter);
3765 	ENA_LOCK_UNLOCK();
3766 
3767 	/* Release metrics task */
3768 	while (taskqueue_cancel(adapter->metrics_tq, &adapter->metrics_task, NULL))
3769 		taskqueue_drain(adapter->metrics_tq, &adapter->metrics_task);
3770 	taskqueue_free(adapter->metrics_tq);
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();
3778 	ena_down(adapter);
3779 	ena_destroy_device(adapter, true);
3780 	ENA_LOCK_UNLOCK();
3781 
3782 	/* Restore unregistered sysctl queue nodes. */
3783 	ena_sysctl_update_queue_node_nb(adapter, adapter->num_io_queues,
3784 	    adapter->max_num_io_queues);
3785 
3786 #ifdef DEV_NETMAP
3787 	netmap_detach(adapter->ifp);
3788 #endif /* DEV_NETMAP */
3789 
3790 	ena_free_stats(adapter);
3791 
3792 	rc = ena_free_rx_dma_tag(adapter);
3793 	if (unlikely(rc != 0))
3794 		ena_log(adapter->pdev, WARN,
3795 		    "Unmapped RX DMA tag associations\n");
3796 
3797 	rc = ena_free_tx_dma_tag(adapter);
3798 	if (unlikely(rc != 0))
3799 		ena_log(adapter->pdev, WARN,
3800 		    "Unmapped TX DMA tag associations\n");
3801 
3802 	ena_free_irqs(adapter);
3803 
3804 	ena_free_pci_resources(adapter);
3805 
3806 	if (adapter->rss_indir != NULL)
3807 		free(adapter->rss_indir, M_DEVBUF);
3808 
3809 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter)))
3810 		ena_com_rss_destroy(ena_dev);
3811 
3812 	ena_com_delete_host_info(ena_dev);
3813 
3814 	if_free(adapter->ifp);
3815 
3816 	free(ena_dev->bus, M_DEVBUF);
3817 
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 		ena_log(adapter->pdev, INFO, "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 		ena_log(adapter->pdev, INFO, "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
3857 ena_notification(void *adapter_data, 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(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3863 	    adapter->ena_dev, "Invalid group(%x) expected %x\n",
3864 	    aenq_e->aenq_common_desc.group, 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 		ena_log(adapter->pdev, ERR,
3874 		    "Invalid aenq notification link state %d\n",
3875 		    aenq_e->aenq_common_desc.syndrome);
3876 	}
3877 }
3878 
3879 static void
3880 ena_lock_init(void *arg)
3881 {
3882 	ENA_LOCK_INIT();
3883 }
3884 SYSINIT(ena_lock_init, SI_SUB_LOCK, SI_ORDER_FIRST, ena_lock_init, NULL);
3885 
3886 static void
3887 ena_lock_uninit(void *arg)
3888 {
3889 	ENA_LOCK_DESTROY();
3890 }
3891 SYSUNINIT(ena_lock_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, ena_lock_uninit, NULL);
3892 
3893 /**
3894  * This handler will called for unknown event group or unimplemented handlers
3895  **/
3896 static void
3897 unimplemented_aenq_handler(void *adapter_data,
3898     struct ena_admin_aenq_entry *aenq_e)
3899 {
3900 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3901 
3902 	ena_log(adapter->pdev, ERR,
3903 	    "Unknown event was received or event with unimplemented handler\n");
3904 }
3905 
3906 static struct ena_aenq_handlers aenq_handlers = {
3907     .handlers = {
3908 	    [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3909 	    [ENA_ADMIN_NOTIFICATION] = ena_notification,
3910 	    [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3911     },
3912     .unimplemented_handler = unimplemented_aenq_handler
3913 };
3914 
3915 /*********************************************************************
3916  *  FreeBSD Device Interface Entry Points
3917  *********************************************************************/
3918 
3919 static device_method_t ena_methods[] = { /* Device interface */
3920 	DEVMETHOD(device_probe, ena_probe),
3921 	DEVMETHOD(device_attach, ena_attach),
3922 	DEVMETHOD(device_detach, ena_detach), DEVMETHOD_END
3923 };
3924 
3925 static driver_t ena_driver = {
3926 	"ena",
3927 	ena_methods,
3928 	sizeof(struct ena_adapter),
3929 };
3930 
3931 DRIVER_MODULE(ena, pci, ena_driver, 0, 0);
3932 MODULE_PNP_INFO("U16:vendor;U16:device", pci, ena, ena_vendor_info_array,
3933     nitems(ena_vendor_info_array) - 1);
3934 MODULE_DEPEND(ena, pci, 1, 1, 1);
3935 MODULE_DEPEND(ena, ether, 1, 1, 1);
3936 #ifdef DEV_NETMAP
3937 MODULE_DEPEND(ena, netmap, 1, 1, 1);
3938 #endif /* DEV_NETMAP */
3939 
3940 /*********************************************************************/
3941