xref: /freebsd/sys/dev/ena/ena.c (revision ce6a89e27cd190313be39bb479880aeda4778436)
1 /*-
2  * BSD LICENSE
3  *
4  * Copyright (c) 2015-2019 Amazon.com, Inc. or its affiliates.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  *
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
23  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
24  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
28  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/bus.h>
36 #include <sys/endian.h>
37 #include <sys/kernel.h>
38 #include <sys/kthread.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/rman.h>
43 #include <sys/smp.h>
44 #include <sys/socket.h>
45 #include <sys/sockio.h>
46 #include <sys/sysctl.h>
47 #include <sys/taskqueue.h>
48 #include <sys/time.h>
49 #include <sys/eventhandler.h>
50 
51 #include <machine/bus.h>
52 #include <machine/resource.h>
53 #include <machine/in_cksum.h>
54 
55 #include <net/bpf.h>
56 #include <net/ethernet.h>
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_arp.h>
60 #include <net/if_dl.h>
61 #include <net/if_media.h>
62 #include <net/if_types.h>
63 #include <net/if_vlan_var.h>
64 
65 #include <netinet/in_systm.h>
66 #include <netinet/in.h>
67 #include <netinet/if_ether.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip6.h>
70 #include <netinet/tcp.h>
71 #include <netinet/udp.h>
72 
73 #include <dev/pci/pcivar.h>
74 #include <dev/pci/pcireg.h>
75 
76 #include <vm/vm.h>
77 #include <vm/pmap.h>
78 
79 #include "ena_datapath.h"
80 #include "ena.h"
81 #include "ena_sysctl.h"
82 
83 #ifdef DEV_NETMAP
84 #include "ena_netmap.h"
85 #endif /* DEV_NETMAP */
86 
87 /*********************************************************
88  *  Function prototypes
89  *********************************************************/
90 static int	ena_probe(device_t);
91 static void	ena_intr_msix_mgmnt(void *);
92 static void	ena_free_pci_resources(struct ena_adapter *);
93 static int	ena_change_mtu(if_t, int);
94 static inline void ena_alloc_counters(counter_u64_t *, int);
95 static inline void ena_free_counters(counter_u64_t *, int);
96 static inline void ena_reset_counters(counter_u64_t *, int);
97 static void	ena_init_io_rings_common(struct ena_adapter *,
98     struct ena_ring *, uint16_t);
99 static void	ena_init_io_rings(struct ena_adapter *);
100 static void	ena_free_io_ring_resources(struct ena_adapter *, unsigned int);
101 static void	ena_free_all_io_rings_resources(struct ena_adapter *);
102 static int	ena_setup_tx_dma_tag(struct ena_adapter *);
103 static int	ena_free_tx_dma_tag(struct ena_adapter *);
104 static int	ena_setup_rx_dma_tag(struct ena_adapter *);
105 static int	ena_free_rx_dma_tag(struct ena_adapter *);
106 static void	ena_release_all_tx_dmamap(struct ena_ring *);
107 static int	ena_setup_tx_resources(struct ena_adapter *, int);
108 static void	ena_free_tx_resources(struct ena_adapter *, int);
109 static int	ena_setup_all_tx_resources(struct ena_adapter *);
110 static void	ena_free_all_tx_resources(struct ena_adapter *);
111 static int	ena_setup_rx_resources(struct ena_adapter *, unsigned int);
112 static void	ena_free_rx_resources(struct ena_adapter *, unsigned int);
113 static int	ena_setup_all_rx_resources(struct ena_adapter *);
114 static void	ena_free_all_rx_resources(struct ena_adapter *);
115 static inline int ena_alloc_rx_mbuf(struct ena_adapter *, struct ena_ring *,
116     struct ena_rx_buffer *);
117 static void	ena_free_rx_mbuf(struct ena_adapter *, struct ena_ring *,
118     struct ena_rx_buffer *);
119 static void	ena_free_rx_bufs(struct ena_adapter *, unsigned int);
120 static void	ena_refill_all_rx_bufs(struct ena_adapter *);
121 static void	ena_free_all_rx_bufs(struct ena_adapter *);
122 static void	ena_free_tx_bufs(struct ena_adapter *, unsigned int);
123 static void	ena_free_all_tx_bufs(struct ena_adapter *);
124 static void	ena_destroy_all_tx_queues(struct ena_adapter *);
125 static void	ena_destroy_all_rx_queues(struct ena_adapter *);
126 static void	ena_destroy_all_io_queues(struct ena_adapter *);
127 static int	ena_create_io_queues(struct ena_adapter *);
128 static int	ena_handle_msix(void *);
129 static int	ena_enable_msix(struct ena_adapter *);
130 static void	ena_setup_mgmnt_intr(struct ena_adapter *);
131 static int	ena_setup_io_intr(struct ena_adapter *);
132 static int	ena_request_mgmnt_irq(struct ena_adapter *);
133 static int	ena_request_io_irq(struct ena_adapter *);
134 static void	ena_free_mgmnt_irq(struct ena_adapter *);
135 static void	ena_free_io_irq(struct ena_adapter *);
136 static void	ena_free_irqs(struct ena_adapter*);
137 static void	ena_disable_msix(struct ena_adapter *);
138 static void	ena_unmask_all_io_irqs(struct ena_adapter *);
139 static int	ena_rss_configure(struct ena_adapter *);
140 static int	ena_up_complete(struct ena_adapter *);
141 static uint64_t	ena_get_counter(if_t, ift_counter);
142 static int	ena_media_change(if_t);
143 static void	ena_media_status(if_t, struct ifmediareq *);
144 static void	ena_init(void *);
145 static int	ena_ioctl(if_t, u_long, caddr_t);
146 static int	ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *);
147 static void	ena_update_host_info(struct ena_admin_host_info *, if_t);
148 static void	ena_update_hwassist(struct ena_adapter *);
149 static int	ena_setup_ifnet(device_t, struct ena_adapter *,
150     struct ena_com_dev_get_features_ctx *);
151 static int	ena_enable_wc(struct resource *);
152 static int	ena_set_queues_placement_policy(device_t, struct ena_com_dev *,
153     struct ena_admin_feature_llq_desc *, struct ena_llq_configurations *);
154 static int	ena_calc_io_queue_num(struct ena_adapter *,
155     struct ena_com_dev_get_features_ctx *);
156 static int	ena_calc_queue_size(struct ena_adapter *,
157     struct ena_calc_queue_size_ctx *);
158 static int	ena_handle_updated_queues(struct ena_adapter *,
159     struct ena_com_dev_get_features_ctx *);
160 static int	ena_rss_init_default(struct ena_adapter *);
161 static void	ena_rss_init_default_deferred(void *);
162 static void	ena_config_host_info(struct ena_com_dev *, device_t);
163 static int	ena_attach(device_t);
164 static int	ena_detach(device_t);
165 static int	ena_device_init(struct ena_adapter *, device_t,
166     struct ena_com_dev_get_features_ctx *, int *);
167 static int	ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *,
168     int);
169 static void ena_update_on_link_change(void *, struct ena_admin_aenq_entry *);
170 static void	unimplemented_aenq_handler(void *,
171     struct ena_admin_aenq_entry *);
172 static void	ena_timer_service(void *);
173 
174 static char ena_version[] = DEVICE_NAME DRV_MODULE_NAME " v" DRV_MODULE_VERSION;
175 
176 static ena_vendor_info_t ena_vendor_info_array[] = {
177     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0},
178     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_PF, 0},
179     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0},
180     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_VF, 0},
181     /* Last entry */
182     { 0, 0, 0 }
183 };
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,
200     ena_mem_handle_t *dma , int mapflags)
201 {
202 	struct ena_adapter* adapter = device_get_softc(dmadev);
203 	uint32_t maxsize;
204 	uint64_t dma_space_addr;
205 	int error;
206 
207 	maxsize = ((size - 1) / PAGE_SIZE + 1) * PAGE_SIZE;
208 
209 	dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
210 	if (unlikely(dma_space_addr == 0))
211 		dma_space_addr = BUS_SPACE_MAXADDR;
212 
213 	error = bus_dma_tag_create(bus_get_dma_tag(dmadev), /* parent */
214 	    8, 0,	      /* alignment, bounds 		*/
215 	    dma_space_addr,   /* lowaddr of exclusion window	*/
216 	    BUS_SPACE_MAXADDR,/* highaddr of exclusion window	*/
217 	    NULL, NULL,	      /* filter, filterarg 		*/
218 	    maxsize,	      /* maxsize 			*/
219 	    1,		      /* nsegments 			*/
220 	    maxsize,	      /* maxsegsize 			*/
221 	    BUS_DMA_ALLOCNOW, /* flags 				*/
222 	    NULL,	      /* lockfunc 			*/
223 	    NULL,	      /* lockarg 			*/
224 	    &dma->tag);
225 	if (unlikely(error != 0)) {
226 		ena_trace(ENA_ALERT, "bus_dma_tag_create failed: %d\n", error);
227 		goto fail_tag;
228 	}
229 
230 	error = bus_dmamem_alloc(dma->tag, (void**) &dma->vaddr,
231 	    BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->map);
232 	if (unlikely(error != 0)) {
233 		ena_trace(ENA_ALERT, "bus_dmamem_alloc(%ju) failed: %d\n",
234 		    (uintmax_t)size, error);
235 		goto fail_map_create;
236 	}
237 
238 	dma->paddr = 0;
239 	error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr,
240 	    size, ena_dmamap_callback, &dma->paddr, mapflags);
241 	if (unlikely((error != 0) || (dma->paddr == 0))) {
242 		ena_trace(ENA_ALERT, ": bus_dmamap_load failed: %d\n", error);
243 		goto fail_map_load;
244 	}
245 
246 	bus_dmamap_sync(dma->tag, dma->map,
247 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
248 
249 	return (0);
250 
251 fail_map_load:
252 	bus_dmamem_free(dma->tag, dma->vaddr, dma->map);
253 fail_map_create:
254 	bus_dma_tag_destroy(dma->tag);
255 fail_tag:
256 	dma->tag = NULL;
257 	dma->vaddr = NULL;
258 	dma->paddr = 0;
259 
260 	return (error);
261 }
262 
263 static void
264 ena_free_pci_resources(struct ena_adapter *adapter)
265 {
266 	device_t pdev = adapter->pdev;
267 
268 	if (adapter->memory != NULL) {
269 		bus_release_resource(pdev, SYS_RES_MEMORY,
270 		    PCIR_BAR(ENA_MEM_BAR), adapter->memory);
271 	}
272 
273 	if (adapter->registers != NULL) {
274 		bus_release_resource(pdev, SYS_RES_MEMORY,
275 		    PCIR_BAR(ENA_REG_BAR), adapter->registers);
276 	}
277 }
278 
279 static int
280 ena_probe(device_t dev)
281 {
282 	ena_vendor_info_t *ent;
283 	char		adapter_name[60];
284 	uint16_t	pci_vendor_id = 0;
285 	uint16_t	pci_device_id = 0;
286 
287 	pci_vendor_id = pci_get_vendor(dev);
288 	pci_device_id = pci_get_device(dev);
289 
290 	ent = ena_vendor_info_array;
291 	while (ent->vendor_id != 0) {
292 		if ((pci_vendor_id == ent->vendor_id) &&
293 		    (pci_device_id == ent->device_id)) {
294 			ena_trace(ENA_DBG, "vendor=%x device=%x\n",
295 			    pci_vendor_id, pci_device_id);
296 
297 			sprintf(adapter_name, DEVICE_DESC);
298 			device_set_desc_copy(dev, adapter_name);
299 			return (BUS_PROBE_DEFAULT);
300 		}
301 
302 		ent++;
303 
304 	}
305 
306 	return (ENXIO);
307 }
308 
309 static int
310 ena_change_mtu(if_t ifp, int new_mtu)
311 {
312 	struct ena_adapter *adapter = if_getsoftc(ifp);
313 	int rc;
314 
315 	if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
316 		device_printf(adapter->pdev, "Invalid MTU setting. "
317 		    "new_mtu: %d max mtu: %d min mtu: %d\n",
318 		    new_mtu, adapter->max_mtu, ENA_MIN_MTU);
319 		return (EINVAL);
320 	}
321 
322 	rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
323 	if (likely(rc == 0)) {
324 		ena_trace(ENA_DBG, "set MTU to %d\n", new_mtu);
325 		if_setmtu(ifp, new_mtu);
326 	} else {
327 		device_printf(adapter->pdev, "Failed to set MTU to %d\n",
328 		    new_mtu);
329 	}
330 
331 	return (rc);
332 }
333 
334 static inline void
335 ena_alloc_counters(counter_u64_t *begin, int size)
336 {
337 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
338 
339 	for (; begin < end; ++begin)
340 		*begin = counter_u64_alloc(M_WAITOK);
341 }
342 
343 static inline void
344 ena_free_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 		counter_u64_free(*begin);
350 }
351 
352 static inline void
353 ena_reset_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_zero(*begin);
359 }
360 
361 static void
362 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
363     uint16_t qid)
364 {
365 
366 	ring->qid = qid;
367 	ring->adapter = adapter;
368 	ring->ena_dev = adapter->ena_dev;
369 	ring->first_interrupt = false;
370 	ring->no_interrupt_event_cnt = 0;
371 }
372 
373 static void
374 ena_init_io_rings(struct ena_adapter *adapter)
375 {
376 	struct ena_com_dev *ena_dev;
377 	struct ena_ring *txr, *rxr;
378 	struct ena_que *que;
379 	int i;
380 
381 	ena_dev = adapter->ena_dev;
382 
383 	for (i = 0; i < adapter->num_queues; i++) {
384 		txr = &adapter->tx_ring[i];
385 		rxr = &adapter->rx_ring[i];
386 
387 		/* TX/RX common ring state */
388 		ena_init_io_rings_common(adapter, txr, i);
389 		ena_init_io_rings_common(adapter, rxr, i);
390 
391 		/* TX specific ring state */
392 		txr->ring_size = adapter->tx_ring_size;
393 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
394 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
395 		txr->smoothed_interval =
396 		    ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
397 
398 		/* Allocate a buf ring */
399 		txr->buf_ring_size = adapter->buf_ring_size;
400 		txr->br = buf_ring_alloc(txr->buf_ring_size, M_DEVBUF,
401 		    M_WAITOK, &txr->ring_mtx);
402 
403 		/* Alloc TX statistics. */
404 		ena_alloc_counters((counter_u64_t *)&txr->tx_stats,
405 		    sizeof(txr->tx_stats));
406 
407 		/* RX specific ring state */
408 		rxr->ring_size = adapter->rx_ring_size;
409 		rxr->smoothed_interval =
410 		    ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
411 
412 		/* Alloc RX statistics. */
413 		ena_alloc_counters((counter_u64_t *)&rxr->rx_stats,
414 		    sizeof(rxr->rx_stats));
415 
416 		/* Initialize locks */
417 		snprintf(txr->mtx_name, nitems(txr->mtx_name), "%s:tx(%d)",
418 		    device_get_nameunit(adapter->pdev), i);
419 		snprintf(rxr->mtx_name, nitems(rxr->mtx_name), "%s:rx(%d)",
420 		    device_get_nameunit(adapter->pdev), i);
421 
422 		mtx_init(&txr->ring_mtx, txr->mtx_name, NULL, MTX_DEF);
423 
424 		que = &adapter->que[i];
425 		que->adapter = adapter;
426 		que->id = i;
427 		que->tx_ring = txr;
428 		que->rx_ring = rxr;
429 
430 		txr->que = que;
431 		rxr->que = que;
432 
433 		rxr->empty_rx_queue = 0;
434 	}
435 }
436 
437 static void
438 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
439 {
440 	struct ena_ring *txr = &adapter->tx_ring[qid];
441 	struct ena_ring *rxr = &adapter->rx_ring[qid];
442 
443 	ena_free_counters((counter_u64_t *)&txr->tx_stats,
444 	    sizeof(txr->tx_stats));
445 	ena_free_counters((counter_u64_t *)&rxr->rx_stats,
446 	    sizeof(rxr->rx_stats));
447 
448 	ENA_RING_MTX_LOCK(txr);
449 	drbr_free(txr->br, M_DEVBUF);
450 	ENA_RING_MTX_UNLOCK(txr);
451 
452 	mtx_destroy(&txr->ring_mtx);
453 }
454 
455 static void
456 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
457 {
458 	int i;
459 
460 	for (i = 0; i < adapter->num_queues; i++)
461 		ena_free_io_ring_resources(adapter, i);
462 
463 }
464 
465 static int
466 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
467 {
468 	int ret;
469 
470 	/* Create DMA tag for Tx buffers */
471 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
472 	    1, 0,				  /* alignment, bounds 	     */
473 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
474 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
475 	    NULL, NULL,				  /* filter, filterarg 	     */
476 	    ENA_TSO_MAXSIZE,			  /* maxsize 		     */
477 	    adapter->max_tx_sgl_size - 1,	  /* nsegments 		     */
478 	    ENA_TSO_MAXSIZE,			  /* maxsegsize 	     */
479 	    0,					  /* flags 		     */
480 	    NULL,				  /* lockfunc 		     */
481 	    NULL,				  /* lockfuncarg 	     */
482 	    &adapter->tx_buf_tag);
483 
484 	return (ret);
485 }
486 
487 static int
488 ena_free_tx_dma_tag(struct ena_adapter *adapter)
489 {
490 	int ret;
491 
492 	ret = bus_dma_tag_destroy(adapter->tx_buf_tag);
493 
494 	if (likely(ret == 0))
495 		adapter->tx_buf_tag = NULL;
496 
497 	return (ret);
498 }
499 
500 static int
501 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
502 {
503 	int ret;
504 
505 	/* Create DMA tag for Rx buffers*/
506 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent   */
507 	    1, 0,				  /* alignment, bounds 	     */
508 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
509 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
510 	    NULL, NULL,				  /* filter, filterarg 	     */
511 	    MJUM16BYTES,			  /* maxsize 		     */
512 	    adapter->max_rx_sgl_size,		  /* nsegments 		     */
513 	    MJUM16BYTES,			  /* maxsegsize 	     */
514 	    0,					  /* flags 		     */
515 	    NULL,				  /* lockfunc 		     */
516 	    NULL,				  /* lockarg 		     */
517 	    &adapter->rx_buf_tag);
518 
519 	return (ret);
520 }
521 
522 static int
523 ena_free_rx_dma_tag(struct ena_adapter *adapter)
524 {
525 	int ret;
526 
527 	ret = bus_dma_tag_destroy(adapter->rx_buf_tag);
528 
529 	if (likely(ret == 0))
530 		adapter->rx_buf_tag = NULL;
531 
532 	return (ret);
533 }
534 
535 static void
536 ena_release_all_tx_dmamap(struct ena_ring *tx_ring)
537 {
538 	struct ena_adapter *adapter = tx_ring->adapter;
539 	struct ena_tx_buffer *tx_info;
540 	bus_dma_tag_t tx_tag = adapter->tx_buf_tag;;
541 	int i;
542 #ifdef DEV_NETMAP
543 	struct ena_netmap_tx_info *nm_info;
544 	int j;
545 #endif /* DEV_NETMAP */
546 
547 	for (i = 0; i < tx_ring->ring_size; ++i) {
548 		tx_info = &tx_ring->tx_buffer_info[i];
549 #ifdef DEV_NETMAP
550 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
551 			nm_info = &tx_info->nm_info;
552 			for (j = 0; j < ENA_PKT_MAX_BUFS; ++j) {
553 				if (nm_info->map_seg[j] != NULL) {
554 					bus_dmamap_destroy(tx_tag,
555 					    nm_info->map_seg[j]);
556 					nm_info->map_seg[j] = NULL;
557 				}
558 			}
559 		}
560 #endif /* DEV_NETMAP */
561 		if (tx_info->dmamap != NULL) {
562 			bus_dmamap_destroy(tx_tag, tx_info->dmamap);
563 			tx_info->dmamap = NULL;
564 		}
565 	}
566 }
567 
568 /**
569  * ena_setup_tx_resources - allocate Tx resources (Descriptors)
570  * @adapter: network interface device structure
571  * @qid: queue index
572  *
573  * Returns 0 on success, otherwise on failure.
574  **/
575 static int
576 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
577 {
578 	struct ena_que *que = &adapter->que[qid];
579 	struct ena_ring *tx_ring = que->tx_ring;
580 	int size, i, err;
581 #ifdef DEV_NETMAP
582 	bus_dmamap_t *map;
583 	int j;
584 
585 	ena_netmap_reset_tx_ring(adapter, qid);
586 #endif /* DEV_NETMAP */
587 
588 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
589 
590 	tx_ring->tx_buffer_info = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
591 	if (unlikely(tx_ring->tx_buffer_info == NULL))
592 		return (ENOMEM);
593 
594 	size = sizeof(uint16_t) * tx_ring->ring_size;
595 	tx_ring->free_tx_ids = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
596 	if (unlikely(tx_ring->free_tx_ids == NULL))
597 		goto err_buf_info_free;
598 
599 	size = tx_ring->tx_max_header_size;
600 	tx_ring->push_buf_intermediate_buf = malloc(size, M_DEVBUF,
601 	    M_NOWAIT | M_ZERO);
602 	if (unlikely(tx_ring->push_buf_intermediate_buf == NULL))
603 		goto err_tx_ids_free;
604 
605 	/* Req id stack for TX OOO completions */
606 	for (i = 0; i < tx_ring->ring_size; i++)
607 		tx_ring->free_tx_ids[i] = i;
608 
609 	/* Reset TX statistics. */
610 	ena_reset_counters((counter_u64_t *)&tx_ring->tx_stats,
611 	    sizeof(tx_ring->tx_stats));
612 
613 	tx_ring->next_to_use = 0;
614 	tx_ring->next_to_clean = 0;
615 	tx_ring->acum_pkts = 0;
616 
617 	/* Make sure that drbr is empty */
618 	ENA_RING_MTX_LOCK(tx_ring);
619 	drbr_flush(adapter->ifp, tx_ring->br);
620 	ENA_RING_MTX_UNLOCK(tx_ring);
621 
622 	/* ... and create the buffer DMA maps */
623 	for (i = 0; i < tx_ring->ring_size; i++) {
624 		err = bus_dmamap_create(adapter->tx_buf_tag, 0,
625 		    &tx_ring->tx_buffer_info[i].dmamap);
626 		if (unlikely(err != 0)) {
627 			ena_trace(ENA_ALERT,
628 			    "Unable to create Tx DMA map for buffer %d\n",
629 			    i);
630 			goto err_map_release;
631 		}
632 
633 #ifdef DEV_NETMAP
634 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
635 			map = tx_ring->tx_buffer_info[i].nm_info.map_seg;
636 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
637 				err = bus_dmamap_create(adapter->tx_buf_tag, 0,
638 				    &map[j]);
639 				if (unlikely(err != 0)) {
640 					ena_trace(ENA_ALERT, "Unable to create "
641 					    "Tx DMA for buffer %d %d\n", i, j);
642 					goto err_map_release;
643 				}
644 			}
645 		}
646 #endif /* DEV_NETMAP */
647 	}
648 
649 	/* Allocate taskqueues */
650 	TASK_INIT(&tx_ring->enqueue_task, 0, ena_deferred_mq_start, tx_ring);
651 	tx_ring->enqueue_tq = taskqueue_create_fast("ena_tx_enque", M_NOWAIT,
652 	    taskqueue_thread_enqueue, &tx_ring->enqueue_tq);
653 	if (unlikely(tx_ring->enqueue_tq == NULL)) {
654 		ena_trace(ENA_ALERT,
655 		    "Unable to create taskqueue for enqueue task\n");
656 		i = tx_ring->ring_size;
657 		goto err_map_release;
658 	}
659 
660 	tx_ring->running = true;
661 
662 	taskqueue_start_threads(&tx_ring->enqueue_tq, 1, PI_NET,
663 	    "%s txeq %d", device_get_nameunit(adapter->pdev), que->cpu);
664 
665 	return (0);
666 
667 err_map_release:
668 	ena_release_all_tx_dmamap(tx_ring);
669 err_tx_ids_free:
670 	free(tx_ring->free_tx_ids, M_DEVBUF);
671 	tx_ring->free_tx_ids = NULL;
672 err_buf_info_free:
673 	free(tx_ring->tx_buffer_info, M_DEVBUF);
674 	tx_ring->tx_buffer_info = NULL;
675 
676 	return (ENOMEM);
677 }
678 
679 /**
680  * ena_free_tx_resources - Free Tx Resources per Queue
681  * @adapter: network interface device structure
682  * @qid: queue index
683  *
684  * Free all transmit software resources
685  **/
686 static void
687 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
688 {
689 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
690 #ifdef DEV_NETMAP
691 	struct ena_netmap_tx_info *nm_info;
692 	int j;
693 #endif /* DEV_NETMAP */
694 
695 	while (taskqueue_cancel(tx_ring->enqueue_tq, &tx_ring->enqueue_task,
696 	    NULL))
697 		taskqueue_drain(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
698 
699 	taskqueue_free(tx_ring->enqueue_tq);
700 
701 	ENA_RING_MTX_LOCK(tx_ring);
702 	/* Flush buffer ring, */
703 	drbr_flush(adapter->ifp, tx_ring->br);
704 
705 	/* Free buffer DMA maps, */
706 	for (int i = 0; i < tx_ring->ring_size; i++) {
707 		bus_dmamap_sync(adapter->tx_buf_tag,
708 		    tx_ring->tx_buffer_info[i].dmamap, BUS_DMASYNC_POSTWRITE);
709 		bus_dmamap_unload(adapter->tx_buf_tag,
710 		    tx_ring->tx_buffer_info[i].dmamap);
711 		bus_dmamap_destroy(adapter->tx_buf_tag,
712 		    tx_ring->tx_buffer_info[i].dmamap);
713 
714 #ifdef DEV_NETMAP
715 		if (adapter->ifp->if_capenable & IFCAP_NETMAP) {
716 			nm_info = &tx_ring->tx_buffer_info[i].nm_info;
717 			for (j = 0; j < ENA_PKT_MAX_BUFS; j++) {
718 				if (nm_info->socket_buf_idx[j] != 0) {
719 					bus_dmamap_sync(adapter->tx_buf_tag,
720 					    nm_info->map_seg[j],
721 					    BUS_DMASYNC_POSTWRITE);
722 					ena_netmap_unload(adapter,
723 					    nm_info->map_seg[j]);
724 				}
725 				bus_dmamap_destroy(adapter->tx_buf_tag,
726 				    nm_info->map_seg[j]);
727 				nm_info->socket_buf_idx[j] = 0;
728 			}
729 		}
730 #endif /* DEV_NETMAP */
731 
732 		m_freem(tx_ring->tx_buffer_info[i].mbuf);
733 		tx_ring->tx_buffer_info[i].mbuf = NULL;
734 	}
735 	ENA_RING_MTX_UNLOCK(tx_ring);
736 
737 	/* And free allocated memory. */
738 	free(tx_ring->tx_buffer_info, M_DEVBUF);
739 	tx_ring->tx_buffer_info = NULL;
740 
741 	free(tx_ring->free_tx_ids, M_DEVBUF);
742 	tx_ring->free_tx_ids = NULL;
743 
744 	ENA_MEM_FREE(adapter->ena_dev->dmadev,
745 	    tx_ring->push_buf_intermediate_buf);
746 	tx_ring->push_buf_intermediate_buf = NULL;
747 }
748 
749 /**
750  * ena_setup_all_tx_resources - allocate all queues Tx resources
751  * @adapter: network interface device structure
752  *
753  * Returns 0 on success, otherwise on failure.
754  **/
755 static int
756 ena_setup_all_tx_resources(struct ena_adapter *adapter)
757 {
758 	int i, rc;
759 
760 	for (i = 0; i < adapter->num_queues; i++) {
761 		rc = ena_setup_tx_resources(adapter, i);
762 		if (rc != 0) {
763 			device_printf(adapter->pdev,
764 			    "Allocation for Tx Queue %u failed\n", i);
765 			goto err_setup_tx;
766 		}
767 	}
768 
769 	return (0);
770 
771 err_setup_tx:
772 	/* Rewind the index freeing the rings as we go */
773 	while (i--)
774 		ena_free_tx_resources(adapter, i);
775 	return (rc);
776 }
777 
778 /**
779  * ena_free_all_tx_resources - Free Tx Resources for All Queues
780  * @adapter: network interface device structure
781  *
782  * Free all transmit software resources
783  **/
784 static void
785 ena_free_all_tx_resources(struct ena_adapter *adapter)
786 {
787 	int i;
788 
789 	for (i = 0; i < adapter->num_queues; i++)
790 		ena_free_tx_resources(adapter, i);
791 }
792 
793 /**
794  * ena_setup_rx_resources - allocate Rx resources (Descriptors)
795  * @adapter: network interface device structure
796  * @qid: queue index
797  *
798  * Returns 0 on success, otherwise on failure.
799  **/
800 static int
801 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
802 {
803 	struct ena_que *que = &adapter->que[qid];
804 	struct ena_ring *rx_ring = que->rx_ring;
805 	int size, err, i;
806 
807 	size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
808 
809 #ifdef DEV_NETMAP
810 	ena_netmap_reset_rx_ring(adapter, qid);
811 	rx_ring->initialized = false;
812 #endif /* DEV_NETMAP */
813 
814 	/*
815 	 * Alloc extra element so in rx path
816 	 * we can always prefetch rx_info + 1
817 	 */
818 	size += sizeof(struct ena_rx_buffer);
819 
820 	rx_ring->rx_buffer_info = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
821 
822 	size = sizeof(uint16_t) * rx_ring->ring_size;
823 	rx_ring->free_rx_ids = malloc(size, M_DEVBUF, M_WAITOK);
824 
825 	for (i = 0; i < rx_ring->ring_size; i++)
826 		rx_ring->free_rx_ids[i] = i;
827 
828 	/* Reset RX statistics. */
829 	ena_reset_counters((counter_u64_t *)&rx_ring->rx_stats,
830 	    sizeof(rx_ring->rx_stats));
831 
832 	rx_ring->next_to_clean = 0;
833 	rx_ring->next_to_use = 0;
834 
835 	/* ... and create the buffer DMA maps */
836 	for (i = 0; i < rx_ring->ring_size; i++) {
837 		err = bus_dmamap_create(adapter->rx_buf_tag, 0,
838 		    &(rx_ring->rx_buffer_info[i].map));
839 		if (err != 0) {
840 			ena_trace(ENA_ALERT,
841 			    "Unable to create Rx DMA map for buffer %d\n", i);
842 			goto err_buf_info_unmap;
843 		}
844 	}
845 
846 	/* Create LRO for the ring */
847 	if ((adapter->ifp->if_capenable & IFCAP_LRO) != 0) {
848 		int err = tcp_lro_init(&rx_ring->lro);
849 		if (err != 0) {
850 			device_printf(adapter->pdev,
851 			    "LRO[%d] Initialization failed!\n", qid);
852 		} else {
853 			ena_trace(ENA_INFO,
854 			    "RX Soft LRO[%d] Initialized\n", qid);
855 			rx_ring->lro.ifp = adapter->ifp;
856 		}
857 	}
858 
859 	return (0);
860 
861 err_buf_info_unmap:
862 	while (i--) {
863 		bus_dmamap_destroy(adapter->rx_buf_tag,
864 		    rx_ring->rx_buffer_info[i].map);
865 	}
866 
867 	free(rx_ring->free_rx_ids, M_DEVBUF);
868 	rx_ring->free_rx_ids = NULL;
869 	free(rx_ring->rx_buffer_info, M_DEVBUF);
870 	rx_ring->rx_buffer_info = NULL;
871 	return (ENOMEM);
872 }
873 
874 /**
875  * ena_free_rx_resources - Free Rx Resources
876  * @adapter: network interface device structure
877  * @qid: queue index
878  *
879  * Free all receive software resources
880  **/
881 static void
882 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
883 {
884 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
885 
886 	/* Free buffer DMA maps, */
887 	for (int i = 0; i < rx_ring->ring_size; i++) {
888 		bus_dmamap_sync(adapter->rx_buf_tag,
889 		    rx_ring->rx_buffer_info[i].map, BUS_DMASYNC_POSTREAD);
890 		m_freem(rx_ring->rx_buffer_info[i].mbuf);
891 		rx_ring->rx_buffer_info[i].mbuf = NULL;
892 		bus_dmamap_unload(adapter->rx_buf_tag,
893 		    rx_ring->rx_buffer_info[i].map);
894 		bus_dmamap_destroy(adapter->rx_buf_tag,
895 		    rx_ring->rx_buffer_info[i].map);
896 	}
897 
898 	/* free LRO resources, */
899 	tcp_lro_free(&rx_ring->lro);
900 
901 	/* free allocated memory */
902 	free(rx_ring->rx_buffer_info, M_DEVBUF);
903 	rx_ring->rx_buffer_info = NULL;
904 
905 	free(rx_ring->free_rx_ids, M_DEVBUF);
906 	rx_ring->free_rx_ids = NULL;
907 }
908 
909 /**
910  * ena_setup_all_rx_resources - allocate all queues Rx resources
911  * @adapter: network interface device structure
912  *
913  * Returns 0 on success, otherwise on failure.
914  **/
915 static int
916 ena_setup_all_rx_resources(struct ena_adapter *adapter)
917 {
918 	int i, rc = 0;
919 
920 	for (i = 0; i < adapter->num_queues; i++) {
921 		rc = ena_setup_rx_resources(adapter, i);
922 		if (rc != 0) {
923 			device_printf(adapter->pdev,
924 			    "Allocation for Rx Queue %u failed\n", i);
925 			goto err_setup_rx;
926 		}
927 	}
928 	return (0);
929 
930 err_setup_rx:
931 	/* rewind the index freeing the rings as we go */
932 	while (i--)
933 		ena_free_rx_resources(adapter, i);
934 	return (rc);
935 }
936 
937 /**
938  * ena_free_all_rx_resources - Free Rx resources for all queues
939  * @adapter: network interface device structure
940  *
941  * Free all receive software resources
942  **/
943 static void
944 ena_free_all_rx_resources(struct ena_adapter *adapter)
945 {
946 	int i;
947 
948 	for (i = 0; i < adapter->num_queues; i++)
949 		ena_free_rx_resources(adapter, i);
950 }
951 
952 static inline int
953 ena_alloc_rx_mbuf(struct ena_adapter *adapter,
954     struct ena_ring *rx_ring, struct ena_rx_buffer *rx_info)
955 {
956 	struct ena_com_buf *ena_buf;
957 	bus_dma_segment_t segs[1];
958 	int nsegs, error;
959 	int mlen;
960 
961 	/* if previous allocated frag is not used */
962 	if (unlikely(rx_info->mbuf != NULL))
963 		return (0);
964 
965 	/* Get mbuf using UMA allocator */
966 	rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM16BYTES);
967 
968 	if (unlikely(rx_info->mbuf == NULL)) {
969 		counter_u64_add(rx_ring->rx_stats.mjum_alloc_fail, 1);
970 		rx_info->mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
971 		if (unlikely(rx_info->mbuf == NULL)) {
972 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
973 			return (ENOMEM);
974 		}
975 		mlen = MCLBYTES;
976 	} else {
977 		mlen = MJUM16BYTES;
978 	}
979 	/* Set mbuf length*/
980 	rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = mlen;
981 
982 	/* Map packets for DMA */
983 	ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
984 	    "Using tag %p for buffers' DMA mapping, mbuf %p len: %d\n",
985 	    adapter->rx_buf_tag,rx_info->mbuf, rx_info->mbuf->m_len);
986 	error = bus_dmamap_load_mbuf_sg(adapter->rx_buf_tag, rx_info->map,
987 	    rx_info->mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
988 	if (unlikely((error != 0) || (nsegs != 1))) {
989 		ena_trace(ENA_WARNING, "failed to map mbuf, error: %d, "
990 		    "nsegs: %d\n", error, nsegs);
991 		counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
992 		goto exit;
993 
994 	}
995 
996 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map, BUS_DMASYNC_PREREAD);
997 
998 	ena_buf = &rx_info->ena_buf;
999 	ena_buf->paddr = segs[0].ds_addr;
1000 	ena_buf->len = mlen;
1001 
1002 	ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
1003 	    "ALLOC RX BUF: mbuf %p, rx_info %p, len %d, paddr %#jx\n",
1004 	    rx_info->mbuf, rx_info,ena_buf->len, (uintmax_t)ena_buf->paddr);
1005 
1006 	return (0);
1007 
1008 exit:
1009 	m_freem(rx_info->mbuf);
1010 	rx_info->mbuf = NULL;
1011 	return (EFAULT);
1012 }
1013 
1014 static void
1015 ena_free_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
1016     struct ena_rx_buffer *rx_info)
1017 {
1018 
1019 	if (rx_info->mbuf == NULL) {
1020 		ena_trace(ENA_WARNING, "Trying to free unallocated buffer\n");
1021 		return;
1022 	}
1023 
1024 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map,
1025 	    BUS_DMASYNC_POSTREAD);
1026 	bus_dmamap_unload(adapter->rx_buf_tag, rx_info->map);
1027 	m_freem(rx_info->mbuf);
1028 	rx_info->mbuf = NULL;
1029 }
1030 
1031 /**
1032  * ena_refill_rx_bufs - Refills ring with descriptors
1033  * @rx_ring: the ring which we want to feed with free descriptors
1034  * @num: number of descriptors to refill
1035  * Refills the ring with newly allocated DMA-mapped mbufs for receiving
1036  **/
1037 int
1038 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
1039 {
1040 	struct ena_adapter *adapter = rx_ring->adapter;
1041 	uint16_t next_to_use, req_id;
1042 	uint32_t i;
1043 	int rc;
1044 
1045 	ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC, "refill qid: %d\n",
1046 	    rx_ring->qid);
1047 
1048 	next_to_use = rx_ring->next_to_use;
1049 
1050 	for (i = 0; i < num; i++) {
1051 		struct ena_rx_buffer *rx_info;
1052 
1053 		ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC,
1054 		    "RX buffer - next to use: %d\n", next_to_use);
1055 
1056 		req_id = rx_ring->free_rx_ids[next_to_use];
1057 		rx_info = &rx_ring->rx_buffer_info[req_id];
1058 #ifdef DEV_NETMAP
1059 		if (ena_rx_ring_in_netmap(adapter, rx_ring->qid))
1060 			rc = ena_netmap_alloc_rx_slot(adapter, rx_ring, rx_info);
1061 		else
1062 #endif /* DEV_NETMAP */
1063 			rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1064 		if (unlikely(rc != 0)) {
1065 			ena_trace(ENA_WARNING,
1066 			    "failed to alloc buffer for rx queue %d\n",
1067 			    rx_ring->qid);
1068 			break;
1069 		}
1070 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1071 		    &rx_info->ena_buf, req_id);
1072 		if (unlikely(rc != 0)) {
1073 			ena_trace(ENA_WARNING,
1074 			    "failed to add buffer for rx queue %d\n",
1075 			    rx_ring->qid);
1076 			break;
1077 		}
1078 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1079 		    rx_ring->ring_size);
1080 	}
1081 
1082 	if (unlikely(i < num)) {
1083 		counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1084 		ena_trace(ENA_WARNING,
1085 		     "refilled rx qid %d with only %d mbufs (from %d)\n",
1086 		     rx_ring->qid, i, num);
1087 	}
1088 
1089 	if (likely(i != 0)) {
1090 		wmb();
1091 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1092 	}
1093 	rx_ring->next_to_use = next_to_use;
1094 	return (i);
1095 }
1096 
1097 static void
1098 ena_free_rx_bufs(struct ena_adapter *adapter, unsigned int qid)
1099 {
1100 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1101 	unsigned int i;
1102 
1103 	for (i = 0; i < rx_ring->ring_size; i++) {
1104 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1105 
1106 		if (rx_info->mbuf != NULL)
1107 			ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1108 #ifdef DEV_NETMAP
1109 		if (((if_getflags(adapter->ifp) & IFF_DYING) == 0) &&
1110 		    (adapter->ifp->if_capenable & IFCAP_NETMAP)) {
1111 			if (rx_info->netmap_buf_idx != 0)
1112 				ena_netmap_free_rx_slot(adapter, rx_ring,
1113 				    rx_info);
1114 		}
1115 #endif /* DEV_NETMAP */
1116 	}
1117 }
1118 
1119 /**
1120  * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1121  * @adapter: network interface device structure
1122  *
1123  */
1124 static void
1125 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1126 {
1127 	struct ena_ring *rx_ring;
1128 	int i, rc, bufs_num;
1129 
1130 	for (i = 0; i < adapter->num_queues; i++) {
1131 		rx_ring = &adapter->rx_ring[i];
1132 		bufs_num = rx_ring->ring_size - 1;
1133 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1134 		if (unlikely(rc != bufs_num))
1135 			ena_trace(ENA_WARNING, "refilling Queue %d failed. "
1136 			    "Allocated %d buffers from: %d\n", i, rc, bufs_num);
1137 #ifdef DEV_NETMAP
1138 		rx_ring->initialized = true;
1139 #endif /* DEV_NETMAP */
1140 	}
1141 }
1142 
1143 static void
1144 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1145 {
1146 	int i;
1147 
1148 	for (i = 0; i < adapter->num_queues; i++)
1149 		ena_free_rx_bufs(adapter, i);
1150 }
1151 
1152 /**
1153  * ena_free_tx_bufs - Free Tx Buffers per Queue
1154  * @adapter: network interface device structure
1155  * @qid: queue index
1156  **/
1157 static void
1158 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1159 {
1160 	bool print_once = true;
1161 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1162 
1163 	ENA_RING_MTX_LOCK(tx_ring);
1164 	for (int i = 0; i < tx_ring->ring_size; i++) {
1165 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1166 
1167 		if (tx_info->mbuf == NULL)
1168 			continue;
1169 
1170 		if (print_once) {
1171 			device_printf(adapter->pdev,
1172 			    "free uncompleted tx mbuf qid %d idx 0x%x\n",
1173 			    qid, i);
1174 			print_once = false;
1175 		} else {
1176 			ena_trace(ENA_DBG,
1177 			    "free uncompleted tx mbuf qid %d idx 0x%x\n",
1178 			     qid, i);
1179 		}
1180 
1181 		bus_dmamap_sync(adapter->tx_buf_tag, tx_info->dmamap,
1182 		    BUS_DMASYNC_POSTWRITE);
1183 		bus_dmamap_unload(adapter->tx_buf_tag, tx_info->dmamap);
1184 
1185 		m_free(tx_info->mbuf);
1186 		tx_info->mbuf = NULL;
1187 	}
1188 	ENA_RING_MTX_UNLOCK(tx_ring);
1189 }
1190 
1191 static void
1192 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1193 {
1194 
1195 	for (int i = 0; i < adapter->num_queues; i++)
1196 		ena_free_tx_bufs(adapter, i);
1197 }
1198 
1199 static void
1200 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1201 {
1202 	uint16_t ena_qid;
1203 	int i;
1204 
1205 	for (i = 0; i < adapter->num_queues; i++) {
1206 		ena_qid = ENA_IO_TXQ_IDX(i);
1207 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1208 	}
1209 }
1210 
1211 static void
1212 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1213 {
1214 	uint16_t ena_qid;
1215 	int i;
1216 
1217 	for (i = 0; i < adapter->num_queues; i++) {
1218 		ena_qid = ENA_IO_RXQ_IDX(i);
1219 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1220 	}
1221 }
1222 
1223 static void
1224 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1225 {
1226 	struct ena_que *queue;
1227 	int i;
1228 
1229 	for (i = 0; i < adapter->num_queues; i++) {
1230 		queue = &adapter->que[i];
1231 		while (taskqueue_cancel(queue->cleanup_tq,
1232 		    &queue->cleanup_task, NULL))
1233 			taskqueue_drain(queue->cleanup_tq,
1234 			    &queue->cleanup_task);
1235 		taskqueue_free(queue->cleanup_tq);
1236 	}
1237 
1238 	ena_destroy_all_tx_queues(adapter);
1239 	ena_destroy_all_rx_queues(adapter);
1240 }
1241 
1242 static int
1243 ena_create_io_queues(struct ena_adapter *adapter)
1244 {
1245 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1246 	struct ena_com_create_io_ctx ctx;
1247 	struct ena_ring *ring;
1248 	struct ena_que *queue;
1249 	uint16_t ena_qid;
1250 	uint32_t msix_vector;
1251 	int rc, i;
1252 
1253 	/* Create TX queues */
1254 	for (i = 0; i < adapter->num_queues; i++) {
1255 		msix_vector = ENA_IO_IRQ_IDX(i);
1256 		ena_qid = ENA_IO_TXQ_IDX(i);
1257 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1258 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1259 		ctx.queue_size = adapter->tx_ring_size;
1260 		ctx.msix_vector = msix_vector;
1261 		ctx.qid = ena_qid;
1262 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1263 		if (rc != 0) {
1264 			device_printf(adapter->pdev,
1265 			    "Failed to create io TX queue #%d rc: %d\n", i, rc);
1266 			goto err_tx;
1267 		}
1268 		ring = &adapter->tx_ring[i];
1269 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1270 		    &ring->ena_com_io_sq,
1271 		    &ring->ena_com_io_cq);
1272 		if (rc != 0) {
1273 			device_printf(adapter->pdev,
1274 			    "Failed to get TX queue handlers. TX queue num"
1275 			    " %d rc: %d\n", i, rc);
1276 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1277 			goto err_tx;
1278 		}
1279 	}
1280 
1281 	/* Create RX queues */
1282 	for (i = 0; i < adapter->num_queues; i++) {
1283 		msix_vector = ENA_IO_IRQ_IDX(i);
1284 		ena_qid = ENA_IO_RXQ_IDX(i);
1285 		ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1286 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1287 		ctx.queue_size = adapter->rx_ring_size;
1288 		ctx.msix_vector = msix_vector;
1289 		ctx.qid = ena_qid;
1290 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1291 		if (unlikely(rc != 0)) {
1292 			device_printf(adapter->pdev,
1293 			    "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1294 			goto err_rx;
1295 		}
1296 
1297 		ring = &adapter->rx_ring[i];
1298 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1299 		    &ring->ena_com_io_sq,
1300 		    &ring->ena_com_io_cq);
1301 		if (unlikely(rc != 0)) {
1302 			device_printf(adapter->pdev,
1303 			    "Failed to get RX queue handlers. RX queue num"
1304 			    " %d rc: %d\n", i, rc);
1305 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1306 			goto err_rx;
1307 		}
1308 	}
1309 
1310 	for (i = 0; i < adapter->num_queues; i++) {
1311 		queue = &adapter->que[i];
1312 
1313 		NET_TASK_INIT(&queue->cleanup_task, 0, ena_cleanup, queue);
1314 		queue->cleanup_tq = taskqueue_create_fast("ena cleanup",
1315 		    M_WAITOK, taskqueue_thread_enqueue, &queue->cleanup_tq);
1316 
1317 		taskqueue_start_threads(&queue->cleanup_tq, 1, PI_NET,
1318 		    "%s queue %d cleanup",
1319 		    device_get_nameunit(adapter->pdev), i);
1320 	}
1321 
1322 	return (0);
1323 
1324 err_rx:
1325 	while (i--)
1326 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1327 	i = adapter->num_queues;
1328 err_tx:
1329 	while (i--)
1330 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1331 
1332 	return (ENXIO);
1333 }
1334 
1335 /*********************************************************************
1336  *
1337  *  MSIX & Interrupt Service routine
1338  *
1339  **********************************************************************/
1340 
1341 /**
1342  * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1343  * @arg: interrupt number
1344  **/
1345 static void
1346 ena_intr_msix_mgmnt(void *arg)
1347 {
1348 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1349 
1350 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1351 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter)))
1352 		ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1353 }
1354 
1355 /**
1356  * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1357  * @arg: queue
1358  **/
1359 static int
1360 ena_handle_msix(void *arg)
1361 {
1362 	struct ena_que *queue = arg;
1363 	struct ena_adapter *adapter = queue->adapter;
1364 	if_t ifp = adapter->ifp;
1365 
1366 	if (unlikely((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0))
1367 		return (FILTER_STRAY);
1368 
1369 	taskqueue_enqueue(queue->cleanup_tq, &queue->cleanup_task);
1370 
1371 	return (FILTER_HANDLED);
1372 }
1373 
1374 static int
1375 ena_enable_msix(struct ena_adapter *adapter)
1376 {
1377 	device_t dev = adapter->pdev;
1378 	int msix_vecs, msix_req;
1379 	int i, rc = 0;
1380 
1381 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1382 		device_printf(dev, "Error, MSI-X is already enabled\n");
1383 		return (EINVAL);
1384 	}
1385 
1386 	/* Reserved the max msix vectors we might need */
1387 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->num_queues);
1388 
1389 	adapter->msix_entries = malloc(msix_vecs * sizeof(struct msix_entry),
1390 	    M_DEVBUF, M_WAITOK | M_ZERO);
1391 
1392 	ena_trace(ENA_DBG, "trying to enable MSI-X, vectors: %d\n", msix_vecs);
1393 
1394 	for (i = 0; i < msix_vecs; i++) {
1395 		adapter->msix_entries[i].entry = i;
1396 		/* Vectors must start from 1 */
1397 		adapter->msix_entries[i].vector = i + 1;
1398 	}
1399 
1400 	msix_req = msix_vecs;
1401 	rc = pci_alloc_msix(dev, &msix_vecs);
1402 	if (unlikely(rc != 0)) {
1403 		device_printf(dev,
1404 		    "Failed to enable MSIX, vectors %d rc %d\n", msix_vecs, rc);
1405 
1406 		rc = ENOSPC;
1407 		goto err_msix_free;
1408 	}
1409 
1410 	if (msix_vecs != msix_req) {
1411 		if (msix_vecs == ENA_ADMIN_MSIX_VEC) {
1412 			device_printf(dev,
1413 			    "Not enough number of MSI-x allocated: %d\n",
1414 			    msix_vecs);
1415 			pci_release_msi(dev);
1416 			rc = ENOSPC;
1417 			goto err_msix_free;
1418 		}
1419 		device_printf(dev, "Enable only %d MSI-x (out of %d), reduce "
1420 		    "the number of queues\n", msix_vecs, msix_req);
1421 		adapter->num_queues = msix_vecs - ENA_ADMIN_MSIX_VEC;
1422 	}
1423 
1424 	adapter->msix_vecs = msix_vecs;
1425 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1426 
1427 	return (0);
1428 
1429 err_msix_free:
1430 	free(adapter->msix_entries, M_DEVBUF);
1431 	adapter->msix_entries = NULL;
1432 
1433 	return (rc);
1434 }
1435 
1436 static void
1437 ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1438 {
1439 
1440 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1441 	    ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1442 	    device_get_nameunit(adapter->pdev));
1443 	/*
1444 	 * Handler is NULL on purpose, it will be set
1445 	 * when mgmnt interrupt is acquired
1446 	 */
1447 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler = NULL;
1448 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1449 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1450 	    adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
1451 }
1452 
1453 static int
1454 ena_setup_io_intr(struct ena_adapter *adapter)
1455 {
1456 	static int last_bind_cpu = -1;
1457 	int irq_idx;
1458 
1459 	if (adapter->msix_entries == NULL)
1460 		return (EINVAL);
1461 
1462 	for (int i = 0; i < adapter->num_queues; i++) {
1463 		irq_idx = ENA_IO_IRQ_IDX(i);
1464 
1465 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1466 		    "%s-TxRx-%d", device_get_nameunit(adapter->pdev), i);
1467 		adapter->irq_tbl[irq_idx].handler = ena_handle_msix;
1468 		adapter->irq_tbl[irq_idx].data = &adapter->que[i];
1469 		adapter->irq_tbl[irq_idx].vector =
1470 		    adapter->msix_entries[irq_idx].vector;
1471 		ena_trace(ENA_INFO | ENA_IOQ, "ena_setup_io_intr vector: %d\n",
1472 		    adapter->msix_entries[irq_idx].vector);
1473 
1474 		/*
1475 		 * We want to bind rings to the corresponding cpu
1476 		 * using something similar to the RSS round-robin technique.
1477 		 */
1478 		if (unlikely(last_bind_cpu < 0))
1479 			last_bind_cpu = CPU_FIRST();
1480 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1481 		    last_bind_cpu;
1482 		last_bind_cpu = CPU_NEXT(last_bind_cpu);
1483 	}
1484 
1485 	return (0);
1486 }
1487 
1488 static int
1489 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1490 {
1491 	struct ena_irq *irq;
1492 	unsigned long flags;
1493 	int rc, rcc;
1494 
1495 	flags = RF_ACTIVE | RF_SHAREABLE;
1496 
1497 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1498 	irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1499 	    &irq->vector, flags);
1500 
1501 	if (unlikely(irq->res == NULL)) {
1502 		device_printf(adapter->pdev, "could not allocate "
1503 		    "irq vector: %d\n", irq->vector);
1504 		return (ENXIO);
1505 	}
1506 
1507 	rc = bus_setup_intr(adapter->pdev, irq->res,
1508 	    INTR_TYPE_NET | INTR_MPSAFE, NULL, ena_intr_msix_mgmnt,
1509 	    irq->data, &irq->cookie);
1510 	if (unlikely(rc != 0)) {
1511 		device_printf(adapter->pdev, "failed to register "
1512 		    "interrupt handler for irq %ju: %d\n",
1513 		    rman_get_start(irq->res), rc);
1514 		goto err_res_free;
1515 	}
1516 	irq->requested = true;
1517 
1518 	return (rc);
1519 
1520 err_res_free:
1521 	ena_trace(ENA_INFO | ENA_ADMQ, "releasing resource for irq %d\n",
1522 	    irq->vector);
1523 	rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1524 	    irq->vector, irq->res);
1525 	if (unlikely(rcc != 0))
1526 		device_printf(adapter->pdev, "dev has no parent while "
1527 		    "releasing res for irq: %d\n", irq->vector);
1528 	irq->res = NULL;
1529 
1530 	return (rc);
1531 }
1532 
1533 static int
1534 ena_request_io_irq(struct ena_adapter *adapter)
1535 {
1536 	struct ena_irq *irq;
1537 	unsigned long flags = 0;
1538 	int rc = 0, i, rcc;
1539 
1540 	if (unlikely(!ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter))) {
1541 		device_printf(adapter->pdev,
1542 		    "failed to request I/O IRQ: MSI-X is not enabled\n");
1543 		return (EINVAL);
1544 	} else {
1545 		flags = RF_ACTIVE | RF_SHAREABLE;
1546 	}
1547 
1548 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1549 		irq = &adapter->irq_tbl[i];
1550 
1551 		if (unlikely(irq->requested))
1552 			continue;
1553 
1554 		irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1555 		    &irq->vector, flags);
1556 		if (unlikely(irq->res == NULL)) {
1557 			rc = ENOMEM;
1558 			device_printf(adapter->pdev, "could not allocate "
1559 			    "irq vector: %d\n", irq->vector);
1560 			goto err;
1561 		}
1562 
1563 		rc = bus_setup_intr(adapter->pdev, irq->res,
1564 		    INTR_TYPE_NET | INTR_MPSAFE, irq->handler, NULL,
1565 		    irq->data, &irq->cookie);
1566 		 if (unlikely(rc != 0)) {
1567 			device_printf(adapter->pdev, "failed to register "
1568 			    "interrupt handler for irq %ju: %d\n",
1569 			    rman_get_start(irq->res), rc);
1570 			goto err;
1571 		}
1572 		irq->requested = true;
1573 
1574 		ena_trace(ENA_INFO, "queue %d - cpu %d\n",
1575 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
1576 	}
1577 
1578 	return (rc);
1579 
1580 err:
1581 
1582 	for (; i >= ENA_IO_IRQ_FIRST_IDX; i--) {
1583 		irq = &adapter->irq_tbl[i];
1584 		rcc = 0;
1585 
1586 		/* Once we entered err: section and irq->requested is true we
1587 		   free both intr and resources */
1588 		if (irq->requested)
1589 			rcc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
1590 		if (unlikely(rcc != 0))
1591 			device_printf(adapter->pdev, "could not release"
1592 			    " irq: %d, error: %d\n", irq->vector, rcc);
1593 
1594 		/* If we entred err: section without irq->requested set we know
1595 		   it was bus_alloc_resource_any() that needs cleanup, provided
1596 		   res is not NULL. In case res is NULL no work in needed in
1597 		   this iteration */
1598 		rcc = 0;
1599 		if (irq->res != NULL) {
1600 			rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1601 			    irq->vector, irq->res);
1602 		}
1603 		if (unlikely(rcc != 0))
1604 			device_printf(adapter->pdev, "dev has no parent while "
1605 			    "releasing res for irq: %d\n", irq->vector);
1606 		irq->requested = false;
1607 		irq->res = NULL;
1608 	}
1609 
1610 	return (rc);
1611 }
1612 
1613 static void
1614 ena_free_mgmnt_irq(struct ena_adapter *adapter)
1615 {
1616 	struct ena_irq *irq;
1617 	int rc;
1618 
1619 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1620 	if (irq->requested) {
1621 		ena_trace(ENA_INFO | ENA_ADMQ, "tear down irq: %d\n",
1622 		    irq->vector);
1623 		rc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
1624 		if (unlikely(rc != 0))
1625 			device_printf(adapter->pdev, "failed to tear "
1626 			    "down irq: %d\n", irq->vector);
1627 		irq->requested = 0;
1628 	}
1629 
1630 	if (irq->res != NULL) {
1631 		ena_trace(ENA_INFO | ENA_ADMQ, "release resource irq: %d\n",
1632 		    irq->vector);
1633 		rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1634 		    irq->vector, irq->res);
1635 		irq->res = NULL;
1636 		if (unlikely(rc != 0))
1637 			device_printf(adapter->pdev, "dev has no parent while "
1638 			    "releasing res for irq: %d\n", irq->vector);
1639 	}
1640 }
1641 
1642 static void
1643 ena_free_io_irq(struct ena_adapter *adapter)
1644 {
1645 	struct ena_irq *irq;
1646 	int rc;
1647 
1648 	for (int i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1649 		irq = &adapter->irq_tbl[i];
1650 		if (irq->requested) {
1651 			ena_trace(ENA_INFO | ENA_IOQ, "tear down irq: %d\n",
1652 			    irq->vector);
1653 			rc = bus_teardown_intr(adapter->pdev, irq->res,
1654 			    irq->cookie);
1655 			if (unlikely(rc != 0)) {
1656 				device_printf(adapter->pdev, "failed to tear "
1657 				    "down irq: %d\n", irq->vector);
1658 			}
1659 			irq->requested = 0;
1660 		}
1661 
1662 		if (irq->res != NULL) {
1663 			ena_trace(ENA_INFO | ENA_IOQ, "release resource irq: %d\n",
1664 			    irq->vector);
1665 			rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1666 			    irq->vector, irq->res);
1667 			irq->res = NULL;
1668 			if (unlikely(rc != 0)) {
1669 				device_printf(adapter->pdev, "dev has no parent"
1670 				    " while releasing res for irq: %d\n",
1671 				    irq->vector);
1672 			}
1673 		}
1674 	}
1675 }
1676 
1677 static void
1678 ena_free_irqs(struct ena_adapter* adapter)
1679 {
1680 
1681 	ena_free_io_irq(adapter);
1682 	ena_free_mgmnt_irq(adapter);
1683 	ena_disable_msix(adapter);
1684 }
1685 
1686 static void
1687 ena_disable_msix(struct ena_adapter *adapter)
1688 {
1689 
1690 	if (ENA_FLAG_ISSET(ENA_FLAG_MSIX_ENABLED, adapter)) {
1691 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_MSIX_ENABLED, adapter);
1692 		pci_release_msi(adapter->pdev);
1693 	}
1694 
1695 	adapter->msix_vecs = 0;
1696 	if (adapter->msix_entries != NULL)
1697 		free(adapter->msix_entries, M_DEVBUF);
1698 	adapter->msix_entries = NULL;
1699 }
1700 
1701 static void
1702 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
1703 {
1704 	struct ena_com_io_cq* io_cq;
1705 	struct ena_eth_io_intr_reg intr_reg;
1706 	uint16_t ena_qid;
1707 	int i;
1708 
1709 	/* Unmask interrupts for all queues */
1710 	for (i = 0; i < adapter->num_queues; i++) {
1711 		ena_qid = ENA_IO_TXQ_IDX(i);
1712 		io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1713 		ena_com_update_intr_reg(&intr_reg, 0, 0, true);
1714 		ena_com_unmask_intr(io_cq, &intr_reg);
1715 	}
1716 }
1717 
1718 /* Configure the Rx forwarding */
1719 static int
1720 ena_rss_configure(struct ena_adapter *adapter)
1721 {
1722 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1723 	int rc;
1724 
1725 	/* Set indirect table */
1726 	rc = ena_com_indirect_table_set(ena_dev);
1727 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
1728 		return (rc);
1729 
1730 	/* Configure hash function (if supported) */
1731 	rc = ena_com_set_hash_function(ena_dev);
1732 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
1733 		return (rc);
1734 
1735 	/* Configure hash inputs (if supported) */
1736 	rc = ena_com_set_hash_ctrl(ena_dev);
1737 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
1738 		return (rc);
1739 
1740 	return (0);
1741 }
1742 
1743 static int
1744 ena_up_complete(struct ena_adapter *adapter)
1745 {
1746 	int rc;
1747 
1748 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter))) {
1749 		rc = ena_rss_configure(adapter);
1750 		if (rc != 0)
1751 			return (rc);
1752 	}
1753 
1754 	rc = ena_change_mtu(adapter->ifp, adapter->ifp->if_mtu);
1755 	if (unlikely(rc != 0))
1756 		return (rc);
1757 
1758 	ena_refill_all_rx_bufs(adapter);
1759 	ena_reset_counters((counter_u64_t *)&adapter->hw_stats,
1760 	    sizeof(adapter->hw_stats));
1761 
1762 	return (0);
1763 }
1764 
1765 int
1766 ena_up(struct ena_adapter *adapter)
1767 {
1768 	int rc = 0;
1769 
1770 	if (unlikely(device_is_attached(adapter->pdev) == 0)) {
1771 		device_printf(adapter->pdev, "device is not attached!\n");
1772 		return (ENXIO);
1773 	}
1774 
1775 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter)) {
1776 		device_printf(adapter->pdev, "device is going UP\n");
1777 
1778 		/* setup interrupts for IO queues */
1779 		rc = ena_setup_io_intr(adapter);
1780 		if (unlikely(rc != 0)) {
1781 			ena_trace(ENA_ALERT, "error setting up IO interrupt\n");
1782 			goto error;
1783 		}
1784 		rc = ena_request_io_irq(adapter);
1785 		if (unlikely(rc != 0)) {
1786 			ena_trace(ENA_ALERT, "err_req_irq\n");
1787 			goto error;
1788 		}
1789 
1790 		/* allocate transmit descriptors */
1791 		rc = ena_setup_all_tx_resources(adapter);
1792 		if (unlikely(rc != 0)) {
1793 			ena_trace(ENA_ALERT, "err_setup_tx\n");
1794 			goto err_setup_tx;
1795 		}
1796 
1797 		/* allocate receive descriptors */
1798 		rc = ena_setup_all_rx_resources(adapter);
1799 		if (unlikely(rc != 0)) {
1800 			ena_trace(ENA_ALERT, "err_setup_rx\n");
1801 			goto err_setup_rx;
1802 		}
1803 
1804 		/* create IO queues for Rx & Tx */
1805 		rc = ena_create_io_queues(adapter);
1806 		if (unlikely(rc != 0)) {
1807 			ena_trace(ENA_ALERT,
1808 			    "create IO queues failed\n");
1809 			goto err_io_que;
1810 		}
1811 
1812 		if (ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter))
1813 			if_link_state_change(adapter->ifp, LINK_STATE_UP);
1814 
1815 		rc = ena_up_complete(adapter);
1816 		if (unlikely(rc != 0))
1817 			goto err_up_complete;
1818 
1819 		counter_u64_add(adapter->dev_stats.interface_up, 1);
1820 
1821 		ena_update_hwassist(adapter);
1822 
1823 		if_setdrvflagbits(adapter->ifp, IFF_DRV_RUNNING,
1824 		    IFF_DRV_OACTIVE);
1825 
1826 		/* Activate timer service only if the device is running.
1827 		 * If this flag is not set, it means that the driver is being
1828 		 * reset and timer service will be activated afterwards.
1829 		 */
1830 		if (ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter)) {
1831 			callout_reset_sbt(&adapter->timer_service, SBT_1S,
1832 			    SBT_1S, ena_timer_service, (void *)adapter, 0);
1833 		}
1834 
1835 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP, adapter);
1836 
1837 		ena_unmask_all_io_irqs(adapter);
1838 	}
1839 
1840 	return (0);
1841 
1842 err_up_complete:
1843 	ena_destroy_all_io_queues(adapter);
1844 err_io_que:
1845 	ena_free_all_rx_resources(adapter);
1846 err_setup_rx:
1847 	ena_free_all_tx_resources(adapter);
1848 err_setup_tx:
1849 	ena_free_io_irq(adapter);
1850 error:
1851 	return (rc);
1852 }
1853 
1854 static uint64_t
1855 ena_get_counter(if_t ifp, ift_counter cnt)
1856 {
1857 	struct ena_adapter *adapter;
1858 	struct ena_hw_stats *stats;
1859 
1860 	adapter = if_getsoftc(ifp);
1861 	stats = &adapter->hw_stats;
1862 
1863 	switch (cnt) {
1864 	case IFCOUNTER_IPACKETS:
1865 		return (counter_u64_fetch(stats->rx_packets));
1866 	case IFCOUNTER_OPACKETS:
1867 		return (counter_u64_fetch(stats->tx_packets));
1868 	case IFCOUNTER_IBYTES:
1869 		return (counter_u64_fetch(stats->rx_bytes));
1870 	case IFCOUNTER_OBYTES:
1871 		return (counter_u64_fetch(stats->tx_bytes));
1872 	case IFCOUNTER_IQDROPS:
1873 		return (counter_u64_fetch(stats->rx_drops));
1874 	default:
1875 		return (if_get_counter_default(ifp, cnt));
1876 	}
1877 }
1878 
1879 static int
1880 ena_media_change(if_t ifp)
1881 {
1882 	/* Media Change is not supported by firmware */
1883 	return (0);
1884 }
1885 
1886 static void
1887 ena_media_status(if_t ifp, struct ifmediareq *ifmr)
1888 {
1889 	struct ena_adapter *adapter = if_getsoftc(ifp);
1890 	ena_trace(ENA_DBG, "enter\n");
1891 
1892 	mtx_lock(&adapter->global_mtx);
1893 
1894 	ifmr->ifm_status = IFM_AVALID;
1895 	ifmr->ifm_active = IFM_ETHER;
1896 
1897 	if (!ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter)) {
1898 		mtx_unlock(&adapter->global_mtx);
1899 		ena_trace(ENA_INFO, "Link is down\n");
1900 		return;
1901 	}
1902 
1903 	ifmr->ifm_status |= IFM_ACTIVE;
1904 	ifmr->ifm_active |= IFM_UNKNOWN | IFM_FDX;
1905 
1906 	mtx_unlock(&adapter->global_mtx);
1907 }
1908 
1909 static void
1910 ena_init(void *arg)
1911 {
1912 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1913 
1914 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter)) {
1915 		sx_xlock(&adapter->ioctl_sx);
1916 		ena_up(adapter);
1917 		sx_unlock(&adapter->ioctl_sx);
1918 	}
1919 }
1920 
1921 static int
1922 ena_ioctl(if_t ifp, u_long command, caddr_t data)
1923 {
1924 	struct ena_adapter *adapter;
1925 	struct ifreq *ifr;
1926 	int rc;
1927 
1928 	adapter = ifp->if_softc;
1929 	ifr = (struct ifreq *)data;
1930 
1931 	/*
1932 	 * Acquiring lock to prevent from running up and down routines parallel.
1933 	 */
1934 	rc = 0;
1935 	switch (command) {
1936 	case SIOCSIFMTU:
1937 		if (ifp->if_mtu == ifr->ifr_mtu)
1938 			break;
1939 		sx_xlock(&adapter->ioctl_sx);
1940 		ena_down(adapter);
1941 
1942 		ena_change_mtu(ifp, ifr->ifr_mtu);
1943 
1944 		rc = ena_up(adapter);
1945 		sx_unlock(&adapter->ioctl_sx);
1946 		break;
1947 
1948 	case SIOCSIFFLAGS:
1949 		if ((ifp->if_flags & IFF_UP) != 0) {
1950 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1951 				if ((ifp->if_flags & (IFF_PROMISC |
1952 				    IFF_ALLMULTI)) != 0) {
1953 					device_printf(adapter->pdev,
1954 					    "ioctl promisc/allmulti\n");
1955 				}
1956 			} else {
1957 				sx_xlock(&adapter->ioctl_sx);
1958 				rc = ena_up(adapter);
1959 				sx_unlock(&adapter->ioctl_sx);
1960 			}
1961 		} else {
1962 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1963 				sx_xlock(&adapter->ioctl_sx);
1964 				ena_down(adapter);
1965 				sx_unlock(&adapter->ioctl_sx);
1966 			}
1967 		}
1968 		break;
1969 
1970 	case SIOCADDMULTI:
1971 	case SIOCDELMULTI:
1972 		break;
1973 
1974 	case SIOCSIFMEDIA:
1975 	case SIOCGIFMEDIA:
1976 		rc = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
1977 		break;
1978 
1979 	case SIOCSIFCAP:
1980 		{
1981 			int reinit = 0;
1982 
1983 			if (ifr->ifr_reqcap != ifp->if_capenable) {
1984 				ifp->if_capenable = ifr->ifr_reqcap;
1985 				reinit = 1;
1986 			}
1987 
1988 			if ((reinit != 0) &&
1989 			    ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)) {
1990 				sx_xlock(&adapter->ioctl_sx);
1991 				ena_down(adapter);
1992 				rc = ena_up(adapter);
1993 				sx_unlock(&adapter->ioctl_sx);
1994 			}
1995 		}
1996 
1997 		break;
1998 	default:
1999 		rc = ether_ioctl(ifp, command, data);
2000 		break;
2001 	}
2002 
2003 	return (rc);
2004 }
2005 
2006 static int
2007 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2008 {
2009 	int caps = 0;
2010 
2011 	if ((feat->offload.tx &
2012 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2013 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2014 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)) != 0)
2015 		caps |= IFCAP_TXCSUM;
2016 
2017 	if ((feat->offload.tx &
2018 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2019 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)) != 0)
2020 		caps |= IFCAP_TXCSUM_IPV6;
2021 
2022 	if ((feat->offload.tx &
2023 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0)
2024 		caps |= IFCAP_TSO4;
2025 
2026 	if ((feat->offload.tx &
2027 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK) != 0)
2028 		caps |= IFCAP_TSO6;
2029 
2030 	if ((feat->offload.rx_supported &
2031 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2032 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK)) != 0)
2033 		caps |= IFCAP_RXCSUM;
2034 
2035 	if ((feat->offload.rx_supported &
2036 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK) != 0)
2037 		caps |= IFCAP_RXCSUM_IPV6;
2038 
2039 	caps |= IFCAP_LRO | IFCAP_JUMBO_MTU;
2040 
2041 	return (caps);
2042 }
2043 
2044 static void
2045 ena_update_host_info(struct ena_admin_host_info *host_info, if_t ifp)
2046 {
2047 
2048 	host_info->supported_network_features[0] =
2049 	    (uint32_t)if_getcapabilities(ifp);
2050 }
2051 
2052 static void
2053 ena_update_hwassist(struct ena_adapter *adapter)
2054 {
2055 	if_t ifp = adapter->ifp;
2056 	uint32_t feat = adapter->tx_offload_cap;
2057 	int cap = if_getcapenable(ifp);
2058 	int flags = 0;
2059 
2060 	if_clearhwassist(ifp);
2061 
2062 	if ((cap & IFCAP_TXCSUM) != 0) {
2063 		if ((feat &
2064 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK) != 0)
2065 			flags |= CSUM_IP;
2066 		if ((feat &
2067 		    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2068 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)) != 0)
2069 			flags |= CSUM_IP_UDP | CSUM_IP_TCP;
2070 	}
2071 
2072 	if ((cap & IFCAP_TXCSUM_IPV6) != 0)
2073 		flags |= CSUM_IP6_UDP | CSUM_IP6_TCP;
2074 
2075 	if ((cap & IFCAP_TSO4) != 0)
2076 		flags |= CSUM_IP_TSO;
2077 
2078 	if ((cap & IFCAP_TSO6) != 0)
2079 		flags |= CSUM_IP6_TSO;
2080 
2081 	if_sethwassistbits(ifp, flags, 0);
2082 }
2083 
2084 static int
2085 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2086     struct ena_com_dev_get_features_ctx *feat)
2087 {
2088 	if_t ifp;
2089 	int caps = 0;
2090 
2091 	ifp = adapter->ifp = if_gethandle(IFT_ETHER);
2092 	if (unlikely(ifp == NULL)) {
2093 		ena_trace(ENA_ALERT, "can not allocate ifnet structure\n");
2094 		return (ENXIO);
2095 	}
2096 	if_initname(ifp, device_get_name(pdev), device_get_unit(pdev));
2097 	if_setdev(ifp, pdev);
2098 	if_setsoftc(ifp, adapter);
2099 
2100 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
2101 	if_setinitfn(ifp, ena_init);
2102 	if_settransmitfn(ifp, ena_mq_start);
2103 	if_setqflushfn(ifp, ena_qflush);
2104 	if_setioctlfn(ifp, ena_ioctl);
2105 	if_setgetcounterfn(ifp, ena_get_counter);
2106 
2107 	if_setsendqlen(ifp, adapter->tx_ring_size);
2108 	if_setsendqready(ifp);
2109 	if_setmtu(ifp, ETHERMTU);
2110 	if_setbaudrate(ifp, 0);
2111 	/* Zeroize capabilities... */
2112 	if_setcapabilities(ifp, 0);
2113 	if_setcapenable(ifp, 0);
2114 	/* check hardware support */
2115 	caps = ena_get_dev_offloads(feat);
2116 	/* ... and set them */
2117 	if_setcapabilitiesbit(ifp, caps, 0);
2118 
2119 	/* TSO parameters */
2120 	ifp->if_hw_tsomax = ENA_TSO_MAXSIZE -
2121 	    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2122 	ifp->if_hw_tsomaxsegcount = adapter->max_tx_sgl_size - 1;
2123 	ifp->if_hw_tsomaxsegsize = ENA_TSO_MAXSIZE;
2124 
2125 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
2126 	if_setcapenable(ifp, if_getcapabilities(ifp));
2127 
2128 	/*
2129 	 * Specify the media types supported by this adapter and register
2130 	 * callbacks to update media and link information
2131 	 */
2132 	ifmedia_init(&adapter->media, IFM_IMASK,
2133 	    ena_media_change, ena_media_status);
2134 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2135 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2136 
2137 	ether_ifattach(ifp, adapter->mac_addr);
2138 
2139 	return (0);
2140 }
2141 
2142 void
2143 ena_down(struct ena_adapter *adapter)
2144 {
2145 	int rc;
2146 
2147 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter)) {
2148 		device_printf(adapter->pdev, "device is going DOWN\n");
2149 
2150 		callout_drain(&adapter->timer_service);
2151 
2152 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP, adapter);
2153 		if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE,
2154 		    IFF_DRV_RUNNING);
2155 
2156 		ena_free_io_irq(adapter);
2157 
2158 		if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter)) {
2159 			rc = ena_com_dev_reset(adapter->ena_dev,
2160 			    adapter->reset_reason);
2161 			if (unlikely(rc != 0))
2162 				device_printf(adapter->pdev,
2163 				    "Device reset failed\n");
2164 		}
2165 
2166 		ena_destroy_all_io_queues(adapter);
2167 
2168 		ena_free_all_tx_bufs(adapter);
2169 		ena_free_all_rx_bufs(adapter);
2170 		ena_free_all_tx_resources(adapter);
2171 		ena_free_all_rx_resources(adapter);
2172 
2173 		counter_u64_add(adapter->dev_stats.interface_down, 1);
2174 	}
2175 }
2176 
2177 static int
2178 ena_calc_io_queue_num(struct ena_adapter *adapter,
2179     struct ena_com_dev_get_features_ctx *get_feat_ctx)
2180 {
2181 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2182 	int io_tx_sq_num, io_tx_cq_num, io_rx_num, io_queue_num;
2183 
2184 	/* Regular queues capabilities */
2185 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2186 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2187 		    &get_feat_ctx->max_queue_ext.max_queue_ext;
2188 		io_rx_num = min_t(int, max_queue_ext->max_rx_sq_num,
2189 			max_queue_ext->max_rx_cq_num);
2190 
2191 		io_tx_sq_num = max_queue_ext->max_tx_sq_num;
2192 		io_tx_cq_num = max_queue_ext->max_tx_cq_num;
2193 	} else {
2194 		struct ena_admin_queue_feature_desc *max_queues =
2195 		    &get_feat_ctx->max_queues;
2196 		io_tx_sq_num = max_queues->max_sq_num;
2197 		io_tx_cq_num = max_queues->max_cq_num;
2198 		io_rx_num = min_t(int, io_tx_sq_num, io_tx_cq_num);
2199 	}
2200 
2201 	/* In case of LLQ use the llq fields for the tx SQ/CQ */
2202 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
2203 		io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
2204 
2205 	io_queue_num = min_t(int, mp_ncpus, ENA_MAX_NUM_IO_QUEUES);
2206 	io_queue_num = min_t(int, io_queue_num, io_rx_num);
2207 	io_queue_num = min_t(int, io_queue_num, io_tx_sq_num);
2208 	io_queue_num = min_t(int, io_queue_num, io_tx_cq_num);
2209 	/* 1 IRQ for for mgmnt and 1 IRQ for each TX/RX pair */
2210 	io_queue_num = min_t(int, io_queue_num,
2211 	    pci_msix_count(adapter->pdev) - 1);
2212 
2213 	return (io_queue_num);
2214 }
2215 
2216 static int
2217 ena_enable_wc(struct resource *res)
2218 {
2219 #if defined(__i386) || defined(__amd64) || defined(__aarch64__)
2220 	vm_offset_t va;
2221 	vm_size_t len;
2222 	int rc;
2223 
2224 	va = (vm_offset_t)rman_get_virtual(res);
2225 	len = rman_get_size(res);
2226 	/* Enable write combining */
2227 	rc = pmap_change_attr(va, len, VM_MEMATTR_WRITE_COMBINING);
2228 	if (unlikely(rc != 0)) {
2229 		ena_trace(ENA_ALERT, "pmap_change_attr failed, %d\n", rc);
2230 		return (rc);
2231 	}
2232 
2233 	return (0);
2234 #endif
2235 	return (EOPNOTSUPP);
2236 }
2237 
2238 static int
2239 ena_set_queues_placement_policy(device_t pdev, struct ena_com_dev *ena_dev,
2240     struct ena_admin_feature_llq_desc *llq,
2241     struct ena_llq_configurations *llq_default_configurations)
2242 {
2243 	struct ena_adapter *adapter = device_get_softc(pdev);
2244 	int rc, rid;
2245 	uint32_t llq_feature_mask;
2246 
2247 	llq_feature_mask = 1 << ENA_ADMIN_LLQ;
2248 	if (!(ena_dev->supported_features & llq_feature_mask)) {
2249 		device_printf(pdev,
2250 		    "LLQ is not supported. Fallback to host mode policy.\n");
2251 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2252 		return (0);
2253 	}
2254 
2255 	rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
2256 	if (unlikely(rc != 0)) {
2257 		device_printf(pdev, "Failed to configure the device mode. "
2258 		    "Fallback to host mode policy.\n");
2259 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2260 		return (0);
2261 	}
2262 
2263 	/* Nothing to config, exit */
2264 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
2265 		return (0);
2266 
2267 	/* Try to allocate resources for LLQ bar */
2268 	rid = PCIR_BAR(ENA_MEM_BAR);
2269 	adapter->memory = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
2270 	    &rid, RF_ACTIVE);
2271 	if (unlikely(adapter->memory == NULL)) {
2272 		device_printf(pdev, "unable to allocate LLQ bar resource. "
2273 		    "Fallback to host mode policy.\n");
2274 		ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2275 		return (0);
2276 	}
2277 
2278 	/* Enable write combining for better LLQ performance */
2279 	rc = ena_enable_wc(adapter->memory);
2280 	if (unlikely(rc != 0)) {
2281 		device_printf(pdev, "failed to enable write combining.\n");
2282 		return (rc);
2283 	}
2284 
2285 	/*
2286 	 * Save virtual address of the device's memory region
2287 	 * for the ena_com layer.
2288 	 */
2289 	ena_dev->mem_bar = rman_get_virtual(adapter->memory);
2290 
2291 	return (0);
2292 }
2293 
2294 static inline
2295 void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
2296 {
2297 	llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
2298 	llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
2299 	llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
2300 	llq_config->llq_num_decs_before_header =
2301 	    ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
2302 	llq_config->llq_ring_entry_size_value = 128;
2303 }
2304 
2305 static int
2306 ena_calc_queue_size(struct ena_adapter *adapter,
2307     struct ena_calc_queue_size_ctx *ctx)
2308 {
2309 	struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
2310 	struct ena_com_dev *ena_dev = ctx->ena_dev;
2311 	uint32_t tx_queue_size = ENA_DEFAULT_RING_SIZE;
2312 	uint32_t rx_queue_size = adapter->rx_ring_size;
2313 
2314 	if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
2315 		struct ena_admin_queue_ext_feature_fields *max_queue_ext =
2316 		    &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
2317 		rx_queue_size = min_t(uint32_t, rx_queue_size,
2318 		    max_queue_ext->max_rx_cq_depth);
2319 		rx_queue_size = min_t(uint32_t, rx_queue_size,
2320 		    max_queue_ext->max_rx_sq_depth);
2321 		tx_queue_size = min_t(uint32_t, tx_queue_size,
2322 		    max_queue_ext->max_tx_cq_depth);
2323 
2324 		if (ena_dev->tx_mem_queue_type ==
2325 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2326 			tx_queue_size = min_t(uint32_t, tx_queue_size,
2327 			    llq->max_llq_depth);
2328 		else
2329 			tx_queue_size = min_t(uint32_t, tx_queue_size,
2330 			    max_queue_ext->max_tx_sq_depth);
2331 
2332 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2333 		    max_queue_ext->max_per_packet_rx_descs);
2334 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2335 		    max_queue_ext->max_per_packet_tx_descs);
2336 	} else {
2337 		struct ena_admin_queue_feature_desc *max_queues =
2338 		    &ctx->get_feat_ctx->max_queues;
2339 		rx_queue_size = min_t(uint32_t, rx_queue_size,
2340 		    max_queues->max_cq_depth);
2341 		rx_queue_size = min_t(uint32_t, rx_queue_size,
2342 		    max_queues->max_sq_depth);
2343 		tx_queue_size = min_t(uint32_t, tx_queue_size,
2344 		    max_queues->max_cq_depth);
2345 
2346 		if (ena_dev->tx_mem_queue_type ==
2347 		    ENA_ADMIN_PLACEMENT_POLICY_DEV)
2348 			tx_queue_size = min_t(uint32_t, tx_queue_size,
2349 			    llq->max_llq_depth);
2350 		else
2351 			tx_queue_size = min_t(uint32_t, tx_queue_size,
2352 			    max_queues->max_sq_depth);
2353 
2354 		ctx->max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2355 		    max_queues->max_packet_tx_descs);
2356 		ctx->max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
2357 		    max_queues->max_packet_rx_descs);
2358 	}
2359 
2360 	/* round down to the nearest power of 2 */
2361 	rx_queue_size = 1 << (fls(rx_queue_size) - 1);
2362 	tx_queue_size = 1 << (fls(tx_queue_size) - 1);
2363 
2364 	if (unlikely(rx_queue_size == 0 || tx_queue_size == 0)) {
2365 		device_printf(ctx->pdev, "Invalid queue size\n");
2366 		return (EFAULT);
2367 	}
2368 
2369 	ctx->rx_queue_size = rx_queue_size;
2370 	ctx->tx_queue_size = tx_queue_size;
2371 
2372 	return (0);
2373 }
2374 
2375 static int
2376 ena_handle_updated_queues(struct ena_adapter *adapter,
2377     struct ena_com_dev_get_features_ctx *get_feat_ctx)
2378 {
2379 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2380 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
2381 	device_t pdev = adapter->pdev;
2382 	bool are_queues_changed = false;
2383 	int io_queue_num, rc;
2384 
2385 	calc_queue_ctx.ena_dev = ena_dev;
2386 	calc_queue_ctx.get_feat_ctx = get_feat_ctx;
2387 	calc_queue_ctx.pdev = pdev;
2388 
2389 	io_queue_num = ena_calc_io_queue_num(adapter, get_feat_ctx);
2390 	rc = ena_calc_queue_size(adapter, &calc_queue_ctx);
2391 	if (unlikely(rc != 0 || io_queue_num <= 0))
2392 		return EFAULT;
2393 
2394 	if (adapter->tx_ring->buf_ring_size != adapter->buf_ring_size)
2395 		are_queues_changed = true;
2396 
2397 	if (unlikely(adapter->tx_ring_size > calc_queue_ctx.tx_queue_size ||
2398 	    adapter->rx_ring_size > calc_queue_ctx.rx_queue_size)) {
2399 		device_printf(pdev,
2400 		    "Not enough resources to allocate requested queue sizes "
2401 		    "(TX,RX)=(%d,%d), falling back to queue sizes "
2402 		    "(TX,RX)=(%d,%d)\n",
2403 		    adapter->tx_ring_size,
2404 		    adapter->rx_ring_size,
2405 		    calc_queue_ctx.tx_queue_size,
2406 		    calc_queue_ctx.rx_queue_size);
2407 		adapter->tx_ring_size = calc_queue_ctx.tx_queue_size;
2408 		adapter->rx_ring_size = calc_queue_ctx.rx_queue_size;
2409 		adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
2410 		adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
2411 		are_queues_changed = true;
2412 	}
2413 
2414 	if (unlikely(adapter->num_queues > io_queue_num)) {
2415 		device_printf(pdev,
2416 		    "Not enough resources to allocate %d queues, "
2417 		    "falling back to %d queues\n",
2418 		    adapter->num_queues, io_queue_num);
2419 		adapter->num_queues = io_queue_num;
2420 		if (ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter)) {
2421 			ena_com_rss_destroy(ena_dev);
2422 			rc = ena_rss_init_default(adapter);
2423 			if (unlikely(rc != 0) && (rc != EOPNOTSUPP)) {
2424 				device_printf(pdev, "Cannot init RSS rc: %d\n",
2425 				    rc);
2426 				return (rc);
2427 			}
2428 		}
2429 		are_queues_changed = true;
2430 	}
2431 
2432 	if (unlikely(are_queues_changed)) {
2433 		ena_free_all_io_rings_resources(adapter);
2434 		ena_init_io_rings(adapter);
2435 	}
2436 
2437 	return (0);
2438 }
2439 
2440 static int
2441 ena_rss_init_default(struct ena_adapter *adapter)
2442 {
2443 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2444 	device_t dev = adapter->pdev;
2445 	int qid, rc, i;
2446 
2447 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
2448 	if (unlikely(rc != 0)) {
2449 		device_printf(dev, "Cannot init indirect table\n");
2450 		return (rc);
2451 	}
2452 
2453 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
2454 		qid = i % adapter->num_queues;
2455 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
2456 		    ENA_IO_RXQ_IDX(qid));
2457 		if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
2458 			device_printf(dev, "Cannot fill indirect table\n");
2459 			goto err_rss_destroy;
2460 		}
2461 	}
2462 
2463 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
2464 	    ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
2465 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
2466 		device_printf(dev, "Cannot fill hash function\n");
2467 		goto err_rss_destroy;
2468 	}
2469 
2470 	rc = ena_com_set_default_hash_ctrl(ena_dev);
2471 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
2472 		device_printf(dev, "Cannot fill hash control\n");
2473 		goto err_rss_destroy;
2474 	}
2475 
2476 	return (0);
2477 
2478 err_rss_destroy:
2479 	ena_com_rss_destroy(ena_dev);
2480 	return (rc);
2481 }
2482 
2483 static void
2484 ena_rss_init_default_deferred(void *arg)
2485 {
2486 	struct ena_adapter *adapter;
2487 	devclass_t dc;
2488 	int max;
2489 	int rc;
2490 
2491 	dc = devclass_find("ena");
2492 	if (unlikely(dc == NULL)) {
2493 		ena_trace(ENA_ALERT, "No devclass ena\n");
2494 		return;
2495 	}
2496 
2497 	max = devclass_get_maxunit(dc);
2498 	while (max-- >= 0) {
2499 		adapter = devclass_get_softc(dc, max);
2500 		if (adapter != NULL) {
2501 			rc = ena_rss_init_default(adapter);
2502 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_RSS_ACTIVE, adapter);
2503 			if (unlikely(rc != 0)) {
2504 				device_printf(adapter->pdev,
2505 				    "WARNING: RSS was not properly initialized,"
2506 				    " it will affect bandwidth\n");
2507 				ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_RSS_ACTIVE, adapter);
2508 			}
2509 		}
2510 	}
2511 }
2512 SYSINIT(ena_rss_init, SI_SUB_KICK_SCHEDULER, SI_ORDER_SECOND, ena_rss_init_default_deferred, NULL);
2513 
2514 static void
2515 ena_config_host_info(struct ena_com_dev *ena_dev, device_t dev)
2516 {
2517 	struct ena_admin_host_info *host_info;
2518 	uintptr_t rid;
2519 	int rc;
2520 
2521 	/* Allocate only the host info */
2522 	rc = ena_com_allocate_host_info(ena_dev);
2523 	if (unlikely(rc != 0)) {
2524 		ena_trace(ENA_ALERT, "Cannot allocate host info\n");
2525 		return;
2526 	}
2527 
2528 	host_info = ena_dev->host_attr.host_info;
2529 
2530 	if (pci_get_id(dev, PCI_ID_RID, &rid) == 0)
2531 		host_info->bdf = rid;
2532 	host_info->os_type = ENA_ADMIN_OS_FREEBSD;
2533 	host_info->kernel_ver = osreldate;
2534 
2535 	sprintf(host_info->kernel_ver_str, "%d", osreldate);
2536 	host_info->os_dist = 0;
2537 	strncpy(host_info->os_dist_str, osrelease,
2538 	    sizeof(host_info->os_dist_str) - 1);
2539 
2540 	host_info->driver_version =
2541 		(DRV_MODULE_VER_MAJOR) |
2542 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2543 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2544 	host_info->num_cpus = mp_ncpus;
2545 
2546 	rc = ena_com_set_host_attributes(ena_dev);
2547 	if (unlikely(rc != 0)) {
2548 		if (rc == EOPNOTSUPP)
2549 			ena_trace(ENA_WARNING, "Cannot set host attributes\n");
2550 		else
2551 			ena_trace(ENA_ALERT, "Cannot set host attributes\n");
2552 
2553 		goto err;
2554 	}
2555 
2556 	return;
2557 
2558 err:
2559 	ena_com_delete_host_info(ena_dev);
2560 }
2561 
2562 static int
2563 ena_device_init(struct ena_adapter *adapter, device_t pdev,
2564     struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
2565 {
2566 	struct ena_com_dev* ena_dev = adapter->ena_dev;
2567 	bool readless_supported;
2568 	uint32_t aenq_groups;
2569 	int dma_width;
2570 	int rc;
2571 
2572 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
2573 	if (unlikely(rc != 0)) {
2574 		device_printf(pdev, "failed to init mmio read less\n");
2575 		return (rc);
2576 	}
2577 
2578 	/*
2579 	 * The PCIe configuration space revision id indicate if mmio reg
2580 	 * read is disabled
2581 	 */
2582 	readless_supported = !(pci_get_revid(pdev) & ENA_MMIO_DISABLE_REG_READ);
2583 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2584 
2585 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2586 	if (unlikely(rc != 0)) {
2587 		device_printf(pdev, "Can not reset device\n");
2588 		goto err_mmio_read_less;
2589 	}
2590 
2591 	rc = ena_com_validate_version(ena_dev);
2592 	if (unlikely(rc != 0)) {
2593 		device_printf(pdev, "device version is too low\n");
2594 		goto err_mmio_read_less;
2595 	}
2596 
2597 	dma_width = ena_com_get_dma_width(ena_dev);
2598 	if (unlikely(dma_width < 0)) {
2599 		device_printf(pdev, "Invalid dma width value %d", dma_width);
2600 		rc = dma_width;
2601 		goto err_mmio_read_less;
2602 	}
2603 	adapter->dma_width = dma_width;
2604 
2605 	/* ENA admin level init */
2606 	rc = ena_com_admin_init(ena_dev, &aenq_handlers);
2607 	if (unlikely(rc != 0)) {
2608 		device_printf(pdev,
2609 		    "Can not initialize ena admin queue with device\n");
2610 		goto err_mmio_read_less;
2611 	}
2612 
2613 	/*
2614 	 * To enable the msix interrupts the driver needs to know the number
2615 	 * of queues. So the driver uses polling mode to retrieve this
2616 	 * information
2617 	 */
2618 	ena_com_set_admin_polling_mode(ena_dev, true);
2619 
2620 	ena_config_host_info(ena_dev, pdev);
2621 
2622 	/* Get Device Attributes */
2623 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2624 	if (unlikely(rc != 0)) {
2625 		device_printf(pdev,
2626 		    "Cannot get attribute for ena device rc: %d\n", rc);
2627 		goto err_admin_init;
2628 	}
2629 
2630 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2631 	    BIT(ENA_ADMIN_FATAL_ERROR) |
2632 	    BIT(ENA_ADMIN_WARNING) |
2633 	    BIT(ENA_ADMIN_NOTIFICATION) |
2634 	    BIT(ENA_ADMIN_KEEP_ALIVE);
2635 
2636 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
2637 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2638 	if (unlikely(rc != 0)) {
2639 		device_printf(pdev, "Cannot configure aenq groups rc: %d\n", rc);
2640 		goto err_admin_init;
2641 	}
2642 
2643 	*wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2644 
2645 	return (0);
2646 
2647 err_admin_init:
2648 	ena_com_delete_host_info(ena_dev);
2649 	ena_com_admin_destroy(ena_dev);
2650 err_mmio_read_less:
2651 	ena_com_mmio_reg_read_request_destroy(ena_dev);
2652 
2653 	return (rc);
2654 }
2655 
2656 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
2657     int io_vectors)
2658 {
2659 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2660 	int rc;
2661 
2662 	rc = ena_enable_msix(adapter);
2663 	if (unlikely(rc != 0)) {
2664 		device_printf(adapter->pdev, "Error with MSI-X enablement\n");
2665 		return (rc);
2666 	}
2667 
2668 	ena_setup_mgmnt_intr(adapter);
2669 
2670 	rc = ena_request_mgmnt_irq(adapter);
2671 	if (unlikely(rc != 0)) {
2672 		device_printf(adapter->pdev, "Cannot setup mgmnt queue intr\n");
2673 		goto err_disable_msix;
2674 	}
2675 
2676 	ena_com_set_admin_polling_mode(ena_dev, false);
2677 
2678 	ena_com_admin_aenq_enable(ena_dev);
2679 
2680 	return (0);
2681 
2682 err_disable_msix:
2683 	ena_disable_msix(adapter);
2684 
2685 	return (rc);
2686 }
2687 
2688 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
2689 static void ena_keep_alive_wd(void *adapter_data,
2690     struct ena_admin_aenq_entry *aenq_e)
2691 {
2692 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
2693 	struct ena_admin_aenq_keep_alive_desc *desc;
2694 	sbintime_t stime;
2695 	uint64_t rx_drops;
2696 
2697 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
2698 
2699 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
2700 	counter_u64_zero(adapter->hw_stats.rx_drops);
2701 	counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
2702 
2703 	stime = getsbinuptime();
2704 	atomic_store_rel_64(&adapter->keep_alive_timestamp, stime);
2705 }
2706 
2707 /* Check for keep alive expiration */
2708 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
2709 {
2710 	sbintime_t timestamp, time;
2711 
2712 	if (adapter->wd_active == 0)
2713 		return;
2714 
2715 	if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2716 		return;
2717 
2718 	timestamp = atomic_load_acq_64(&adapter->keep_alive_timestamp);
2719 	time = getsbinuptime() - timestamp;
2720 	if (unlikely(time > adapter->keep_alive_timeout)) {
2721 		device_printf(adapter->pdev,
2722 		    "Keep alive watchdog timeout.\n");
2723 		counter_u64_add(adapter->dev_stats.wd_expired, 1);
2724 		if (likely(!ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
2725 			adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
2726 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
2727 		}
2728 	}
2729 }
2730 
2731 /* Check if admin queue is enabled */
2732 static void check_for_admin_com_state(struct ena_adapter *adapter)
2733 {
2734 	if (unlikely(ena_com_get_admin_running_state(adapter->ena_dev) ==
2735 	    false)) {
2736 		device_printf(adapter->pdev,
2737 		    "ENA admin queue is not in running state!\n");
2738 		counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
2739 		if (likely(!ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
2740 			adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
2741 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
2742 		}
2743 	}
2744 }
2745 
2746 static int
2747 check_for_rx_interrupt_queue(struct ena_adapter *adapter,
2748     struct ena_ring *rx_ring)
2749 {
2750 	if (likely(rx_ring->first_interrupt))
2751 		return (0);
2752 
2753 	if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
2754 		return (0);
2755 
2756 	rx_ring->no_interrupt_event_cnt++;
2757 
2758 	if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
2759 		device_printf(adapter->pdev, "Potential MSIX issue on Rx side "
2760 		    "Queue = %d. Reset the device\n", rx_ring->qid);
2761 		if (likely(!ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
2762 			adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2763 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
2764 		}
2765 		return (EIO);
2766 	}
2767 
2768 	return (0);
2769 }
2770 
2771 static int
2772 check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
2773     struct ena_ring *tx_ring)
2774 {
2775 	struct bintime curtime, time;
2776 	struct ena_tx_buffer *tx_buf;
2777 	sbintime_t time_offset;
2778 	uint32_t missed_tx = 0;
2779 	int i, rc = 0;
2780 
2781 	getbinuptime(&curtime);
2782 
2783 	for (i = 0; i < tx_ring->ring_size; i++) {
2784 		tx_buf = &tx_ring->tx_buffer_info[i];
2785 
2786 		if (bintime_isset(&tx_buf->timestamp) == 0)
2787 			continue;
2788 
2789 		time = curtime;
2790 		bintime_sub(&time, &tx_buf->timestamp);
2791 		time_offset = bttosbt(time);
2792 
2793 		if (unlikely(!tx_ring->first_interrupt &&
2794 		    time_offset > 2 * adapter->missing_tx_timeout)) {
2795 			/*
2796 			 * If after graceful period interrupt is still not
2797 			 * received, we schedule a reset.
2798 			 */
2799 			device_printf(adapter->pdev,
2800 			    "Potential MSIX issue on Tx side Queue = %d. "
2801 			    "Reset the device\n", tx_ring->qid);
2802 			if (likely(!ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET,
2803 			    adapter))) {
2804 				adapter->reset_reason =
2805 				    ENA_REGS_RESET_MISS_INTERRUPT;
2806 				ENA_FLAG_SET_ATOMIC(ENA_FLAG_TRIGGER_RESET,
2807 				    adapter);
2808 			}
2809 			return (EIO);
2810 		}
2811 
2812 		/* Check again if packet is still waiting */
2813 		if (unlikely(time_offset > adapter->missing_tx_timeout)) {
2814 
2815 			if (!tx_buf->print_once)
2816 				ena_trace(ENA_WARNING, "Found a Tx that wasn't "
2817 				    "completed on time, qid %d, index %d.\n",
2818 				    tx_ring->qid, i);
2819 
2820 			tx_buf->print_once = true;
2821 			missed_tx++;
2822 		}
2823 	}
2824 
2825 	if (unlikely(missed_tx > adapter->missing_tx_threshold)) {
2826 		device_printf(adapter->pdev,
2827 		    "The number of lost tx completion is above the threshold "
2828 		    "(%d > %d). Reset the device\n",
2829 		    missed_tx, adapter->missing_tx_threshold);
2830 		if (likely(!ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
2831 			adapter->reset_reason = ENA_REGS_RESET_MISS_TX_CMPL;
2832 			ENA_FLAG_SET_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
2833 		}
2834 		rc = EIO;
2835 	}
2836 
2837 	counter_u64_add(tx_ring->tx_stats.missing_tx_comp, missed_tx);
2838 
2839 	return (rc);
2840 }
2841 
2842 /*
2843  * Check for TX which were not completed on time.
2844  * Timeout is defined by "missing_tx_timeout".
2845  * Reset will be performed if number of incompleted
2846  * transactions exceeds "missing_tx_threshold".
2847  */
2848 static void
2849 check_for_missing_completions(struct ena_adapter *adapter)
2850 {
2851 	struct ena_ring *tx_ring;
2852 	struct ena_ring *rx_ring;
2853 	int i, budget, rc;
2854 
2855 	/* Make sure the driver doesn't turn the device in other process */
2856 	rmb();
2857 
2858 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2859 		return;
2860 
2861 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
2862 		return;
2863 
2864 	if (adapter->missing_tx_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2865 		return;
2866 
2867 	budget = adapter->missing_tx_max_queues;
2868 
2869 	for (i = adapter->next_monitored_tx_qid; i < adapter->num_queues; i++) {
2870 		tx_ring = &adapter->tx_ring[i];
2871 		rx_ring = &adapter->rx_ring[i];
2872 
2873 		rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
2874 		if (unlikely(rc != 0))
2875 			return;
2876 
2877 		rc = check_for_rx_interrupt_queue(adapter, rx_ring);
2878 		if (unlikely(rc != 0))
2879 			return;
2880 
2881 		budget--;
2882 		if (budget == 0) {
2883 			i++;
2884 			break;
2885 		}
2886 	}
2887 
2888 	adapter->next_monitored_tx_qid = i % adapter->num_queues;
2889 }
2890 
2891 /* trigger rx cleanup after 2 consecutive detections */
2892 #define EMPTY_RX_REFILL 2
2893 /* For the rare case where the device runs out of Rx descriptors and the
2894  * msix handler failed to refill new Rx descriptors (due to a lack of memory
2895  * for example).
2896  * This case will lead to a deadlock:
2897  * The device won't send interrupts since all the new Rx packets will be dropped
2898  * The msix handler won't allocate new Rx descriptors so the device won't be
2899  * able to send new packets.
2900  *
2901  * When such a situation is detected - execute rx cleanup task in another thread
2902  */
2903 static void
2904 check_for_empty_rx_ring(struct ena_adapter *adapter)
2905 {
2906 	struct ena_ring *rx_ring;
2907 	int i, refill_required;
2908 
2909 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
2910 		return;
2911 
2912 	if (ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))
2913 		return;
2914 
2915 	for (i = 0; i < adapter->num_queues; i++) {
2916 		rx_ring = &adapter->rx_ring[i];
2917 
2918 		refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
2919 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
2920 			rx_ring->empty_rx_queue++;
2921 
2922 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL)	{
2923 				counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
2924 				    1);
2925 
2926 				device_printf(adapter->pdev,
2927 				    "trigger refill for ring %d\n", i);
2928 
2929 				taskqueue_enqueue(rx_ring->que->cleanup_tq,
2930 				    &rx_ring->que->cleanup_task);
2931 				rx_ring->empty_rx_queue = 0;
2932 			}
2933 		} else {
2934 			rx_ring->empty_rx_queue = 0;
2935 		}
2936 	}
2937 }
2938 
2939 static void ena_update_hints(struct ena_adapter *adapter,
2940 			     struct ena_admin_ena_hw_hints *hints)
2941 {
2942 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2943 
2944 	if (hints->admin_completion_tx_timeout)
2945 		ena_dev->admin_queue.completion_timeout =
2946 		    hints->admin_completion_tx_timeout * 1000;
2947 
2948 	if (hints->mmio_read_timeout)
2949 		/* convert to usec */
2950 		ena_dev->mmio_read.reg_read_to =
2951 		    hints->mmio_read_timeout * 1000;
2952 
2953 	if (hints->missed_tx_completion_count_threshold_to_reset)
2954 		adapter->missing_tx_threshold =
2955 		    hints->missed_tx_completion_count_threshold_to_reset;
2956 
2957 	if (hints->missing_tx_completion_timeout) {
2958 		if (hints->missing_tx_completion_timeout ==
2959 		     ENA_HW_HINTS_NO_TIMEOUT)
2960 			adapter->missing_tx_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2961 		else
2962 			adapter->missing_tx_timeout =
2963 			    SBT_1MS * hints->missing_tx_completion_timeout;
2964 	}
2965 
2966 	if (hints->driver_watchdog_timeout) {
2967 		if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2968 			adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2969 		else
2970 			adapter->keep_alive_timeout =
2971 			    SBT_1MS * hints->driver_watchdog_timeout;
2972 	}
2973 }
2974 
2975 static void
2976 ena_timer_service(void *data)
2977 {
2978 	struct ena_adapter *adapter = (struct ena_adapter *)data;
2979 	struct ena_admin_host_info *host_info =
2980 	    adapter->ena_dev->host_attr.host_info;
2981 
2982 	check_for_missing_keep_alive(adapter);
2983 
2984 	check_for_admin_com_state(adapter);
2985 
2986 	check_for_missing_completions(adapter);
2987 
2988 	check_for_empty_rx_ring(adapter);
2989 
2990 	if (host_info != NULL)
2991 		ena_update_host_info(host_info, adapter->ifp);
2992 
2993 	if (unlikely(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
2994 		device_printf(adapter->pdev, "Trigger reset is on\n");
2995 		taskqueue_enqueue(adapter->reset_tq, &adapter->reset_task);
2996 		return;
2997 	}
2998 
2999 	/*
3000 	 * Schedule another timeout one second from now.
3001 	 */
3002 	callout_schedule_sbt(&adapter->timer_service, SBT_1S, SBT_1S, 0);
3003 }
3004 
3005 void
3006 ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3007 {
3008 	if_t ifp = adapter->ifp;
3009 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3010 	bool dev_up;
3011 
3012 	if (!ENA_FLAG_ISSET(ENA_FLAG_DEVICE_RUNNING, adapter))
3013 		return;
3014 
3015 	if_link_state_change(ifp, LINK_STATE_DOWN);
3016 
3017 	callout_drain(&adapter->timer_service);
3018 
3019 	dev_up = ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter);
3020 	if (dev_up)
3021 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3022 	else
3023 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter);
3024 
3025 	if (!graceful)
3026 		ena_com_set_admin_running_state(ena_dev, false);
3027 
3028 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP, adapter))
3029 		ena_down(adapter);
3030 
3031 	/*
3032 	 * Stop the device from sending AENQ events (if the device was up, and
3033 	 * the trigger reset was on, ena_down already performs device reset)
3034 	 */
3035 	if (!(ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter) && dev_up))
3036 		ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3037 
3038 	ena_free_mgmnt_irq(adapter);
3039 
3040 	ena_disable_msix(adapter);
3041 
3042 	ena_com_abort_admin_commands(ena_dev);
3043 
3044 	ena_com_wait_for_abort_completion(ena_dev);
3045 
3046 	ena_com_admin_destroy(ena_dev);
3047 
3048 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3049 
3050 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3051 
3052 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_TRIGGER_RESET, adapter);
3053 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3054 }
3055 
3056 static int
3057 ena_device_validate_params(struct ena_adapter *adapter,
3058     struct ena_com_dev_get_features_ctx *get_feat_ctx)
3059 {
3060 
3061 	if (memcmp(get_feat_ctx->dev_attr.mac_addr, adapter->mac_addr,
3062 	    ETHER_ADDR_LEN) != 0) {
3063 		device_printf(adapter->pdev,
3064 		    "Error, mac address are different\n");
3065 		return (EINVAL);
3066 	}
3067 
3068 	if (get_feat_ctx->dev_attr.max_mtu < if_getmtu(adapter->ifp)) {
3069 		device_printf(adapter->pdev,
3070 		    "Error, device max mtu is smaller than ifp MTU\n");
3071 		return (EINVAL);
3072 	}
3073 
3074 	return 0;
3075 }
3076 
3077 int
3078 ena_restore_device(struct ena_adapter *adapter)
3079 {
3080 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3081 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3082 	if_t ifp = adapter->ifp;
3083 	device_t dev = adapter->pdev;
3084 	int wd_active;
3085 	int rc;
3086 
3087 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3088 
3089 	rc = ena_device_init(adapter, dev, &get_feat_ctx, &wd_active);
3090 	if (rc != 0) {
3091 		device_printf(dev, "Cannot initialize device\n");
3092 		goto err;
3093 	}
3094 	/*
3095 	 * Only enable WD if it was enabled before reset, so it won't override
3096 	 * value set by the user by the sysctl.
3097 	 */
3098 	if (adapter->wd_active != 0)
3099 		adapter->wd_active = wd_active;
3100 
3101 	rc = ena_device_validate_params(adapter, &get_feat_ctx);
3102 	if (rc != 0) {
3103 		device_printf(dev, "Validation of device parameters failed\n");
3104 		goto err_device_destroy;
3105 	}
3106 
3107 	rc = ena_handle_updated_queues(adapter, &get_feat_ctx);
3108 	if (rc != 0)
3109 		goto err_device_destroy;
3110 
3111 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3112 	/* Make sure we don't have a race with AENQ Links state handler */
3113 	if (ENA_FLAG_ISSET(ENA_FLAG_LINK_UP, adapter))
3114 		if_link_state_change(ifp, LINK_STATE_UP);
3115 
3116 	rc = ena_enable_msix_and_set_admin_interrupts(adapter,
3117 	    adapter->num_queues);
3118 	if (rc != 0) {
3119 		device_printf(dev, "Enable MSI-X failed\n");
3120 		goto err_device_destroy;
3121 	}
3122 
3123 	/* If the interface was up before the reset bring it up */
3124 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter)) {
3125 		rc = ena_up(adapter);
3126 		if (rc != 0) {
3127 			device_printf(dev, "Failed to create I/O queues\n");
3128 			goto err_disable_msix;
3129 		}
3130 	}
3131 
3132 	/* Indicate that device is running again and ready to work */
3133 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3134 
3135 	if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter)) {
3136 		/*
3137 		 * As the AENQ handlers weren't executed during reset because
3138 		 * the flag ENA_FLAG_DEVICE_RUNNING was turned off, the
3139 		 * timestamp must be updated again That will prevent next reset
3140 		 * caused by missing keep alive.
3141 		 */
3142 		adapter->keep_alive_timestamp = getsbinuptime();
3143 		callout_reset_sbt(&adapter->timer_service, SBT_1S, SBT_1S,
3144 		    ena_timer_service, (void *)adapter, 0);
3145 	}
3146 
3147 	device_printf(dev,
3148 	    "Device reset completed successfully, Driver info: %s\n", ena_version);
3149 
3150 	return (rc);
3151 
3152 err_disable_msix:
3153 	ena_free_mgmnt_irq(adapter);
3154 	ena_disable_msix(adapter);
3155 err_device_destroy:
3156 	ena_com_abort_admin_commands(ena_dev);
3157 	ena_com_wait_for_abort_completion(ena_dev);
3158 	ena_com_admin_destroy(ena_dev);
3159 	ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3160 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3161 err:
3162 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3163 	ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_ONGOING_RESET, adapter);
3164 	device_printf(dev, "Reset attempt failed. Can not reset the device\n");
3165 
3166 	return (rc);
3167 }
3168 
3169 static void
3170 ena_reset_task(void *arg, int pending)
3171 {
3172 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3173 
3174 	if (unlikely(!ENA_FLAG_ISSET(ENA_FLAG_TRIGGER_RESET, adapter))) {
3175 		device_printf(adapter->pdev,
3176 		    "device reset scheduled but trigger_reset is off\n");
3177 		return;
3178 	}
3179 
3180 	sx_xlock(&adapter->ioctl_sx);
3181 	ena_destroy_device(adapter, false);
3182 	ena_restore_device(adapter);
3183 	sx_unlock(&adapter->ioctl_sx);
3184 }
3185 
3186 /**
3187  * ena_attach - Device Initialization Routine
3188  * @pdev: device information struct
3189  *
3190  * Returns 0 on success, otherwise on failure.
3191  *
3192  * ena_attach initializes an adapter identified by a device structure.
3193  * The OS initialization, configuring of the adapter private structure,
3194  * and a hardware reset occur.
3195  **/
3196 static int
3197 ena_attach(device_t pdev)
3198 {
3199 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3200 	struct ena_llq_configurations llq_config;
3201 	struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
3202 	static int version_printed;
3203 	struct ena_adapter *adapter;
3204 	struct ena_com_dev *ena_dev = NULL;
3205 	const char *queue_type_str;
3206 	int io_queue_num;
3207 	int rid, rc;
3208 
3209 	adapter = device_get_softc(pdev);
3210 	adapter->pdev = pdev;
3211 
3212 	mtx_init(&adapter->global_mtx, "ENA global mtx", NULL, MTX_DEF);
3213 	sx_init(&adapter->ioctl_sx, "ENA ioctl sx");
3214 
3215 	/* Set up the timer service */
3216 	callout_init_mtx(&adapter->timer_service, &adapter->global_mtx, 0);
3217 	adapter->keep_alive_timeout = DEFAULT_KEEP_ALIVE_TO;
3218 	adapter->missing_tx_timeout = DEFAULT_TX_CMP_TO;
3219 	adapter->missing_tx_max_queues = DEFAULT_TX_MONITORED_QUEUES;
3220 	adapter->missing_tx_threshold = DEFAULT_TX_CMP_THRESHOLD;
3221 
3222 	if (version_printed++ == 0)
3223 		device_printf(pdev, "%s\n", ena_version);
3224 
3225 	/* Allocate memory for ena_dev structure */
3226 	ena_dev = malloc(sizeof(struct ena_com_dev), M_DEVBUF,
3227 	    M_WAITOK | M_ZERO);
3228 
3229 	adapter->ena_dev = ena_dev;
3230 	ena_dev->dmadev = pdev;
3231 
3232 	rid = PCIR_BAR(ENA_REG_BAR);
3233 	adapter->memory = NULL;
3234 	adapter->registers = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
3235 	    &rid, RF_ACTIVE);
3236 	if (unlikely(adapter->registers == NULL)) {
3237 		device_printf(pdev,
3238 		    "unable to allocate bus resource: registers!\n");
3239 		rc = ENOMEM;
3240 		goto err_dev_free;
3241 	}
3242 
3243 	ena_dev->bus = malloc(sizeof(struct ena_bus), M_DEVBUF,
3244 	    M_WAITOK | M_ZERO);
3245 
3246 	/* Store register resources */
3247 	((struct ena_bus*)(ena_dev->bus))->reg_bar_t =
3248 	    rman_get_bustag(adapter->registers);
3249 	((struct ena_bus*)(ena_dev->bus))->reg_bar_h =
3250 	    rman_get_bushandle(adapter->registers);
3251 
3252 	if (unlikely(((struct ena_bus*)(ena_dev->bus))->reg_bar_h == 0)) {
3253 		device_printf(pdev, "failed to pmap registers bar\n");
3254 		rc = ENXIO;
3255 		goto err_bus_free;
3256 	}
3257 
3258 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3259 
3260 	/* Initially clear all the flags */
3261 	ENA_FLAG_ZERO(adapter);
3262 
3263 	/* Device initialization */
3264 	rc = ena_device_init(adapter, pdev, &get_feat_ctx, &adapter->wd_active);
3265 	if (unlikely(rc != 0)) {
3266 		device_printf(pdev, "ENA device init failed! (err: %d)\n", rc);
3267 		rc = ENXIO;
3268 		goto err_bus_free;
3269 	}
3270 
3271 	set_default_llq_configurations(&llq_config);
3272 
3273 	rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
3274 	     &llq_config);
3275 	if (unlikely(rc != 0)) {
3276 		device_printf(pdev, "failed to set placement policy\n");
3277 		goto err_com_free;
3278 	}
3279 
3280 	if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3281 		queue_type_str = "Regular";
3282 	else
3283 		queue_type_str = "Low Latency";
3284 	device_printf(pdev, "Placement policy: %s\n", queue_type_str);
3285 
3286 	adapter->keep_alive_timestamp = getsbinuptime();
3287 
3288 	adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3289 
3290 	memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3291 	    ETHER_ADDR_LEN);
3292 
3293 	calc_queue_ctx.ena_dev = ena_dev;
3294 	calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
3295 	calc_queue_ctx.pdev = pdev;
3296 
3297 	/* calculate IO queue number to create */
3298 	io_queue_num = ena_calc_io_queue_num(adapter, &get_feat_ctx);
3299 
3300 	ENA_ASSERT(io_queue_num > 0, "Invalid queue number: %d\n",
3301 	    io_queue_num);
3302 	adapter->num_queues = io_queue_num;
3303 
3304 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
3305 	// Set the requested Rx ring size
3306 	adapter->rx_ring_size = ENA_DEFAULT_RING_SIZE;
3307 	/* calculatre ring sizes */
3308 	rc = ena_calc_queue_size(adapter, &calc_queue_ctx);
3309 	if (unlikely((rc != 0) || (io_queue_num <= 0))) {
3310 		rc = EFAULT;
3311 		goto err_com_free;
3312 	}
3313 
3314 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3315 
3316 	adapter->tx_ring_size = calc_queue_ctx.tx_queue_size;
3317 	adapter->rx_ring_size = calc_queue_ctx.rx_queue_size;
3318 
3319 	adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
3320 	adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
3321 
3322 	adapter->buf_ring_size = ENA_DEFAULT_BUF_RING_SIZE;
3323 
3324 	/* set up dma tags for rx and tx buffers */
3325 	rc = ena_setup_tx_dma_tag(adapter);
3326 	if (unlikely(rc != 0)) {
3327 		device_printf(pdev, "Failed to create TX DMA tag\n");
3328 		goto err_com_free;
3329 	}
3330 
3331 	rc = ena_setup_rx_dma_tag(adapter);
3332 	if (unlikely(rc != 0)) {
3333 		device_printf(pdev, "Failed to create RX DMA tag\n");
3334 		goto err_tx_tag_free;
3335 	}
3336 
3337 	/* initialize rings basic information */
3338 	device_printf(pdev,
3339 	    "Creating %d io queues. Rx queue size: %d, Tx queue size: %d\n",
3340 	    io_queue_num,
3341 	    calc_queue_ctx.rx_queue_size,
3342 	    calc_queue_ctx.tx_queue_size);
3343 	ena_init_io_rings(adapter);
3344 
3345 	rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3346 	if (unlikely(rc != 0)) {
3347 		device_printf(pdev,
3348 		    "Failed to enable and set the admin interrupts\n");
3349 		goto err_io_free;
3350 	}
3351 
3352 	/* setup network interface */
3353 	rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
3354 	if (unlikely(rc != 0)) {
3355 		device_printf(pdev, "Error with network interface setup\n");
3356 		goto err_msix_free;
3357 	}
3358 
3359 	/* Initialize reset task queue */
3360 	TASK_INIT(&adapter->reset_task, 0, ena_reset_task, adapter);
3361 	adapter->reset_tq = taskqueue_create("ena_reset_enqueue",
3362 	    M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->reset_tq);
3363 	taskqueue_start_threads(&adapter->reset_tq, 1, PI_NET,
3364 	    "%s rstq", device_get_nameunit(adapter->pdev));
3365 
3366 	/* Initialize statistics */
3367 	ena_alloc_counters((counter_u64_t *)&adapter->dev_stats,
3368 	    sizeof(struct ena_stats_dev));
3369 	ena_alloc_counters((counter_u64_t *)&adapter->hw_stats,
3370 	    sizeof(struct ena_hw_stats));
3371 	ena_sysctl_add_nodes(adapter);
3372 
3373 #ifdef DEV_NETMAP
3374 	rc = ena_netmap_attach(adapter);
3375 	if (rc != 0) {
3376 		device_printf(pdev, "netmap attach failed: %d\n", rc);
3377 		goto err_detach;
3378 	}
3379 #endif /* DEV_NETMAP */
3380 
3381 	/* Tell the stack that the interface is not active */
3382 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
3383 	ENA_FLAG_SET_ATOMIC(ENA_FLAG_DEVICE_RUNNING, adapter);
3384 
3385 	return (0);
3386 
3387 #ifdef DEV_NETMAP
3388 err_detach:
3389 	ether_ifdetach(adapter->ifp);
3390 #endif /* DEV_NETMAP */
3391 err_msix_free:
3392 	ena_com_dev_reset(adapter->ena_dev, ENA_REGS_RESET_INIT_ERR);
3393 	ena_free_mgmnt_irq(adapter);
3394 	ena_disable_msix(adapter);
3395 err_io_free:
3396 	ena_free_all_io_rings_resources(adapter);
3397 	ena_free_rx_dma_tag(adapter);
3398 err_tx_tag_free:
3399 	ena_free_tx_dma_tag(adapter);
3400 err_com_free:
3401 	ena_com_admin_destroy(ena_dev);
3402 	ena_com_delete_host_info(ena_dev);
3403 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3404 err_bus_free:
3405 	free(ena_dev->bus, M_DEVBUF);
3406 	ena_free_pci_resources(adapter);
3407 err_dev_free:
3408 	free(ena_dev, M_DEVBUF);
3409 
3410 	return (rc);
3411 }
3412 
3413 /**
3414  * ena_detach - Device Removal Routine
3415  * @pdev: device information struct
3416  *
3417  * ena_detach is called by the device subsystem to alert the driver
3418  * that it should release a PCI device.
3419  **/
3420 static int
3421 ena_detach(device_t pdev)
3422 {
3423 	struct ena_adapter *adapter = device_get_softc(pdev);
3424 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3425 	int rc;
3426 
3427 	/* Make sure VLANS are not using driver */
3428 	if (adapter->ifp->if_vlantrunk != NULL) {
3429 		device_printf(adapter->pdev ,"VLAN is in use, detach first\n");
3430 		return (EBUSY);
3431 	}
3432 
3433 	ether_ifdetach(adapter->ifp);
3434 
3435 	/* Free reset task and callout */
3436 	callout_drain(&adapter->timer_service);
3437 	while (taskqueue_cancel(adapter->reset_tq, &adapter->reset_task, NULL))
3438 		taskqueue_drain(adapter->reset_tq, &adapter->reset_task);
3439 	taskqueue_free(adapter->reset_tq);
3440 
3441 	sx_xlock(&adapter->ioctl_sx);
3442 	ena_down(adapter);
3443 	ena_destroy_device(adapter, true);
3444 	sx_unlock(&adapter->ioctl_sx);
3445 
3446 #ifdef DEV_NETMAP
3447 	netmap_detach(adapter->ifp);
3448 #endif /* DEV_NETMAP */
3449 
3450 	ena_free_all_io_rings_resources(adapter);
3451 
3452 	ena_free_counters((counter_u64_t *)&adapter->hw_stats,
3453 	    sizeof(struct ena_hw_stats));
3454 	ena_free_counters((counter_u64_t *)&adapter->dev_stats,
3455 	    sizeof(struct ena_stats_dev));
3456 
3457 	rc = ena_free_rx_dma_tag(adapter);
3458 	if (unlikely(rc != 0))
3459 		device_printf(adapter->pdev,
3460 		    "Unmapped RX DMA tag associations\n");
3461 
3462 	rc = ena_free_tx_dma_tag(adapter);
3463 	if (unlikely(rc != 0))
3464 		device_printf(adapter->pdev,
3465 		    "Unmapped TX DMA tag associations\n");
3466 
3467 	ena_free_irqs(adapter);
3468 
3469 	ena_free_pci_resources(adapter);
3470 
3471 	if (likely(ENA_FLAG_ISSET(ENA_FLAG_RSS_ACTIVE, adapter)))
3472 		ena_com_rss_destroy(ena_dev);
3473 
3474 	ena_com_delete_host_info(ena_dev);
3475 
3476 	mtx_destroy(&adapter->global_mtx);
3477 	sx_destroy(&adapter->ioctl_sx);
3478 
3479 	if_free(adapter->ifp);
3480 
3481 	if (ena_dev->bus != NULL)
3482 		free(ena_dev->bus, M_DEVBUF);
3483 
3484 	if (ena_dev != NULL)
3485 		free(ena_dev, M_DEVBUF);
3486 
3487 	return (bus_generic_detach(pdev));
3488 }
3489 
3490 /******************************************************************************
3491  ******************************** AENQ Handlers *******************************
3492  *****************************************************************************/
3493 /**
3494  * ena_update_on_link_change:
3495  * Notify the network interface about the change in link status
3496  **/
3497 static void
3498 ena_update_on_link_change(void *adapter_data,
3499     struct ena_admin_aenq_entry *aenq_e)
3500 {
3501 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3502 	struct ena_admin_aenq_link_change_desc *aenq_desc;
3503 	int status;
3504 	if_t ifp;
3505 
3506 	aenq_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
3507 	ifp = adapter->ifp;
3508 	status = aenq_desc->flags &
3509 	    ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3510 
3511 	if (status != 0) {
3512 		device_printf(adapter->pdev, "link is UP\n");
3513 		ENA_FLAG_SET_ATOMIC(ENA_FLAG_LINK_UP, adapter);
3514 		if (!ENA_FLAG_ISSET(ENA_FLAG_ONGOING_RESET, adapter))
3515 			if_link_state_change(ifp, LINK_STATE_UP);
3516 	} else {
3517 		device_printf(adapter->pdev, "link is DOWN\n");
3518 		if_link_state_change(ifp, LINK_STATE_DOWN);
3519 		ENA_FLAG_CLEAR_ATOMIC(ENA_FLAG_LINK_UP, adapter);
3520 	}
3521 }
3522 
3523 static void ena_notification(void *adapter_data,
3524     struct ena_admin_aenq_entry *aenq_e)
3525 {
3526 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3527 	struct ena_admin_ena_hw_hints *hints;
3528 
3529 	ENA_WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3530 	    "Invalid group(%x) expected %x\n",	aenq_e->aenq_common_desc.group,
3531 	    ENA_ADMIN_NOTIFICATION);
3532 
3533 	switch (aenq_e->aenq_common_desc.syndrom) {
3534 	case ENA_ADMIN_UPDATE_HINTS:
3535 		hints =
3536 		    (struct ena_admin_ena_hw_hints *)(&aenq_e->inline_data_w4);
3537 		ena_update_hints(adapter, hints);
3538 		break;
3539 	default:
3540 		device_printf(adapter->pdev,
3541 		    "Invalid aenq notification link state %d\n",
3542 		    aenq_e->aenq_common_desc.syndrom);
3543 	}
3544 }
3545 
3546 /**
3547  * This handler will called for unknown event group or unimplemented handlers
3548  **/
3549 static void
3550 unimplemented_aenq_handler(void *adapter_data,
3551     struct ena_admin_aenq_entry *aenq_e)
3552 {
3553 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3554 
3555 	device_printf(adapter->pdev,
3556 	    "Unknown event was received or event with unimplemented handler\n");
3557 }
3558 
3559 static struct ena_aenq_handlers aenq_handlers = {
3560     .handlers = {
3561 	    [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3562 	    [ENA_ADMIN_NOTIFICATION] = ena_notification,
3563 	    [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3564     },
3565     .unimplemented_handler = unimplemented_aenq_handler
3566 };
3567 
3568 /*********************************************************************
3569  *  FreeBSD Device Interface Entry Points
3570  *********************************************************************/
3571 
3572 static device_method_t ena_methods[] = {
3573     /* Device interface */
3574     DEVMETHOD(device_probe, ena_probe),
3575     DEVMETHOD(device_attach, ena_attach),
3576     DEVMETHOD(device_detach, ena_detach),
3577     DEVMETHOD_END
3578 };
3579 
3580 static driver_t ena_driver = {
3581     "ena", ena_methods, sizeof(struct ena_adapter),
3582 };
3583 
3584 devclass_t ena_devclass;
3585 DRIVER_MODULE(ena, pci, ena_driver, ena_devclass, 0, 0);
3586 MODULE_PNP_INFO("U16:vendor;U16:device", pci, ena, ena_vendor_info_array,
3587     nitems(ena_vendor_info_array) - 1);
3588 MODULE_DEPEND(ena, pci, 1, 1, 1);
3589 MODULE_DEPEND(ena, ether, 1, 1, 1);
3590 #ifdef DEV_NETMAP
3591 MODULE_DEPEND(ena, netmap, 1, 1, 1);
3592 #endif /* DEV_NETMAP */
3593 
3594 /*********************************************************************/
3595