xref: /freebsd/sys/dev/ena/ena.c (revision 3cfadb28c3f6f21cb937f7331bfa066c0edda29c)
1 /*-
2  * BSD LICENSE
3  *
4  * Copyright (c) 2015-2017 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/rss_config.h>
63 #include <net/if_types.h>
64 #include <net/if_vlan_var.h>
65 
66 #include <netinet/in_rss.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in.h>
69 #include <netinet/if_ether.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet/tcp.h>
73 #include <netinet/udp.h>
74 
75 #include <dev/pci/pcivar.h>
76 #include <dev/pci/pcireg.h>
77 
78 #include "ena.h"
79 #include "ena_sysctl.h"
80 
81 /*********************************************************
82  *  Function prototypes
83  *********************************************************/
84 static int	ena_probe(device_t);
85 static void	ena_intr_msix_mgmnt(void *);
86 static int	ena_allocate_pci_resources(struct ena_adapter*);
87 static void	ena_free_pci_resources(struct ena_adapter *);
88 static int	ena_change_mtu(if_t, int);
89 static inline void ena_alloc_counters(counter_u64_t *, int);
90 static inline void ena_free_counters(counter_u64_t *, int);
91 static inline void ena_reset_counters(counter_u64_t *, int);
92 static void	ena_init_io_rings_common(struct ena_adapter *,
93     struct ena_ring *, uint16_t);
94 static void	ena_init_io_rings(struct ena_adapter *);
95 static void	ena_free_io_ring_resources(struct ena_adapter *, unsigned int);
96 static void	ena_free_all_io_rings_resources(struct ena_adapter *);
97 static int	ena_setup_tx_dma_tag(struct ena_adapter *);
98 static int	ena_free_tx_dma_tag(struct ena_adapter *);
99 static int	ena_setup_rx_dma_tag(struct ena_adapter *);
100 static int	ena_free_rx_dma_tag(struct ena_adapter *);
101 static int	ena_setup_tx_resources(struct ena_adapter *, int);
102 static void	ena_free_tx_resources(struct ena_adapter *, int);
103 static int	ena_setup_all_tx_resources(struct ena_adapter *);
104 static void	ena_free_all_tx_resources(struct ena_adapter *);
105 static inline int validate_rx_req_id(struct ena_ring *, uint16_t);
106 static int	ena_setup_rx_resources(struct ena_adapter *, unsigned int);
107 static void	ena_free_rx_resources(struct ena_adapter *, unsigned int);
108 static int	ena_setup_all_rx_resources(struct ena_adapter *);
109 static void	ena_free_all_rx_resources(struct ena_adapter *);
110 static inline int ena_alloc_rx_mbuf(struct ena_adapter *, struct ena_ring *,
111     struct ena_rx_buffer *);
112 static void	ena_free_rx_mbuf(struct ena_adapter *, struct ena_ring *,
113     struct ena_rx_buffer *);
114 static int	ena_refill_rx_bufs(struct ena_ring *, uint32_t);
115 static void	ena_free_rx_bufs(struct ena_adapter *, unsigned int);
116 static void	ena_refill_all_rx_bufs(struct ena_adapter *);
117 static void	ena_free_all_rx_bufs(struct ena_adapter *);
118 static void	ena_free_tx_bufs(struct ena_adapter *, unsigned int);
119 static void	ena_free_all_tx_bufs(struct ena_adapter *);
120 static void	ena_destroy_all_tx_queues(struct ena_adapter *);
121 static void	ena_destroy_all_rx_queues(struct ena_adapter *);
122 static void	ena_destroy_all_io_queues(struct ena_adapter *);
123 static int	ena_create_io_queues(struct ena_adapter *);
124 static int	ena_tx_cleanup(struct ena_ring *);
125 static void	ena_deferred_rx_cleanup(void *, int);
126 static int	ena_rx_cleanup(struct ena_ring *);
127 static inline int validate_tx_req_id(struct ena_ring *, uint16_t);
128 static void	ena_rx_hash_mbuf(struct ena_ring *, struct ena_com_rx_ctx *,
129     struct mbuf *);
130 static struct mbuf* ena_rx_mbuf(struct ena_ring *, struct ena_com_rx_buf_info *,
131     struct ena_com_rx_ctx *, uint16_t *);
132 static inline void ena_rx_checksum(struct ena_ring *, struct ena_com_rx_ctx *,
133     struct mbuf *);
134 static void	ena_handle_msix(void *);
135 static int	ena_enable_msix(struct ena_adapter *);
136 static void	ena_setup_mgmnt_intr(struct ena_adapter *);
137 static void	ena_setup_io_intr(struct ena_adapter *);
138 static int	ena_request_mgmnt_irq(struct ena_adapter *);
139 static int	ena_request_io_irq(struct ena_adapter *);
140 static void	ena_free_mgmnt_irq(struct ena_adapter *);
141 static void	ena_free_io_irq(struct ena_adapter *);
142 static void	ena_free_irqs(struct ena_adapter*);
143 static void	ena_disable_msix(struct ena_adapter *);
144 static void	ena_unmask_all_io_irqs(struct ena_adapter *);
145 static int	ena_rss_configure(struct ena_adapter *);
146 static int	ena_up_complete(struct ena_adapter *);
147 static int	ena_up(struct ena_adapter *);
148 static void	ena_down(struct ena_adapter *);
149 static uint64_t	ena_get_counter(if_t, ift_counter);
150 static int	ena_media_change(if_t);
151 static void	ena_media_status(if_t, struct ifmediareq *);
152 static void	ena_init(void *);
153 static int	ena_ioctl(if_t, u_long, caddr_t);
154 static int	ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *);
155 static void	ena_update_host_info(struct ena_admin_host_info *, if_t);
156 static void	ena_update_hwassist(struct ena_adapter *);
157 static int	ena_setup_ifnet(device_t, struct ena_adapter *,
158     struct ena_com_dev_get_features_ctx *);
159 static void	ena_tx_csum(struct ena_com_tx_ctx *, struct mbuf *);
160 static int	ena_check_and_collapse_mbuf(struct ena_ring *tx_ring,
161     struct mbuf **mbuf);
162 static int	ena_xmit_mbuf(struct ena_ring *, struct mbuf **);
163 static void	ena_start_xmit(struct ena_ring *);
164 static int	ena_mq_start(if_t, struct mbuf *);
165 static void	ena_deferred_mq_start(void *, int);
166 static void	ena_qflush(if_t);
167 static int	ena_calc_io_queue_num(struct ena_adapter *,
168     struct ena_com_dev_get_features_ctx *);
169 static int	ena_calc_queue_size(struct ena_adapter *, uint16_t *,
170     uint16_t *, struct ena_com_dev_get_features_ctx *);
171 static int	ena_rss_init_default(struct ena_adapter *);
172 static void	ena_rss_init_default_deferred(void *);
173 static void	ena_config_host_info(struct ena_com_dev *);
174 static int	ena_attach(device_t);
175 static int	ena_detach(device_t);
176 static int	ena_device_init(struct ena_adapter *, device_t,
177     struct ena_com_dev_get_features_ctx *, int *);
178 static int	ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *,
179     int);
180 static void ena_update_on_link_change(void *, struct ena_admin_aenq_entry *);
181 static void	unimplemented_aenq_handler(void *,
182     struct ena_admin_aenq_entry *);
183 static void	ena_timer_service(void *);
184 
185 static char ena_version[] = DEVICE_NAME DRV_MODULE_NAME " v" DRV_MODULE_VERSION;
186 
187 static SYSCTL_NODE(_hw, OID_AUTO, ena, CTLFLAG_RD, 0, "ENA driver parameters");
188 
189 /*
190  * Tuneable number of buffers in the buf-ring (drbr)
191  */
192 static int ena_buf_ring_size = 4096;
193 SYSCTL_INT(_hw_ena, OID_AUTO, buf_ring_size, CTLFLAG_RWTUN,
194     &ena_buf_ring_size, 0, "Size of the bufring");
195 
196 /*
197  * Logging level for changing verbosity of the output
198  */
199 int ena_log_level = ENA_ALERT | ENA_WARNING;
200 SYSCTL_INT(_hw_ena, OID_AUTO, log_level, CTLFLAG_RWTUN,
201     &ena_log_level, 0, "Logging level indicating verbosity of the logs");
202 
203 static ena_vendor_info_t ena_vendor_info_array[] = {
204     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0},
205     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_PF, 0},
206     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0},
207     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_VF, 0},
208     /* Last entry */
209     { 0, 0, 0 }
210 };
211 
212 /*
213  * Contains pointers to event handlers, e.g. link state chage.
214  */
215 static struct ena_aenq_handlers aenq_handlers;
216 
217 void
218 ena_dmamap_callback(void *arg, bus_dma_segment_t *segs, int nseg, int error)
219 {
220 	if (error != 0)
221 		return;
222 	*(bus_addr_t *) arg = segs[0].ds_addr;
223 }
224 
225 int
226 ena_dma_alloc(device_t dmadev, bus_size_t size,
227     ena_mem_handle_t *dma , int mapflags)
228 {
229 	struct ena_adapter* adapter = device_get_softc(dmadev);
230 	uint32_t maxsize;
231 	uint64_t dma_space_addr;
232 	int error;
233 
234 	maxsize = ((size - 1) / PAGE_SIZE + 1) * PAGE_SIZE;
235 
236 	dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
237 	if (unlikely(dma_space_addr == 0))
238 		dma_space_addr = BUS_SPACE_MAXADDR;
239 
240 	error = bus_dma_tag_create(bus_get_dma_tag(dmadev), /* parent */
241 	    8, 0,	      /* alignment, bounds 		*/
242 	    dma_space_addr,   /* lowaddr of exclusion window	*/
243 	    BUS_SPACE_MAXADDR,/* highaddr of exclusion window	*/
244 	    NULL, NULL,	      /* filter, filterarg 		*/
245 	    maxsize,	      /* maxsize 			*/
246 	    1,		      /* nsegments 			*/
247 	    maxsize,	      /* maxsegsize 			*/
248 	    BUS_DMA_ALLOCNOW, /* flags 				*/
249 	    NULL,	      /* lockfunc 			*/
250 	    NULL,	      /* lockarg 			*/
251 	    &dma->tag);
252 	if (unlikely(error != 0)) {
253 		ena_trace(ENA_ALERT, "bus_dma_tag_create failed: %d\n", error);
254 		goto fail_tag;
255 	}
256 
257 	error = bus_dmamem_alloc(dma->tag, (void**) &dma->vaddr,
258 	    BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->map);
259 	if (unlikely(error != 0)) {
260 		ena_trace(ENA_ALERT, "bus_dmamem_alloc(%ju) failed: %d\n",
261 		    (uintmax_t)size, error);
262 		goto fail_map_create;
263 	}
264 
265 	dma->paddr = 0;
266 	error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr,
267 	    size, ena_dmamap_callback, &dma->paddr, mapflags);
268 	if (unlikely((error != 0) || (dma->paddr == 0))) {
269 		ena_trace(ENA_ALERT, ": bus_dmamap_load failed: %d\n", error);
270 		goto fail_map_load;
271 	}
272 
273 	return (0);
274 
275 fail_map_load:
276 	bus_dmamem_free(dma->tag, dma->vaddr, dma->map);
277 fail_map_create:
278 	bus_dma_tag_destroy(dma->tag);
279 fail_tag:
280 	dma->tag = NULL;
281 
282 	return (error);
283 }
284 
285 static int
286 ena_allocate_pci_resources(struct ena_adapter* adapter)
287 {
288 	device_t pdev = adapter->pdev;
289 	int rid;
290 
291 	rid = PCIR_BAR(ENA_REG_BAR);
292 	adapter->memory = NULL;
293 	adapter->registers = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
294 	    &rid, RF_ACTIVE);
295 	if (unlikely(adapter->registers == NULL)) {
296 		device_printf(pdev, "Unable to allocate bus resource: "
297 		    "registers\n");
298 		return (ENXIO);
299 	}
300 
301 	return (0);
302 }
303 
304 static void
305 ena_free_pci_resources(struct ena_adapter *adapter)
306 {
307 	device_t pdev = adapter->pdev;
308 
309 	if (adapter->memory != NULL) {
310 		bus_release_resource(pdev, SYS_RES_MEMORY,
311 		    PCIR_BAR(ENA_MEM_BAR), adapter->memory);
312 	}
313 
314 	if (adapter->registers != NULL) {
315 		bus_release_resource(pdev, SYS_RES_MEMORY,
316 		    PCIR_BAR(ENA_REG_BAR), adapter->registers);
317 	}
318 }
319 
320 static int
321 ena_probe(device_t dev)
322 {
323 	ena_vendor_info_t *ent;
324 	char		adapter_name[60];
325 	uint16_t	pci_vendor_id = 0;
326 	uint16_t	pci_device_id = 0;
327 
328 	pci_vendor_id = pci_get_vendor(dev);
329 	pci_device_id = pci_get_device(dev);
330 
331 	ent = ena_vendor_info_array;
332 	while (ent->vendor_id != 0) {
333 		if ((pci_vendor_id == ent->vendor_id) &&
334 		    (pci_device_id == ent->device_id)) {
335 			ena_trace(ENA_DBG, "vendor=%x device=%x ",
336 			    pci_vendor_id, pci_device_id);
337 
338 			sprintf(adapter_name, DEVICE_DESC);
339 			device_set_desc_copy(dev, adapter_name);
340 			return (BUS_PROBE_DEFAULT);
341 		}
342 
343 		ent++;
344 
345 	}
346 
347 	return (ENXIO);
348 }
349 
350 static int
351 ena_change_mtu(if_t ifp, int new_mtu)
352 {
353 	struct ena_adapter *adapter = if_getsoftc(ifp);
354 	int rc;
355 
356 	if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
357 		device_printf(adapter->pdev, "Invalid MTU setting. "
358 		    "new_mtu: %d max mtu: %d min mtu: %d\n",
359 		    new_mtu, adapter->max_mtu, ENA_MIN_MTU);
360 		return (EINVAL);
361 	}
362 
363 	rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
364 	if (likely(rc == 0)) {
365 		ena_trace(ENA_DBG, "set MTU to %d\n", new_mtu);
366 		if_setmtu(ifp, new_mtu);
367 	} else {
368 		device_printf(adapter->pdev, "Failed to set MTU to %d\n",
369 		    new_mtu);
370 	}
371 
372 	return (rc);
373 }
374 
375 static inline void
376 ena_alloc_counters(counter_u64_t *begin, int size)
377 {
378 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
379 
380 	for (; begin < end; ++begin)
381 		*begin = counter_u64_alloc(M_WAITOK);
382 }
383 
384 static inline void
385 ena_free_counters(counter_u64_t *begin, int size)
386 {
387 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
388 
389 	for (; begin < end; ++begin)
390 		counter_u64_free(*begin);
391 }
392 
393 static inline void
394 ena_reset_counters(counter_u64_t *begin, int size)
395 {
396 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
397 
398 	for (; begin < end; ++begin)
399 		counter_u64_zero(*begin);
400 }
401 
402 static void
403 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
404     uint16_t qid)
405 {
406 
407 	ring->qid = qid;
408 	ring->adapter = adapter;
409 	ring->ena_dev = adapter->ena_dev;
410 }
411 
412 static void
413 ena_init_io_rings(struct ena_adapter *adapter)
414 {
415 	struct ena_com_dev *ena_dev;
416 	struct ena_ring *txr, *rxr;
417 	struct ena_que *que;
418 	int i;
419 
420 	ena_dev = adapter->ena_dev;
421 
422 	for (i = 0; i < adapter->num_queues; i++) {
423 		txr = &adapter->tx_ring[i];
424 		rxr = &adapter->rx_ring[i];
425 
426 		/* TX/RX common ring state */
427 		ena_init_io_rings_common(adapter, txr, i);
428 		ena_init_io_rings_common(adapter, rxr, i);
429 
430 		/* TX specific ring state */
431 		txr->ring_size = adapter->tx_ring_size;
432 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
433 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
434 		txr->smoothed_interval =
435 		    ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
436 
437 		/* Allocate a buf ring */
438 		txr->br = buf_ring_alloc(ena_buf_ring_size, M_DEVBUF,
439 		    M_WAITOK, &txr->ring_mtx);
440 
441 		/* Alloc TX statistics. */
442 		ena_alloc_counters((counter_u64_t *)&txr->tx_stats,
443 		    sizeof(txr->tx_stats));
444 
445 		/* RX specific ring state */
446 		rxr->ring_size = adapter->rx_ring_size;
447 		rxr->smoothed_interval =
448 		    ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
449 
450 		/* Alloc RX statistics. */
451 		ena_alloc_counters((counter_u64_t *)&rxr->rx_stats,
452 		    sizeof(rxr->rx_stats));
453 
454 		/* Initialize locks */
455 		snprintf(txr->mtx_name, nitems(txr->mtx_name), "%s:tx(%d)",
456 		    device_get_nameunit(adapter->pdev), i);
457 		snprintf(rxr->mtx_name, nitems(rxr->mtx_name), "%s:rx(%d)",
458 		    device_get_nameunit(adapter->pdev), i);
459 
460 		mtx_init(&txr->ring_mtx, txr->mtx_name, NULL, MTX_DEF);
461 		mtx_init(&rxr->ring_mtx, rxr->mtx_name, NULL, MTX_DEF);
462 
463 		que = &adapter->que[i];
464 		que->adapter = adapter;
465 		que->id = i;
466 		que->tx_ring = txr;
467 		que->rx_ring = rxr;
468 
469 		txr->que = que;
470 		rxr->que = que;
471 
472 		rxr->empty_rx_queue = 0;
473 	}
474 }
475 
476 static void
477 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
478 {
479 	struct ena_ring *txr = &adapter->tx_ring[qid];
480 	struct ena_ring *rxr = &adapter->rx_ring[qid];
481 
482 	ena_free_counters((counter_u64_t *)&txr->tx_stats,
483 	    sizeof(txr->tx_stats));
484 	ena_free_counters((counter_u64_t *)&rxr->rx_stats,
485 	    sizeof(rxr->rx_stats));
486 
487 	ENA_RING_MTX_LOCK(txr);
488 	drbr_free(txr->br, M_DEVBUF);
489 	ENA_RING_MTX_UNLOCK(txr);
490 
491 	mtx_destroy(&txr->ring_mtx);
492 	mtx_destroy(&rxr->ring_mtx);
493 }
494 
495 static void
496 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
497 {
498 	int i;
499 
500 	for (i = 0; i < adapter->num_queues; i++)
501 		ena_free_io_ring_resources(adapter, i);
502 
503 }
504 
505 static int
506 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
507 {
508 	int ret;
509 
510 	/* Create DMA tag for Tx buffers */
511 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
512 	    1, 0,				  /* alignment, bounds 	     */
513 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
514 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
515 	    NULL, NULL,				  /* filter, filterarg 	     */
516 	    ENA_TSO_MAXSIZE,			  /* maxsize 		     */
517 	    adapter->max_tx_sgl_size - 1,	  /* nsegments 		     */
518 	    ENA_TSO_MAXSIZE,			  /* maxsegsize 	     */
519 	    0,					  /* flags 		     */
520 	    NULL,				  /* lockfunc 		     */
521 	    NULL,				  /* lockfuncarg 	     */
522 	    &adapter->tx_buf_tag);
523 
524 	return (ret);
525 }
526 
527 static int
528 ena_free_tx_dma_tag(struct ena_adapter *adapter)
529 {
530 	int ret;
531 
532 	ret = bus_dma_tag_destroy(adapter->tx_buf_tag);
533 
534 	if (likely(ret == 0))
535 		adapter->tx_buf_tag = NULL;
536 
537 	return (ret);
538 }
539 
540 static int
541 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
542 {
543 	int ret;
544 
545 	/* Create DMA tag for Rx buffers*/
546 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent   */
547 	    1, 0,				  /* alignment, bounds 	     */
548 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
549 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
550 	    NULL, NULL,				  /* filter, filterarg 	     */
551 	    MJUM16BYTES,			  /* maxsize 		     */
552 	    1,					  /* nsegments 		     */
553 	    MJUM16BYTES,			  /* maxsegsize 	     */
554 	    0,					  /* flags 		     */
555 	    NULL,				  /* lockfunc 		     */
556 	    NULL,				  /* lockarg 		     */
557 	    &adapter->rx_buf_tag);
558 
559 	return (ret);
560 }
561 
562 static int
563 ena_free_rx_dma_tag(struct ena_adapter *adapter)
564 {
565 	int ret;
566 
567 	ret = bus_dma_tag_destroy(adapter->rx_buf_tag);
568 
569 	if (likely(ret == 0))
570 		adapter->rx_buf_tag = NULL;
571 
572 	return (ret);
573 }
574 
575 /**
576  * ena_setup_tx_resources - allocate Tx resources (Descriptors)
577  * @adapter: network interface device structure
578  * @qid: queue index
579  *
580  * Returns 0 on success, otherwise on failure.
581  **/
582 static int
583 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
584 {
585 	struct ena_que *que = &adapter->que[qid];
586 	struct ena_ring *tx_ring = que->tx_ring;
587 	int size, i, err;
588 #ifdef	RSS
589 	cpuset_t cpu_mask;
590 #endif
591 
592 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
593 
594 	tx_ring->tx_buffer_info = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
595 	if (unlikely(tx_ring->tx_buffer_info == NULL))
596 		return (ENOMEM);
597 
598 	size = sizeof(uint16_t) * tx_ring->ring_size;
599 	tx_ring->free_tx_ids = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
600 	if (unlikely(tx_ring->free_tx_ids == NULL))
601 		goto err_buf_info_free;
602 
603 	/* Req id stack for TX OOO completions */
604 	for (i = 0; i < tx_ring->ring_size; i++)
605 		tx_ring->free_tx_ids[i] = i;
606 
607 	/* Reset TX statistics. */
608 	ena_reset_counters((counter_u64_t *)&tx_ring->tx_stats,
609 	    sizeof(tx_ring->tx_stats));
610 
611 	tx_ring->next_to_use = 0;
612 	tx_ring->next_to_clean = 0;
613 
614 	/* Make sure that drbr is empty */
615 	ENA_RING_MTX_LOCK(tx_ring);
616 	drbr_flush(adapter->ifp, tx_ring->br);
617 	ENA_RING_MTX_UNLOCK(tx_ring);
618 
619 	/* ... and create the buffer DMA maps */
620 	for (i = 0; i < tx_ring->ring_size; i++) {
621 		err = bus_dmamap_create(adapter->tx_buf_tag, 0,
622 		    &tx_ring->tx_buffer_info[i].map);
623 		if (unlikely(err != 0)) {
624 			ena_trace(ENA_ALERT,
625 			     "Unable to create Tx DMA map for buffer %d\n", i);
626 			goto err_buf_info_unmap;
627 		}
628 	}
629 
630 	/* Allocate taskqueues */
631 	TASK_INIT(&tx_ring->enqueue_task, 0, ena_deferred_mq_start, tx_ring);
632 	tx_ring->enqueue_tq = taskqueue_create_fast("ena_tx_enque", M_NOWAIT,
633 	    taskqueue_thread_enqueue, &tx_ring->enqueue_tq);
634 	if (unlikely(tx_ring->enqueue_tq == NULL)) {
635 		ena_trace(ENA_ALERT,
636 		    "Unable to create taskqueue for enqueue task\n");
637 		i = tx_ring->ring_size;
638 		goto err_buf_info_unmap;
639 	}
640 
641 	/* RSS set cpu for thread */
642 #ifdef RSS
643 	CPU_SETOF(que->cpu, &cpu_mask);
644 	taskqueue_start_threads_cpuset(&tx_ring->enqueue_tq, 1, PI_NET,
645 	    &cpu_mask, "%s tx_ring enq (bucket %d)",
646 	    device_get_nameunit(adapter->pdev), que->cpu);
647 #else /* RSS */
648 	taskqueue_start_threads(&tx_ring->enqueue_tq, 1, PI_NET,
649 	    "%s txeq %d", device_get_nameunit(adapter->pdev), que->cpu);
650 #endif /* RSS */
651 
652 	return (0);
653 
654 err_buf_info_unmap:
655 	while (i--) {
656 		bus_dmamap_destroy(adapter->tx_buf_tag,
657 		    tx_ring->tx_buffer_info[i].map);
658 	}
659 	free(tx_ring->free_tx_ids, M_DEVBUF);
660 	tx_ring->free_tx_ids = NULL;
661 err_buf_info_free:
662 	free(tx_ring->tx_buffer_info, M_DEVBUF);
663 	tx_ring->tx_buffer_info = NULL;
664 
665 	return (ENOMEM);
666 }
667 
668 /**
669  * ena_free_tx_resources - Free Tx Resources per Queue
670  * @adapter: network interface device structure
671  * @qid: queue index
672  *
673  * Free all transmit software resources
674  **/
675 static void
676 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
677 {
678 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
679 
680 	while (taskqueue_cancel(tx_ring->enqueue_tq, &tx_ring->enqueue_task,
681 	    NULL))
682 		taskqueue_drain(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
683 
684 	taskqueue_free(tx_ring->enqueue_tq);
685 
686 	ENA_RING_MTX_LOCK(tx_ring);
687 	/* Flush buffer ring, */
688 	drbr_flush(adapter->ifp, tx_ring->br);
689 
690 	/* Free buffer DMA maps, */
691 	for (int i = 0; i < tx_ring->ring_size; i++) {
692 		m_freem(tx_ring->tx_buffer_info[i].mbuf);
693 		tx_ring->tx_buffer_info[i].mbuf = NULL;
694 		bus_dmamap_unload(adapter->tx_buf_tag,
695 		    tx_ring->tx_buffer_info[i].map);
696 		bus_dmamap_destroy(adapter->tx_buf_tag,
697 		    tx_ring->tx_buffer_info[i].map);
698 	}
699 	ENA_RING_MTX_UNLOCK(tx_ring);
700 
701 	/* And free allocated memory. */
702 	free(tx_ring->tx_buffer_info, M_DEVBUF);
703 	tx_ring->tx_buffer_info = NULL;
704 
705 	free(tx_ring->free_tx_ids, M_DEVBUF);
706 	tx_ring->free_tx_ids = NULL;
707 }
708 
709 /**
710  * ena_setup_all_tx_resources - allocate all queues Tx resources
711  * @adapter: network interface device structure
712  *
713  * Returns 0 on success, otherwise on failure.
714  **/
715 static int
716 ena_setup_all_tx_resources(struct ena_adapter *adapter)
717 {
718 	int i, rc;
719 
720 	for (i = 0; i < adapter->num_queues; i++) {
721 		rc = ena_setup_tx_resources(adapter, i);
722 		if (rc != 0) {
723 			device_printf(adapter->pdev,
724 			    "Allocation for Tx Queue %u failed\n", i);
725 			goto err_setup_tx;
726 		}
727 	}
728 
729 	return (0);
730 
731 err_setup_tx:
732 	/* Rewind the index freeing the rings as we go */
733 	while (i--)
734 		ena_free_tx_resources(adapter, i);
735 	return (rc);
736 }
737 
738 /**
739  * ena_free_all_tx_resources - Free Tx Resources for All Queues
740  * @adapter: network interface device structure
741  *
742  * Free all transmit software resources
743  **/
744 static void
745 ena_free_all_tx_resources(struct ena_adapter *adapter)
746 {
747 	int i;
748 
749 	for (i = 0; i < adapter->num_queues; i++)
750 		ena_free_tx_resources(adapter, i);
751 }
752 
753 static inline int
754 validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
755 {
756 	if (likely(req_id < rx_ring->ring_size))
757 		return (0);
758 
759 	device_printf(rx_ring->adapter->pdev, "Invalid rx req_id: %hu\n",
760 	    req_id);
761 	counter_u64_add(rx_ring->rx_stats.bad_req_id, 1);
762 
763 	/* Trigger device reset */
764 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
765 	rx_ring->adapter->trigger_reset = true;
766 
767 	return (EFAULT);
768 }
769 
770 /**
771  * ena_setup_rx_resources - allocate Rx resources (Descriptors)
772  * @adapter: network interface device structure
773  * @qid: queue index
774  *
775  * Returns 0 on success, otherwise on failure.
776  **/
777 static int
778 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
779 {
780 	struct ena_que *que = &adapter->que[qid];
781 	struct ena_ring *rx_ring = que->rx_ring;
782 	int size, err, i;
783 #ifdef	RSS
784 	cpuset_t cpu_mask;
785 #endif
786 
787 	size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
788 
789 	/*
790 	 * Alloc extra element so in rx path
791 	 * we can always prefetch rx_info + 1
792 	 */
793 	size += sizeof(struct ena_rx_buffer);
794 
795 	rx_ring->rx_buffer_info = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
796 
797 	size = sizeof(uint16_t) * rx_ring->ring_size;
798 	rx_ring->free_rx_ids = malloc(size, M_DEVBUF, M_WAITOK);
799 
800 	for (i = 0; i < rx_ring->ring_size; i++)
801 		rx_ring->free_rx_ids[i] = i;
802 
803 	/* Reset RX statistics. */
804 	ena_reset_counters((counter_u64_t *)&rx_ring->rx_stats,
805 	    sizeof(rx_ring->rx_stats));
806 
807 	rx_ring->next_to_clean = 0;
808 	rx_ring->next_to_use = 0;
809 
810 	/* ... and create the buffer DMA maps */
811 	for (i = 0; i < rx_ring->ring_size; i++) {
812 		err = bus_dmamap_create(adapter->rx_buf_tag, 0,
813 		    &(rx_ring->rx_buffer_info[i].map));
814 		if (err != 0) {
815 			ena_trace(ENA_ALERT,
816 			    "Unable to create Rx DMA map for buffer %d\n", i);
817 			goto err_buf_info_unmap;
818 		}
819 	}
820 
821 	/* Create LRO for the ring */
822 	if ((adapter->ifp->if_capenable & IFCAP_LRO) != 0) {
823 		int err = tcp_lro_init(&rx_ring->lro);
824 		if (err != 0) {
825 			device_printf(adapter->pdev,
826 			    "LRO[%d] Initialization failed!\n", qid);
827 		} else {
828 			ena_trace(ENA_INFO,
829 			    "RX Soft LRO[%d] Initialized\n", qid);
830 			rx_ring->lro.ifp = adapter->ifp;
831 		}
832 	}
833 
834 	/* Allocate taskqueues */
835 	TASK_INIT(&rx_ring->cmpl_task, 0, ena_deferred_rx_cleanup, rx_ring);
836 	rx_ring->cmpl_tq = taskqueue_create_fast("ena RX completion", M_WAITOK,
837 	    taskqueue_thread_enqueue, &rx_ring->cmpl_tq);
838 
839 	/* RSS set cpu for thread */
840 #ifdef RSS
841 	CPU_SETOF(que->cpu, &cpu_mask);
842 	taskqueue_start_threads_cpuset(&rx_ring->cmpl_tq, 1, PI_NET, &cpu_mask,
843 	    "%s rx_ring cmpl (bucket %d)",
844 	    device_get_nameunit(adapter->pdev), que->cpu);
845 #else
846 	taskqueue_start_threads(&rx_ring->cmpl_tq, 1, PI_NET,
847 	    "%s rx_ring cmpl %d", device_get_nameunit(adapter->pdev), que->cpu);
848 #endif
849 
850 	return (0);
851 
852 err_buf_info_unmap:
853 	while (i--) {
854 		bus_dmamap_destroy(adapter->rx_buf_tag,
855 		    rx_ring->rx_buffer_info[i].map);
856 	}
857 
858 	free(rx_ring->free_rx_ids, M_DEVBUF);
859 	rx_ring->free_rx_ids = NULL;
860 	free(rx_ring->rx_buffer_info, M_DEVBUF);
861 	rx_ring->rx_buffer_info = NULL;
862 	return (ENOMEM);
863 }
864 
865 /**
866  * ena_free_rx_resources - Free Rx Resources
867  * @adapter: network interface device structure
868  * @qid: queue index
869  *
870  * Free all receive software resources
871  **/
872 static void
873 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
874 {
875 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
876 
877 	while (taskqueue_cancel(rx_ring->cmpl_tq, &rx_ring->cmpl_task, NULL) != 0)
878 		taskqueue_drain(rx_ring->cmpl_tq, &rx_ring->cmpl_task);
879 
880 	taskqueue_free(rx_ring->cmpl_tq);
881 
882 	/* Free buffer DMA maps, */
883 	for (int i = 0; i < rx_ring->ring_size; i++) {
884 		m_freem(rx_ring->rx_buffer_info[i].mbuf);
885 		rx_ring->rx_buffer_info[i].mbuf = NULL;
886 		bus_dmamap_unload(adapter->rx_buf_tag,
887 		    rx_ring->rx_buffer_info[i].map);
888 		bus_dmamap_destroy(adapter->rx_buf_tag,
889 		    rx_ring->rx_buffer_info[i].map);
890 	}
891 
892 	/* free LRO resources, */
893 	tcp_lro_free(&rx_ring->lro);
894 
895 	/* free allocated memory */
896 	free(rx_ring->rx_buffer_info, M_DEVBUF);
897 	rx_ring->rx_buffer_info = NULL;
898 
899 	free(rx_ring->free_rx_ids, M_DEVBUF);
900 	rx_ring->free_rx_ids = NULL;
901 }
902 
903 /**
904  * ena_setup_all_rx_resources - allocate all queues Rx resources
905  * @adapter: network interface device structure
906  *
907  * Returns 0 on success, otherwise on failure.
908  **/
909 static int
910 ena_setup_all_rx_resources(struct ena_adapter *adapter)
911 {
912 	int i, rc = 0;
913 
914 	for (i = 0; i < adapter->num_queues; i++) {
915 		rc = ena_setup_rx_resources(adapter, i);
916 		if (rc != 0) {
917 			device_printf(adapter->pdev,
918 			    "Allocation for Rx Queue %u failed\n", i);
919 			goto err_setup_rx;
920 		}
921 	}
922 	return (0);
923 
924 err_setup_rx:
925 	/* rewind the index freeing the rings as we go */
926 	while (i--)
927 		ena_free_rx_resources(adapter, i);
928 	return (rc);
929 }
930 
931 /**
932  * ena_free_all_rx_resources - Free Rx resources for all queues
933  * @adapter: network interface device structure
934  *
935  * Free all receive software resources
936  **/
937 static void
938 ena_free_all_rx_resources(struct ena_adapter *adapter)
939 {
940 	int i;
941 
942 	for (i = 0; i < adapter->num_queues; i++)
943 		ena_free_rx_resources(adapter, i);
944 }
945 
946 static inline int
947 ena_alloc_rx_mbuf(struct ena_adapter *adapter,
948     struct ena_ring *rx_ring, struct ena_rx_buffer *rx_info)
949 {
950 	struct ena_com_buf *ena_buf;
951 	bus_dma_segment_t segs[1];
952 	int nsegs, error;
953 
954 	/* if previous allocated frag is not used */
955 	if (unlikely(rx_info->mbuf != NULL))
956 		return (0);
957 
958 	/* Get mbuf using UMA allocator */
959 	rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM16BYTES);
960 
961 	if (unlikely(rx_info->mbuf == NULL)) {
962 		counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
963 		return (ENOMEM);
964 	}
965 	/* Set mbuf length*/
966 	rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = MJUM16BYTES;
967 
968 	/* Map packets for DMA */
969 	ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
970 	    "Using tag %p for buffers' DMA mapping, mbuf %p len: %d",
971 	    adapter->rx_buf_tag,rx_info->mbuf, rx_info->mbuf->m_len);
972 	error = bus_dmamap_load_mbuf_sg(adapter->rx_buf_tag, rx_info->map,
973 	    rx_info->mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
974 	if (unlikely((error != 0) || (nsegs != 1))) {
975 		ena_trace(ENA_WARNING, "failed to map mbuf, error: %d, "
976 		    "nsegs: %d\n", error, nsegs);
977 		counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
978 		goto exit;
979 
980 	}
981 
982 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map, BUS_DMASYNC_PREREAD);
983 
984 	ena_buf = &rx_info->ena_buf;
985 	ena_buf->paddr = segs[0].ds_addr;
986 	ena_buf->len = MJUM16BYTES;
987 
988 	ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
989 	    "ALLOC RX BUF: mbuf %p, rx_info %p, len %d, paddr %#jx\n",
990 	    rx_info->mbuf, rx_info,ena_buf->len, (uintmax_t)ena_buf->paddr);
991 
992 	return (0);
993 
994 exit:
995 	m_freem(rx_info->mbuf);
996 	rx_info->mbuf = NULL;
997 	return (EFAULT);
998 }
999 
1000 static void
1001 ena_free_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
1002     struct ena_rx_buffer *rx_info)
1003 {
1004 
1005 	if (rx_info->mbuf == NULL) {
1006 		ena_trace(ENA_WARNING, "Trying to free unallocated buffer\n");
1007 		return;
1008 	}
1009 
1010 	bus_dmamap_unload(adapter->rx_buf_tag, rx_info->map);
1011 	m_freem(rx_info->mbuf);
1012 	rx_info->mbuf = NULL;
1013 }
1014 
1015 /**
1016  * ena_refill_rx_bufs - Refills ring with descriptors
1017  * @rx_ring: the ring which we want to feed with free descriptors
1018  * @num: number of descriptors to refill
1019  * Refills the ring with newly allocated DMA-mapped mbufs for receiving
1020  **/
1021 static int
1022 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
1023 {
1024 	struct ena_adapter *adapter = rx_ring->adapter;
1025 	uint16_t next_to_use, req_id;
1026 	uint32_t i;
1027 	int rc;
1028 
1029 	ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC, "refill qid: %d",
1030 	    rx_ring->qid);
1031 
1032 	next_to_use = rx_ring->next_to_use;
1033 
1034 	for (i = 0; i < num; i++) {
1035 		struct ena_rx_buffer *rx_info;
1036 
1037 		ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC,
1038 		    "RX buffer - next to use: %d", next_to_use);
1039 
1040 		req_id = rx_ring->free_rx_ids[next_to_use];
1041 		rc = validate_rx_req_id(rx_ring, req_id);
1042 		if (unlikely(rc != 0))
1043 			break;
1044 
1045 		rx_info = &rx_ring->rx_buffer_info[req_id];
1046 
1047 		rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1048 		if (unlikely(rc != 0)) {
1049 			ena_trace(ENA_WARNING,
1050 			    "failed to alloc buffer for rx queue %d\n",
1051 			    rx_ring->qid);
1052 			break;
1053 		}
1054 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1055 		    &rx_info->ena_buf, req_id);
1056 		if (unlikely(rc != 0)) {
1057 			ena_trace(ENA_WARNING,
1058 			    "failed to add buffer for rx queue %d\n",
1059 			    rx_ring->qid);
1060 			break;
1061 		}
1062 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1063 		    rx_ring->ring_size);
1064 	}
1065 
1066 	if (unlikely(i < num)) {
1067 		counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1068 		ena_trace(ENA_WARNING,
1069 		     "refilled rx qid %d with only %d mbufs (from %d)\n",
1070 		     rx_ring->qid, i, num);
1071 	}
1072 
1073 	if (likely(i != 0)) {
1074 		wmb();
1075 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1076 	}
1077 	rx_ring->next_to_use = next_to_use;
1078 	return (i);
1079 }
1080 
1081 static void
1082 ena_free_rx_bufs(struct ena_adapter *adapter, unsigned int qid)
1083 {
1084 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1085 	unsigned int i;
1086 
1087 	for (i = 0; i < rx_ring->ring_size; i++) {
1088 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1089 
1090 		if (rx_info->mbuf != NULL)
1091 			ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1092 	}
1093 }
1094 
1095 /**
1096  * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1097  * @adapter: network interface device structure
1098  *
1099  */
1100 static void
1101 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1102 {
1103 	struct ena_ring *rx_ring;
1104 	int i, rc, bufs_num;
1105 
1106 	for (i = 0; i < adapter->num_queues; i++) {
1107 		rx_ring = &adapter->rx_ring[i];
1108 		bufs_num = rx_ring->ring_size - 1;
1109 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1110 
1111 		if (unlikely(rc != bufs_num))
1112 			ena_trace(ENA_WARNING, "refilling Queue %d failed. "
1113 			    "Allocated %d buffers from: %d\n", i, rc, bufs_num);
1114 	}
1115 }
1116 
1117 static void
1118 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1119 {
1120 	int i;
1121 
1122 	for (i = 0; i < adapter->num_queues; i++)
1123 		ena_free_rx_bufs(adapter, i);
1124 }
1125 
1126 /**
1127  * ena_free_tx_bufs - Free Tx Buffers per Queue
1128  * @adapter: network interface device structure
1129  * @qid: queue index
1130  **/
1131 static void
1132 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1133 {
1134 	bool print_once = true;
1135 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1136 
1137 	ENA_RING_MTX_LOCK(tx_ring);
1138 	for (int i = 0; i < tx_ring->ring_size; i++) {
1139 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1140 
1141 		if (tx_info->mbuf == NULL)
1142 			continue;
1143 
1144 		if (print_once) {
1145 			device_printf(adapter->pdev,
1146 			    "free uncompleted tx mbuf qid %d idx 0x%x",
1147 			    qid, i);
1148 			print_once = false;
1149 		} else {
1150 			ena_trace(ENA_DBG,
1151 			    "free uncompleted tx mbuf qid %d idx 0x%x",
1152 			     qid, i);
1153 		}
1154 
1155 		bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
1156 		m_free(tx_info->mbuf);
1157 		tx_info->mbuf = NULL;
1158 	}
1159 	ENA_RING_MTX_UNLOCK(tx_ring);
1160 }
1161 
1162 static void
1163 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1164 {
1165 
1166 	for (int i = 0; i < adapter->num_queues; i++)
1167 		ena_free_tx_bufs(adapter, i);
1168 }
1169 
1170 static void
1171 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1172 {
1173 	uint16_t ena_qid;
1174 	int i;
1175 
1176 	for (i = 0; i < adapter->num_queues; i++) {
1177 		ena_qid = ENA_IO_TXQ_IDX(i);
1178 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1179 	}
1180 }
1181 
1182 static void
1183 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1184 {
1185 	uint16_t ena_qid;
1186 	int i;
1187 
1188 	for (i = 0; i < adapter->num_queues; i++) {
1189 		ena_qid = ENA_IO_RXQ_IDX(i);
1190 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1191 	}
1192 }
1193 
1194 static void
1195 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1196 {
1197 	ena_destroy_all_tx_queues(adapter);
1198 	ena_destroy_all_rx_queues(adapter);
1199 }
1200 
1201 static inline int
1202 validate_tx_req_id(struct ena_ring *tx_ring, uint16_t req_id)
1203 {
1204 	struct ena_adapter *adapter = tx_ring->adapter;
1205 	struct ena_tx_buffer *tx_info = NULL;
1206 
1207 	if (likely(req_id < tx_ring->ring_size)) {
1208 		tx_info = &tx_ring->tx_buffer_info[req_id];
1209 		if (tx_info->mbuf != NULL)
1210 			return (0);
1211 	}
1212 
1213 	if (tx_info->mbuf == NULL)
1214 		device_printf(adapter->pdev,
1215 		    "tx_info doesn't have valid mbuf\n");
1216 	else
1217 		device_printf(adapter->pdev, "Invalid req_id: %hu\n", req_id);
1218 
1219 	counter_u64_add(tx_ring->tx_stats.bad_req_id, 1);
1220 
1221 	return (EFAULT);
1222 }
1223 
1224 static int
1225 ena_create_io_queues(struct ena_adapter *adapter)
1226 {
1227 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1228 	struct ena_com_create_io_ctx ctx;
1229 	struct ena_ring *ring;
1230 	uint16_t ena_qid;
1231 	uint32_t msix_vector;
1232 	int rc, i;
1233 
1234 	/* Create TX queues */
1235 	for (i = 0; i < adapter->num_queues; i++) {
1236 		msix_vector = ENA_IO_IRQ_IDX(i);
1237 		ena_qid = ENA_IO_TXQ_IDX(i);
1238 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1239 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1240 		ctx.queue_size = adapter->tx_ring_size;
1241 		ctx.msix_vector = msix_vector;
1242 		ctx.qid = ena_qid;
1243 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1244 		if (rc != 0) {
1245 			device_printf(adapter->pdev,
1246 			    "Failed to create io TX queue #%d rc: %d\n", i, rc);
1247 			goto err_tx;
1248 		}
1249 		ring = &adapter->tx_ring[i];
1250 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1251 		    &ring->ena_com_io_sq,
1252 		    &ring->ena_com_io_cq);
1253 		if (rc != 0) {
1254 			device_printf(adapter->pdev,
1255 			    "Failed to get TX queue handlers. TX queue num"
1256 			    " %d rc: %d\n", i, rc);
1257 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1258 			goto err_tx;
1259 		}
1260 	}
1261 
1262 	/* Create RX queues */
1263 	for (i = 0; i < adapter->num_queues; i++) {
1264 		msix_vector = ENA_IO_IRQ_IDX(i);
1265 		ena_qid = ENA_IO_RXQ_IDX(i);
1266 		ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1267 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1268 		ctx.queue_size = adapter->rx_ring_size;
1269 		ctx.msix_vector = msix_vector;
1270 		ctx.qid = ena_qid;
1271 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1272 		if (unlikely(rc != 0)) {
1273 			device_printf(adapter->pdev,
1274 			    "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1275 			goto err_rx;
1276 		}
1277 
1278 		ring = &adapter->rx_ring[i];
1279 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1280 		    &ring->ena_com_io_sq,
1281 		    &ring->ena_com_io_cq);
1282 		if (unlikely(rc != 0)) {
1283 			device_printf(adapter->pdev,
1284 			    "Failed to get RX queue handlers. RX queue num"
1285 			    " %d rc: %d\n", i, rc);
1286 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1287 			goto err_rx;
1288 		}
1289 	}
1290 
1291 	return (0);
1292 
1293 err_rx:
1294 	while (i--)
1295 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1296 	i = adapter->num_queues;
1297 err_tx:
1298 	while (i--)
1299 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1300 
1301 	return (ENXIO);
1302 }
1303 
1304 /**
1305  * ena_tx_cleanup - clear sent packets and corresponding descriptors
1306  * @tx_ring: ring for which we want to clean packets
1307  *
1308  * Once packets are sent, we ask the device in a loop for no longer used
1309  * descriptors. We find the related mbuf chain in a map (index in an array)
1310  * and free it, then update ring state.
1311  * This is performed in "endless" loop, updating ring pointers every
1312  * TX_COMMIT. The first check of free descriptor is performed before the actual
1313  * loop, then repeated at the loop end.
1314  **/
1315 static int
1316 ena_tx_cleanup(struct ena_ring *tx_ring)
1317 {
1318 	struct ena_adapter *adapter;
1319 	struct ena_com_io_cq* io_cq;
1320 	uint16_t next_to_clean;
1321 	uint16_t req_id;
1322 	uint16_t ena_qid;
1323 	unsigned int total_done = 0;
1324 	int rc;
1325 	int commit = TX_COMMIT;
1326 	int budget = TX_BUDGET;
1327 	int work_done;
1328 
1329 	adapter = tx_ring->que->adapter;
1330 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
1331 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1332 	next_to_clean = tx_ring->next_to_clean;
1333 
1334 	do {
1335 		struct ena_tx_buffer *tx_info;
1336 		struct mbuf *mbuf;
1337 
1338 		rc = ena_com_tx_comp_req_id_get(io_cq, &req_id);
1339 		if (unlikely(rc != 0))
1340 			break;
1341 
1342 		rc = validate_tx_req_id(tx_ring, req_id);
1343 		if (unlikely(rc != 0))
1344 			break;
1345 
1346 		tx_info = &tx_ring->tx_buffer_info[req_id];
1347 
1348 		mbuf = tx_info->mbuf;
1349 
1350 		tx_info->mbuf = NULL;
1351 		bintime_clear(&tx_info->timestamp);
1352 
1353 		if (likely(tx_info->num_of_bufs != 0)) {
1354 			/* Map is no longer required */
1355 			bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
1356 		}
1357 
1358 		ena_trace(ENA_DBG | ENA_TXPTH, "tx: q %d mbuf %p completed",
1359 		    tx_ring->qid, mbuf);
1360 
1361 		m_freem(mbuf);
1362 
1363 		total_done += tx_info->tx_descs;
1364 
1365 		tx_ring->free_tx_ids[next_to_clean] = req_id;
1366 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1367 		    tx_ring->ring_size);
1368 
1369 		if (unlikely(--commit == 0)) {
1370 			commit = TX_COMMIT;
1371 			/* update ring state every TX_COMMIT descriptor */
1372 			tx_ring->next_to_clean = next_to_clean;
1373 			ena_com_comp_ack(
1374 			    &adapter->ena_dev->io_sq_queues[ena_qid],
1375 			    total_done);
1376 			ena_com_update_dev_comp_head(io_cq);
1377 			total_done = 0;
1378 		}
1379 	} while (likely(--budget));
1380 
1381 	work_done = TX_BUDGET - budget;
1382 
1383 	ena_trace(ENA_DBG | ENA_TXPTH, "tx: q %d done. total pkts: %d",
1384 	tx_ring->qid, work_done);
1385 
1386 	/* If there is still something to commit update ring state */
1387 	if (likely(commit != TX_COMMIT)) {
1388 		tx_ring->next_to_clean = next_to_clean;
1389 		ena_com_comp_ack(&adapter->ena_dev->io_sq_queues[ena_qid],
1390 		    total_done);
1391 		ena_com_update_dev_comp_head(io_cq);
1392 	}
1393 
1394 	taskqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
1395 
1396 	return (work_done);
1397 }
1398 
1399 static void
1400 ena_rx_hash_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1401     struct mbuf *mbuf)
1402 {
1403 	struct ena_adapter *adapter = rx_ring->adapter;
1404 
1405 	if (likely(adapter->rss_support)) {
1406 		mbuf->m_pkthdr.flowid = ena_rx_ctx->hash;
1407 
1408 		if (ena_rx_ctx->frag &&
1409 		    (ena_rx_ctx->l3_proto != ENA_ETH_IO_L3_PROTO_UNKNOWN)) {
1410 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1411 			return;
1412 		}
1413 
1414 		switch (ena_rx_ctx->l3_proto) {
1415 		case ENA_ETH_IO_L3_PROTO_IPV4:
1416 			switch (ena_rx_ctx->l4_proto) {
1417 			case ENA_ETH_IO_L4_PROTO_TCP:
1418 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV4);
1419 				break;
1420 			case ENA_ETH_IO_L4_PROTO_UDP:
1421 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV4);
1422 				break;
1423 			default:
1424 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV4);
1425 			}
1426 			break;
1427 		case ENA_ETH_IO_L3_PROTO_IPV6:
1428 			switch (ena_rx_ctx->l4_proto) {
1429 			case ENA_ETH_IO_L4_PROTO_TCP:
1430 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV6);
1431 				break;
1432 			case ENA_ETH_IO_L4_PROTO_UDP:
1433 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV6);
1434 				break;
1435 			default:
1436 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV6);
1437 			}
1438 			break;
1439 		case ENA_ETH_IO_L3_PROTO_UNKNOWN:
1440 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1441 			break;
1442 		default:
1443 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1444 		}
1445 	} else {
1446 		mbuf->m_pkthdr.flowid = rx_ring->qid;
1447 		M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1448 	}
1449 }
1450 
1451 /**
1452  * ena_rx_mbuf - assemble mbuf from descriptors
1453  * @rx_ring: ring for which we want to clean packets
1454  * @ena_bufs: buffer info
1455  * @ena_rx_ctx: metadata for this packet(s)
1456  * @next_to_clean: ring pointer, will be updated only upon success
1457  *
1458  **/
1459 static struct mbuf*
1460 ena_rx_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_buf_info *ena_bufs,
1461     struct ena_com_rx_ctx *ena_rx_ctx, uint16_t *next_to_clean)
1462 {
1463 	struct mbuf *mbuf;
1464 	struct ena_rx_buffer *rx_info;
1465 	struct ena_adapter *adapter;
1466 	unsigned int descs = ena_rx_ctx->descs;
1467 	uint16_t ntc, len, req_id, buf = 0;
1468 
1469 	ntc = *next_to_clean;
1470 	adapter = rx_ring->adapter;
1471 	rx_info = &rx_ring->rx_buffer_info[ntc];
1472 
1473 	if (unlikely(rx_info->mbuf == NULL)) {
1474 		device_printf(adapter->pdev, "NULL mbuf in rx_info");
1475 		return (NULL);
1476 	}
1477 
1478 	len = ena_bufs[buf].len;
1479 	req_id = ena_bufs[buf].req_id;
1480 	rx_info = &rx_ring->rx_buffer_info[req_id];
1481 
1482 	ena_trace(ENA_DBG | ENA_RXPTH, "rx_info %p, mbuf %p, paddr %jx",
1483 	    rx_info, rx_info->mbuf, (uintmax_t)rx_info->ena_buf.paddr);
1484 
1485 	mbuf = rx_info->mbuf;
1486 	mbuf->m_flags |= M_PKTHDR;
1487 	mbuf->m_pkthdr.len = len;
1488 	mbuf->m_len = len;
1489 	mbuf->m_pkthdr.rcvif = rx_ring->que->adapter->ifp;
1490 
1491 	/* Fill mbuf with hash key and it's interpretation for optimization */
1492 	ena_rx_hash_mbuf(rx_ring, ena_rx_ctx, mbuf);
1493 
1494 	ena_trace(ENA_DBG | ENA_RXPTH, "rx mbuf 0x%p, flags=0x%x, len: %d",
1495 	    mbuf, mbuf->m_flags, mbuf->m_pkthdr.len);
1496 
1497 	/* DMA address is not needed anymore, unmap it */
1498 	bus_dmamap_unload(rx_ring->adapter->rx_buf_tag, rx_info->map);
1499 
1500 	rx_info->mbuf = NULL;
1501 	rx_ring->free_rx_ids[ntc] = req_id;
1502 	ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1503 
1504 	/*
1505 	 * While we have more than 1 descriptors for one rcvd packet, append
1506 	 * other mbufs to the main one
1507 	 */
1508 	while (--descs) {
1509 		++buf;
1510 		len = ena_bufs[buf].len;
1511 		req_id = ena_bufs[buf].req_id;
1512 		rx_info = &rx_ring->rx_buffer_info[req_id];
1513 
1514 		if (unlikely(rx_info->mbuf == NULL)) {
1515 			device_printf(adapter->pdev, "NULL mbuf in rx_info");
1516 			/*
1517 			 * If one of the required mbufs was not allocated yet,
1518 			 * we can break there.
1519 			 * All earlier used descriptors will be reallocated
1520 			 * later and not used mbufs can be reused.
1521 			 * The next_to_clean pointer will not be updated in case
1522 			 * of an error, so caller should advance it manually
1523 			 * in error handling routine to keep it up to date
1524 			 * with hw ring.
1525 			 */
1526 			m_freem(mbuf);
1527 			return (NULL);
1528 		}
1529 
1530 		if (unlikely(m_append(mbuf, len, rx_info->mbuf->m_data) == 0)) {
1531 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
1532 			ena_trace(ENA_WARNING, "Failed to append Rx mbuf %p",
1533 			    mbuf);
1534 		}
1535 
1536 		ena_trace(ENA_DBG | ENA_RXPTH,
1537 		    "rx mbuf updated. len %d", mbuf->m_pkthdr.len);
1538 
1539 		/* Free already appended mbuf, it won't be useful anymore */
1540 		bus_dmamap_unload(rx_ring->adapter->rx_buf_tag, rx_info->map);
1541 		m_freem(rx_info->mbuf);
1542 		rx_info->mbuf = NULL;
1543 
1544 		rx_ring->free_rx_ids[ntc] = req_id;
1545 		ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1546 	}
1547 
1548 	*next_to_clean = ntc;
1549 
1550 	return (mbuf);
1551 }
1552 
1553 /**
1554  * ena_rx_checksum - indicate in mbuf if hw indicated a good cksum
1555  **/
1556 static inline void
1557 ena_rx_checksum(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1558     struct mbuf *mbuf)
1559 {
1560 
1561 	/* if IP and error */
1562 	if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1563 	    ena_rx_ctx->l3_csum_err)) {
1564 		/* ipv4 checksum error */
1565 		mbuf->m_pkthdr.csum_flags = 0;
1566 		counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1567 		ena_trace(ENA_DBG, "RX IPv4 header checksum error");
1568 		return;
1569 	}
1570 
1571 	/* if TCP/UDP */
1572 	if ((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1573 	    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)) {
1574 		if (ena_rx_ctx->l4_csum_err) {
1575 			/* TCP/UDP checksum error */
1576 			mbuf->m_pkthdr.csum_flags = 0;
1577 			counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1578 			ena_trace(ENA_DBG, "RX L4 checksum error");
1579 		} else {
1580 			mbuf->m_pkthdr.csum_flags = CSUM_IP_CHECKED;
1581 			mbuf->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1582 		}
1583 	}
1584 }
1585 
1586 static void
1587 ena_deferred_rx_cleanup(void *arg, int pending)
1588 {
1589 	struct ena_ring *rx_ring = arg;
1590 	int budget = CLEAN_BUDGET;
1591 
1592 	ENA_RING_MTX_LOCK(rx_ring);
1593 	/*
1594 	 * If deferred task was executed, perform cleanup of all awaiting
1595 	 * descs (or until given budget is depleted to avoid infinite loop).
1596 	 */
1597 	while (likely(budget--)) {
1598 		if (ena_rx_cleanup(rx_ring) == 0)
1599 			break;
1600 	}
1601 	ENA_RING_MTX_UNLOCK(rx_ring);
1602 }
1603 
1604 /**
1605  * ena_rx_cleanup - handle rx irq
1606  * @arg: ring for which irq is being handled
1607  **/
1608 static int
1609 ena_rx_cleanup(struct ena_ring *rx_ring)
1610 {
1611 	struct ena_adapter *adapter;
1612 	struct mbuf *mbuf;
1613 	struct ena_com_rx_ctx ena_rx_ctx;
1614 	struct ena_com_io_cq* io_cq;
1615 	struct ena_com_io_sq* io_sq;
1616 	if_t ifp;
1617 	uint16_t ena_qid;
1618 	uint16_t next_to_clean;
1619 	uint32_t refill_required;
1620 	uint32_t refill_threshold;
1621 	uint32_t do_if_input = 0;
1622 	unsigned int qid;
1623 	int rc, i;
1624 	int budget = RX_BUDGET;
1625 
1626 	adapter = rx_ring->que->adapter;
1627 	ifp = adapter->ifp;
1628 	qid = rx_ring->que->id;
1629 	ena_qid = ENA_IO_RXQ_IDX(qid);
1630 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1631 	io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
1632 	next_to_clean = rx_ring->next_to_clean;
1633 
1634 	ena_trace(ENA_DBG, "rx: qid %d", qid);
1635 
1636 	do {
1637 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1638 		ena_rx_ctx.max_bufs = adapter->max_rx_sgl_size;
1639 		ena_rx_ctx.descs = 0;
1640 		rc = ena_com_rx_pkt(io_cq, io_sq, &ena_rx_ctx);
1641 
1642 		if (unlikely(rc != 0))
1643 			goto error;
1644 
1645 		if (unlikely(ena_rx_ctx.descs == 0))
1646 			break;
1647 
1648 		ena_trace(ENA_DBG | ENA_RXPTH, "rx: q %d got packet from ena. "
1649 		    "descs #: %d l3 proto %d l4 proto %d hash: %x",
1650 		    rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1651 		    ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1652 
1653 		/* Receive mbuf from the ring */
1654 		mbuf = ena_rx_mbuf(rx_ring, rx_ring->ena_bufs,
1655 		    &ena_rx_ctx, &next_to_clean);
1656 
1657 		/* Exit if we failed to retrieve a buffer */
1658 		if (unlikely(mbuf == NULL)) {
1659 			for (i = 0; i < ena_rx_ctx.descs; ++i) {
1660 				rx_ring->free_rx_ids[next_to_clean] =
1661 				    rx_ring->ena_bufs[i].req_id;
1662 				next_to_clean =
1663 				    ENA_RX_RING_IDX_NEXT(next_to_clean,
1664 				    rx_ring->ring_size);
1665 
1666 			}
1667 			break;
1668 		}
1669 
1670 		if (((ifp->if_capenable & IFCAP_RXCSUM) != 0) ||
1671 		    ((ifp->if_capenable & IFCAP_RXCSUM_IPV6) != 0)) {
1672 			ena_rx_checksum(rx_ring, &ena_rx_ctx, mbuf);
1673 		}
1674 
1675 		counter_enter();
1676 		counter_u64_add_protected(rx_ring->rx_stats.bytes,
1677 		    mbuf->m_pkthdr.len);
1678 		counter_u64_add_protected(adapter->hw_stats.rx_bytes,
1679 		    mbuf->m_pkthdr.len);
1680 		counter_exit();
1681 		/*
1682 		 * LRO is only for IP/TCP packets and TCP checksum of the packet
1683 		 * should be computed by hardware.
1684 		 */
1685 		do_if_input = 1;
1686 		if (((ifp->if_capenable & IFCAP_LRO) != 0)  &&
1687 		    ((mbuf->m_pkthdr.csum_flags & CSUM_IP_VALID) != 0) &&
1688 		    (ena_rx_ctx.l4_proto == ENA_ETH_IO_L4_PROTO_TCP)) {
1689 			/*
1690 			 * Send to the stack if:
1691 			 *  - LRO not enabled, or
1692 			 *  - no LRO resources, or
1693 			 *  - lro enqueue fails
1694 			 */
1695 			if ((rx_ring->lro.lro_cnt != 0) &&
1696 			    (tcp_lro_rx(&rx_ring->lro, mbuf, 0) == 0))
1697 					do_if_input = 0;
1698 		}
1699 		if (do_if_input != 0) {
1700 			ena_trace(ENA_DBG | ENA_RXPTH,
1701 			    "calling if_input() with mbuf %p", mbuf);
1702 			(*ifp->if_input)(ifp, mbuf);
1703 		}
1704 
1705 		counter_enter();
1706 		counter_u64_add_protected(rx_ring->rx_stats.cnt, 1);
1707 		counter_u64_add_protected(adapter->hw_stats.rx_packets, 1);
1708 		counter_exit();
1709 	} while (--budget);
1710 
1711 	rx_ring->next_to_clean = next_to_clean;
1712 
1713 	refill_required = ena_com_free_desc(io_sq);
1714 	refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
1715 
1716 	if (refill_required > refill_threshold) {
1717 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1718 		ena_refill_rx_bufs(rx_ring, refill_required);
1719 	}
1720 
1721 	tcp_lro_flush_all(&rx_ring->lro);
1722 
1723 	return (RX_BUDGET - budget);
1724 
1725 error:
1726 	counter_u64_add(rx_ring->rx_stats.bad_desc_num, 1);
1727 	return (RX_BUDGET - budget);
1728 }
1729 
1730 /*********************************************************************
1731  *
1732  *  MSIX & Interrupt Service routine
1733  *
1734  **********************************************************************/
1735 
1736 /**
1737  * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1738  * @arg: interrupt number
1739  **/
1740 static void
1741 ena_intr_msix_mgmnt(void *arg)
1742 {
1743 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1744 
1745 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1746 	if (likely(adapter->running))
1747 		ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1748 }
1749 
1750 /**
1751  * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1752  * @arg: interrupt number
1753  **/
1754 static void
1755 ena_handle_msix(void *arg)
1756 {
1757 	struct ena_que	*que = arg;
1758 	struct ena_adapter *adapter = que->adapter;
1759 	if_t ifp = adapter->ifp;
1760 	struct ena_ring *tx_ring;
1761 	struct ena_ring *rx_ring;
1762 	struct ena_com_io_cq* io_cq;
1763 	struct ena_eth_io_intr_reg intr_reg;
1764 	int qid, ena_qid;
1765 	int txc, rxc, i;
1766 
1767 	if (unlikely((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0))
1768 		return;
1769 
1770 	ena_trace(ENA_DBG, "MSI-X TX/RX routine");
1771 
1772 	tx_ring = que->tx_ring;
1773 	rx_ring = que->rx_ring;
1774 	qid = que->id;
1775 	ena_qid = ENA_IO_TXQ_IDX(qid);
1776 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1777 
1778 	for (i = 0; i < CLEAN_BUDGET; ++i) {
1779 		/*
1780 		 * If lock cannot be acquired, then deferred cleanup task was
1781 		 * being executed and rx ring is being cleaned up in
1782 		 * another thread.
1783 		 */
1784 		if (likely(ENA_RING_MTX_TRYLOCK(rx_ring) != 0)) {
1785 			rxc = ena_rx_cleanup(rx_ring);
1786 			ENA_RING_MTX_UNLOCK(rx_ring);
1787 		} else {
1788 			rxc = 0;
1789 		}
1790 
1791 		/* Protection from calling ena_tx_cleanup from ena_start_xmit */
1792 		ENA_RING_MTX_LOCK(tx_ring);
1793 		txc = ena_tx_cleanup(tx_ring);
1794 		ENA_RING_MTX_UNLOCK(tx_ring);
1795 
1796 		if (unlikely((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0))
1797 			return;
1798 
1799 		if ((txc != TX_BUDGET) && (rxc != RX_BUDGET))
1800 		       break;
1801 	}
1802 
1803 	/* Signal that work is done and unmask interrupt */
1804 	ena_com_update_intr_reg(&intr_reg,
1805 	    RX_IRQ_INTERVAL,
1806 	    TX_IRQ_INTERVAL,
1807 	    true);
1808 	ena_com_unmask_intr(io_cq, &intr_reg);
1809 }
1810 
1811 static int
1812 ena_enable_msix(struct ena_adapter *adapter)
1813 {
1814 	device_t dev = adapter->pdev;
1815 	int msix_vecs, msix_req;
1816 	int i, rc = 0;
1817 
1818 	/* Reserved the max msix vectors we might need */
1819 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->num_queues);
1820 
1821 	adapter->msix_entries = malloc(msix_vecs * sizeof(struct msix_entry),
1822 	    M_DEVBUF, M_WAITOK | M_ZERO);
1823 
1824 	ena_trace(ENA_DBG, "trying to enable MSI-X, vectors: %d", msix_vecs);
1825 
1826 	for (i = 0; i < msix_vecs; i++) {
1827 		adapter->msix_entries[i].entry = i;
1828 		/* Vectors must start from 1 */
1829 		adapter->msix_entries[i].vector = i + 1;
1830 	}
1831 
1832 	msix_req = msix_vecs;
1833 	rc = pci_alloc_msix(dev, &msix_vecs);
1834 	if (unlikely(rc != 0)) {
1835 		device_printf(dev,
1836 		    "Failed to enable MSIX, vectors %d rc %d\n", msix_vecs, rc);
1837 
1838 		rc = ENOSPC;
1839 		goto err_msix_free;
1840 	}
1841 
1842 	if (msix_vecs != msix_req) {
1843 		device_printf(dev, "Enable only %d MSI-x (out of %d), reduce "
1844 		    "the number of queues\n", msix_vecs, msix_req);
1845 		adapter->num_queues = msix_vecs - ENA_ADMIN_MSIX_VEC;
1846 	}
1847 
1848 	adapter->msix_vecs = msix_vecs;
1849 	adapter->msix_enabled = true;
1850 
1851 	return (0);
1852 
1853 err_msix_free:
1854 	free(adapter->msix_entries, M_DEVBUF);
1855 	adapter->msix_entries = NULL;
1856 
1857 	return (rc);
1858 }
1859 
1860 static void
1861 ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1862 {
1863 
1864 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1865 	    ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1866 	    device_get_nameunit(adapter->pdev));
1867 	/*
1868 	 * Handler is NULL on purpose, it will be set
1869 	 * when mgmnt interrupt is acquired
1870 	 */
1871 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler = NULL;
1872 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1873 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1874 	    adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
1875 }
1876 
1877 static void
1878 ena_setup_io_intr(struct ena_adapter *adapter)
1879 {
1880 	static int last_bind_cpu = -1;
1881 	int irq_idx;
1882 
1883 	for (int i = 0; i < adapter->num_queues; i++) {
1884 		irq_idx = ENA_IO_IRQ_IDX(i);
1885 
1886 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1887 		    "%s-TxRx-%d", device_get_nameunit(adapter->pdev), i);
1888 		adapter->irq_tbl[irq_idx].handler = ena_handle_msix;
1889 		adapter->irq_tbl[irq_idx].data = &adapter->que[i];
1890 		adapter->irq_tbl[irq_idx].vector =
1891 		    adapter->msix_entries[irq_idx].vector;
1892 		ena_trace(ENA_INFO | ENA_IOQ, "ena_setup_io_intr vector: %d\n",
1893 		    adapter->msix_entries[irq_idx].vector);
1894 #ifdef	RSS
1895 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1896 		    rss_getcpu(i % rss_getnumbuckets());
1897 #else
1898 		/*
1899 		 * We still want to bind rings to the corresponding cpu
1900 		 * using something similar to the RSS round-robin technique.
1901 		 */
1902 		if (unlikely(last_bind_cpu < 0))
1903 			last_bind_cpu = CPU_FIRST();
1904 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1905 		    last_bind_cpu;
1906 		last_bind_cpu = CPU_NEXT(last_bind_cpu);
1907 #endif
1908 	}
1909 }
1910 
1911 static int
1912 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1913 {
1914 	struct ena_irq *irq;
1915 	unsigned long flags;
1916 	int rc, rcc;
1917 
1918 	flags = RF_ACTIVE | RF_SHAREABLE;
1919 
1920 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1921 	irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1922 	    &irq->vector, flags);
1923 
1924 	if (unlikely(irq->res == NULL)) {
1925 		device_printf(adapter->pdev, "could not allocate "
1926 		    "irq vector: %d\n", irq->vector);
1927 		return (ENXIO);
1928 	}
1929 
1930 	rc = bus_activate_resource(adapter->pdev, SYS_RES_IRQ,
1931 	    irq->vector, irq->res);
1932 	if (unlikely(rc != 0)) {
1933 		device_printf(adapter->pdev, "could not activate "
1934 		    "irq vector: %d\n", irq->vector);
1935 		goto err_res_free;
1936 	}
1937 
1938 	rc = bus_setup_intr(adapter->pdev, irq->res,
1939 	    INTR_TYPE_NET | INTR_MPSAFE, NULL, ena_intr_msix_mgmnt,
1940 	    irq->data, &irq->cookie);
1941 	if (unlikely(rc != 0)) {
1942 		device_printf(adapter->pdev, "failed to register "
1943 		    "interrupt handler for irq %ju: %d\n",
1944 		    rman_get_start(irq->res), rc);
1945 		goto err_res_free;
1946 	}
1947 	irq->requested = true;
1948 
1949 	return (rc);
1950 
1951 err_res_free:
1952 	ena_trace(ENA_INFO | ENA_ADMQ, "releasing resource for irq %d\n",
1953 	    irq->vector);
1954 	rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
1955 	    irq->vector, irq->res);
1956 	if (unlikely(rcc != 0))
1957 		device_printf(adapter->pdev, "dev has no parent while "
1958 		    "releasing res for irq: %d\n", irq->vector);
1959 	irq->res = NULL;
1960 
1961 	return (rc);
1962 }
1963 
1964 static int
1965 ena_request_io_irq(struct ena_adapter *adapter)
1966 {
1967 	struct ena_irq *irq;
1968 	unsigned long flags = 0;
1969 	int rc = 0, i, rcc;
1970 
1971 	if (unlikely(adapter->msix_enabled == 0)) {
1972 		device_printf(adapter->pdev,
1973 		    "failed to request I/O IRQ: MSI-X is not enabled\n");
1974 		return (EINVAL);
1975 	} else {
1976 		flags = RF_ACTIVE | RF_SHAREABLE;
1977 	}
1978 
1979 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1980 		irq = &adapter->irq_tbl[i];
1981 
1982 		if (unlikely(irq->requested))
1983 			continue;
1984 
1985 		irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1986 		    &irq->vector, flags);
1987 		if (unlikely(irq->res == NULL)) {
1988 			device_printf(adapter->pdev, "could not allocate "
1989 			    "irq vector: %d\n", irq->vector);
1990 			goto err;
1991 		}
1992 
1993 		rc = bus_setup_intr(adapter->pdev, irq->res,
1994 		    INTR_TYPE_NET | INTR_MPSAFE, NULL,
1995 		    irq->handler, irq->data, &irq->cookie);
1996 		 if (unlikely(rc != 0)) {
1997 			device_printf(adapter->pdev, "failed to register "
1998 			    "interrupt handler for irq %ju: %d\n",
1999 			    rman_get_start(irq->res), rc);
2000 			goto err;
2001 		}
2002 		irq->requested = true;
2003 
2004 #ifdef	RSS
2005 		ena_trace(ENA_INFO, "queue %d - RSS bucket %d\n",
2006 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
2007 #else
2008 		ena_trace(ENA_INFO, "queue %d - cpu %d\n",
2009 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
2010 #endif
2011 	}
2012 
2013 	return (rc);
2014 
2015 err:
2016 
2017 	for (; i >= ENA_IO_IRQ_FIRST_IDX; i--) {
2018 		irq = &adapter->irq_tbl[i];
2019 		rcc = 0;
2020 
2021 		/* Once we entered err: section and irq->requested is true we
2022 		   free both intr and resources */
2023 		if (irq->requested)
2024 			rcc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
2025 		if (unlikely(rcc != 0))
2026 			device_printf(adapter->pdev, "could not release"
2027 			    " irq: %d, error: %d\n", irq->vector, rcc);
2028 
2029 		/* If we entred err: section without irq->requested set we know
2030 		   it was bus_alloc_resource_any() that needs cleanup, provided
2031 		   res is not NULL. In case res is NULL no work in needed in
2032 		   this iteration */
2033 		rcc = 0;
2034 		if (irq->res != NULL) {
2035 			rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2036 			    irq->vector, irq->res);
2037 		}
2038 		if (unlikely(rcc != 0))
2039 			device_printf(adapter->pdev, "dev has no parent while "
2040 			    "releasing res for irq: %d\n", irq->vector);
2041 		irq->requested = false;
2042 		irq->res = NULL;
2043 	}
2044 
2045 	return (rc);
2046 }
2047 
2048 static void
2049 ena_free_mgmnt_irq(struct ena_adapter *adapter)
2050 {
2051 	struct ena_irq *irq;
2052 	int rc;
2053 
2054 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2055 	if (irq->requested) {
2056 		ena_trace(ENA_INFO | ENA_ADMQ, "tear down irq: %d\n",
2057 		    irq->vector);
2058 		rc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
2059 		if (unlikely(rc != 0))
2060 			device_printf(adapter->pdev, "failed to tear "
2061 			    "down irq: %d\n", irq->vector);
2062 		irq->requested = 0;
2063 	}
2064 
2065 	if (irq->res != NULL) {
2066 		ena_trace(ENA_INFO | ENA_ADMQ, "release resource irq: %d\n",
2067 		    irq->vector);
2068 		rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2069 		    irq->vector, irq->res);
2070 		irq->res = NULL;
2071 		if (unlikely(rc != 0))
2072 			device_printf(adapter->pdev, "dev has no parent while "
2073 			    "releasing res for irq: %d\n", irq->vector);
2074 	}
2075 }
2076 
2077 static void
2078 ena_free_io_irq(struct ena_adapter *adapter)
2079 {
2080 	struct ena_irq *irq;
2081 	int rc;
2082 
2083 	for (int i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
2084 		irq = &adapter->irq_tbl[i];
2085 		if (irq->requested) {
2086 			ena_trace(ENA_INFO | ENA_IOQ, "tear down irq: %d\n",
2087 			    irq->vector);
2088 			rc = bus_teardown_intr(adapter->pdev, irq->res,
2089 			    irq->cookie);
2090 			if (unlikely(rc != 0)) {
2091 				device_printf(adapter->pdev, "failed to tear "
2092 				    "down irq: %d\n", irq->vector);
2093 			}
2094 			irq->requested = 0;
2095 		}
2096 
2097 		if (irq->res != NULL) {
2098 			ena_trace(ENA_INFO | ENA_IOQ, "release resource irq: %d\n",
2099 			    irq->vector);
2100 			rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2101 			    irq->vector, irq->res);
2102 			irq->res = NULL;
2103 			if (unlikely(rc != 0)) {
2104 				device_printf(adapter->pdev, "dev has no parent"
2105 				    " while releasing res for irq: %d\n",
2106 				    irq->vector);
2107 			}
2108 		}
2109 	}
2110 }
2111 
2112 static void
2113 ena_free_irqs(struct ena_adapter* adapter)
2114 {
2115 
2116 	ena_free_io_irq(adapter);
2117 	ena_free_mgmnt_irq(adapter);
2118 	ena_disable_msix(adapter);
2119 }
2120 
2121 static void
2122 ena_disable_msix(struct ena_adapter *adapter)
2123 {
2124 
2125 	pci_release_msi(adapter->pdev);
2126 
2127 	adapter->msix_vecs = 0;
2128 	free(adapter->msix_entries, M_DEVBUF);
2129 	adapter->msix_entries = NULL;
2130 }
2131 
2132 static void
2133 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
2134 {
2135 	struct ena_com_io_cq* io_cq;
2136 	struct ena_eth_io_intr_reg intr_reg;
2137 	uint16_t ena_qid;
2138 	int i;
2139 
2140 	/* Unmask interrupts for all queues */
2141 	for (i = 0; i < adapter->num_queues; i++) {
2142 		ena_qid = ENA_IO_TXQ_IDX(i);
2143 		io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
2144 		ena_com_update_intr_reg(&intr_reg, 0, 0, true);
2145 		ena_com_unmask_intr(io_cq, &intr_reg);
2146 	}
2147 }
2148 
2149 /* Configure the Rx forwarding */
2150 static int
2151 ena_rss_configure(struct ena_adapter *adapter)
2152 {
2153 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2154 	int rc;
2155 
2156 	/* Set indirect table */
2157 	rc = ena_com_indirect_table_set(ena_dev);
2158 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2159 		return (rc);
2160 
2161 	/* Configure hash function (if supported) */
2162 	rc = ena_com_set_hash_function(ena_dev);
2163 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2164 		return (rc);
2165 
2166 	/* Configure hash inputs (if supported) */
2167 	rc = ena_com_set_hash_ctrl(ena_dev);
2168 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2169 		return (rc);
2170 
2171 	return (0);
2172 }
2173 
2174 static int
2175 ena_up_complete(struct ena_adapter *adapter)
2176 {
2177 	int rc;
2178 
2179 	if (likely(adapter->rss_support)) {
2180 		rc = ena_rss_configure(adapter);
2181 		if (rc != 0)
2182 			return (rc);
2183 	}
2184 
2185 	rc = ena_change_mtu(adapter->ifp, adapter->ifp->if_mtu);
2186 	if (unlikely(rc != 0))
2187 		return (rc);
2188 
2189 	ena_refill_all_rx_bufs(adapter);
2190 	ena_reset_counters((counter_u64_t *)&adapter->hw_stats,
2191 	    sizeof(adapter->hw_stats));
2192 
2193 	return (0);
2194 }
2195 
2196 static int
2197 ena_up(struct ena_adapter *adapter)
2198 {
2199 	int rc = 0;
2200 
2201 	if (unlikely(device_is_attached(adapter->pdev) == 0)) {
2202 		device_printf(adapter->pdev, "device is not attached!\n");
2203 		return (ENXIO);
2204 	}
2205 
2206 	if (unlikely(!adapter->running)) {
2207 		device_printf(adapter->pdev, "device is not running!\n");
2208 		return (ENXIO);
2209 	}
2210 
2211 	if (!adapter->up) {
2212 		device_printf(adapter->pdev, "device is going UP\n");
2213 
2214 		/* setup interrupts for IO queues */
2215 		ena_setup_io_intr(adapter);
2216 		rc = ena_request_io_irq(adapter);
2217 		if (unlikely(rc != 0)) {
2218 			ena_trace(ENA_ALERT, "err_req_irq");
2219 			goto err_req_irq;
2220 		}
2221 
2222 		/* allocate transmit descriptors */
2223 		rc = ena_setup_all_tx_resources(adapter);
2224 		if (unlikely(rc != 0)) {
2225 			ena_trace(ENA_ALERT, "err_setup_tx");
2226 			goto err_setup_tx;
2227 		}
2228 
2229 		/* allocate receive descriptors */
2230 		rc = ena_setup_all_rx_resources(adapter);
2231 		if (unlikely(rc != 0)) {
2232 			ena_trace(ENA_ALERT, "err_setup_rx");
2233 			goto err_setup_rx;
2234 		}
2235 
2236 		/* create IO queues for Rx & Tx */
2237 		rc = ena_create_io_queues(adapter);
2238 		if (unlikely(rc != 0)) {
2239 			ena_trace(ENA_ALERT,
2240 			    "create IO queues failed");
2241 			goto err_io_que;
2242 		}
2243 
2244 		if (unlikely(adapter->link_status))
2245 			if_link_state_change(adapter->ifp, LINK_STATE_UP);
2246 
2247 		rc = ena_up_complete(adapter);
2248 		if (unlikely(rc != 0))
2249 			goto err_up_complete;
2250 
2251 		counter_u64_add(adapter->dev_stats.interface_up, 1);
2252 
2253 		ena_update_hwassist(adapter);
2254 
2255 		if_setdrvflagbits(adapter->ifp, IFF_DRV_RUNNING,
2256 		    IFF_DRV_OACTIVE);
2257 
2258 		callout_reset_sbt(&adapter->timer_service, SBT_1S, SBT_1S,
2259 		    ena_timer_service, (void *)adapter, 0);
2260 
2261 		adapter->up = true;
2262 
2263 		ena_unmask_all_io_irqs(adapter);
2264 	}
2265 
2266 	return (0);
2267 
2268 err_up_complete:
2269 	ena_destroy_all_io_queues(adapter);
2270 err_io_que:
2271 	ena_free_all_rx_resources(adapter);
2272 err_setup_rx:
2273 	ena_free_all_tx_resources(adapter);
2274 err_setup_tx:
2275 	ena_free_io_irq(adapter);
2276 err_req_irq:
2277 	return (rc);
2278 }
2279 
2280 static uint64_t
2281 ena_get_counter(if_t ifp, ift_counter cnt)
2282 {
2283 	struct ena_adapter *adapter;
2284 	struct ena_hw_stats *stats;
2285 
2286 	adapter = if_getsoftc(ifp);
2287 	stats = &adapter->hw_stats;
2288 
2289 	switch (cnt) {
2290 	case IFCOUNTER_IPACKETS:
2291 		return (counter_u64_fetch(stats->rx_packets));
2292 	case IFCOUNTER_OPACKETS:
2293 		return (counter_u64_fetch(stats->tx_packets));
2294 	case IFCOUNTER_IBYTES:
2295 		return (counter_u64_fetch(stats->rx_bytes));
2296 	case IFCOUNTER_OBYTES:
2297 		return (counter_u64_fetch(stats->tx_bytes));
2298 	case IFCOUNTER_IQDROPS:
2299 		return (counter_u64_fetch(stats->rx_drops));
2300 	default:
2301 		return (if_get_counter_default(ifp, cnt));
2302 	}
2303 }
2304 
2305 static int
2306 ena_media_change(if_t ifp)
2307 {
2308 	/* Media Change is not supported by firmware */
2309 	return (0);
2310 }
2311 
2312 static void
2313 ena_media_status(if_t ifp, struct ifmediareq *ifmr)
2314 {
2315 	struct ena_adapter *adapter = if_getsoftc(ifp);
2316 	ena_trace(ENA_DBG, "enter");
2317 
2318 	mtx_lock(&adapter->global_mtx);
2319 
2320 	ifmr->ifm_status = IFM_AVALID;
2321 	ifmr->ifm_active = IFM_ETHER;
2322 
2323 	if (!adapter->link_status) {
2324 		mtx_unlock(&adapter->global_mtx);
2325 		ena_trace(ENA_INFO, "link_status = false");
2326 		return;
2327 	}
2328 
2329 	ifmr->ifm_status |= IFM_ACTIVE;
2330 	ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
2331 
2332 	mtx_unlock(&adapter->global_mtx);
2333 }
2334 
2335 static void
2336 ena_init(void *arg)
2337 {
2338 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
2339 
2340 	if (!adapter->up) {
2341 		sx_xlock(&adapter->ioctl_sx);
2342 		ena_up(adapter);
2343 		sx_unlock(&adapter->ioctl_sx);
2344 	}
2345 }
2346 
2347 static int
2348 ena_ioctl(if_t ifp, u_long command, caddr_t data)
2349 {
2350 	struct ena_adapter *adapter;
2351 	struct ifreq *ifr;
2352 	int rc;
2353 
2354 	adapter = ifp->if_softc;
2355 	ifr = (struct ifreq *)data;
2356 
2357 	/*
2358 	 * Acquiring lock to prevent from running up and down routines parallel.
2359 	 */
2360 	rc = 0;
2361 	switch (command) {
2362 	case SIOCSIFMTU:
2363 		sx_xlock(&adapter->ioctl_sx);
2364 		ena_down(adapter);
2365 
2366 		ena_change_mtu(ifp, ifr->ifr_mtu);
2367 
2368 		rc = ena_up(adapter);
2369 		sx_unlock(&adapter->ioctl_sx);
2370 		break;
2371 
2372 	case SIOCSIFFLAGS:
2373 		if ((ifp->if_flags & IFF_UP) != 0) {
2374 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2375 				if ((ifp->if_flags & (IFF_PROMISC |
2376 				    IFF_ALLMULTI)) != 0) {
2377 					device_printf(adapter->pdev,
2378 					    "ioctl promisc/allmulti\n");
2379 				}
2380 			} else {
2381 				sx_xlock(&adapter->ioctl_sx);
2382 				rc = ena_up(adapter);
2383 				sx_unlock(&adapter->ioctl_sx);
2384 			}
2385 		} else {
2386 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2387 				sx_xlock(&adapter->ioctl_sx);
2388 				ena_down(adapter);
2389 				sx_unlock(&adapter->ioctl_sx);
2390 			}
2391 		}
2392 		break;
2393 
2394 	case SIOCADDMULTI:
2395 	case SIOCDELMULTI:
2396 		break;
2397 
2398 	case SIOCSIFMEDIA:
2399 	case SIOCGIFMEDIA:
2400 		rc = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
2401 		break;
2402 
2403 	case SIOCSIFCAP:
2404 		{
2405 			int reinit = 0;
2406 
2407 			if (ifr->ifr_reqcap != ifp->if_capenable) {
2408 				ifp->if_capenable = ifr->ifr_reqcap;
2409 				reinit = 1;
2410 			}
2411 
2412 			if ((reinit != 0) &&
2413 			    ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)) {
2414 				sx_xlock(&adapter->ioctl_sx);
2415 				ena_down(adapter);
2416 				rc = ena_up(adapter);
2417 				sx_unlock(&adapter->ioctl_sx);
2418 			}
2419 		}
2420 
2421 		break;
2422 	default:
2423 		rc = ether_ioctl(ifp, command, data);
2424 		break;
2425 	}
2426 
2427 	return (rc);
2428 }
2429 
2430 static int
2431 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2432 {
2433 	int caps = 0;
2434 
2435 	if ((feat->offload.tx &
2436 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2437 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2438 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)) != 0)
2439 		caps |= IFCAP_TXCSUM;
2440 
2441 	if ((feat->offload.tx &
2442 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2443 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)) != 0)
2444 		caps |= IFCAP_TXCSUM_IPV6;
2445 
2446 	if ((feat->offload.tx &
2447 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0)
2448 		caps |= IFCAP_TSO4;
2449 
2450 	if ((feat->offload.tx &
2451 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK) != 0)
2452 		caps |= IFCAP_TSO6;
2453 
2454 	if ((feat->offload.rx_supported &
2455 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2456 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK)) != 0)
2457 		caps |= IFCAP_RXCSUM;
2458 
2459 	if ((feat->offload.rx_supported &
2460 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK) != 0)
2461 		caps |= IFCAP_RXCSUM_IPV6;
2462 
2463 	caps |= IFCAP_LRO | IFCAP_JUMBO_MTU;
2464 
2465 	return (caps);
2466 }
2467 
2468 static void
2469 ena_update_host_info(struct ena_admin_host_info *host_info, if_t ifp)
2470 {
2471 
2472 	host_info->supported_network_features[0] =
2473 	    (uint32_t)if_getcapabilities(ifp);
2474 }
2475 
2476 static void
2477 ena_update_hwassist(struct ena_adapter *adapter)
2478 {
2479 	if_t ifp = adapter->ifp;
2480 	uint32_t feat = adapter->tx_offload_cap;
2481 	int cap = if_getcapenable(ifp);
2482 	int flags = 0;
2483 
2484 	if_clearhwassist(ifp);
2485 
2486 	if ((cap & IFCAP_TXCSUM) != 0) {
2487 		if ((feat &
2488 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK) != 0)
2489 			flags |= CSUM_IP;
2490 		if ((feat &
2491 		    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2492 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)) != 0)
2493 			flags |= CSUM_IP_UDP | CSUM_IP_TCP;
2494 	}
2495 
2496 	if ((cap & IFCAP_TXCSUM_IPV6) != 0)
2497 		flags |= CSUM_IP6_UDP | CSUM_IP6_TCP;
2498 
2499 	if ((cap & IFCAP_TSO4) != 0)
2500 		flags |= CSUM_IP_TSO;
2501 
2502 	if ((cap & IFCAP_TSO6) != 0)
2503 		flags |= CSUM_IP6_TSO;
2504 
2505 	if_sethwassistbits(ifp, flags, 0);
2506 }
2507 
2508 static int
2509 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2510     struct ena_com_dev_get_features_ctx *feat)
2511 {
2512 	if_t ifp;
2513 	int caps = 0;
2514 
2515 	ifp = adapter->ifp = if_gethandle(IFT_ETHER);
2516 	if (unlikely(ifp == NULL)) {
2517 		ena_trace(ENA_ALERT, "can not allocate ifnet structure\n");
2518 		return (ENXIO);
2519 	}
2520 	if_initname(ifp, device_get_name(pdev), device_get_unit(pdev));
2521 	if_setdev(ifp, pdev);
2522 	if_setsoftc(ifp, adapter);
2523 
2524 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
2525 	if_setinitfn(ifp, ena_init);
2526 	if_settransmitfn(ifp, ena_mq_start);
2527 	if_setqflushfn(ifp, ena_qflush);
2528 	if_setioctlfn(ifp, ena_ioctl);
2529 	if_setgetcounterfn(ifp, ena_get_counter);
2530 
2531 	if_setsendqlen(ifp, adapter->tx_ring_size);
2532 	if_setsendqready(ifp);
2533 	if_setmtu(ifp, ETHERMTU);
2534 	if_setbaudrate(ifp, 0);
2535 	/* Zeroize capabilities... */
2536 	if_setcapabilities(ifp, 0);
2537 	if_setcapenable(ifp, 0);
2538 	/* check hardware support */
2539 	caps = ena_get_dev_offloads(feat);
2540 	/* ... and set them */
2541 	if_setcapabilitiesbit(ifp, caps, 0);
2542 
2543 	/* TSO parameters */
2544 	ifp->if_hw_tsomax = ENA_TSO_MAXSIZE -
2545 	    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2546 	ifp->if_hw_tsomaxsegcount = adapter->max_tx_sgl_size - 1;
2547 	ifp->if_hw_tsomaxsegsize = ENA_TSO_MAXSIZE;
2548 
2549 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
2550 	if_setcapenable(ifp, if_getcapabilities(ifp));
2551 
2552 	/*
2553 	 * Specify the media types supported by this adapter and register
2554 	 * callbacks to update media and link information
2555 	 */
2556 	ifmedia_init(&adapter->media, IFM_IMASK,
2557 	    ena_media_change, ena_media_status);
2558 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2559 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2560 
2561 	ether_ifattach(ifp, adapter->mac_addr);
2562 
2563 	return (0);
2564 }
2565 
2566 static void
2567 ena_down(struct ena_adapter *adapter)
2568 {
2569 	int rc;
2570 
2571 	if (adapter->up) {
2572 		device_printf(adapter->pdev, "device is going DOWN\n");
2573 
2574 		callout_drain(&adapter->timer_service);
2575 
2576 		adapter->up = false;
2577 		if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE,
2578 		    IFF_DRV_RUNNING);
2579 
2580 		ena_free_io_irq(adapter);
2581 
2582 		if (adapter->trigger_reset) {
2583 			rc = ena_com_dev_reset(adapter->ena_dev,
2584 			    adapter->reset_reason);
2585 			if (unlikely(rc != 0))
2586 				device_printf(adapter->pdev,
2587 				    "Device reset failed\n");
2588 		}
2589 
2590 		ena_destroy_all_io_queues(adapter);
2591 
2592 		ena_free_all_tx_bufs(adapter);
2593 		ena_free_all_rx_bufs(adapter);
2594 		ena_free_all_tx_resources(adapter);
2595 		ena_free_all_rx_resources(adapter);
2596 
2597 		counter_u64_add(adapter->dev_stats.interface_down, 1);
2598 	}
2599 }
2600 
2601 static void
2602 ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct mbuf *mbuf)
2603 {
2604 	struct ena_com_tx_meta *ena_meta;
2605 	struct ether_vlan_header *eh;
2606 	u32 mss;
2607 	bool offload;
2608 	uint16_t etype;
2609 	int ehdrlen;
2610 	struct ip *ip;
2611 	int iphlen;
2612 	struct tcphdr *th;
2613 
2614 	offload = false;
2615 	ena_meta = &ena_tx_ctx->ena_meta;
2616 	mss = mbuf->m_pkthdr.tso_segsz;
2617 
2618 	if (mss != 0)
2619 		offload = true;
2620 
2621 	if ((mbuf->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2622 		offload = true;
2623 
2624 	if ((mbuf->m_pkthdr.csum_flags & CSUM_OFFLOAD) != 0)
2625 		offload = true;
2626 
2627 	if (!offload) {
2628 		ena_tx_ctx->meta_valid = 0;
2629 		return;
2630 	}
2631 
2632 	/* Determine where frame payload starts. */
2633 	eh = mtod(mbuf, struct ether_vlan_header *);
2634 	if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2635 		etype = ntohs(eh->evl_proto);
2636 		ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
2637 	} else {
2638 		etype = ntohs(eh->evl_encap_proto);
2639 		ehdrlen = ETHER_HDR_LEN;
2640 	}
2641 
2642 	ip = (struct ip *)(mbuf->m_data + ehdrlen);
2643 	iphlen = ip->ip_hl << 2;
2644 	th = (struct tcphdr *)((caddr_t)ip + iphlen);
2645 
2646 	if ((mbuf->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2647 		ena_tx_ctx->l3_csum_enable = 1;
2648 	}
2649 	if ((mbuf->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2650 		ena_tx_ctx->tso_enable = 1;
2651 		ena_meta->l4_hdr_len = (th->th_off);
2652 	}
2653 
2654 	switch (etype) {
2655 	case ETHERTYPE_IP:
2656 		ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2657 		if ((ip->ip_off & htons(IP_DF)) != 0)
2658 			ena_tx_ctx->df = 1;
2659 		break;
2660 	case ETHERTYPE_IPV6:
2661 		ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2662 
2663 	default:
2664 		break;
2665 	}
2666 
2667 	if (ip->ip_p == IPPROTO_TCP) {
2668 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2669 		if ((mbuf->m_pkthdr.csum_flags &
2670 		    (CSUM_IP_TCP | CSUM_IP6_TCP)) != 0)
2671 			ena_tx_ctx->l4_csum_enable = 1;
2672 		else
2673 			ena_tx_ctx->l4_csum_enable = 0;
2674 	} else if (ip->ip_p == IPPROTO_UDP) {
2675 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2676 		if ((mbuf->m_pkthdr.csum_flags &
2677 		    (CSUM_IP_UDP | CSUM_IP6_UDP)) != 0)
2678 			ena_tx_ctx->l4_csum_enable = 1;
2679 		else
2680 			ena_tx_ctx->l4_csum_enable = 0;
2681 	} else {
2682 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
2683 		ena_tx_ctx->l4_csum_enable = 0;
2684 	}
2685 
2686 	ena_meta->mss = mss;
2687 	ena_meta->l3_hdr_len = iphlen;
2688 	ena_meta->l3_hdr_offset = ehdrlen;
2689 	ena_tx_ctx->meta_valid = 1;
2690 }
2691 
2692 static int
2693 ena_check_and_collapse_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2694 {
2695 	struct ena_adapter *adapter;
2696 	struct mbuf *collapsed_mbuf;
2697 	int num_frags;
2698 
2699 	adapter = tx_ring->adapter;
2700 	num_frags = ena_mbuf_count(*mbuf);
2701 
2702 	/* One segment must be reserved for configuration descriptor. */
2703 	if (num_frags < adapter->max_tx_sgl_size)
2704 		return (0);
2705 	counter_u64_add(tx_ring->tx_stats.collapse, 1);
2706 
2707 	collapsed_mbuf = m_collapse(*mbuf, M_NOWAIT,
2708 	    adapter->max_tx_sgl_size - 1);
2709 	if (unlikely(collapsed_mbuf == NULL)) {
2710 		counter_u64_add(tx_ring->tx_stats.collapse_err, 1);
2711 		return (ENOMEM);
2712 	}
2713 
2714 	/* If mbuf was collapsed succesfully, original mbuf is released. */
2715 	*mbuf = collapsed_mbuf;
2716 
2717 	return (0);
2718 }
2719 
2720 static int
2721 ena_xmit_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2722 {
2723 	struct ena_adapter *adapter;
2724 	struct ena_tx_buffer *tx_info;
2725 	struct ena_com_tx_ctx ena_tx_ctx;
2726 	struct ena_com_dev *ena_dev;
2727 	struct ena_com_buf *ena_buf;
2728 	struct ena_com_io_sq* io_sq;
2729 	bus_dma_segment_t segs[ENA_BUS_DMA_SEGS];
2730 	void *push_hdr;
2731 	uint16_t next_to_use;
2732 	uint16_t req_id;
2733 	uint16_t push_len;
2734 	uint16_t ena_qid;
2735 	uint32_t len, nsegs, header_len;
2736 	int i, rc;
2737 	int nb_hw_desc;
2738 
2739 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2740 	adapter = tx_ring->que->adapter;
2741 	ena_dev = adapter->ena_dev;
2742 	io_sq = &ena_dev->io_sq_queues[ena_qid];
2743 
2744 	rc = ena_check_and_collapse_mbuf(tx_ring, mbuf);
2745 	if (unlikely(rc != 0)) {
2746 		ena_trace(ENA_WARNING,
2747 		    "Failed to collapse mbuf! err: %d", rc);
2748 		return (rc);
2749 	}
2750 
2751 	next_to_use = tx_ring->next_to_use;
2752 	req_id = tx_ring->free_tx_ids[next_to_use];
2753 	tx_info = &tx_ring->tx_buffer_info[req_id];
2754 
2755 	tx_info->mbuf = *mbuf;
2756 	tx_info->num_of_bufs = 0;
2757 
2758 	ena_buf = tx_info->bufs;
2759 	len = (*mbuf)->m_len;
2760 
2761 	ena_trace(ENA_DBG | ENA_TXPTH, "Tx: %d bytes", (*mbuf)->m_pkthdr.len);
2762 
2763 	push_len = 0;
2764 	header_len = min_t(uint32_t, len, tx_ring->tx_max_header_size);
2765 	push_hdr = NULL;
2766 
2767 	rc = bus_dmamap_load_mbuf_sg(adapter->tx_buf_tag, tx_info->map,
2768 	    *mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
2769 
2770 	if (unlikely((rc != 0) || (nsegs == 0))) {
2771 		ena_trace(ENA_WARNING,
2772 		    "dmamap load failed! err: %d nsegs: %d", rc, nsegs);
2773 		counter_u64_add(tx_ring->tx_stats.dma_mapping_err, 1);
2774 		tx_info->mbuf = NULL;
2775 		if (rc == ENOMEM)
2776 			return (ENA_COM_NO_MEM);
2777 		else
2778 			return (ENA_COM_INVAL);
2779 	}
2780 
2781 	for (i = 0; i < nsegs; i++) {
2782 		ena_buf->len = segs[i].ds_len;
2783 		ena_buf->paddr = segs[i].ds_addr;
2784 		ena_buf++;
2785 	}
2786 	tx_info->num_of_bufs = nsegs;
2787 
2788 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2789 	ena_tx_ctx.ena_bufs = tx_info->bufs;
2790 	ena_tx_ctx.push_header = push_hdr;
2791 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2792 	ena_tx_ctx.req_id = req_id;
2793 	ena_tx_ctx.header_len = header_len;
2794 
2795 	/* Set flags and meta data */
2796 	ena_tx_csum(&ena_tx_ctx, *mbuf);
2797 	/* Prepare the packet's descriptors and send them to device */
2798 	rc = ena_com_prepare_tx(io_sq, &ena_tx_ctx, &nb_hw_desc);
2799 	if (unlikely(rc != 0)) {
2800 		device_printf(adapter->pdev, "failed to prepare tx bufs\n");
2801 		counter_u64_add(tx_ring->tx_stats.prepare_ctx_err, 1);
2802 		goto dma_error;
2803 	}
2804 
2805 	counter_enter();
2806 	counter_u64_add_protected(tx_ring->tx_stats.cnt, 1);
2807 	counter_u64_add_protected(tx_ring->tx_stats.bytes,
2808 	    (*mbuf)->m_pkthdr.len);
2809 
2810 	counter_u64_add_protected(adapter->hw_stats.tx_packets, 1);
2811 	counter_u64_add_protected(adapter->hw_stats.tx_bytes,
2812 	    (*mbuf)->m_pkthdr.len);
2813 	counter_exit();
2814 
2815 	tx_info->tx_descs = nb_hw_desc;
2816 	getbinuptime(&tx_info->timestamp);
2817 	tx_info->print_once = true;
2818 
2819 	tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2820 	    tx_ring->ring_size);
2821 
2822 	bus_dmamap_sync(adapter->tx_buf_tag, tx_info->map,
2823 	    BUS_DMASYNC_PREWRITE);
2824 
2825 	return (0);
2826 
2827 dma_error:
2828 	tx_info->mbuf = NULL;
2829 	bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
2830 
2831 	return (rc);
2832 }
2833 
2834 static void
2835 ena_start_xmit(struct ena_ring *tx_ring)
2836 {
2837 	struct mbuf *mbuf;
2838 	struct ena_adapter *adapter = tx_ring->adapter;
2839 	struct ena_com_io_sq* io_sq;
2840 	int ena_qid;
2841 	int acum_pkts = 0;
2842 	int ret = 0;
2843 
2844 	if (unlikely((if_getdrvflags(adapter->ifp) & IFF_DRV_RUNNING) == 0))
2845 		return;
2846 
2847 	if (unlikely(!adapter->link_status))
2848 		return;
2849 
2850 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2851 	io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
2852 
2853 	while ((mbuf = drbr_peek(adapter->ifp, tx_ring->br)) != NULL) {
2854 		ena_trace(ENA_DBG | ENA_TXPTH, "\ndequeued mbuf %p with flags %#x and"
2855 		    " header csum flags %#jx",
2856 		    mbuf, mbuf->m_flags, (uint64_t)mbuf->m_pkthdr.csum_flags);
2857 
2858 		if (unlikely(!ena_com_sq_have_enough_space(io_sq,
2859 		    ENA_TX_CLEANUP_THRESHOLD)))
2860 			ena_tx_cleanup(tx_ring);
2861 
2862 		if (unlikely((ret = ena_xmit_mbuf(tx_ring, &mbuf)) != 0)) {
2863 			if (ret == ENA_COM_NO_MEM) {
2864 				drbr_putback(adapter->ifp, tx_ring->br, mbuf);
2865 			} else if (ret == ENA_COM_NO_SPACE) {
2866 				drbr_putback(adapter->ifp, tx_ring->br, mbuf);
2867 			} else {
2868 				m_freem(mbuf);
2869 				drbr_advance(adapter->ifp, tx_ring->br);
2870 			}
2871 
2872 			break;
2873 		}
2874 
2875 		drbr_advance(adapter->ifp, tx_ring->br);
2876 
2877 		if (unlikely((if_getdrvflags(adapter->ifp) &
2878 		    IFF_DRV_RUNNING) == 0))
2879 			return;
2880 
2881 		acum_pkts++;
2882 
2883 		BPF_MTAP(adapter->ifp, mbuf);
2884 
2885 		if (unlikely(acum_pkts == DB_THRESHOLD)) {
2886 			acum_pkts = 0;
2887 			wmb();
2888 			/* Trigger the dma engine */
2889 			ena_com_write_sq_doorbell(io_sq);
2890 			counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2891 		}
2892 
2893 	}
2894 
2895 	if (likely(acum_pkts != 0)) {
2896 		wmb();
2897 		/* Trigger the dma engine */
2898 		ena_com_write_sq_doorbell(io_sq);
2899 		counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2900 	}
2901 
2902 	if (!ena_com_sq_have_enough_space(io_sq, ENA_TX_CLEANUP_THRESHOLD))
2903 		ena_tx_cleanup(tx_ring);
2904 }
2905 
2906 static void
2907 ena_deferred_mq_start(void *arg, int pending)
2908 {
2909 	struct ena_ring *tx_ring = (struct ena_ring *)arg;
2910 	struct ifnet *ifp = tx_ring->adapter->ifp;
2911 
2912 	while (!drbr_empty(ifp, tx_ring->br) &&
2913 	    (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2914 		ENA_RING_MTX_LOCK(tx_ring);
2915 		ena_start_xmit(tx_ring);
2916 		ENA_RING_MTX_UNLOCK(tx_ring);
2917 	}
2918 }
2919 
2920 static int
2921 ena_mq_start(if_t ifp, struct mbuf *m)
2922 {
2923 	struct ena_adapter *adapter = ifp->if_softc;
2924 	struct ena_ring *tx_ring;
2925 	int ret, is_drbr_empty;
2926 	uint32_t i;
2927 
2928 	if (unlikely((if_getdrvflags(adapter->ifp) & IFF_DRV_RUNNING) == 0))
2929 		return (ENODEV);
2930 
2931 	/* Which queue to use */
2932 	/*
2933 	 * If everything is setup correctly, it should be the
2934 	 * same bucket that the current CPU we're on is.
2935 	 * It should improve performance.
2936 	 */
2937 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) {
2938 #ifdef	RSS
2939 		if (rss_hash2bucket(m->m_pkthdr.flowid,
2940 		    M_HASHTYPE_GET(m), &i) == 0) {
2941 			i = i % adapter->num_queues;
2942 
2943 		} else
2944 #endif
2945 		{
2946 			i = m->m_pkthdr.flowid % adapter->num_queues;
2947 		}
2948 	} else {
2949 		i = curcpu % adapter->num_queues;
2950 	}
2951 	tx_ring = &adapter->tx_ring[i];
2952 
2953 	/* Check if drbr is empty before putting packet */
2954 	is_drbr_empty = drbr_empty(ifp, tx_ring->br);
2955 	ret = drbr_enqueue(ifp, tx_ring->br, m);
2956 	if (unlikely(ret != 0)) {
2957 		taskqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
2958 		return (ret);
2959 	}
2960 
2961 	if ((is_drbr_empty != 0) && (ENA_RING_MTX_TRYLOCK(tx_ring) != 0)) {
2962 		ena_start_xmit(tx_ring);
2963 		ENA_RING_MTX_UNLOCK(tx_ring);
2964 	} else {
2965 		taskqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
2966 	}
2967 
2968 	return (0);
2969 }
2970 
2971 static void
2972 ena_qflush(if_t ifp)
2973 {
2974 	struct ena_adapter *adapter = ifp->if_softc;
2975 	struct ena_ring *tx_ring = adapter->tx_ring;
2976 	int i;
2977 
2978 	for(i = 0; i < adapter->num_queues; ++i, ++tx_ring)
2979 		if (!drbr_empty(ifp, tx_ring->br)) {
2980 			ENA_RING_MTX_LOCK(tx_ring);
2981 			drbr_flush(ifp, tx_ring->br);
2982 			ENA_RING_MTX_UNLOCK(tx_ring);
2983 		}
2984 
2985 	if_qflush(ifp);
2986 }
2987 
2988 static int
2989 ena_calc_io_queue_num(struct ena_adapter *adapter,
2990     struct ena_com_dev_get_features_ctx *get_feat_ctx)
2991 {
2992 	int io_sq_num, io_cq_num, io_queue_num;
2993 
2994 	io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2995 	io_cq_num = get_feat_ctx->max_queues.max_cq_num;
2996 
2997 	io_queue_num = min_t(int, mp_ncpus, ENA_MAX_NUM_IO_QUEUES);
2998 	io_queue_num = min_t(int, io_queue_num, io_sq_num);
2999 	io_queue_num = min_t(int, io_queue_num, io_cq_num);
3000 	/* 1 IRQ for for mgmnt and 1 IRQ for each TX/RX pair */
3001 	io_queue_num = min_t(int, io_queue_num,
3002 	    pci_msix_count(adapter->pdev) - 1);
3003 #ifdef	RSS
3004 	io_queue_num = min_t(int, io_queue_num, rss_getnumbuckets());
3005 #endif
3006 
3007 	return (io_queue_num);
3008 }
3009 
3010 static int
3011 ena_calc_queue_size(struct ena_adapter *adapter, uint16_t *max_tx_sgl_size,
3012     uint16_t *max_rx_sgl_size, struct ena_com_dev_get_features_ctx *feat)
3013 {
3014 	uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
3015 	uint32_t v;
3016 	uint32_t q;
3017 
3018 	queue_size = min_t(uint32_t, queue_size,
3019 	    feat->max_queues.max_cq_depth);
3020 	queue_size = min_t(uint32_t, queue_size,
3021 	    feat->max_queues.max_sq_depth);
3022 
3023 	/* round down to the nearest power of 2 */
3024 	v = queue_size;
3025 	while (v != 0) {
3026 		if (powerof2(queue_size) != 0)
3027 			break;
3028 		v /= 2;
3029 		q = rounddown2(queue_size, v);
3030 		if (q != 0) {
3031 			queue_size = q;
3032 			break;
3033 		}
3034 	}
3035 
3036 	if (unlikely(queue_size == 0)) {
3037 		device_printf(adapter->pdev, "Invalid queue size\n");
3038 		return (ENA_COM_FAULT);
3039 	}
3040 
3041 	*max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3042 	    feat->max_queues.max_packet_tx_descs);
3043 	*max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3044 	    feat->max_queues.max_packet_rx_descs);
3045 
3046 	return (queue_size);
3047 }
3048 
3049 static int
3050 ena_rss_init_default(struct ena_adapter *adapter)
3051 {
3052 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3053 	device_t dev = adapter->pdev;
3054 	int qid, rc, i;
3055 
3056 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3057 	if (unlikely(rc != 0)) {
3058 		device_printf(dev, "Cannot init indirect table\n");
3059 		return (rc);
3060 	}
3061 
3062 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3063 #ifdef	RSS
3064 		qid = rss_get_indirection_to_bucket(i);
3065 		qid = qid % adapter->num_queues;
3066 #else
3067 		qid = i % adapter->num_queues;
3068 #endif
3069 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3070 		    ENA_IO_RXQ_IDX(qid));
3071 		if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3072 			device_printf(dev, "Cannot fill indirect table\n");
3073 			goto err_rss_destroy;
3074 		}
3075 	}
3076 
3077 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3078 	    ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3079 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3080 		device_printf(dev, "Cannot fill hash function\n");
3081 		goto err_rss_destroy;
3082 	}
3083 
3084 	rc = ena_com_set_default_hash_ctrl(ena_dev);
3085 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3086 		device_printf(dev, "Cannot fill hash control\n");
3087 		goto err_rss_destroy;
3088 	}
3089 
3090 	return (0);
3091 
3092 err_rss_destroy:
3093 	ena_com_rss_destroy(ena_dev);
3094 	return (rc);
3095 }
3096 
3097 static void
3098 ena_rss_init_default_deferred(void *arg)
3099 {
3100 	struct ena_adapter *adapter;
3101 	devclass_t dc;
3102 	int max;
3103 	int rc;
3104 
3105 	dc = devclass_find("ena");
3106 	if (unlikely(dc == NULL)) {
3107 		ena_trace(ENA_ALERT, "No devclass ena\n");
3108 		return;
3109 	}
3110 
3111 	max = devclass_get_maxunit(dc);
3112 	while (max-- >= 0) {
3113 		adapter = devclass_get_softc(dc, max);
3114 		if (adapter != NULL) {
3115 			rc = ena_rss_init_default(adapter);
3116 			adapter->rss_support = true;
3117 			if (unlikely(rc != 0)) {
3118 				device_printf(adapter->pdev,
3119 				    "WARNING: RSS was not properly initialized,"
3120 				    " it will affect bandwidth\n");
3121 				adapter->rss_support = false;
3122 			}
3123 		}
3124 	}
3125 }
3126 SYSINIT(ena_rss_init, SI_SUB_KICK_SCHEDULER, SI_ORDER_SECOND, ena_rss_init_default_deferred, NULL);
3127 
3128 static void
3129 ena_config_host_info(struct ena_com_dev *ena_dev)
3130 {
3131 	struct ena_admin_host_info *host_info;
3132 	int rc;
3133 
3134 	/* Allocate only the host info */
3135 	rc = ena_com_allocate_host_info(ena_dev);
3136 	if (unlikely(rc != 0)) {
3137 		ena_trace(ENA_ALERT, "Cannot allocate host info\n");
3138 		return;
3139 	}
3140 
3141 	host_info = ena_dev->host_attr.host_info;
3142 
3143 	host_info->os_type = ENA_ADMIN_OS_FREEBSD;
3144 	host_info->kernel_ver = osreldate;
3145 
3146 	sprintf(host_info->kernel_ver_str, "%d", osreldate);
3147 	host_info->os_dist = 0;
3148 	strncpy(host_info->os_dist_str, osrelease,
3149 	    sizeof(host_info->os_dist_str) - 1);
3150 
3151 	host_info->driver_version =
3152 		(DRV_MODULE_VER_MAJOR) |
3153 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3154 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
3155 
3156 	rc = ena_com_set_host_attributes(ena_dev);
3157 	if (unlikely(rc != 0)) {
3158 		if (rc == EOPNOTSUPP)
3159 			ena_trace(ENA_WARNING, "Cannot set host attributes\n");
3160 		else
3161 			ena_trace(ENA_ALERT, "Cannot set host attributes\n");
3162 
3163 		goto err;
3164 	}
3165 
3166 	return;
3167 
3168 err:
3169 	ena_com_delete_host_info(ena_dev);
3170 }
3171 
3172 static int
3173 ena_device_init(struct ena_adapter *adapter, device_t pdev,
3174     struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
3175 {
3176 	struct ena_com_dev* ena_dev = adapter->ena_dev;
3177 	bool readless_supported;
3178 	uint32_t aenq_groups;
3179 	int dma_width;
3180 	int rc;
3181 
3182 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
3183 	if (unlikely(rc != 0)) {
3184 		device_printf(pdev, "failed to init mmio read less\n");
3185 		return (rc);
3186 	}
3187 
3188 	/*
3189 	 * The PCIe configuration space revision id indicate if mmio reg
3190 	 * read is disabled
3191 	 */
3192 	readless_supported = !(pci_get_revid(pdev) & ENA_MMIO_DISABLE_REG_READ);
3193 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3194 
3195 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3196 	if (unlikely(rc != 0)) {
3197 		device_printf(pdev, "Can not reset device\n");
3198 		goto err_mmio_read_less;
3199 	}
3200 
3201 	rc = ena_com_validate_version(ena_dev);
3202 	if (unlikely(rc != 0)) {
3203 		device_printf(pdev, "device version is too low\n");
3204 		goto err_mmio_read_less;
3205 	}
3206 
3207 	dma_width = ena_com_get_dma_width(ena_dev);
3208 	if (unlikely(dma_width < 0)) {
3209 		device_printf(pdev, "Invalid dma width value %d", dma_width);
3210 		rc = dma_width;
3211 		goto err_mmio_read_less;
3212 	}
3213 	adapter->dma_width = dma_width;
3214 
3215 	/* ENA admin level init */
3216 	rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
3217 	if (unlikely(rc != 0)) {
3218 		device_printf(pdev,
3219 		    "Can not initialize ena admin queue with device\n");
3220 		goto err_mmio_read_less;
3221 	}
3222 
3223 	/*
3224 	 * To enable the msix interrupts the driver needs to know the number
3225 	 * of queues. So the driver uses polling mode to retrieve this
3226 	 * information
3227 	 */
3228 	ena_com_set_admin_polling_mode(ena_dev, true);
3229 
3230 	ena_config_host_info(ena_dev);
3231 
3232 	/* Get Device Attributes */
3233 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3234 	if (unlikely(rc != 0)) {
3235 		device_printf(pdev,
3236 		    "Cannot get attribute for ena device rc: %d\n", rc);
3237 		goto err_admin_init;
3238 	}
3239 
3240 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3241 	    BIT(ENA_ADMIN_FATAL_ERROR) |
3242 	    BIT(ENA_ADMIN_WARNING) |
3243 	    BIT(ENA_ADMIN_NOTIFICATION) |
3244 	    BIT(ENA_ADMIN_KEEP_ALIVE);
3245 
3246 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
3247 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3248 	if (unlikely(rc != 0)) {
3249 		device_printf(pdev, "Cannot configure aenq groups rc: %d\n", rc);
3250 		goto err_admin_init;
3251 	}
3252 
3253 	*wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3254 
3255 	return (0);
3256 
3257 err_admin_init:
3258 	ena_com_delete_host_info(ena_dev);
3259 	ena_com_admin_destroy(ena_dev);
3260 err_mmio_read_less:
3261 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3262 
3263 	return (rc);
3264 }
3265 
3266 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
3267     int io_vectors)
3268 {
3269 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3270 	int rc;
3271 
3272 	rc = ena_enable_msix(adapter);
3273 	if (unlikely(rc != 0)) {
3274 		device_printf(adapter->pdev, "Error with MSI-X enablement\n");
3275 		return (rc);
3276 	}
3277 
3278 	ena_setup_mgmnt_intr(adapter);
3279 
3280 	rc = ena_request_mgmnt_irq(adapter);
3281 	if (unlikely(rc != 0)) {
3282 		device_printf(adapter->pdev, "Cannot setup mgmnt queue intr\n");
3283 		goto err_disable_msix;
3284 	}
3285 
3286 	ena_com_set_admin_polling_mode(ena_dev, false);
3287 
3288 	ena_com_admin_aenq_enable(ena_dev);
3289 
3290 	return (0);
3291 
3292 err_disable_msix:
3293 	ena_disable_msix(adapter);
3294 
3295 	return (rc);
3296 }
3297 
3298 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
3299 static void ena_keep_alive_wd(void *adapter_data,
3300     struct ena_admin_aenq_entry *aenq_e)
3301 {
3302 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3303 	struct ena_admin_aenq_keep_alive_desc *desc;
3304 	sbintime_t stime;
3305 	uint64_t rx_drops;
3306 
3307 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3308 
3309 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
3310 	counter_u64_zero(adapter->hw_stats.rx_drops);
3311 	counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
3312 
3313 	stime = getsbinuptime();
3314 	atomic_store_rel_64(&adapter->keep_alive_timestamp, stime);
3315 }
3316 
3317 /* Check for keep alive expiration */
3318 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3319 {
3320 	sbintime_t timestamp, time;
3321 
3322 	if (adapter->wd_active == 0)
3323 		return;
3324 
3325 	if (likely(adapter->keep_alive_timeout == 0))
3326 		return;
3327 
3328 	timestamp = atomic_load_acq_64(&adapter->keep_alive_timestamp);
3329 	time = getsbinuptime() - timestamp;
3330 	if (unlikely(time > adapter->keep_alive_timeout)) {
3331 		device_printf(adapter->pdev,
3332 		    "Keep alive watchdog timeout.\n");
3333 		counter_u64_add(adapter->dev_stats.wd_expired, 1);
3334 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3335 		adapter->trigger_reset = true;
3336 	}
3337 }
3338 
3339 /* Check if admin queue is enabled */
3340 static void check_for_admin_com_state(struct ena_adapter *adapter)
3341 {
3342 	if (unlikely(ena_com_get_admin_running_state(adapter->ena_dev) ==
3343 	    false)) {
3344 		device_printf(adapter->pdev,
3345 		    "ENA admin queue is not in running state!\n");
3346 		counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
3347 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3348 		adapter->trigger_reset = true;
3349 	}
3350 }
3351 
3352 static int
3353 check_missing_comp_in_queue(struct ena_adapter *adapter,
3354     struct ena_ring *tx_ring)
3355 {
3356 	struct bintime curtime, time;
3357 	struct ena_tx_buffer *tx_buf;
3358 	uint32_t missed_tx = 0;
3359 	int i;
3360 
3361 	getbinuptime(&curtime);
3362 
3363 	for (i = 0; i < tx_ring->ring_size; i++) {
3364 		tx_buf = &tx_ring->tx_buffer_info[i];
3365 
3366 		if (bintime_isset(&tx_buf->timestamp) == 0)
3367 			continue;
3368 
3369 		time = curtime;
3370 		bintime_sub(&time, &tx_buf->timestamp);
3371 
3372 		/* Check again if packet is still waiting */
3373 		if (unlikely(bttosbt(time) > adapter->missing_tx_timeout)) {
3374 
3375 			if (!tx_buf->print_once)
3376 				ena_trace(ENA_WARNING, "Found a Tx that wasn't "
3377 				    "completed on time, qid %d, index %d.\n",
3378 				    tx_ring->qid, i);
3379 
3380 			tx_buf->print_once = true;
3381 			missed_tx++;
3382 			counter_u64_add(tx_ring->tx_stats.missing_tx_comp, 1);
3383 
3384 			if (unlikely(missed_tx >
3385 			    adapter->missing_tx_threshold)) {
3386 				device_printf(adapter->pdev,
3387 				    "The number of lost tx completion "
3388 				    "is above the threshold (%d > %d). "
3389 				    "Reset the device\n",
3390 				    missed_tx, adapter->missing_tx_threshold);
3391 				adapter->reset_reason =
3392 				    ENA_REGS_RESET_MISS_TX_CMPL;
3393 				adapter->trigger_reset = true;
3394 				return (EIO);
3395 			}
3396 		}
3397 	}
3398 
3399 	return (0);
3400 }
3401 
3402 /*
3403  * Check for TX which were not completed on time.
3404  * Timeout is defined by "missing_tx_timeout".
3405  * Reset will be performed if number of incompleted
3406  * transactions exceeds "missing_tx_threshold".
3407  */
3408 static void
3409 check_for_missing_tx_completions(struct ena_adapter *adapter)
3410 {
3411 	struct ena_ring *tx_ring;
3412 	int i, budget, rc;
3413 
3414 	/* Make sure the driver doesn't turn the device in other process */
3415 	rmb();
3416 
3417 	if (!adapter->up)
3418 		return;
3419 
3420 	if (adapter->trigger_reset)
3421 		return;
3422 
3423 	if (adapter->missing_tx_timeout == 0)
3424 		return;
3425 
3426 	budget = adapter->missing_tx_max_queues;
3427 
3428 	for (i = adapter->next_monitored_tx_qid; i < adapter->num_queues; i++) {
3429 		tx_ring = &adapter->tx_ring[i];
3430 
3431 		rc = check_missing_comp_in_queue(adapter, tx_ring);
3432 		if (unlikely(rc != 0))
3433 			return;
3434 
3435 		budget--;
3436 		if (budget == 0) {
3437 			i++;
3438 			break;
3439 		}
3440 	}
3441 
3442 	adapter->next_monitored_tx_qid = i % adapter->num_queues;
3443 }
3444 
3445 /* trigger deferred rx cleanup after 2 consecutive detections */
3446 #define EMPTY_RX_REFILL 2
3447 /* For the rare case where the device runs out of Rx descriptors and the
3448  * msix handler failed to refill new Rx descriptors (due to a lack of memory
3449  * for example).
3450  * This case will lead to a deadlock:
3451  * The device won't send interrupts since all the new Rx packets will be dropped
3452  * The msix handler won't allocate new Rx descriptors so the device won't be
3453  * able to send new packets.
3454  *
3455  * When such a situation is detected - execute rx cleanup task in another thread
3456  */
3457 static void
3458 check_for_empty_rx_ring(struct ena_adapter *adapter)
3459 {
3460 	struct ena_ring *rx_ring;
3461 	int i, refill_required;
3462 
3463 	if (!adapter->up)
3464 		return;
3465 
3466 	if (adapter->trigger_reset)
3467 		return;
3468 
3469 	for (i = 0; i < adapter->num_queues; i++) {
3470 		rx_ring = &adapter->rx_ring[i];
3471 
3472 		refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
3473 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3474 			rx_ring->empty_rx_queue++;
3475 
3476 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL)	{
3477 				counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
3478 				    1);
3479 
3480 				device_printf(adapter->pdev,
3481 				    "trigger refill for ring %d\n", i);
3482 
3483 				taskqueue_enqueue(rx_ring->cmpl_tq,
3484 				    &rx_ring->cmpl_task);
3485 				rx_ring->empty_rx_queue = 0;
3486 			}
3487 		} else {
3488 			rx_ring->empty_rx_queue = 0;
3489 		}
3490 	}
3491 }
3492 
3493 static void
3494 ena_timer_service(void *data)
3495 {
3496 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3497 	struct ena_admin_host_info *host_info =
3498 	    adapter->ena_dev->host_attr.host_info;
3499 
3500 	check_for_missing_keep_alive(adapter);
3501 
3502 	check_for_admin_com_state(adapter);
3503 
3504 	check_for_missing_tx_completions(adapter);
3505 
3506 	check_for_empty_rx_ring(adapter);
3507 
3508 	if (host_info != NULL)
3509 		ena_update_host_info(host_info, adapter->ifp);
3510 
3511 	if (unlikely(adapter->trigger_reset)) {
3512 		device_printf(adapter->pdev, "Trigger reset is on\n");
3513 		taskqueue_enqueue(adapter->reset_tq, &adapter->reset_task);
3514 		return;
3515 	}
3516 
3517 	/*
3518 	 * Schedule another timeout one second from now.
3519 	 */
3520 	callout_schedule_sbt(&adapter->timer_service, SBT_1S, SBT_1S, 0);
3521 }
3522 
3523 static void
3524 ena_reset_task(void *arg, int pending)
3525 {
3526 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3527 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3528 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3529 	bool dev_up;
3530 	int rc;
3531 
3532 	if (unlikely(!adapter->trigger_reset)) {
3533 		device_printf(adapter->pdev,
3534 		    "device reset scheduled but trigger_reset is off\n");
3535 		return;
3536 	}
3537 
3538 	sx_xlock(&adapter->ioctl_sx);
3539 
3540 	callout_drain(&adapter->timer_service);
3541 
3542 	dev_up = adapter->up;
3543 
3544 	ena_com_set_admin_running_state(ena_dev, false);
3545 	ena_down(adapter);
3546 	ena_free_mgmnt_irq(adapter);
3547 	ena_disable_msix(adapter);
3548 	ena_com_abort_admin_commands(ena_dev);
3549 	ena_com_wait_for_abort_completion(ena_dev);
3550 	ena_com_admin_destroy(ena_dev);
3551 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3552 
3553 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3554 	adapter->trigger_reset = false;
3555 
3556 	/* Finished destroy part. Restart the device */
3557 	rc = ena_device_init(adapter, adapter->pdev, &get_feat_ctx,
3558 	    &adapter->wd_active);
3559 	if (unlikely(rc != 0)) {
3560 		device_printf(adapter->pdev,
3561 		    "ENA device init failed! (err: %d)\n", rc);
3562 		goto err_dev_free;
3563 	}
3564 
3565 	rc = ena_enable_msix_and_set_admin_interrupts(adapter,
3566 	    adapter->num_queues);
3567 	if (unlikely(rc != 0)) {
3568 		device_printf(adapter->pdev, "Enable MSI-X failed\n");
3569 		goto err_com_free;
3570 	}
3571 
3572 	/* If the interface was up before the reset bring it up */
3573 	if (dev_up) {
3574 		rc = ena_up(adapter);
3575 		if (unlikely(rc != 0)) {
3576 			device_printf(adapter->pdev,
3577 			    "Failed to create I/O queues\n");
3578 			goto err_msix_free;
3579 		}
3580 	}
3581 
3582 	callout_reset_sbt(&adapter->timer_service, SBT_1S, SBT_1S,
3583 	    ena_timer_service, (void *)adapter, 0);
3584 
3585 	sx_unlock(&adapter->ioctl_sx);
3586 
3587 	return;
3588 
3589 err_msix_free:
3590 	ena_free_mgmnt_irq(adapter);
3591 	ena_disable_msix(adapter);
3592 err_com_free:
3593 	ena_com_admin_destroy(ena_dev);
3594 err_dev_free:
3595 	device_printf(adapter->pdev, "ENA reset failed!\n");
3596 	adapter->running = false;
3597 	sx_unlock(&adapter->ioctl_sx);
3598 }
3599 
3600 /**
3601  * ena_attach - Device Initialization Routine
3602  * @pdev: device information struct
3603  *
3604  * Returns 0 on success, otherwise on failure.
3605  *
3606  * ena_attach initializes an adapter identified by a device structure.
3607  * The OS initialization, configuring of the adapter private structure,
3608  * and a hardware reset occur.
3609  **/
3610 static int
3611 ena_attach(device_t pdev)
3612 {
3613 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3614 	static int version_printed;
3615 	struct ena_adapter *adapter;
3616 	struct ena_com_dev *ena_dev = NULL;
3617 	uint16_t tx_sgl_size = 0;
3618 	uint16_t rx_sgl_size = 0;
3619 	int io_queue_num;
3620 	int queue_size;
3621 	int rc;
3622 	adapter = device_get_softc(pdev);
3623 	adapter->pdev = pdev;
3624 
3625 	mtx_init(&adapter->global_mtx, "ENA global mtx", NULL, MTX_DEF);
3626 	sx_init(&adapter->ioctl_sx, "ENA ioctl sx");
3627 
3628 	/* Set up the timer service */
3629 	callout_init_mtx(&adapter->timer_service, &adapter->global_mtx, 0);
3630 	adapter->keep_alive_timeout = DEFAULT_KEEP_ALIVE_TO;
3631 	adapter->missing_tx_timeout = DEFAULT_TX_CMP_TO;
3632 	adapter->missing_tx_max_queues = DEFAULT_TX_MONITORED_QUEUES;
3633 	adapter->missing_tx_threshold = DEFAULT_TX_CMP_THRESHOLD;
3634 
3635 	if (version_printed++ == 0)
3636 		device_printf(pdev, "%s\n", ena_version);
3637 
3638 	rc = ena_allocate_pci_resources(adapter);
3639 	if (unlikely(rc != 0)) {
3640 		device_printf(pdev, "PCI resource allocation failed!\n");
3641 		ena_free_pci_resources(adapter);
3642 		return (rc);
3643 	}
3644 
3645 	/* Allocate memory for ena_dev structure */
3646 	ena_dev = malloc(sizeof(struct ena_com_dev), M_DEVBUF,
3647 	    M_WAITOK | M_ZERO);
3648 
3649 	adapter->ena_dev = ena_dev;
3650 	ena_dev->dmadev = pdev;
3651 	ena_dev->bus = malloc(sizeof(struct ena_bus), M_DEVBUF,
3652 	    M_WAITOK | M_ZERO);
3653 
3654 	/* Store register resources */
3655 	((struct ena_bus*)(ena_dev->bus))->reg_bar_t =
3656 	    rman_get_bustag(adapter->registers);
3657 	((struct ena_bus*)(ena_dev->bus))->reg_bar_h =
3658 	    rman_get_bushandle(adapter->registers);
3659 
3660 	if (unlikely(((struct ena_bus*)(ena_dev->bus))->reg_bar_h == 0)) {
3661 		device_printf(pdev, "failed to pmap registers bar\n");
3662 		rc = ENXIO;
3663 		goto err_bus_free;
3664 	}
3665 
3666 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3667 
3668 	/* Device initialization */
3669 	rc = ena_device_init(adapter, pdev, &get_feat_ctx, &adapter->wd_active);
3670 	if (unlikely(rc != 0)) {
3671 		device_printf(pdev, "ENA device init failed! (err: %d)\n", rc);
3672 		rc = ENXIO;
3673 		goto err_bus_free;
3674 	}
3675 
3676 	adapter->keep_alive_timestamp = getsbinuptime();
3677 
3678 	adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3679 
3680 	/* Set for sure that interface is not up */
3681 	adapter->up = false;
3682 
3683 	memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3684 	    ETHER_ADDR_LEN);
3685 
3686 	/* calculate IO queue number to create */
3687 	io_queue_num = ena_calc_io_queue_num(adapter, &get_feat_ctx);
3688 
3689 	ENA_ASSERT(io_queue_num > 0, "Invalid queue number: %d\n",
3690 	    io_queue_num);
3691 	adapter->num_queues = io_queue_num;
3692 
3693 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
3694 
3695 	/* calculatre ring sizes */
3696 	queue_size = ena_calc_queue_size(adapter,&tx_sgl_size,
3697 	    &rx_sgl_size, &get_feat_ctx);
3698 	if (unlikely((queue_size <= 0) || (io_queue_num <= 0))) {
3699 		rc = ENA_COM_FAULT;
3700 		goto err_com_free;
3701 	}
3702 
3703 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3704 
3705 	adapter->tx_ring_size = queue_size;
3706 	adapter->rx_ring_size = queue_size;
3707 
3708 	adapter->max_tx_sgl_size = tx_sgl_size;
3709 	adapter->max_rx_sgl_size = rx_sgl_size;
3710 
3711 	/* set up dma tags for rx and tx buffers */
3712 	rc = ena_setup_tx_dma_tag(adapter);
3713 	if (unlikely(rc != 0)) {
3714 		device_printf(pdev, "Failed to create TX DMA tag\n");
3715 		goto err_com_free;
3716 	}
3717 
3718 	rc = ena_setup_rx_dma_tag(adapter);
3719 	if (unlikely(rc != 0)) {
3720 		device_printf(pdev, "Failed to create RX DMA tag\n");
3721 		goto err_tx_tag_free;
3722 	}
3723 
3724 	/* initialize rings basic information */
3725 	device_printf(pdev, "initalize %d io queues\n", io_queue_num);
3726 	ena_init_io_rings(adapter);
3727 
3728 	/* setup network interface */
3729 	rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
3730 	if (unlikely(rc != 0)) {
3731 		device_printf(pdev, "Error with network interface setup\n");
3732 		goto err_io_free;
3733 	}
3734 
3735 	rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3736 	if (unlikely(rc != 0)) {
3737 		device_printf(pdev,
3738 		    "Failed to enable and set the admin interrupts\n");
3739 		goto err_ifp_free;
3740 	}
3741 
3742 	/* Initialize reset task queue */
3743 	TASK_INIT(&adapter->reset_task, 0, ena_reset_task, adapter);
3744 	adapter->reset_tq = taskqueue_create("ena_reset_enqueue",
3745 	    M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->reset_tq);
3746 	taskqueue_start_threads(&adapter->reset_tq, 1, PI_NET,
3747 	    "%s rstq", device_get_nameunit(adapter->pdev));
3748 
3749 	/* Initialize statistics */
3750 	ena_alloc_counters((counter_u64_t *)&adapter->dev_stats,
3751 	    sizeof(struct ena_stats_dev));
3752 	ena_alloc_counters((counter_u64_t *)&adapter->hw_stats,
3753 	    sizeof(struct ena_hw_stats));
3754 	ena_sysctl_add_nodes(adapter);
3755 
3756 	/* Tell the stack that the interface is not active */
3757 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
3758 
3759 	adapter->running = true;
3760 	return (0);
3761 
3762 err_ifp_free:
3763 	if_detach(adapter->ifp);
3764 	if_free(adapter->ifp);
3765 err_io_free:
3766 	ena_free_all_io_rings_resources(adapter);
3767 	ena_free_rx_dma_tag(adapter);
3768 err_tx_tag_free:
3769 	ena_free_tx_dma_tag(adapter);
3770 err_com_free:
3771 	ena_com_admin_destroy(ena_dev);
3772 	ena_com_delete_host_info(ena_dev);
3773 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3774 err_bus_free:
3775 	free(ena_dev->bus, M_DEVBUF);
3776 	free(ena_dev, M_DEVBUF);
3777 	ena_free_pci_resources(adapter);
3778 
3779 	return (rc);
3780 }
3781 
3782 /**
3783  * ena_detach - Device Removal Routine
3784  * @pdev: device information struct
3785  *
3786  * ena_detach is called by the device subsystem to alert the driver
3787  * that it should release a PCI device.
3788  **/
3789 static int
3790 ena_detach(device_t pdev)
3791 {
3792 	struct ena_adapter *adapter = device_get_softc(pdev);
3793 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3794 	int rc;
3795 
3796 	/* Make sure VLANS are not using driver */
3797 	if (adapter->ifp->if_vlantrunk != NULL) {
3798 		device_printf(adapter->pdev ,"VLAN is in use, detach first\n");
3799 		return (EBUSY);
3800 	}
3801 
3802 	/* Free reset task and callout */
3803 	callout_drain(&adapter->timer_service);
3804 	while (taskqueue_cancel(adapter->reset_tq, &adapter->reset_task, NULL))
3805 		taskqueue_drain(adapter->reset_tq, &adapter->reset_task);
3806 	taskqueue_free(adapter->reset_tq);
3807 
3808 	sx_xlock(&adapter->ioctl_sx);
3809 	ena_down(adapter);
3810 	sx_unlock(&adapter->ioctl_sx);
3811 
3812 	if (adapter->ifp != NULL) {
3813 		ether_ifdetach(adapter->ifp);
3814 		if_free(adapter->ifp);
3815 	}
3816 
3817 	ena_free_all_io_rings_resources(adapter);
3818 
3819 	ena_free_counters((counter_u64_t *)&adapter->hw_stats,
3820 	    sizeof(struct ena_hw_stats));
3821 	ena_free_counters((counter_u64_t *)&adapter->dev_stats,
3822 	    sizeof(struct ena_stats_dev));
3823 
3824 	if (likely(adapter->rss_support))
3825 		ena_com_rss_destroy(ena_dev);
3826 
3827 	rc = ena_free_rx_dma_tag(adapter);
3828 	if (unlikely(rc != 0))
3829 		device_printf(adapter->pdev,
3830 		    "Unmapped RX DMA tag associations\n");
3831 
3832 	rc = ena_free_tx_dma_tag(adapter);
3833 	if (unlikely(rc != 0))
3834 		device_printf(adapter->pdev,
3835 		    "Unmapped TX DMA tag associations\n");
3836 
3837 	/* Reset the device only if the device is running. */
3838 	if (adapter->running)
3839 		ena_com_dev_reset(ena_dev, adapter->reset_reason);
3840 
3841 	ena_com_delete_host_info(ena_dev);
3842 
3843 	ena_free_irqs(adapter);
3844 
3845 	ena_com_abort_admin_commands(ena_dev);
3846 
3847 	ena_com_wait_for_abort_completion(ena_dev);
3848 
3849 	ena_com_admin_destroy(ena_dev);
3850 
3851 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3852 
3853 	ena_free_pci_resources(adapter);
3854 
3855 	mtx_destroy(&adapter->global_mtx);
3856 	sx_destroy(&adapter->ioctl_sx);
3857 
3858 	if (ena_dev->bus != NULL)
3859 		free(ena_dev->bus, M_DEVBUF);
3860 
3861 	if (ena_dev != NULL)
3862 		free(ena_dev, M_DEVBUF);
3863 
3864 	return (bus_generic_detach(pdev));
3865 }
3866 
3867 /******************************************************************************
3868  ******************************** AENQ Handlers *******************************
3869  *****************************************************************************/
3870 /**
3871  * ena_update_on_link_change:
3872  * Notify the network interface about the change in link status
3873  **/
3874 static void
3875 ena_update_on_link_change(void *adapter_data,
3876     struct ena_admin_aenq_entry *aenq_e)
3877 {
3878 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3879 	struct ena_admin_aenq_link_change_desc *aenq_desc;
3880 	int status;
3881 	if_t ifp;
3882 
3883 	aenq_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
3884 	ifp = adapter->ifp;
3885 	status = aenq_desc->flags &
3886 	    ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3887 
3888 	if (status != 0) {
3889 		device_printf(adapter->pdev, "link is UP\n");
3890 		if_link_state_change(ifp, LINK_STATE_UP);
3891 	} else if (status == 0) {
3892 		device_printf(adapter->pdev, "link is DOWN\n");
3893 		if_link_state_change(ifp, LINK_STATE_DOWN);
3894 	} else {
3895 		device_printf(adapter->pdev, "invalid value recvd\n");
3896 		BUG();
3897 	}
3898 
3899 	adapter->link_status = status;
3900 }
3901 
3902 /**
3903  * This handler will called for unknown event group or unimplemented handlers
3904  **/
3905 static void
3906 unimplemented_aenq_handler(void *data,
3907     struct ena_admin_aenq_entry *aenq_e)
3908 {
3909 	return;
3910 }
3911 
3912 static struct ena_aenq_handlers aenq_handlers = {
3913     .handlers = {
3914 	    [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3915 	    [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3916     },
3917     .unimplemented_handler = unimplemented_aenq_handler
3918 };
3919 
3920 /*********************************************************************
3921  *  FreeBSD Device Interface Entry Points
3922  *********************************************************************/
3923 
3924 static device_method_t ena_methods[] = {
3925     /* Device interface */
3926     DEVMETHOD(device_probe, ena_probe),
3927     DEVMETHOD(device_attach, ena_attach),
3928     DEVMETHOD(device_detach, ena_detach),
3929     DEVMETHOD_END
3930 };
3931 
3932 static driver_t ena_driver = {
3933     "ena", ena_methods, sizeof(struct ena_adapter),
3934 };
3935 
3936 devclass_t ena_devclass;
3937 DRIVER_MODULE(ena, pci, ena_driver, ena_devclass, 0, 0);
3938 MODULE_DEPEND(ena, pci, 1, 1, 1);
3939 MODULE_DEPEND(ena, ether, 1, 1, 1);
3940 
3941 /*********************************************************************/
3942