xref: /freebsd/sys/dev/ena/ena.c (revision 7d2544e60eec556c8fa02d2af61b804fee1d7136)
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 static ena_vendor_info_t ena_vendor_info_array[] = {
198     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0},
199     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_PF, 0},
200     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0},
201     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_VF, 0},
202     /* Last entry */
203     { 0, 0, 0 }
204 };
205 
206 /*
207  * Contains pointers to event handlers, e.g. link state chage.
208  */
209 static struct ena_aenq_handlers aenq_handlers;
210 
211 void
212 ena_dmamap_callback(void *arg, bus_dma_segment_t *segs, int nseg, int error)
213 {
214 	if (error)
215 		return;
216 	*(bus_addr_t *) arg = segs[0].ds_addr;
217 	return;
218 }
219 
220 int
221 ena_dma_alloc(device_t dmadev, bus_size_t size,
222     ena_mem_handle_t *dma , int mapflags)
223 {
224 	struct ena_adapter* adapter = device_get_softc(dmadev);
225 	uint32_t maxsize = ((size - 1)/PAGE_SIZE + 1) * PAGE_SIZE;
226 	uint64_t dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
227 	int error;
228 
229 	if (dma_space_addr == 0)
230 		dma_space_addr = BUS_SPACE_MAXADDR;
231 	error = bus_dma_tag_create(bus_get_dma_tag(dmadev), /* parent */
232 	    8, 0,	      /* alignment, bounds 		*/
233 	    dma_space_addr,   /* lowaddr of exclusion window	*/
234 	    BUS_SPACE_MAXADDR,/* highaddr of exclusion window	*/
235 	    NULL, NULL,	      /* filter, filterarg 		*/
236 	    maxsize,	      /* maxsize 			*/
237 	    1,		      /* nsegments 			*/
238 	    maxsize,	      /* maxsegsize 			*/
239 	    BUS_DMA_ALLOCNOW, /* flags 				*/
240 	    NULL,	      /* lockfunc 			*/
241 	    NULL,	      /* lockarg 			*/
242 	    &dma->tag);
243 	if (error) {
244 		device_printf(dmadev,
245 		"%s: bus_dma_tag_create failed: %d\n",
246 		__func__, error);
247 		goto fail_tag;
248 	}
249 
250 	error = bus_dmamem_alloc(dma->tag, (void**) &dma->vaddr,
251 	    BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->map);
252 	if (error) {
253 		device_printf(dmadev,
254 		"%s: bus_dmamem_alloc(%ju) failed: %d\n",
255 		__func__, (uintmax_t)size, error);
256 		goto fail_map_create;
257 	}
258 
259 	dma->paddr = 0;
260 	error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr,
261 	    size, ena_dmamap_callback, &dma->paddr, mapflags);
262 	if (error || dma->paddr == 0) {
263 		device_printf(dmadev,
264 		"%s: bus_dmamap_load failed: %d\n",
265 		__func__, error);
266 		goto fail_map_load;
267 	}
268 
269 	return (0);
270 
271 fail_map_load:
272 	bus_dmamem_free(dma->tag, dma->vaddr, dma->map);
273 fail_map_create:
274 	bus_dma_tag_destroy(dma->tag);
275 fail_tag:
276 	dma->tag = NULL;
277 
278 	return (error);
279 }
280 
281 static int
282 ena_allocate_pci_resources(struct ena_adapter* adapter)
283 {
284 	device_t pdev = adapter->pdev;
285 	int rid;
286 
287 	rid = PCIR_BAR(ENA_REG_BAR);
288 	adapter->memory = NULL;
289 	adapter->registers = bus_alloc_resource_any(pdev, SYS_RES_MEMORY,
290 	    &rid, RF_ACTIVE);
291 	if (adapter->registers == NULL) {
292 		device_printf(pdev, "Unable to allocate bus resource: "
293 		    "registers\n");
294 		return (ENXIO);
295 	}
296 
297 	return (0);
298 }
299 
300 static void
301 ena_free_pci_resources(struct ena_adapter *adapter)
302 {
303 	device_t pdev = adapter->pdev;
304 
305 	if (adapter->memory != NULL) {
306 		bus_release_resource(pdev, SYS_RES_MEMORY,
307 		    PCIR_BAR(ENA_MEM_BAR), adapter->memory);
308 	}
309 
310 	if (adapter->registers != NULL) {
311 		bus_release_resource(pdev, SYS_RES_MEMORY,
312 		    PCIR_BAR(ENA_REG_BAR), adapter->registers);
313 	}
314 
315 	return;
316 }
317 
318 static int
319 ena_probe(device_t dev)
320 {
321 	ena_vendor_info_t *ent;
322 	char		adapter_name[60];
323 	uint16_t	pci_vendor_id = 0;
324 	uint16_t	pci_device_id = 0;
325 
326 	pci_vendor_id = pci_get_vendor(dev);
327 	pci_device_id = pci_get_device(dev);
328 
329 	ent = ena_vendor_info_array;
330 	while (ent->vendor_id != 0) {
331 		if ((pci_vendor_id == ent->vendor_id) &&
332 		    (pci_device_id == ent->device_id)) {
333 			ena_trace(ENA_DBG, "vendor=%x device=%x ",
334 			    pci_vendor_id, pci_device_id);
335 
336 			sprintf(adapter_name, DEVICE_DESC);
337 			device_set_desc_copy(dev, adapter_name);
338 			return (BUS_PROBE_DEFAULT);
339 		}
340 
341 		ent++;
342 
343 	}
344 
345 	return (ENXIO);
346 }
347 
348 static int
349 ena_change_mtu(if_t ifp, int new_mtu)
350 {
351 	struct ena_adapter *adapter = if_getsoftc(ifp);
352 	struct ena_com_dev_get_features_ctx get_feat_ctx;
353 	int rc, old_mtu, max_frame;
354 
355 	rc = ena_com_get_dev_attr_feat(adapter->ena_dev, &get_feat_ctx);
356 	if (rc) {
357 		device_printf(adapter->pdev,
358 		    "Cannot get attribute for ena device\n");
359 		return (ENXIO);
360 	}
361 
362 	/* Save old MTU in case of fail */
363 	old_mtu = if_getmtu(ifp);
364 
365 	/* Change MTU and calculate max frame */
366 	if_setmtu(ifp, new_mtu);
367 	max_frame = ETHER_MAX_FRAME(ifp, ETHERTYPE_VLAN, 1);
368 
369 	if ((new_mtu < ENA_MIN_FRAME_LEN) ||
370 	    (new_mtu > get_feat_ctx.dev_attr.max_mtu) ||
371 	    (max_frame > ENA_MAX_FRAME_LEN)) {
372 		device_printf(adapter->pdev, "Invalid MTU setting. "
373 		    "new_mtu: %d\n", new_mtu);
374 		goto error;
375 	}
376 
377 	rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
378 	if (rc != 0)
379 		goto error;
380 
381 	return (0);
382 error:
383 	if_setmtu(ifp, old_mtu);
384 	return (EINVAL);
385 }
386 
387 static inline void
388 ena_alloc_counters(counter_u64_t *begin, int size)
389 {
390 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
391 
392 	for (; begin < end; ++begin)
393 		*begin = counter_u64_alloc(M_WAITOK);
394 }
395 
396 static inline void
397 ena_free_counters(counter_u64_t *begin, int size)
398 {
399 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
400 
401 	for (; begin < end; ++begin)
402 		counter_u64_free(*begin);
403 }
404 
405 static inline void
406 ena_reset_counters(counter_u64_t *begin, int size)
407 {
408 	counter_u64_t *end = (counter_u64_t *)((char *)begin + size);
409 
410 	for (; begin < end; ++begin)
411 		counter_u64_zero(*begin);
412 }
413 
414 static void
415 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
416     uint16_t qid)
417 {
418 
419 	ring->qid = qid;
420 	ring->adapter = adapter;
421 	ring->ena_dev = adapter->ena_dev;
422 }
423 
424 static void
425 ena_init_io_rings(struct ena_adapter *adapter)
426 {
427 	struct ena_com_dev *ena_dev;
428 	struct ena_ring *txr, *rxr;
429 	struct ena_que *que;
430 	int i;
431 
432 	ena_dev = adapter->ena_dev;
433 
434 	for (i = 0; i < adapter->num_queues; i++) {
435 		txr = &adapter->tx_ring[i];
436 		rxr = &adapter->rx_ring[i];
437 
438 		/* TX/RX common ring state */
439 		ena_init_io_rings_common(adapter, txr, i);
440 		ena_init_io_rings_common(adapter, rxr, i);
441 
442 		/* TX specific ring state */
443 		txr->ring_size = adapter->tx_ring_size;
444 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
445 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
446 		txr->smoothed_interval =
447 		    ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
448 
449 		/* Allocate a buf ring */
450 		txr->br = buf_ring_alloc(ena_buf_ring_size, M_DEVBUF,
451 		    M_WAITOK, &txr->ring_mtx);
452 
453 		/* Alloc TX statistics. */
454 		ena_alloc_counters((counter_u64_t *)&txr->tx_stats,
455 		    sizeof(txr->tx_stats));
456 
457 		/* RX specific ring state */
458 		rxr->ring_size = adapter->rx_ring_size;
459 		rxr->rx_small_copy_len = adapter->small_copy_len;
460 		rxr->smoothed_interval =
461 		    ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
462 
463 		/* Alloc RX statistics. */
464 		ena_alloc_counters((counter_u64_t *)&rxr->rx_stats,
465 		    sizeof(rxr->rx_stats));
466 
467 		/* Initialize locks */
468 		snprintf(txr->mtx_name, nitems(txr->mtx_name), "%s:tx(%d)",
469 		    device_get_nameunit(adapter->pdev), i);
470 		snprintf(rxr->mtx_name, nitems(rxr->mtx_name), "%s:rx(%d)",
471 		    device_get_nameunit(adapter->pdev), i);
472 
473 		mtx_init(&txr->ring_mtx, txr->mtx_name, NULL, MTX_DEF);
474 		mtx_init(&rxr->ring_mtx, rxr->mtx_name, NULL, MTX_DEF);
475 
476 		que = &adapter->que[i];
477 		que->adapter = adapter;
478 		que->id = i;
479 		que->tx_ring = txr;
480 		que->rx_ring = rxr;
481 
482 		txr->que = que;
483 		rxr->que = que;
484 
485 		rxr->empty_rx_queue = 0;
486 	}
487 }
488 
489 static void
490 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
491 {
492 	struct ena_ring *txr = &adapter->tx_ring[qid];
493 	struct ena_ring *rxr = &adapter->rx_ring[qid];
494 
495 	ena_free_counters((counter_u64_t *)&txr->tx_stats,
496 	    sizeof(txr->tx_stats));
497 	ena_free_counters((counter_u64_t *)&rxr->rx_stats,
498 	    sizeof(rxr->rx_stats));
499 
500 	ENA_RING_MTX_LOCK(txr);
501 	drbr_free(txr->br, M_DEVBUF);
502 	ENA_RING_MTX_UNLOCK(txr);
503 
504 	mtx_destroy(&txr->ring_mtx);
505 	mtx_destroy(&rxr->ring_mtx);
506 }
507 
508 static void
509 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
510 {
511 	int i;
512 
513 	for (i = 0; i < adapter->num_queues; i++)
514 		ena_free_io_ring_resources(adapter, i);
515 
516 }
517 
518 static int
519 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
520 {
521 	int ret;
522 
523 	/* Create DMA tag for Tx buffers */
524 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
525 	    1, 0,				  /* alignment, bounds 	     */
526 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
527 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
528 	    NULL, NULL,				  /* filter, filterarg 	     */
529 	    ENA_TSO_MAXSIZE,			  /* maxsize 		     */
530 	    adapter->max_tx_sgl_size - 1,	  /* nsegments 		     */
531 	    ENA_TSO_MAXSIZE,			  /* maxsegsize 	     */
532 	    0,					  /* flags 		     */
533 	    NULL,				  /* lockfunc 		     */
534 	    NULL,				  /* lockfuncarg 	     */
535 	    &adapter->tx_buf_tag);
536 
537 	if (ret != 0)
538 		device_printf(adapter->pdev, "Unable to create Tx DMA tag\n");
539 
540 	return (ret);
541 }
542 
543 static int
544 ena_free_tx_dma_tag(struct ena_adapter *adapter)
545 {
546 	int ret;
547 
548 	ret = bus_dma_tag_destroy(adapter->tx_buf_tag);
549 
550 	if (ret == 0)
551 		adapter->tx_buf_tag = NULL;
552 
553 	return (ret);
554 }
555 
556 static int
557 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
558 {
559 	int ret;
560 
561 	/* Create DMA tag for Rx buffers*/
562 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent   */
563 	    1, 0,				  /* alignment, bounds 	     */
564 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
565 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
566 	    NULL, NULL,				  /* filter, filterarg 	     */
567 	    MJUM16BYTES,			  /* maxsize 		     */
568 	    1,					  /* nsegments 		     */
569 	    MJUM16BYTES,			  /* maxsegsize 	     */
570 	    0,					  /* flags 		     */
571 	    NULL,				  /* lockfunc 		     */
572 	    NULL,				  /* lockarg 		     */
573 	    &adapter->rx_buf_tag);
574 
575 	if (ret != 0)
576 		device_printf(adapter->pdev, "Unable to create Rx DMA tag\n");
577 
578 	return (ret);
579 }
580 
581 static int
582 ena_free_rx_dma_tag(struct ena_adapter *adapter)
583 {
584 	int ret;
585 
586 	ret = bus_dma_tag_destroy(adapter->rx_buf_tag);
587 
588 	if (ret == 0)
589 		adapter->rx_buf_tag = NULL;
590 
591 	return (ret);
592 }
593 
594 
595 /**
596  * ena_setup_tx_resources - allocate Tx resources (Descriptors)
597  * @adapter: network interface device structure
598  * @qid: queue index
599  *
600  * Returns 0 on success, otherwise on failure.
601  **/
602 static int
603 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
604 {
605 	struct ena_que *que = &adapter->que[qid];
606 	struct ena_ring *tx_ring = que->tx_ring;
607 	int size, i, err;
608 #ifdef	RSS
609 	cpuset_t cpu_mask;
610 #endif
611 
612 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
613 
614 	tx_ring->tx_buffer_info = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
615 	if (tx_ring->tx_buffer_info == NULL)
616 		return (ENOMEM);
617 
618 	size = sizeof(uint16_t) * tx_ring->ring_size;
619 	tx_ring->free_tx_ids = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
620 	if (tx_ring->free_tx_ids == NULL)
621 		goto err_buf_info_free;
622 
623 	/* Req id stack for TX OOO completions */
624 	for (i = 0; i < tx_ring->ring_size; i++)
625 		tx_ring->free_tx_ids[i] = i;
626 
627 	/* Reset TX statistics. */
628 	ena_reset_counters((counter_u64_t *)&tx_ring->tx_stats,
629 	    sizeof(tx_ring->tx_stats));
630 
631 	tx_ring->next_to_use = 0;
632 	tx_ring->next_to_clean = 0;
633 
634 	/* Make sure that drbr is empty */
635 	ENA_RING_MTX_LOCK(tx_ring);
636 	drbr_flush(adapter->ifp, tx_ring->br);
637 	ENA_RING_MTX_UNLOCK(tx_ring);
638 
639 	/* ... and create the buffer DMA maps */
640 	for (i = 0; i < tx_ring->ring_size; i++) {
641 		err = bus_dmamap_create(adapter->tx_buf_tag, 0,
642 		    &tx_ring->tx_buffer_info[i].map);
643 		if (err != 0) {
644 			device_printf(adapter->pdev,
645 			    "Unable to create Tx DMA map for buffer %d\n", i);
646 			goto err_buf_info_unmap;
647 		}
648 	}
649 
650 	/* Allocate taskqueues */
651 	TASK_INIT(&tx_ring->enqueue_task, 0, ena_deferred_mq_start, tx_ring);
652 	tx_ring->enqueue_tq = taskqueue_create_fast("ena_tx_enque", M_NOWAIT,
653 	    taskqueue_thread_enqueue, &tx_ring->enqueue_tq);
654 	if (tx_ring->enqueue_tq == NULL) {
655 		device_printf(adapter->pdev,
656 		    "Unable to create taskqueue for enqueue task\n");
657 		i = tx_ring->ring_size;
658 		goto err_buf_info_unmap;
659 	}
660 
661 	/* RSS set cpu for thread */
662 #ifdef RSS
663 	CPU_SETOF(que->cpu, &cpu_mask);
664 	taskqueue_start_threads_cpuset(&tx_ring->enqueue_tq, 1, PI_NET,
665 	    &cpu_mask, "%s tx_ring enq (bucket %d)",
666 	    device_get_nameunit(adapter->pdev), que->cpu);
667 #else /* RSS */
668 	taskqueue_start_threads(&tx_ring->enqueue_tq, 1, PI_NET,
669 	    "%s txeq %d", device_get_nameunit(adapter->pdev), que->cpu);
670 #endif /* RSS */
671 
672 	return (0);
673 
674 err_buf_info_unmap:
675 	while (i--) {
676 		bus_dmamap_destroy(adapter->tx_buf_tag,
677 		    tx_ring->tx_buffer_info[i].map);
678 	}
679 	free(tx_ring->free_tx_ids, M_DEVBUF);
680 	tx_ring->free_tx_ids = NULL;
681 err_buf_info_free:
682 	free(tx_ring->tx_buffer_info, M_DEVBUF);
683 	tx_ring->tx_buffer_info = NULL;
684 
685 	return (ENOMEM);
686 }
687 
688 /**
689  * ena_free_tx_resources - Free Tx Resources per Queue
690  * @adapter: network interface device structure
691  * @qid: queue index
692  *
693  * Free all transmit software resources
694  **/
695 static void
696 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
697 {
698 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
699 
700 	while (taskqueue_cancel(tx_ring->enqueue_tq, &tx_ring->enqueue_task,
701 	    NULL))
702 		taskqueue_drain(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
703 
704 	taskqueue_free(tx_ring->enqueue_tq);
705 
706 	ENA_RING_MTX_LOCK(tx_ring);
707 	/* Flush buffer ring, */
708 	drbr_flush(adapter->ifp, tx_ring->br);
709 
710 	/* Free buffer DMA maps, */
711 	for (int i = 0; i < tx_ring->ring_size; i++) {
712 		m_freem(tx_ring->tx_buffer_info[i].mbuf);
713 		tx_ring->tx_buffer_info[i].mbuf = NULL;
714 		bus_dmamap_unload(adapter->tx_buf_tag,
715 		    tx_ring->tx_buffer_info[i].map);
716 		bus_dmamap_destroy(adapter->tx_buf_tag,
717 		    tx_ring->tx_buffer_info[i].map);
718 	}
719 	ENA_RING_MTX_UNLOCK(tx_ring);
720 
721 	/* And free allocated memory. */
722 	free(tx_ring->tx_buffer_info, M_DEVBUF);
723 	tx_ring->tx_buffer_info = NULL;
724 
725 	free(tx_ring->free_tx_ids, M_DEVBUF);
726 	tx_ring->free_tx_ids = NULL;
727 }
728 
729 /**
730  * ena_setup_all_tx_resources - allocate all queues Tx resources
731  * @adapter: network interface device structure
732  *
733  * Returns 0 on success, otherwise on failure.
734  **/
735 static int
736 ena_setup_all_tx_resources(struct ena_adapter *adapter)
737 {
738 	int i, rc;
739 
740 	for (i = 0; i < adapter->num_queues; i++) {
741 		rc = ena_setup_tx_resources(adapter, i);
742 		if (rc) {
743 			device_printf(adapter->pdev,
744 			    "Allocation for Tx Queue %u failed\n", i);
745 			goto err_setup_tx;
746 		}
747 	}
748 
749 	return (0);
750 
751 err_setup_tx:
752 	/* Rewind the index freeing the rings as we go */
753 	while (i--)
754 		ena_free_tx_resources(adapter, i);
755 	return (rc);
756 }
757 
758 /**
759  * ena_free_all_tx_resources - Free Tx Resources for All Queues
760  * @adapter: network interface device structure
761  *
762  * Free all transmit software resources
763  **/
764 static void
765 ena_free_all_tx_resources(struct ena_adapter *adapter)
766 {
767 	int i;
768 
769 	for (i = 0; i < adapter->num_queues; i++)
770 		ena_free_tx_resources(adapter, i);
771 
772 	return;
773 }
774 
775 static inline int
776 validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
777 {
778 	if (likely(req_id < rx_ring->ring_size))
779 		return (0);
780 
781 	device_printf(rx_ring->adapter->pdev, "Invalid rx req_id: %hu\n",
782 	    req_id);
783 	counter_u64_add(rx_ring->rx_stats.bad_req_id, 1);
784 
785 	/* Trigger device reset */
786 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
787 	rx_ring->adapter->trigger_reset = true;
788 
789 	return (EFAULT);
790 }
791 
792 /**
793  * ena_setup_rx_resources - allocate Rx resources (Descriptors)
794  * @adapter: network interface device structure
795  * @qid: queue index
796  *
797  * Returns 0 on success, otherwise on failure.
798  **/
799 static int
800 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
801 {
802 	struct ena_que *que = &adapter->que[qid];
803 	struct ena_ring *rx_ring = que->rx_ring;
804 	int size, err, i;
805 #ifdef	RSS
806 	cpuset_t cpu_mask;
807 #endif
808 
809 	size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
810 
811 	/*
812 	 * Alloc extra element so in rx path
813 	 * we can always prefetch rx_info + 1
814 	 */
815 	size += sizeof(struct ena_rx_buffer);
816 
817 	rx_ring->rx_buffer_info = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
818 
819 	size = sizeof(uint16_t) * rx_ring->ring_size;
820 	rx_ring->free_rx_ids = malloc(size, M_DEVBUF, M_WAITOK);
821 
822 	for (i = 0; i < rx_ring->ring_size; i++)
823 		rx_ring->free_rx_ids[i] = i;
824 
825 	/* Reset RX statistics. */
826 	ena_reset_counters((counter_u64_t *)&rx_ring->rx_stats,
827 	    sizeof(rx_ring->rx_stats));
828 
829 	rx_ring->next_to_clean = 0;
830 	rx_ring->next_to_use = 0;
831 
832 	/* ... and create the buffer DMA maps */
833 	for (i = 0; i < rx_ring->ring_size; i++) {
834 		err = bus_dmamap_create(adapter->rx_buf_tag, 0,
835 		    &(rx_ring->rx_buffer_info[i].map));
836 		if (err != 0) {
837 			device_printf(adapter->pdev,
838 			    "Unable to create Rx DMA map for buffer %d\n", i);
839 			goto err_buf_info_unmap;
840 		}
841 	}
842 
843 	/* Create LRO for the ring */
844 	if (adapter->ifp->if_capenable & IFCAP_LRO) {
845 		int err = tcp_lro_init(&rx_ring->lro);
846 		if (err) {
847 			device_printf(adapter->pdev,
848 			    "LRO[%d] Initialization failed!\n", qid);
849 		} else {
850 			ena_trace(ENA_INFO,
851 			    "RX Soft LRO[%d] Initialized\n", qid);
852 			rx_ring->lro.ifp = adapter->ifp;
853 		}
854 	}
855 
856 	/* Allocate taskqueues */
857 	TASK_INIT(&rx_ring->cmpl_task, 0, ena_deferred_rx_cleanup, rx_ring);
858 	rx_ring->cmpl_tq = taskqueue_create_fast("ena RX completion", M_WAITOK,
859 	    taskqueue_thread_enqueue, &rx_ring->cmpl_tq);
860 
861 	/* RSS set cpu for thread */
862 #ifdef RSS
863 	CPU_SETOF(que->cpu, &cpu_mask);
864 	taskqueue_start_threads_cpuset(&rx_ring->cmpl_tq, 1, PI_NET, &cpu_mask,
865 	    "%s rx_ring cmpl (bucket %d)",
866 	    device_get_nameunit(adapter->pdev), que->cpu);
867 #else
868 	taskqueue_start_threads(&rx_ring->cmpl_tq, 1, PI_NET,
869 	    "%s rx_ring cmpl %d", device_get_nameunit(adapter->pdev), que->cpu);
870 #endif
871 
872 	return (0);
873 
874 err_buf_info_unmap:
875 	while (i--) {
876 		bus_dmamap_destroy(adapter->rx_buf_tag,
877 		    rx_ring->rx_buffer_info[i].map);
878 	}
879 
880 	free(rx_ring->free_rx_ids, M_DEVBUF);
881 	rx_ring->free_rx_ids = NULL;
882 	free(rx_ring->rx_buffer_info, M_DEVBUF);
883 	rx_ring->rx_buffer_info = NULL;
884 	ena_trace(ENA_ALERT, "RX resource allocation fail");
885 	return (ENOMEM);
886 }
887 
888 /**
889  * ena_free_rx_resources - Free Rx Resources
890  * @adapter: network interface device structure
891  * @qid: queue index
892  *
893  * Free all receive software resources
894  **/
895 static void
896 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
897 {
898 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
899 
900 	ena_trace(ENA_INFO, "%s qid %d\n", __func__, qid);
901 
902 	while (taskqueue_cancel(rx_ring->cmpl_tq, &rx_ring->cmpl_task, NULL) != 0)
903 		taskqueue_drain(rx_ring->cmpl_tq, &rx_ring->cmpl_task);
904 
905 	taskqueue_free(rx_ring->cmpl_tq);
906 
907 	/* Free buffer DMA maps, */
908 	for (int i = 0; i < rx_ring->ring_size; i++) {
909 		m_freem(rx_ring->rx_buffer_info[i].mbuf);
910 		rx_ring->rx_buffer_info[i].mbuf = NULL;
911 		bus_dmamap_unload(adapter->rx_buf_tag,
912 		    rx_ring->rx_buffer_info[i].map);
913 		bus_dmamap_destroy(adapter->rx_buf_tag,
914 		    rx_ring->rx_buffer_info[i].map);
915 	}
916 
917 	/* free LRO resources, */
918 	tcp_lro_free(&rx_ring->lro);
919 
920 	/* free allocated memory */
921 	free(rx_ring->rx_buffer_info, M_DEVBUF);
922 	rx_ring->rx_buffer_info = NULL;
923 
924 	free(rx_ring->free_rx_ids, M_DEVBUF);
925 	rx_ring->free_rx_ids = NULL;
926 
927 	return;
928 }
929 
930 /**
931  * ena_setup_all_rx_resources - allocate all queues Rx resources
932  * @adapter: network interface device structure
933  *
934  * Returns 0 on success, otherwise on failure.
935  **/
936 static int
937 ena_setup_all_rx_resources(struct ena_adapter *adapter)
938 {
939 	int i, rc = 0;
940 
941 	for (i = 0; i < adapter->num_queues; i++) {
942 		rc = ena_setup_rx_resources(adapter, i);
943 		if (rc) {
944 			device_printf(adapter->pdev,
945 			    "Allocation for Rx Queue %u failed\n", i);
946 			goto err_setup_rx;
947 		}
948 	}
949 	return (0);
950 
951 err_setup_rx:
952 	/* rewind the index freeing the rings as we go */
953 	while (i--)
954 		ena_free_rx_resources(adapter, i);
955 	return (rc);
956 }
957 
958 /**
959  * ena_free_all_rx_resources - Free Rx resources for all queues
960  * @adapter: network interface device structure
961  *
962  * Free all receive software resources
963  **/
964 static void
965 ena_free_all_rx_resources(struct ena_adapter *adapter)
966 {
967 	int i;
968 
969 	for (i = 0; i < adapter->num_queues; i++)
970 		ena_free_rx_resources(adapter, i);
971 
972 	return;
973 }
974 
975 static inline int
976 ena_alloc_rx_mbuf(struct ena_adapter *adapter,
977     struct ena_ring *rx_ring, struct ena_rx_buffer *rx_info)
978 {
979 	struct ena_com_buf *ena_buf;
980 	bus_dma_segment_t segs[1];
981 	int nsegs, error;
982 
983 	/* if previous allocated frag is not used */
984 	if (rx_info->mbuf != NULL)
985 		return (0);
986 
987 	/* Get mbuf using UMA allocator */
988 	rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM16BYTES);
989 
990 	if (rx_info->mbuf == NULL) {
991 		counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
992 		return (ENOMEM);
993 	}
994 	/* Set mbuf length*/
995 	rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = MJUM16BYTES;
996 
997 	/* Map packets for DMA */
998 	ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
999 	    "Using tag %p for buffers' DMA mapping, mbuf %p len: %d",
1000 	    adapter->rx_buf_tag,rx_info->mbuf, rx_info->mbuf->m_len);
1001 	error = bus_dmamap_load_mbuf_sg(adapter->rx_buf_tag, rx_info->map,
1002 	    rx_info->mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
1003 	if (error || (nsegs != 1)) {
1004 		device_printf(adapter->pdev, "failed to map mbuf, error: %d, "
1005 		    "nsegs: %d\n", error, nsegs);
1006 		counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
1007 		goto exit;
1008 
1009 	}
1010 
1011 	bus_dmamap_sync(adapter->rx_buf_tag, rx_info->map, BUS_DMASYNC_PREREAD);
1012 
1013 	ena_buf = &rx_info->ena_buf;
1014 	ena_buf->paddr = segs[0].ds_addr;
1015 	ena_buf->len = MJUM16BYTES;
1016 
1017 	ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
1018 	    "ALLOC RX BUF: mbuf %p, rx_info %p, len %d, paddr %#jx\n",
1019 	    rx_info->mbuf, rx_info,ena_buf->len, (uintmax_t)ena_buf->paddr);
1020 
1021 	return (0);
1022 
1023 exit:
1024 	m_freem(rx_info->mbuf);
1025 	rx_info->mbuf = NULL;
1026 	return (EFAULT);
1027 }
1028 
1029 static void
1030 ena_free_rx_mbuf(struct ena_adapter *adapter, struct ena_ring *rx_ring,
1031     struct ena_rx_buffer *rx_info)
1032 {
1033 
1034 	if (rx_info->mbuf == NULL)
1035 		return;
1036 
1037 	bus_dmamap_unload(adapter->rx_buf_tag, rx_info->map);
1038 	m_freem(rx_info->mbuf);
1039 	rx_info->mbuf = NULL;
1040 
1041 	return;
1042 }
1043 
1044 
1045 /**
1046  * ena_refill_rx_bufs - Refills ring with descriptors
1047  * @rx_ring: the ring which we want to feed with free descriptors
1048  * @num: number of descriptors to refill
1049  * Refills the ring with newly allocated DMA-mapped mbufs for receiving
1050  **/
1051 static int
1052 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
1053 {
1054 	struct ena_adapter *adapter = rx_ring->adapter;
1055 	uint16_t next_to_use, req_id;
1056 	uint32_t i;
1057 	int rc;
1058 
1059 	ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC, "refill qid: %d",
1060 	    rx_ring->qid);
1061 
1062 	next_to_use = rx_ring->next_to_use;
1063 
1064 	for (i = 0; i < num; i++) {
1065 		struct ena_rx_buffer *rx_info;
1066 
1067 		ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC,
1068 		    "RX buffer - next to use: %d", next_to_use);
1069 
1070 		req_id = rx_ring->free_rx_ids[next_to_use];
1071 		rc = validate_rx_req_id(rx_ring, req_id);
1072 		if (unlikely(rc))
1073 			break;
1074 
1075 		rx_info = &rx_ring->rx_buffer_info[req_id];
1076 
1077 		rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1078 		if (rc != 0) {
1079 			device_printf(adapter->pdev,
1080 			    "failed to alloc buffer for rx queue\n");
1081 			break;
1082 		}
1083 		rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1084 		    &rx_info->ena_buf, req_id);
1085 		if (unlikely(rc)) {
1086 			device_printf(adapter->pdev,
1087 			    "failed to add buffer for rx queue %d\n",
1088 			    rx_ring->qid);
1089 			break;
1090 		}
1091 		next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1092 		    rx_ring->ring_size);
1093 	}
1094 
1095 	if (i < num) {
1096 		counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1097 		device_printf(adapter->pdev,
1098 		    "refilled rx queue %d with %d pages only\n",
1099 		    rx_ring->qid, i);
1100 	}
1101 
1102 	if (i != 0) {
1103 		wmb();
1104 		ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1105 	}
1106 	rx_ring->next_to_use = next_to_use;
1107 	return (i);
1108 }
1109 
1110 static void
1111 ena_free_rx_bufs(struct ena_adapter *adapter, unsigned int qid)
1112 {
1113 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1114 	unsigned int i;
1115 
1116 	for (i = 0; i < rx_ring->ring_size; i++) {
1117 		struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1118 
1119 		if (rx_info->mbuf)
1120 			ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1121 	}
1122 
1123 	return;
1124 }
1125 
1126 /**
1127  * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1128  * @adapter: network interface device structure
1129  *
1130  */
1131 static void
1132 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1133 {
1134 	struct ena_ring *rx_ring;
1135 	int i, rc, bufs_num;
1136 
1137 	for (i = 0; i < adapter->num_queues; i++) {
1138 		rx_ring = &adapter->rx_ring[i];
1139 		bufs_num = rx_ring->ring_size - 1;
1140 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1141 
1142 		if (unlikely(rc != bufs_num))
1143 			device_printf(adapter->pdev,
1144 			    "refilling Queue %d failed. allocated %d buffers"
1145 			    " from: %d\n", i, rc, bufs_num);
1146 	}
1147 }
1148 
1149 static void
1150 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1151 {
1152 	int i;
1153 
1154 	for (i = 0; i < adapter->num_queues; i++)
1155 		ena_free_rx_bufs(adapter, i);
1156 	return;
1157 }
1158 
1159 /**
1160  * ena_free_tx_bufs - Free Tx Buffers per Queue
1161  * @adapter: network interface device structure
1162  * @qid: queue index
1163  **/
1164 static void
1165 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1166 {
1167 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1168 
1169 	ENA_RING_MTX_LOCK(tx_ring);
1170 	for (int i = 0; i < tx_ring->ring_size; i++) {
1171 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1172 
1173 		if (tx_info->mbuf == NULL)
1174 			continue;
1175 
1176 		ena_trace(ENA_DBG | ENA_TXPTH | ENA_RSC,
1177 		    "free uncompleted Tx mbufs qid[%d] idx: 0x%x", qid, i);
1178 
1179 		bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
1180 		m_free(tx_info->mbuf);
1181 		tx_info->mbuf = NULL;
1182 	}
1183 	ENA_RING_MTX_UNLOCK(tx_ring);
1184 
1185 	return;
1186 }
1187 
1188 static void
1189 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1190 {
1191 
1192 	for (int i = 0; i < adapter->num_queues; i++)
1193 		ena_free_tx_bufs(adapter, i);
1194 
1195 	return;
1196 }
1197 
1198 static void
1199 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1200 {
1201 	uint16_t ena_qid;
1202 	int i;
1203 
1204 	for (i = 0; i < adapter->num_queues; i++) {
1205 		ena_qid = ENA_IO_TXQ_IDX(i);
1206 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1207 	}
1208 }
1209 
1210 static void
1211 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1212 {
1213 	uint16_t ena_qid;
1214 	int i;
1215 
1216 	for (i = 0; i < adapter->num_queues; i++) {
1217 		ena_qid = ENA_IO_RXQ_IDX(i);
1218 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1219 	}
1220 }
1221 
1222 static void
1223 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1224 {
1225 	ena_destroy_all_tx_queues(adapter);
1226 	ena_destroy_all_rx_queues(adapter);
1227 }
1228 
1229 static inline int
1230 validate_tx_req_id(struct ena_ring *tx_ring, uint16_t req_id)
1231 {
1232 	struct ena_tx_buffer *tx_info = NULL;
1233 
1234 	if (likely(req_id < tx_ring->ring_size)) {
1235 		tx_info = &tx_ring->tx_buffer_info[req_id];
1236 		if (tx_info->mbuf)
1237 			return 0;
1238 	}
1239 
1240 	counter_u64_add(tx_ring->tx_stats.bad_req_id, 1);
1241 
1242 	return (EFAULT);
1243 }
1244 
1245 static int
1246 ena_create_io_queues(struct ena_adapter *adapter)
1247 {
1248 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1249 	struct ena_com_create_io_ctx ctx;
1250 	struct ena_ring *ring;
1251 	uint16_t ena_qid;
1252 	uint32_t msix_vector;
1253 	int rc, i;
1254 
1255 	/* Create TX queues */
1256 	for (i = 0; i < adapter->num_queues; i++) {
1257 		msix_vector = ENA_IO_IRQ_IDX(i);
1258 		ena_qid = ENA_IO_TXQ_IDX(i);
1259 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1260 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1261 		ctx.queue_size = adapter->tx_ring_size;
1262 		ctx.msix_vector = msix_vector;
1263 		ctx.qid = ena_qid;
1264 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1265 		if (rc) {
1266 			device_printf(adapter->pdev,
1267 			    "Failed to create io TX queue #%d rc: %d\n", i, rc);
1268 			goto err_tx;
1269 		}
1270 		ring = &adapter->tx_ring[i];
1271 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1272 		    &ring->ena_com_io_sq,
1273 		    &ring->ena_com_io_cq);
1274 		if (rc) {
1275 			device_printf(adapter->pdev,
1276 			    "Failed to get TX queue handlers. TX queue num"
1277 			    " %d rc: %d\n", i, rc);
1278 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1279 			goto err_tx;
1280 		}
1281 	}
1282 
1283 	/* Create RX queues */
1284 	for (i = 0; i < adapter->num_queues; i++) {
1285 		msix_vector = ENA_IO_IRQ_IDX(i);
1286 		ena_qid = ENA_IO_RXQ_IDX(i);
1287 		ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1288 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1289 		ctx.queue_size = adapter->rx_ring_size;
1290 		ctx.msix_vector = msix_vector;
1291 		ctx.qid = ena_qid;
1292 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1293 		if (rc) {
1294 			device_printf(adapter->pdev,
1295 			    "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1296 			goto err_rx;
1297 		}
1298 
1299 		ring = &adapter->rx_ring[i];
1300 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1301 		    &ring->ena_com_io_sq,
1302 		    &ring->ena_com_io_cq);
1303 		if (rc) {
1304 			device_printf(adapter->pdev,
1305 			    "Failed to get RX queue handlers. RX queue num"
1306 			    " %d rc: %d\n", i, rc);
1307 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1308 			goto err_rx;
1309 		}
1310 	}
1311 
1312 	return (0);
1313 
1314 err_rx:
1315 	while (i--)
1316 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1317 	i = adapter->num_queues;
1318 err_tx:
1319 	while (i--)
1320 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1321 
1322 	return (ENXIO);
1323 }
1324 
1325 /**
1326  * ena_tx_cleanup - clear sent packets and corresponding descriptors
1327  * @tx_ring: ring for which we want to clean packets
1328  *
1329  * Once packets are sent, we ask the device in a loop for no longer used
1330  * descriptors. We find the related mbuf chain in a map (index in an array)
1331  * and free it, then update ring state.
1332  * This is performed in "endless" loop, updating ring pointers every
1333  * TX_COMMIT. The first check of free descriptor is performed before the actual
1334  * loop, then repeated at the loop end.
1335  **/
1336 static int
1337 ena_tx_cleanup(struct ena_ring *tx_ring)
1338 {
1339 	struct ena_adapter *adapter;
1340 	struct ena_com_io_cq* io_cq;
1341 	uint16_t next_to_clean;
1342 	uint16_t req_id;
1343 	uint16_t ena_qid;
1344 	unsigned int total_done = 0;
1345 	int rc;
1346 	int commit = TX_COMMIT;
1347 	int budget = TX_BUDGET;
1348 	int work_done;
1349 
1350 	adapter = tx_ring->que->adapter;
1351 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
1352 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1353 	next_to_clean = tx_ring->next_to_clean;
1354 
1355 	do {
1356 		struct ena_tx_buffer *tx_info;
1357 		struct mbuf *mbuf;
1358 
1359 		rc = ena_com_tx_comp_req_id_get(io_cq, &req_id);
1360 		if (rc != 0)
1361 			break;
1362 
1363 		rc = validate_tx_req_id(tx_ring, req_id);
1364 		if (rc)
1365 			break;
1366 
1367 		tx_info = &tx_ring->tx_buffer_info[req_id];
1368 
1369 		mbuf = tx_info->mbuf;
1370 
1371 		tx_info->mbuf = NULL;
1372 		bintime_clear(&tx_info->timestamp);
1373 
1374 		if (tx_info->num_of_bufs != 0) {
1375 			/* Map is no longer required */
1376 			bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
1377 		}
1378 
1379 		m_freem(mbuf);
1380 
1381 		total_done += tx_info->tx_descs;
1382 
1383 		tx_ring->free_tx_ids[next_to_clean] = req_id;
1384 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1385 		    tx_ring->ring_size);
1386 
1387 		if (--commit == 0) {
1388 			commit = TX_COMMIT;
1389 			/* update ring state every TX_COMMIT descriptor */
1390 			tx_ring->next_to_clean = next_to_clean;
1391 			ena_com_comp_ack(&adapter->ena_dev->io_sq_queues[ena_qid], total_done);
1392 			ena_com_update_dev_comp_head(io_cq);
1393 			total_done = 0;
1394 		}
1395 	} while (--budget);
1396 
1397 	work_done = TX_BUDGET - budget;
1398 
1399 	/* If there is still something to commit update ring state */
1400 	if (commit != TX_COMMIT) {
1401 		tx_ring->next_to_clean = next_to_clean;
1402 		ena_com_comp_ack(&adapter->ena_dev->io_sq_queues[ena_qid], total_done);
1403 		ena_com_update_dev_comp_head(io_cq);
1404 	}
1405 
1406 	taskqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
1407 
1408 	return (work_done);
1409 }
1410 
1411 static void
1412 ena_rx_hash_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1413     struct mbuf *mbuf)
1414 {
1415 	struct ena_adapter *adapter = rx_ring->adapter;
1416 
1417 	if (adapter->rss_support) {
1418 		mbuf->m_pkthdr.flowid = ena_rx_ctx->hash;
1419 
1420 		if (ena_rx_ctx->frag &&
1421 		    ena_rx_ctx->l3_proto != ENA_ETH_IO_L4_PROTO_UNKNOWN) {
1422 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1423 			return;
1424 		}
1425 
1426 		switch (ena_rx_ctx->l3_proto) {
1427 		case ENA_ETH_IO_L3_PROTO_IPV4:
1428 			switch (ena_rx_ctx->l4_proto) {
1429 			case ENA_ETH_IO_L4_PROTO_TCP:
1430 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV4);
1431 				break;
1432 			case ENA_ETH_IO_L4_PROTO_UDP:
1433 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV4);
1434 				break;
1435 			default:
1436 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV4);
1437 			}
1438 			break;
1439 		case ENA_ETH_IO_L3_PROTO_IPV6:
1440 			switch (ena_rx_ctx->l4_proto) {
1441 			case ENA_ETH_IO_L4_PROTO_TCP:
1442 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV6);
1443 				break;
1444 			case ENA_ETH_IO_L4_PROTO_UDP:
1445 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV6);
1446 				break;
1447 			default:
1448 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV6);
1449 			}
1450 			break;
1451 		case ENA_ETH_IO_L3_PROTO_UNKNOWN:
1452 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1453 			break;
1454 		default:
1455 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1456 		}
1457 	} else {
1458 		mbuf->m_pkthdr.flowid = rx_ring->qid;
1459 		M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1460 	}
1461 }
1462 
1463 /**
1464  * ena_rx_mbuf - assemble mbuf from descriptors
1465  * @rx_ring: ring for which we want to clean packets
1466  * @ena_bufs: buffer info
1467  * @ena_rx_ctx: metadata for this packet(s)
1468  * @next_to_clean: ring pointer, will be updated only upon success
1469  *
1470  **/
1471 static struct mbuf*
1472 ena_rx_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_buf_info *ena_bufs,
1473     struct ena_com_rx_ctx *ena_rx_ctx, uint16_t *next_to_clean)
1474 {
1475 	struct mbuf *mbuf;
1476 	struct ena_rx_buffer *rx_info;
1477 	struct ena_adapter *adapter;
1478 	unsigned int descs = ena_rx_ctx->descs;
1479 	uint16_t ntc, len, req_id, buf = 0;
1480 
1481 	ntc = *next_to_clean;
1482 	adapter = rx_ring->adapter;
1483 	rx_info = &rx_ring->rx_buffer_info[ntc];
1484 
1485 	if (unlikely(rx_info->mbuf == NULL)) {
1486 		device_printf(adapter->pdev, "NULL mbuf in rx_info");
1487 		return (NULL);
1488 	}
1489 
1490 	len = ena_bufs[buf].len;
1491 	req_id = ena_bufs[buf].req_id;
1492 	rx_info = &rx_ring->rx_buffer_info[req_id];
1493 
1494 	ena_trace(ENA_DBG | ENA_RXPTH, "rx_info %p, mbuf %p, paddr %jx",
1495 	    rx_info, rx_info->mbuf, (uintmax_t)rx_info->ena_buf.paddr);
1496 
1497 	mbuf = rx_info->mbuf;
1498 	mbuf->m_flags |= M_PKTHDR;
1499 	mbuf->m_pkthdr.len = len;
1500 	mbuf->m_len = len;
1501 	mbuf->m_pkthdr.rcvif = rx_ring->que->adapter->ifp;
1502 
1503 	/* Fill mbuf with hash key and it's interpretation for optimization */
1504 	ena_rx_hash_mbuf(rx_ring, ena_rx_ctx, mbuf);
1505 
1506 	ena_trace(ENA_DBG | ENA_RXPTH, "rx mbuf 0x%p, flags=0x%x, len: %d",
1507 	    mbuf, mbuf->m_flags, mbuf->m_pkthdr.len);
1508 
1509 	/* DMA address is not needed anymore, unmap it */
1510 	bus_dmamap_unload(rx_ring->adapter->rx_buf_tag, rx_info->map);
1511 
1512 	rx_info->mbuf = NULL;
1513 	rx_ring->free_rx_ids[ntc] = req_id;
1514 	ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1515 
1516 	/*
1517 	 * While we have more than 1 descriptors for one rcvd packet, append
1518 	 * other mbufs to the main one
1519 	 */
1520 	while (--descs) {
1521 		++buf;
1522 		len = ena_bufs[buf].len;
1523 		req_id = ena_bufs[buf].req_id;
1524 		rx_info = &rx_ring->rx_buffer_info[req_id];
1525 
1526 		if (unlikely(rx_info->mbuf == NULL)) {
1527 			device_printf(adapter->pdev, "NULL mbuf in rx_info");
1528 			/*
1529 			 * If one of the required mbufs was not allocated yet,
1530 			 * we can break there.
1531 			 * All earlier used descriptors will be reallocated
1532 			 * later and not used mbufs can be reused.
1533 			 * The next_to_clean pointer will not be updated in case
1534 			 * of an error, so caller should advance it manually
1535 			 * in error handling routine to keep it up to date
1536 			 * with hw ring.
1537 			 */
1538 			m_freem(mbuf);
1539 			return (NULL);
1540 		}
1541 
1542 		if (m_append(mbuf, len, rx_info->mbuf->m_data) == 0) {
1543 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
1544 			ena_trace(ENA_WARNING, "Failed to append Rx mbuf %p",
1545 			    mbuf);
1546 		}
1547 		/* Free already appended mbuf, it won't be useful anymore */
1548 		bus_dmamap_unload(rx_ring->adapter->rx_buf_tag, rx_info->map);
1549 		m_freem(rx_info->mbuf);
1550 		rx_info->mbuf = NULL;
1551 
1552 		rx_ring->free_rx_ids[ntc] = req_id;
1553 		ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1554 	}
1555 
1556 	*next_to_clean = ntc;
1557 
1558 	return (mbuf);
1559 }
1560 
1561 /**
1562  * ena_rx_checksum - indicate in mbuf if hw indicated a good cksum
1563  **/
1564 static inline void
1565 ena_rx_checksum(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1566     struct mbuf *mbuf)
1567 {
1568 
1569 	/* if IP and error */
1570 	if ((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1571 	    (ena_rx_ctx->l3_csum_err)) {
1572 		/* ipv4 checksum error */
1573 		mbuf->m_pkthdr.csum_flags = 0;
1574 		counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1575 		return;
1576 	}
1577 
1578 	/* if TCP/UDP */
1579 	if ((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1580 	    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)) {
1581 		if (ena_rx_ctx->l4_csum_err) {
1582 			/* TCP/UDP checksum error */
1583 			mbuf->m_pkthdr.csum_flags = 0;
1584 			counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1585 		} else {
1586 			mbuf->m_pkthdr.csum_flags = CSUM_IP_CHECKED;
1587 			mbuf->m_pkthdr.csum_flags |= CSUM_IP_VALID;
1588 		}
1589 	}
1590 
1591 	return;
1592 }
1593 
1594 static void
1595 ena_deferred_rx_cleanup(void *arg, int pending)
1596 {
1597 	struct ena_ring *rx_ring = arg;
1598 	int budget = CLEAN_BUDGET;
1599 
1600 	ENA_RING_MTX_LOCK(rx_ring);
1601 	/*
1602 	 * If deferred task was executed, perform cleanup of all awaiting
1603 	 * descs (or until given budget is depleted to avoid infinite loop).
1604 	 */
1605 	while (budget--) {
1606 		if (ena_rx_cleanup(rx_ring) == 0)
1607 			break;
1608 	}
1609 	ENA_RING_MTX_UNLOCK(rx_ring);
1610 }
1611 
1612 /**
1613  * ena_rx_cleanup - handle rx irq
1614  * @arg: ring for which irq is being handled
1615  **/
1616 static int
1617 ena_rx_cleanup(struct ena_ring *rx_ring)
1618 {
1619 	struct ena_adapter *adapter;
1620 	struct mbuf *mbuf;
1621 	struct ena_com_rx_ctx ena_rx_ctx;
1622 	struct ena_com_io_cq* io_cq;
1623 	struct ena_com_io_sq* io_sq;
1624 	/* struct ena_eth_io_intr_reg intr_reg; */
1625 	if_t ifp;
1626 	uint16_t ena_qid;
1627 	uint16_t next_to_clean;
1628 	uint32_t refill_required;
1629 	uint32_t refill_threshold;
1630 	uint32_t do_if_input = 0;
1631 	unsigned int qid;
1632 	int rc, i;
1633 	int budget = RX_BUDGET;
1634 
1635 	adapter = rx_ring->que->adapter;
1636 	ifp = adapter->ifp;
1637 	qid = rx_ring->que->id;
1638 	ena_qid = ENA_IO_RXQ_IDX(qid);
1639 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1640 	io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
1641 	next_to_clean = rx_ring->next_to_clean;
1642 
1643 	do {
1644 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1645 		ena_rx_ctx.max_bufs = adapter->max_rx_sgl_size;
1646 		ena_rx_ctx.descs = 0;
1647 		rc = ena_com_rx_pkt(io_cq, io_sq, &ena_rx_ctx);
1648 
1649 		if (unlikely(rc))
1650 			goto error;
1651 
1652 		if (unlikely(ena_rx_ctx.descs == 0))
1653 			break;
1654 
1655 		/* Receive mbuf from the ring */
1656 		mbuf = ena_rx_mbuf(rx_ring, rx_ring->ena_bufs,
1657 		    &ena_rx_ctx, &next_to_clean);
1658 
1659 		/* Exit if we failed to retrieve a buffer */
1660 		if (unlikely(!mbuf)) {
1661 			for (i = 0; i < ena_rx_ctx.descs; ++i) {
1662 				rx_ring->free_rx_ids[next_to_clean] =
1663 				    rx_ring->ena_bufs[i].req_id;
1664 				next_to_clean =
1665 				    ENA_RX_RING_IDX_NEXT(next_to_clean,
1666 				    rx_ring->ring_size);
1667 
1668 			}
1669 			break;
1670 		}
1671 		ena_trace(ENA_DBG | ENA_RXPTH, "Rx: %d bytes",
1672 		    mbuf->m_pkthdr.len);
1673 
1674 		if ((ifp->if_capenable & IFCAP_RXCSUM) ||
1675 		    (ifp->if_capenable & IFCAP_RXCSUM_IPV6)) {
1676 			ena_rx_checksum(rx_ring, &ena_rx_ctx, mbuf);
1677 		}
1678 
1679 		counter_enter();
1680 		counter_u64_add_protected(rx_ring->rx_stats.bytes,
1681 		    mbuf->m_pkthdr.len);
1682 		counter_u64_add_protected(adapter->hw_stats.rx_bytes,
1683 		    mbuf->m_pkthdr.len);
1684 		counter_exit();
1685 		/*
1686 		 * LRO is only for IP/TCP packets and TCP checksum of the packet
1687 		 * should be computed by hardware.
1688 		 */
1689 		do_if_input = 1;
1690 		if ((ifp->if_capenable & IFCAP_LRO) &&
1691 		    (mbuf->m_pkthdr.csum_flags & CSUM_IP_VALID) &&
1692 		    ena_rx_ctx.l4_proto == ENA_ETH_IO_L4_PROTO_TCP) {
1693 			/*
1694 			 * Send to the stack if:
1695 			 *  - LRO not enabled, or
1696 			 *  - no LRO resources, or
1697 			 *  - lro enqueue fails
1698 			 */
1699 			if (rx_ring->lro.lro_cnt != 0 &&
1700 			    tcp_lro_rx(&rx_ring->lro, mbuf, 0) == 0)
1701 					do_if_input = 0;
1702 		}
1703 		if (do_if_input) {
1704 			ena_trace(ENA_DBG | ENA_RXPTH, "calling if_input() with mbuf %p",
1705 			    mbuf);
1706 			(*ifp->if_input)(ifp, mbuf);
1707 		}
1708 
1709 		counter_enter();
1710 		counter_u64_add_protected(rx_ring->rx_stats.cnt, 1);
1711 		counter_u64_add_protected(adapter->hw_stats.rx_packets, 1);
1712 		counter_exit();
1713 	} while (--budget);
1714 
1715 	rx_ring->next_to_clean = next_to_clean;
1716 
1717 	refill_required = ena_com_free_desc(io_sq);
1718 	refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DEVIDER;
1719 
1720 	if (refill_required > refill_threshold) {
1721 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1722 		ena_refill_rx_bufs(rx_ring, refill_required);
1723 	}
1724 
1725 	tcp_lro_flush_all(&rx_ring->lro);
1726 
1727 	return (RX_BUDGET - budget);
1728 
1729 error:
1730 	counter_u64_add(rx_ring->rx_stats.bad_desc_num, 1);
1731 	return (RX_BUDGET - budget);
1732 }
1733 
1734 /*********************************************************************
1735  *
1736  *  MSIX & Interrupt Service routine
1737  *
1738  **********************************************************************/
1739 
1740 /**
1741  * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1742  * @arg: interrupt number
1743  **/
1744 static void
1745 ena_intr_msix_mgmnt(void *arg)
1746 {
1747 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1748 
1749 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1750 	if (likely(adapter->running))
1751 		ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1752 }
1753 
1754 /**
1755  * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1756  * @arg: interrupt number
1757  **/
1758 static void
1759 ena_handle_msix(void *arg)
1760 {
1761 	struct ena_que	*que = arg;
1762 	struct ena_adapter *adapter = que->adapter;
1763 	if_t ifp = adapter->ifp;
1764 	struct ena_ring *tx_ring;
1765 	struct ena_ring *rx_ring;
1766 	struct ena_com_io_cq* io_cq;
1767 	struct ena_eth_io_intr_reg intr_reg;
1768 	int qid, ena_qid;
1769 	int txc, rxc, i;
1770 
1771 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1772 		return;
1773 
1774 	ena_trace(ENA_DBG, "MSI-X TX/RX routine");
1775 
1776 	tx_ring = que->tx_ring;
1777 	rx_ring = que->rx_ring;
1778 	qid = que->id;
1779 	ena_qid = ENA_IO_TXQ_IDX(qid);
1780 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1781 
1782 	for (i = 0; i < CLEAN_BUDGET; ++i) {
1783 		/*
1784 		 * If lock cannot be acquired, then deferred cleanup task was
1785 		 * being executed and rx ring is being cleaned up in
1786 		 * another thread.
1787 		 */
1788 		if (ENA_RING_MTX_TRYLOCK(rx_ring)) {
1789 			rxc = ena_rx_cleanup(rx_ring);
1790 			ENA_RING_MTX_UNLOCK(rx_ring);
1791 		} else {
1792 			rxc = 0;
1793 		}
1794 
1795 		/* Protection from calling ena_tx_cleanup from ena_start_xmit */
1796 		ENA_RING_MTX_LOCK(tx_ring);
1797 		txc = ena_tx_cleanup(tx_ring);
1798 		ENA_RING_MTX_UNLOCK(tx_ring);
1799 
1800 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1801 			return;
1802 
1803 		if (txc != TX_BUDGET && rxc != RX_BUDGET)
1804 		       break;
1805 	}
1806 
1807 	/* Signal that work is done and unmask interrupt */
1808 	ena_com_update_intr_reg(&intr_reg,
1809 	    RX_IRQ_INTERVAL,
1810 	    TX_IRQ_INTERVAL,
1811 	    true);
1812 	ena_com_unmask_intr(io_cq, &intr_reg);
1813 }
1814 
1815 static int
1816 ena_enable_msix(struct ena_adapter *adapter)
1817 {
1818 	device_t dev = adapter->pdev;
1819 	int i, msix_vecs, rc = 0;
1820 
1821 	/* Reserved the max msix vectors we might need */
1822 	msix_vecs = ENA_MAX_MSIX_VEC(adapter->num_queues);
1823 
1824 	adapter->msix_entries = malloc(msix_vecs * sizeof(struct msix_entry),
1825 	    M_DEVBUF, M_WAITOK | M_ZERO);
1826 
1827 	device_printf(dev, "Allocated msix_entries, vectors (cnt: %d)\n",
1828 	    msix_vecs);
1829 
1830 	for (i = 0; i < msix_vecs; i++) {
1831 		adapter->msix_entries[i].entry = i;
1832 		/* Vectors must start from 1 */
1833 		adapter->msix_entries[i].vector = i + 1;
1834 	}
1835 
1836 	rc = pci_alloc_msix(dev, &msix_vecs);
1837 	if (rc != 0) {
1838 		device_printf(dev,
1839 		    "Failed to enable MSIX, vectors %d rc %d\n", msix_vecs, rc);
1840 
1841 		rc = ENOSPC;
1842 		goto err_msix_free;
1843 	}
1844 
1845 	adapter->msix_vecs = msix_vecs;
1846 	adapter->msix_enabled = true;
1847 
1848 	return (0);
1849 
1850 err_msix_free:
1851 	free(adapter->msix_entries, M_DEVBUF);
1852 	adapter->msix_entries = NULL;
1853 
1854 	return (rc);
1855 }
1856 
1857 static void
1858 ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1859 {
1860 
1861 	snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1862 	    ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1863 	    device_get_nameunit(adapter->pdev));
1864 	/*
1865 	 * Handler is NULL on purpose, it will be set
1866 	 * when mgmnt interrupt is acquired
1867 	 */
1868 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler = NULL;
1869 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1870 	adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1871 	    adapter->msix_entries[ENA_MGMNT_IRQ_IDX].vector;
1872 
1873 	return;
1874 }
1875 
1876 static void
1877 ena_setup_io_intr(struct ena_adapter *adapter)
1878 {
1879 	static int last_bind_cpu = -1;
1880 	int irq_idx;
1881 	ena_trace(ENA_DBG, "enter");
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 (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 	return;
1911 }
1912 
1913 static int
1914 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1915 {
1916 	struct ena_irq *irq;
1917 	unsigned long flags;
1918 	int rc, rcc;
1919 
1920 	flags = RF_ACTIVE | RF_SHAREABLE;
1921 
1922 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1923 	irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1924 	    &irq->vector, flags);
1925 
1926 	if (irq->res == NULL) {
1927 		device_printf(adapter->pdev, "could not allocate "
1928 		    "irq vector: %d\n", irq->vector);
1929 		return (ENXIO);
1930 	}
1931 
1932 	if ((rc = bus_activate_resource(adapter->pdev, SYS_RES_IRQ, irq->vector,
1933 	    irq->res)) != 0) {
1934 		device_printf(adapter->pdev, "could not activate "
1935 		    "irq vector: %d\n", irq->vector);
1936 		goto err_res_free;
1937 	}
1938 
1939 	if ((rc = bus_setup_intr(adapter->pdev, irq->res,
1940 	    INTR_TYPE_NET | INTR_MPSAFE, NULL,
1941 	    ena_intr_msix_mgmnt, irq->data, &irq->cookie)) != 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 	device_printf(adapter->pdev, "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 (rcc)
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 (!adapter->msix_enabled) {
1972 		device_printf(adapter->pdev, "failed to request irq\n");
1973 		return (EINVAL);
1974 	} else {
1975 		flags = RF_ACTIVE | RF_SHAREABLE;
1976 	}
1977 
1978 	for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1979 		irq = &adapter->irq_tbl[i];
1980 
1981 		if (irq->requested)
1982 			continue;
1983 
1984 		irq->res = bus_alloc_resource_any(adapter->pdev, SYS_RES_IRQ,
1985 		    &irq->vector, flags);
1986 		if (irq->res == NULL) {
1987 			device_printf(adapter->pdev, "could not allocate "
1988 			    "irq vector: %d\n", irq->vector);
1989 			goto err;
1990 		}
1991 
1992 		if ((rc = bus_setup_intr(adapter->pdev, irq->res,
1993 			    INTR_TYPE_NET | INTR_MPSAFE, NULL, irq->handler,
1994 			    irq->data, &irq->cookie)) != 0) {
1995 			device_printf(adapter->pdev, "failed to register "
1996 			    "interrupt handler for irq %ju: %d\n",
1997 			    rman_get_start(irq->res), rc);
1998 			goto err;
1999 		}
2000 		irq->requested = true;
2001 
2002 #ifdef	RSS
2003 		device_printf(adapter->pdev, "queue %d - RSS bucket %d\n",
2004 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
2005 #else
2006 		device_printf(adapter->pdev, "queue %d - cpu %d\n",
2007 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
2008 #endif
2009 	}
2010 
2011 	return (rc);
2012 
2013 err:
2014 
2015 	for (; i >= ENA_IO_IRQ_FIRST_IDX; i--) {
2016 		irq = &adapter->irq_tbl[i];
2017 		rcc = 0;
2018 
2019 		/* Once we entered err: section and irq->requested is true we
2020 		   free both intr and resources */
2021 		if (irq->requested == true)
2022 			rcc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
2023 		if (rcc)
2024 			device_printf(adapter->pdev, "could not release"
2025 			    " irq: %d, error: %d\n", irq->vector, rcc);
2026 
2027 		/* If we entred err: section without irq->requested set we know
2028 		   it was bus_alloc_resource_any() that needs cleanup, provided
2029 		   res is not NULL. In case res is NULL no work in needed in
2030 		   this iteration */
2031 		rcc = 0;
2032 		if (irq->res != NULL) {
2033 			rcc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2034 			    irq->vector, irq->res);
2035 		}
2036 		if (rcc)
2037 			device_printf(adapter->pdev, "dev has no parent while "
2038 			    "releasing res for irq: %d\n", irq->vector);
2039 		irq->requested = false;
2040 		irq->res = NULL;
2041 	}
2042 
2043 	return (rc);
2044 }
2045 
2046 static void
2047 ena_free_mgmnt_irq(struct ena_adapter *adapter)
2048 {
2049 	struct ena_irq *irq;
2050 	int rc;
2051 
2052 	irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2053 	if (irq->requested) {
2054 		ena_trace(ENA_INFO | ENA_ADMQ, "tear down irq: %d\n",
2055 		    irq->vector);
2056 		rc = bus_teardown_intr(adapter->pdev, irq->res, irq->cookie);
2057 		if (rc)
2058 			device_printf(adapter->pdev, "failed to tear "
2059 			    "down irq: %d\n", irq->vector);
2060 		irq->requested = 0;
2061 	}
2062 
2063 	if (irq->res != NULL) {
2064 		ena_trace(ENA_INFO | ENA_ADMQ, "release resource irq: %d\n",
2065 		    irq->vector);
2066 		rc = bus_release_resource(adapter->pdev, SYS_RES_IRQ,
2067 		    irq->vector, irq->res);
2068 		irq->res = NULL;
2069 		if (rc)
2070 			device_printf(adapter->pdev, "dev has no parent while "
2071 			    "releasing res for irq: %d\n", irq->vector);
2072 	}
2073 
2074 	return;
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 (rc) {
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 (rc) {
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 	return;
2112 }
2113 
2114 static void
2115 ena_free_irqs(struct ena_adapter* adapter)
2116 {
2117 
2118 	ena_free_io_irq(adapter);
2119 	ena_free_mgmnt_irq(adapter);
2120 	ena_disable_msix(adapter);
2121 }
2122 
2123 static void
2124 ena_disable_msix(struct ena_adapter *adapter)
2125 {
2126 
2127 	pci_release_msi(adapter->pdev);
2128 
2129 	adapter->msix_vecs = 0;
2130 	free(adapter->msix_entries, M_DEVBUF);
2131 	adapter->msix_entries = NULL;
2132 }
2133 
2134 static void
2135 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
2136 {
2137 	struct ena_com_io_cq* io_cq;
2138 	struct ena_eth_io_intr_reg intr_reg;
2139 	uint16_t ena_qid;
2140 	int i;
2141 
2142 	/* Unmask interrupts for all queues */
2143 	for (i = 0; i < adapter->num_queues; i++) {
2144 		ena_qid = ENA_IO_TXQ_IDX(i);
2145 		io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
2146 		ena_com_update_intr_reg(&intr_reg, 0, 0, true);
2147 		ena_com_unmask_intr(io_cq, &intr_reg);
2148 	}
2149 }
2150 
2151 /* Configure the Rx forwarding */
2152 static int ena_rss_configure(struct ena_adapter *adapter)
2153 {
2154 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2155 	int rc;
2156 
2157 	/* Set indirect table */
2158 	rc = ena_com_indirect_table_set(ena_dev);
2159 	if (unlikely(rc && rc != EOPNOTSUPP))
2160 		return rc;
2161 
2162 	/* Configure hash function (if supported) */
2163 	rc = ena_com_set_hash_function(ena_dev);
2164 	if (unlikely(rc && (rc != EOPNOTSUPP)))
2165 		return rc;
2166 
2167 	/* Configure hash inputs (if supported) */
2168 	rc = ena_com_set_hash_ctrl(ena_dev);
2169 	if (unlikely(rc && (rc != EOPNOTSUPP)))
2170 		return rc;
2171 
2172 	return 0;
2173 }
2174 
2175 static int
2176 ena_up_complete(struct ena_adapter *adapter)
2177 {
2178 	int rc;
2179 
2180 	if (adapter->rss_support) {
2181 		rc = ena_rss_configure(adapter);
2182 		if (rc)
2183 			return (rc);
2184 	}
2185 
2186 	rc = ena_change_mtu(adapter->ifp, adapter->ifp->if_mtu);
2187 	if (rc != 0)
2188 		return (rc);
2189 
2190 	ena_refill_all_rx_bufs(adapter);
2191 	ena_reset_counters((counter_u64_t *)&adapter->hw_stats,
2192 	    sizeof(adapter->hw_stats));
2193 
2194 	return (0);
2195 }
2196 
2197 static int
2198 ena_up(struct ena_adapter *adapter)
2199 {
2200 	int rc = 0;
2201 
2202 	if (!device_is_attached(adapter->pdev)) {
2203 		device_printf(adapter->pdev, "device is not attached!\n");
2204 		return (ENXIO);
2205 	}
2206 
2207 	if (!adapter->running) {
2208 		device_printf(adapter->pdev, "device is not running!\n");
2209 		return (ENXIO);
2210 	}
2211 
2212 	if (!adapter->up) {
2213 		device_printf(adapter->pdev, "device is going UP\n");
2214 
2215 		/* setup interrupts for IO queues */
2216 		ena_setup_io_intr(adapter);
2217 		rc = ena_request_io_irq(adapter);
2218 		if (rc) {
2219 			ena_trace(ENA_ALERT, "err_req_irq");
2220 			goto err_req_irq;
2221 		}
2222 
2223 		/* allocate transmit descriptors */
2224 		rc = ena_setup_all_tx_resources(adapter);
2225 		if (rc) {
2226 			ena_trace(ENA_ALERT, "err_setup_tx");
2227 			goto err_setup_tx;
2228 		}
2229 
2230 		/* allocate receive descriptors */
2231 		rc = ena_setup_all_rx_resources(adapter);
2232 		if (rc) {
2233 			ena_trace(ENA_ALERT, "err_setup_rx");
2234 			goto err_setup_rx;
2235 		}
2236 
2237 		/* create IO queues for Rx & Tx */
2238 		rc = ena_create_io_queues(adapter);
2239 		if (rc) {
2240 			ena_trace(ENA_ALERT,
2241 			    "create IO queues failed");
2242 			goto err_io_que;
2243 		}
2244 
2245 		if (adapter->link_status)
2246 			if_link_state_change(adapter->ifp, LINK_STATE_UP);
2247 
2248 		rc = ena_up_complete(adapter);
2249 		if (rc)
2250 			goto err_up_complete;
2251 
2252 		counter_u64_add(adapter->dev_stats.interface_up, 1);
2253 
2254 		ena_update_hwassist(adapter);
2255 
2256 		if_setdrvflagbits(adapter->ifp, IFF_DRV_RUNNING,
2257 		    IFF_DRV_OACTIVE);
2258 
2259 		callout_reset_sbt(&adapter->timer_service, SBT_1S, SBT_1S,
2260 		    ena_timer_service, (void *)adapter, 0);
2261 
2262 		adapter->up = true;
2263 
2264 		ena_unmask_all_io_irqs(adapter);
2265 	}
2266 
2267 	return (0);
2268 
2269 err_up_complete:
2270 	ena_destroy_all_io_queues(adapter);
2271 err_io_que:
2272 	ena_free_all_rx_resources(adapter);
2273 err_setup_rx:
2274 	ena_free_all_tx_resources(adapter);
2275 err_setup_tx:
2276 	ena_free_io_irq(adapter);
2277 err_req_irq:
2278 	return (rc);
2279 }
2280 
2281 static uint64_t
2282 ena_get_counter(if_t ifp, ift_counter cnt)
2283 {
2284 	struct ena_adapter *adapter;
2285 	struct ena_hw_stats *stats;
2286 
2287 	adapter = if_getsoftc(ifp);
2288 	stats = &adapter->hw_stats;
2289 
2290 	switch (cnt) {
2291 	case IFCOUNTER_IPACKETS:
2292 		return (counter_u64_fetch(stats->rx_packets));
2293 	case IFCOUNTER_OPACKETS:
2294 		return (counter_u64_fetch(stats->tx_packets));
2295 	case IFCOUNTER_IBYTES:
2296 		return (counter_u64_fetch(stats->rx_bytes));
2297 	case IFCOUNTER_OBYTES:
2298 		return (counter_u64_fetch(stats->tx_bytes));
2299 	case IFCOUNTER_IQDROPS:
2300 		return (counter_u64_fetch(stats->rx_drops));
2301 	default:
2302 		return (if_get_counter_default(ifp, cnt));
2303 	}
2304 }
2305 
2306 static int
2307 ena_media_change(if_t ifp)
2308 {
2309 	/* Media Change is not supported by firmware */
2310 	return (0);
2311 }
2312 
2313 static void
2314 ena_media_status(if_t ifp, struct ifmediareq *ifmr)
2315 {
2316 	struct ena_adapter *adapter = if_getsoftc(ifp);
2317 	ena_trace(ENA_DBG, "enter");
2318 
2319 	ENA_DEV_LOCK;
2320 
2321 	ifmr->ifm_status = IFM_AVALID;
2322 	ifmr->ifm_active = IFM_ETHER;
2323 
2324 	if (!adapter->link_status) {
2325 		ENA_DEV_UNLOCK;
2326 		ena_trace(ENA_WARNING, "link_status = false");
2327 		return;
2328 	}
2329 
2330 	ifmr->ifm_status |= IFM_ACTIVE;
2331 	ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
2332 
2333 	ENA_DEV_UNLOCK;
2334 
2335 	return;
2336 }
2337 
2338 static void
2339 ena_init(void *arg)
2340 {
2341 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
2342 
2343 	if (adapter->up == false) {
2344 		sx_xlock(&adapter->ioctl_sx);
2345 		ena_up(adapter);
2346 		sx_unlock(&adapter->ioctl_sx);
2347 	}
2348 
2349 	return;
2350 }
2351 
2352 static int
2353 ena_ioctl(if_t ifp, u_long command, caddr_t data)
2354 {
2355 	struct ena_adapter *adapter;
2356 	struct ifreq *ifr;
2357 	int rc;
2358 
2359 	adapter = ifp->if_softc;
2360 	ifr = (struct ifreq *)data;
2361 
2362 	/*
2363 	 * Acquiring lock to prevent from running up and down routines parallel.
2364 	 */
2365 	rc = 0;
2366 	switch (command) {
2367 	case SIOCSIFMTU:
2368 		sx_xlock(&adapter->ioctl_sx);
2369 		ena_down(adapter);
2370 
2371 		ena_change_mtu(ifp, ifr->ifr_mtu);
2372 
2373 		rc = ena_up(adapter);
2374 		sx_unlock(&adapter->ioctl_sx);
2375 		break;
2376 
2377 	case SIOCSIFFLAGS:
2378 		if (ifp->if_flags & IFF_UP) {
2379 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2380 				if (ifp->if_flags & (IFF_PROMISC |
2381 				    IFF_ALLMULTI)) {
2382 					device_printf(adapter->pdev,
2383 					    "ioctl promisc/allmulti\n");
2384 				}
2385 			} else {
2386 				sx_xlock(&adapter->ioctl_sx);
2387 				rc = ena_up(adapter);
2388 				sx_unlock(&adapter->ioctl_sx);
2389 			}
2390 		} else {
2391 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2392 				sx_xlock(&adapter->ioctl_sx);
2393 				ena_down(adapter);
2394 				sx_unlock(&adapter->ioctl_sx);
2395 			}
2396 		}
2397 		break;
2398 
2399 	case SIOCADDMULTI:
2400 	case SIOCDELMULTI:
2401 		break;
2402 
2403 	case SIOCSIFMEDIA:
2404 	case SIOCGIFMEDIA:
2405 		rc = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
2406 		break;
2407 
2408 	case SIOCSIFCAP:
2409 		{
2410 			int reinit = 0;
2411 
2412 			if (ifr->ifr_reqcap != ifp->if_capenable) {
2413 				ifp->if_capenable = ifr->ifr_reqcap;
2414 				reinit = 1;
2415 			}
2416 
2417 			if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
2418 				sx_xlock(&adapter->ioctl_sx);
2419 				ena_down(adapter);
2420 				rc = ena_up(adapter);
2421 				sx_unlock(&adapter->ioctl_sx);
2422 			}
2423 		}
2424 
2425 		break;
2426 	default:
2427 		rc = ether_ioctl(ifp, command, data);
2428 		break;
2429 	}
2430 
2431 	return (rc);
2432 }
2433 
2434 static int
2435 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2436 {
2437 	int caps = 0;
2438 
2439 	if (feat->offload.tx &
2440 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2441 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2442 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK))
2443 		caps |= IFCAP_TXCSUM;
2444 
2445 	if (feat->offload.tx &
2446 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2447 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK))
2448 		caps |= IFCAP_TXCSUM_IPV6;
2449 
2450 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
2451 		caps |= IFCAP_TSO4;
2452 
2453 	if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
2454 		caps |= IFCAP_TSO6;
2455 
2456 	if (feat->offload.rx_supported &
2457 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2458 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK))
2459 		caps |= IFCAP_RXCSUM;
2460 
2461 	if (feat->offload.rx_supported &
2462 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
2463 		caps |= IFCAP_RXCSUM_IPV6;
2464 
2465 	caps |= IFCAP_LRO | IFCAP_JUMBO_MTU;
2466 
2467 	return (caps);
2468 }
2469 
2470 static void
2471 ena_update_host_info(struct ena_admin_host_info *host_info, if_t ifp)
2472 {
2473 
2474 	host_info->supported_network_features[0] =
2475 	    (uint32_t)if_getcapabilities(ifp);
2476 }
2477 
2478 static void
2479 ena_update_hwassist(struct ena_adapter *adapter)
2480 {
2481 	if_t ifp = adapter->ifp;
2482 	uint32_t feat = adapter->tx_offload_cap;
2483 	int cap = if_getcapenable(ifp);
2484 	int flags = 0;
2485 
2486 	if_clearhwassist(ifp);
2487 
2488 	if (cap & IFCAP_TXCSUM) {
2489 		if (feat & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)
2490 			flags |= CSUM_IP;
2491 		if (feat &
2492 		    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2493 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK))
2494 			flags |= CSUM_IP_UDP | CSUM_IP_TCP;
2495 	}
2496 
2497 	if (cap & IFCAP_TXCSUM_IPV6)
2498 		flags |= CSUM_IP6_UDP | CSUM_IP6_TCP;
2499 
2500 	if (cap & IFCAP_TSO4)
2501 		flags |= CSUM_IP_TSO;
2502 
2503 	if (cap & IFCAP_TSO6)
2504 		flags |= CSUM_IP6_TSO;
2505 
2506 	if_sethwassistbits(ifp, flags, 0);
2507 }
2508 
2509 static int
2510 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2511     struct ena_com_dev_get_features_ctx *feat)
2512 {
2513 	if_t ifp;
2514 	int caps = 0;
2515 
2516 	ena_trace(ENA_DBG, "enter");
2517 
2518 	ifp = adapter->ifp = if_gethandle(IFT_ETHER);
2519 	if (ifp == 0) {
2520 		device_printf(pdev, "can not allocate ifnet structure\n");
2521 		return (ENXIO);
2522 	}
2523 	if_initname(ifp, device_get_name(pdev), device_get_unit(pdev));
2524 	if_setdev(ifp, pdev);
2525 	if_setsoftc(ifp, adapter);
2526 
2527 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
2528 	if_setinitfn(ifp, ena_init);
2529 	if_settransmitfn(ifp, ena_mq_start);
2530 	if_setqflushfn(ifp, ena_qflush);
2531 	if_setioctlfn(ifp, ena_ioctl);
2532 	if_setgetcounterfn(ifp, ena_get_counter);
2533 
2534 	if_setsendqlen(ifp, adapter->tx_ring_size);
2535 	if_setsendqready(ifp);
2536 	if_setmtu(ifp, ETHERMTU);
2537 	if_setbaudrate(ifp, 0);
2538 	/* Zeroize capabilities... */
2539 	if_setcapabilities(ifp, 0);
2540 	if_setcapenable(ifp, 0);
2541 	/* check hardware support */
2542 	caps = ena_get_dev_offloads(feat);
2543 	/* ... and set them */
2544 	if_setcapabilitiesbit(ifp, caps, 0);
2545 
2546 	/* TSO parameters */
2547 	ifp->if_hw_tsomax = ENA_TSO_MAXSIZE -
2548 	    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2549 	ifp->if_hw_tsomaxsegcount = adapter->max_tx_sgl_size - 1;
2550 	ifp->if_hw_tsomaxsegsize = ENA_TSO_MAXSIZE;
2551 
2552 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
2553 	if_setcapenable(ifp, if_getcapabilities(ifp));
2554 
2555 	/*
2556 	 * Specify the media types supported by this adapter and register
2557 	 * callbacks to update media and link information
2558 	 */
2559 	ifmedia_init(&adapter->media, IFM_IMASK,
2560 	    ena_media_change, ena_media_status);
2561 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2562 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2563 
2564 	ether_ifattach(ifp, adapter->mac_addr);
2565 
2566 	return (0);
2567 }
2568 
2569 static void
2570 ena_down(struct ena_adapter *adapter)
2571 {
2572 	int rc;
2573 
2574 	if (adapter->up) {
2575 		device_printf(adapter->pdev, "device is going DOWN\n");
2576 
2577 		callout_drain(&adapter->timer_service);
2578 
2579 		adapter->up = false;
2580 		if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE,
2581 		    IFF_DRV_RUNNING);
2582 
2583 		ena_free_io_irq(adapter);
2584 
2585 		if (adapter->trigger_reset) {
2586 			rc = ena_com_dev_reset(adapter->ena_dev,
2587 			    adapter->reset_reason);
2588 			if (rc)
2589 				device_printf(adapter->pdev,
2590 				    "Device reset failed\n");
2591 		}
2592 
2593 		ena_destroy_all_io_queues(adapter);
2594 
2595 		ena_free_all_tx_bufs(adapter);
2596 		ena_free_all_rx_bufs(adapter);
2597 		ena_free_all_tx_resources(adapter);
2598 		ena_free_all_rx_resources(adapter);
2599 
2600 		counter_u64_add(adapter->dev_stats.interface_down, 1);
2601 	}
2602 
2603 	return;
2604 }
2605 
2606 static void
2607 ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct mbuf *mbuf)
2608 {
2609 	struct ena_com_tx_meta *ena_meta;
2610 	struct ether_vlan_header *eh;
2611 	u32 mss;
2612 	bool offload;
2613 	uint16_t etype;
2614 	int ehdrlen;
2615 	struct ip *ip;
2616 	int iphlen;
2617 	struct tcphdr *th;
2618 
2619 	offload = false;
2620 	ena_meta = &ena_tx_ctx->ena_meta;
2621 	mss = mbuf->m_pkthdr.tso_segsz;
2622 
2623 	if (mss != 0)
2624 		offload = true;
2625 
2626 	if ((mbuf->m_pkthdr.csum_flags & CSUM_TSO) != 0)
2627 		offload = true;
2628 
2629 	if ((mbuf->m_pkthdr.csum_flags & CSUM_OFFLOAD) != 0)
2630 		offload = true;
2631 
2632 	if (offload == false) {
2633 		ena_tx_ctx->meta_valid = 0;
2634 		return;
2635 	}
2636 
2637 	/* Determine where frame payload starts. */
2638 	eh = mtod(mbuf, struct ether_vlan_header *);
2639 	if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2640 		etype = ntohs(eh->evl_proto);
2641 		ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
2642 	} else {
2643 		etype = ntohs(eh->evl_encap_proto);
2644 		ehdrlen = ETHER_HDR_LEN;
2645 	}
2646 
2647 	ip = (struct ip *)(mbuf->m_data + ehdrlen);
2648 	iphlen = ip->ip_hl << 2;
2649 	th = (struct tcphdr *)((caddr_t)ip + iphlen);
2650 
2651 	if ((mbuf->m_pkthdr.csum_flags & CSUM_IP) != 0) {
2652 		ena_tx_ctx->l3_csum_enable = 1;
2653 	}
2654 	if ((mbuf->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
2655 		ena_tx_ctx->tso_enable = 1;
2656 		ena_meta->l4_hdr_len = (th->th_off);
2657 	}
2658 
2659 	switch (etype) {
2660 	case ETHERTYPE_IP:
2661 		ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2662 		if (ip->ip_off == 0)
2663 			ena_tx_ctx->df = 1;
2664 		break;
2665 	case ETHERTYPE_IPV6:
2666 		ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2667 
2668 	default:
2669 		break;
2670 	}
2671 
2672 	if (ip->ip_p == IPPROTO_TCP) {
2673 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2674 		if (mbuf->m_pkthdr.csum_flags & (CSUM_IP_TCP | CSUM_IP6_TCP))
2675 		    ena_tx_ctx->l4_csum_enable = 1;
2676 		else
2677 		    ena_tx_ctx->l4_csum_enable = 0;
2678 	} else if (ip->ip_p == IPPROTO_UDP) {
2679 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2680 		if (mbuf->m_pkthdr.csum_flags & (CSUM_IP_UDP | CSUM_IP6_UDP))
2681 		    ena_tx_ctx->l4_csum_enable = 1;
2682 		else
2683 		    ena_tx_ctx->l4_csum_enable = 0;
2684 	} else {
2685 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
2686 		ena_tx_ctx->l4_csum_enable = 0;
2687 	}
2688 
2689 	ena_meta->mss = mss;
2690 	ena_meta->l3_hdr_len = iphlen;
2691 	ena_meta->l3_hdr_offset = ehdrlen;
2692 	ena_tx_ctx->meta_valid = 1;
2693 }
2694 
2695 static int
2696 ena_check_and_collapse_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2697 {
2698 	struct ena_adapter *adapter;
2699 	struct mbuf *collapsed_mbuf;
2700 	int num_frags;
2701 
2702 	adapter = tx_ring->adapter;
2703 	num_frags = ena_mbuf_count(*mbuf);
2704 
2705 	/* One segment must be reserved for configuration descriptor. */
2706 	if (num_frags < adapter->max_tx_sgl_size)
2707 		return (0);
2708 	counter_u64_add(tx_ring->tx_stats.collapse, 1);
2709 
2710 	collapsed_mbuf = m_collapse(*mbuf, M_NOWAIT,
2711 	    adapter->max_tx_sgl_size - 1);
2712 	if (collapsed_mbuf == NULL) {
2713 		counter_u64_add(tx_ring->tx_stats.collapse_err, 1);
2714 		return (ENOMEM);
2715 	}
2716 
2717 	/* If mbuf was collapsed succesfully, original mbuf is released. */
2718 	*mbuf = collapsed_mbuf;
2719 
2720 	return (0);
2721 }
2722 
2723 static int
2724 ena_xmit_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2725 {
2726 	struct ena_adapter *adapter;
2727 	struct ena_tx_buffer *tx_info;
2728 	struct ena_com_tx_ctx ena_tx_ctx;
2729 	struct ena_com_dev *ena_dev;
2730 	struct ena_com_buf *ena_buf;
2731 	struct ena_com_io_sq* io_sq;
2732 	bus_dma_segment_t segs[ENA_BUS_DMA_SEGS];
2733 	void *push_hdr;
2734 	uint16_t next_to_use;
2735 	uint16_t req_id;
2736 	uint16_t push_len;
2737 	uint16_t ena_qid;
2738 	uint32_t len, nsegs, header_len;
2739 	int i, rc;
2740 	int nb_hw_desc;
2741 
2742 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2743 	adapter = tx_ring->que->adapter;
2744 	ena_dev = adapter->ena_dev;
2745 	io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
2746 
2747 	ENA_ASSERT(*mbuf, "mbuf is NULL\n");
2748 
2749 	rc = ena_check_and_collapse_mbuf(tx_ring, mbuf);
2750 	if (rc) {
2751 		ena_trace(ENA_WARNING,
2752 		    "Failed to collapse mbuf! err: %d", rc);
2753 		return (rc);
2754 	}
2755 
2756 	next_to_use = tx_ring->next_to_use;
2757 	req_id = tx_ring->free_tx_ids[next_to_use];
2758 	tx_info = &tx_ring->tx_buffer_info[req_id];
2759 
2760 	tx_info->mbuf = *mbuf;
2761 	tx_info->num_of_bufs = 0;
2762 
2763 	ena_buf = tx_info->bufs;
2764 	len = (*mbuf)->m_len;
2765 
2766 	ena_trace(ENA_DBG | ENA_TXPTH, "Tx: %d bytes", (*mbuf)->m_pkthdr.len);
2767 
2768 	push_len = 0;
2769 	header_len = min_t(uint32_t, len, tx_ring->tx_max_header_size);
2770 	push_hdr = NULL;
2771 
2772 	rc = bus_dmamap_load_mbuf_sg(adapter->tx_buf_tag, tx_info->map,
2773 	    *mbuf, segs, &nsegs, BUS_DMA_NOWAIT);
2774 
2775 	if (rc || (nsegs == 0)) {
2776 		ena_trace(ENA_WARNING,
2777 		    "dmamap load failed! err: %d nsegs: %d", rc, nsegs);
2778 		counter_u64_add(tx_ring->tx_stats.dma_mapping_err, 1);
2779 		tx_info->mbuf = NULL;
2780 		if (rc == ENOMEM)
2781 			return (ENA_COM_NO_MEM);
2782 		else
2783 			return (ENA_COM_INVAL);
2784 	}
2785 
2786 	for (i = 0; i < nsegs; i++) {
2787 		ena_buf->len = segs[i].ds_len;
2788 		ena_buf->paddr = segs[i].ds_addr;
2789 		ena_buf++;
2790 	}
2791 	tx_info->num_of_bufs = nsegs;
2792 
2793 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2794 	ena_tx_ctx.ena_bufs = tx_info->bufs;
2795 	ena_tx_ctx.push_header = push_hdr;
2796 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2797 	ena_tx_ctx.req_id = req_id;
2798 	ena_tx_ctx.header_len = header_len;
2799 
2800 	/* Set flags and meta data */
2801 	ena_tx_csum(&ena_tx_ctx, *mbuf);
2802 	/* Prepare the packet's descriptors and send them to device */
2803 	rc = ena_com_prepare_tx(io_sq, &ena_tx_ctx, &nb_hw_desc);
2804 	if (rc != 0) {
2805 		ena_trace(ENA_WARNING, "failed to prepare tx bufs\n");
2806 		counter_enter();
2807 		counter_u64_add_protected(tx_ring->tx_stats.queue_stop, 1);
2808 		counter_u64_add_protected(tx_ring->tx_stats.prepare_ctx_err, 1);
2809 		counter_exit();
2810 		goto dma_error;
2811 	}
2812 
2813 	counter_enter();
2814 	counter_u64_add_protected(tx_ring->tx_stats.cnt, 1);
2815 	counter_u64_add_protected(tx_ring->tx_stats.bytes,  (*mbuf)->m_pkthdr.len);
2816 
2817 	counter_u64_add_protected(adapter->hw_stats.tx_packets, 1);
2818 	counter_u64_add_protected(adapter->hw_stats.tx_bytes,
2819 	    (*mbuf)->m_pkthdr.len);
2820 	counter_exit();
2821 
2822 	tx_info->tx_descs = nb_hw_desc;
2823 	getbinuptime(&tx_info->timestamp);
2824 	tx_info->print_once = true;
2825 
2826 	tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2827 	    tx_ring->ring_size);
2828 
2829 	bus_dmamap_sync(adapter->tx_buf_tag, tx_info->map, BUS_DMASYNC_PREWRITE);
2830 
2831 	return (0);
2832 
2833 dma_error:
2834 	tx_info->mbuf = NULL;
2835 	bus_dmamap_unload(adapter->tx_buf_tag, tx_info->map);
2836 
2837 	return (rc);
2838 }
2839 
2840 static void
2841 ena_start_xmit(struct ena_ring *tx_ring)
2842 {
2843 	struct mbuf *mbuf;
2844 	struct ena_adapter *adapter = tx_ring->adapter;
2845 	struct ena_com_io_sq* io_sq;
2846 	int ena_qid;
2847 	int acum_pkts = 0;
2848 	int ret = 0;
2849 
2850 	if ((adapter->ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2851 		return;
2852 
2853 	if (!adapter->link_status)
2854 		return;
2855 
2856 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2857 	io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
2858 
2859 	while ((mbuf = drbr_peek(adapter->ifp, tx_ring->br)) != NULL) {
2860 		ena_trace(ENA_DBG | ENA_TXPTH, "\ndequeued mbuf %p with flags %#x and"
2861 		    " header csum flags %#jx",
2862 		    mbuf, mbuf->m_flags, mbuf->m_pkthdr.csum_flags);
2863 
2864 		if (!ena_com_sq_have_enough_space(io_sq,
2865 		    ENA_TX_CLEANUP_THRESHOLD))
2866 			ena_tx_cleanup(tx_ring);
2867 
2868 		if ((ret = ena_xmit_mbuf(tx_ring, &mbuf)) != 0) {
2869 			if (ret == ENA_COM_NO_MEM) {
2870 				drbr_putback(adapter->ifp, tx_ring->br, mbuf);
2871 			} else if (ret == ENA_COM_NO_SPACE) {
2872 				drbr_putback(adapter->ifp, tx_ring->br, mbuf);
2873 			} else {
2874 				m_freem(mbuf);
2875 				drbr_advance(adapter->ifp, tx_ring->br);
2876 			}
2877 
2878 			break;
2879 		}
2880 
2881 		drbr_advance(adapter->ifp, tx_ring->br);
2882 
2883 		if ((adapter->ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2884 			return;
2885 
2886 		acum_pkts++;
2887 
2888 		BPF_MTAP(adapter->ifp, mbuf);
2889 
2890 		if (acum_pkts == DB_THRESHOLD) {
2891 			acum_pkts = 0;
2892 			wmb();
2893 			/* Trigger the dma engine */
2894 			ena_com_write_sq_doorbell(io_sq);
2895 			counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2896 		}
2897 
2898 	}
2899 
2900 	if (acum_pkts) {
2901 		wmb();
2902 		/* Trigger the dma engine */
2903 		ena_com_write_sq_doorbell(io_sq);
2904 		counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2905 	}
2906 
2907 	if (!ena_com_sq_have_enough_space(io_sq,
2908 	    ENA_TX_CLEANUP_THRESHOLD))
2909 		ena_tx_cleanup(tx_ring);
2910 }
2911 
2912 static void
2913 ena_deferred_mq_start(void *arg, int pending)
2914 {
2915 	struct ena_ring *tx_ring = (struct ena_ring *)arg;
2916 	struct ifnet *ifp = tx_ring->adapter->ifp;
2917 
2918 	while (drbr_empty(ifp, tx_ring->br) == FALSE &&
2919 	    (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2920 		ENA_RING_MTX_LOCK(tx_ring);
2921 		ena_start_xmit(tx_ring);
2922 		ENA_RING_MTX_UNLOCK(tx_ring);
2923 	}
2924 }
2925 
2926 static int
2927 ena_mq_start(if_t ifp, struct mbuf *m)
2928 {
2929 	struct ena_adapter *adapter = ifp->if_softc;
2930 	struct ena_ring *tx_ring;
2931 	int ret, is_drbr_empty;
2932 	uint32_t i;
2933 
2934 	if ((adapter->ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
2935 		return (ENODEV);
2936 
2937 	/* Which queue to use */
2938 	/*
2939 	 * If everything is setup correctly, it should be the
2940 	 * same bucket that the current CPU we're on is.
2941 	 * It should improve performance.
2942 	 */
2943 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) {
2944 #ifdef	RSS
2945 		if (rss_hash2bucket(m->m_pkthdr.flowid,
2946 		    M_HASHTYPE_GET(m), &i) == 0) {
2947 			i = i % adapter->num_queues;
2948 
2949 		} else
2950 #endif
2951 		{
2952 			i = m->m_pkthdr.flowid % adapter->num_queues;
2953 		}
2954 	} else {
2955 		i = curcpu % adapter->num_queues;
2956 	}
2957 	tx_ring = &adapter->tx_ring[i];
2958 
2959 	/* Check if drbr is empty before putting packet */
2960 	is_drbr_empty = drbr_empty(ifp, tx_ring->br);
2961 	ret = drbr_enqueue(ifp, tx_ring->br, m);
2962 	if (ret) {
2963 		taskqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
2964 		return (ret);
2965 	}
2966 
2967 	if (is_drbr_empty && ENA_RING_MTX_TRYLOCK(tx_ring)) {
2968 		ena_start_xmit(tx_ring);
2969 		ENA_RING_MTX_UNLOCK(tx_ring);
2970 	} else {
2971 		taskqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
2972 	}
2973 
2974 	return (0);
2975 }
2976 
2977 static void
2978 ena_qflush(if_t ifp)
2979 {
2980 	struct ena_adapter *adapter = ifp->if_softc;
2981 	struct ena_ring *tx_ring = adapter->tx_ring;
2982 	int i;
2983 
2984 	for(i = 0; i < adapter->num_queues; ++i, ++tx_ring)
2985 		if (drbr_empty(ifp, tx_ring->br) == FALSE) {
2986 			ENA_RING_MTX_LOCK(tx_ring);
2987 			drbr_flush(ifp, tx_ring->br);
2988 			ENA_RING_MTX_UNLOCK(tx_ring);
2989 		}
2990 
2991 	if_qflush(ifp);
2992 
2993 	return;
2994 }
2995 
2996 static int ena_calc_io_queue_num(struct ena_adapter *adapter,
2997     struct ena_com_dev_get_features_ctx *get_feat_ctx)
2998 {
2999 	int io_sq_num, io_cq_num, io_queue_num;
3000 
3001 	io_sq_num = get_feat_ctx->max_queues.max_sq_num;
3002 	io_cq_num = get_feat_ctx->max_queues.max_sq_num;
3003 
3004 	io_queue_num = min_t(int, mp_ncpus, ENA_MAX_NUM_IO_QUEUES);
3005 	io_queue_num = min_t(int, io_queue_num, io_sq_num);
3006 	io_queue_num = min_t(int, io_queue_num, io_cq_num);
3007 	/* 1 IRQ for for mgmnt and 1 IRQ for each TX/RX pair */
3008 	io_queue_num = min_t(int, io_queue_num,
3009 	    pci_msix_count(adapter->pdev) - 1);
3010 #ifdef	RSS
3011 	io_queue_num = min_t(int, io_queue_num, rss_getnumbuckets());
3012 #endif
3013 
3014 	return io_queue_num;
3015 }
3016 
3017 static int ena_calc_queue_size(struct ena_adapter *adapter,
3018     uint16_t *max_tx_sgl_size,  uint16_t *max_rx_sgl_size,
3019     struct ena_com_dev_get_features_ctx *feat)
3020 {
3021 	uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
3022 	uint32_t v;
3023 	uint32_t q;
3024 
3025 	queue_size = min_t(uint32_t, queue_size,
3026 	    feat->max_queues.max_cq_depth);
3027 	queue_size = min_t(uint32_t, queue_size,
3028 	    feat->max_queues.max_sq_depth);
3029 
3030 	/* round down to the nearest power of 2 */
3031 	v = queue_size;
3032 	while (v != 0) {
3033 		if (powerof2(queue_size))
3034 			break;
3035 		v /= 2;
3036 		q = rounddown2(queue_size, v);
3037 		if (q != 0) {
3038 			queue_size = q;
3039 			break;
3040 		}
3041 	}
3042 
3043 	if (unlikely(!queue_size)) {
3044 		device_printf(adapter->pdev, "Invalid queue size\n");
3045 		return ENA_COM_FAULT;
3046 	}
3047 
3048 	*max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3049 	    feat->max_queues.max_packet_tx_descs);
3050 	*max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3051 	    feat->max_queues.max_packet_rx_descs);
3052 
3053 	return queue_size;
3054 }
3055 
3056 static int ena_rss_init_default(struct ena_adapter *adapter)
3057 {
3058 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3059 	device_t dev = adapter->pdev;
3060 	int qid, rc, i;
3061 
3062 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3063 	if (unlikely(rc)) {
3064 		device_printf(dev, "Cannot init RSS\n");
3065 		return (rc);
3066 	}
3067 
3068 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3069 #ifdef	RSS
3070 		qid = rss_get_indirection_to_bucket(i);
3071 		qid = qid % adapter->num_queues;
3072 #else
3073 		qid = i % adapter->num_queues;
3074 #endif
3075 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3076 						       ENA_IO_RXQ_IDX(qid));
3077 		if (unlikely(rc && (rc != EOPNOTSUPP))) {
3078 			device_printf(dev, "Cannot fill indirect table\n");
3079 			goto err_rss_destroy;
3080 		}
3081 	}
3082 
3083 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3084 					ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3085 	if (unlikely(rc && (rc != EOPNOTSUPP))) {
3086 		device_printf(dev, "Cannot fill hash function\n");
3087 		goto err_rss_destroy;
3088 	}
3089 
3090 	rc = ena_com_set_default_hash_ctrl(ena_dev);
3091 	if (unlikely(rc && (rc != EOPNOTSUPP))) {
3092 		device_printf(dev, "Cannot fill hash control\n");
3093 		goto err_rss_destroy;
3094 	}
3095 
3096 	return (0);
3097 
3098 err_rss_destroy:
3099 	ena_com_rss_destroy(ena_dev);
3100 	return (rc);
3101 }
3102 
3103 static void
3104 ena_rss_init_default_deferred(void *arg)
3105 {
3106 	struct ena_adapter *adapter;
3107 	devclass_t dc;
3108 	int max;
3109 	int rc;
3110 
3111 	dc = devclass_find("ena");
3112 	if (dc == NULL) {
3113 		ena_trace(ENA_DBG, "No devclass ena\n");
3114 		return;
3115 	}
3116 
3117 	max = devclass_get_maxunit(dc);
3118 	while (max-- >= 0) {
3119 		adapter = devclass_get_softc(dc, max);
3120 		if (adapter != NULL) {
3121 			rc = ena_rss_init_default(adapter);
3122 			adapter->rss_support = true;
3123 			if (rc) {
3124 				device_printf(adapter->pdev,
3125 				    "WARNING: RSS was not properly initialized,"
3126 				    " it will affect bandwith\n");
3127 				adapter->rss_support = false;
3128 			}
3129 		}
3130 	}
3131 }
3132 SYSINIT(ena_rss_init, SI_SUB_KICK_SCHEDULER, SI_ORDER_SECOND, ena_rss_init_default_deferred, NULL);
3133 
3134 static void ena_config_host_info(struct ena_com_dev *ena_dev)
3135 {
3136 	struct ena_admin_host_info *host_info;
3137 	int rc;
3138 
3139 	/* Allocate only the host info */
3140 	rc = ena_com_allocate_host_info(ena_dev);
3141 	if (rc) {
3142 		ena_trace(ENA_ALERT, "Cannot allocate host info\n");
3143 		return;
3144 	}
3145 
3146 	host_info = ena_dev->host_attr.host_info;
3147 
3148 	host_info->os_type = ENA_ADMIN_OS_FREEBSD;
3149 	host_info->kernel_ver = osreldate;
3150 
3151 	sprintf(host_info->kernel_ver_str, "%d", osreldate);
3152 	host_info->os_dist = 0;
3153 	strncpy(host_info->os_dist_str, osrelease,
3154 	    sizeof(host_info->os_dist_str) - 1);
3155 
3156 	host_info->driver_version =
3157 		(DRV_MODULE_VER_MAJOR) |
3158 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3159 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
3160 
3161 	rc = ena_com_set_host_attributes(ena_dev);
3162 	if (rc) {
3163 		if (rc == EOPNOTSUPP)
3164 			ena_trace(ENA_WARNING, "Cannot set host attributes\n");
3165 		else
3166 			ena_trace(ENA_ALERT, "Cannot set host attributes\n");
3167 
3168 		goto err;
3169 	}
3170 
3171 	return;
3172 
3173 err:
3174 	ena_com_delete_host_info(ena_dev);
3175 }
3176 
3177 static int
3178 ena_device_init(struct ena_adapter *adapter, device_t pdev,
3179 	struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
3180 {
3181 	struct ena_com_dev* ena_dev = adapter->ena_dev;
3182 	bool readless_supported;
3183 	uint32_t aenq_groups;
3184 	int dma_width;
3185 	int rc;
3186 
3187 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
3188 	if (rc) {
3189 		device_printf(pdev, "failed to init mmio read less\n");
3190 		return rc;
3191 	}
3192 
3193 	/*
3194 	 * The PCIe configuration space revision id indicate if mmio reg
3195 	 * read is disabled
3196 	 */
3197 	readless_supported = !(pci_get_revid(pdev) & ENA_MMIO_DISABLE_REG_READ);
3198 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3199 
3200 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3201 	if (rc) {
3202 		device_printf(pdev, "Can not reset device\n");
3203 		goto err_mmio_read_less;
3204 	}
3205 
3206 	rc = ena_com_validate_version(ena_dev);
3207 	if (rc) {
3208 		device_printf(pdev, "device version is too low\n");
3209 		goto err_mmio_read_less;
3210 	}
3211 
3212 	dma_width = ena_com_get_dma_width(ena_dev);
3213 	if (dma_width < 0) {
3214 		device_printf(pdev, "Invalid dma width value %d", dma_width);
3215 		rc = dma_width;
3216 		goto err_mmio_read_less;
3217 	}
3218 	adapter->dma_width = dma_width;
3219 
3220 	/* ENA admin level init */
3221 	rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
3222 	if (rc) {
3223 		device_printf(pdev,
3224 		    "Can not initialize ena admin queue with device\n");
3225 		goto err_mmio_read_less;
3226 	}
3227 
3228 	/*
3229 	 * To enable the msix interrupts the driver needs to know the number
3230 	 * of queues. So the driver uses polling mode to retrieve this
3231 	 * information
3232 	 */
3233 	ena_com_set_admin_polling_mode(ena_dev, true);
3234 
3235 	ena_config_host_info(ena_dev);
3236 
3237 	/* Get Device Attributes */
3238 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3239 	if (rc) {
3240 		device_printf(pdev,
3241 		    "Cannot get attribute for ena device rc: %d\n", rc);
3242 		goto err_admin_init;
3243 	}
3244 
3245 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3246 	    BIT(ENA_ADMIN_FATAL_ERROR) |
3247 	    BIT(ENA_ADMIN_WARNING) |
3248 	    BIT(ENA_ADMIN_NOTIFICATION) |
3249 	    BIT(ENA_ADMIN_KEEP_ALIVE);
3250 
3251 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
3252 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3253 	if (rc) {
3254 		device_printf(pdev, "Cannot configure aenq groups rc: %d\n", rc);
3255 		goto err_admin_init;
3256 	}
3257 
3258 	*wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3259 
3260 	return 0;
3261 
3262 err_admin_init:
3263 	ena_com_delete_host_info(ena_dev);
3264 	ena_com_admin_destroy(ena_dev);
3265 err_mmio_read_less:
3266 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3267 
3268 	return rc;
3269 }
3270 
3271 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
3272     int io_vectors)
3273 {
3274 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3275 	int rc;
3276 
3277 	rc = ena_enable_msix(adapter);
3278 	if (rc) {
3279 		device_printf(adapter->pdev, "Error with MSI-X enablement\n");
3280 		return rc;
3281 	}
3282 
3283 	ena_setup_mgmnt_intr(adapter);
3284 
3285 	rc = ena_request_mgmnt_irq(adapter);
3286 	if (rc) {
3287 		device_printf(adapter->pdev, "Cannot setup mgmnt queue intr\n");
3288 		goto err_disable_msix;
3289 	}
3290 
3291 	ena_com_set_admin_polling_mode(ena_dev, false);
3292 
3293 	ena_com_admin_aenq_enable(ena_dev);
3294 
3295 	return 0;
3296 
3297 err_disable_msix:
3298 	ena_disable_msix(adapter);
3299 
3300 	return rc;
3301 }
3302 
3303 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
3304 static void ena_keep_alive_wd(void *adapter_data,
3305     struct ena_admin_aenq_entry *aenq_e)
3306 {
3307 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3308 	struct ena_admin_aenq_keep_alive_desc *desc;
3309 	sbintime_t stime;
3310 	uint64_t rx_drops;
3311 
3312 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3313 
3314 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
3315 	counter_u64_zero(adapter->hw_stats.rx_drops);
3316 	counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
3317 
3318 	stime = getsbinuptime();
3319 	atomic_store_rel_64(&adapter->keep_alive_timestamp, stime);
3320 }
3321 
3322 /* Check for keep alive expiration */
3323 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3324 {
3325 	sbintime_t timestamp, time;
3326 
3327 	if (adapter->wd_active == 0)
3328 		return;
3329 
3330 	if (adapter->keep_alive_timeout == 0)
3331 		return;
3332 
3333 	timestamp = atomic_load_acq_64(&adapter->keep_alive_timestamp);
3334 	time = getsbinuptime() - timestamp;
3335 	if (unlikely(time > adapter->keep_alive_timeout)) {
3336 		device_printf(adapter->pdev,
3337 		    "Keep alive watchdog timeout.\n");
3338 		counter_u64_add(adapter->dev_stats.wd_expired, 1);
3339 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3340 		adapter->trigger_reset = true;
3341 	}
3342 }
3343 
3344 /* Check if admin queue is enabled */
3345 static void check_for_admin_com_state(struct ena_adapter *adapter)
3346 {
3347 	if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3348 		device_printf(adapter->pdev,
3349 		    "ENA admin queue is not in running state!\n");
3350 		counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
3351 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3352 		adapter->trigger_reset = true;
3353 	}
3354 }
3355 
3356 static int
3357 check_missing_comp_in_queue(struct ena_adapter *adapter,
3358     struct ena_ring *tx_ring)
3359 {
3360 	struct bintime curtime, time;
3361 	struct ena_tx_buffer *tx_buf;
3362 	uint32_t missed_tx = 0;
3363 	int i;
3364 
3365 	getbinuptime(&curtime);
3366 
3367 	for (i = 0; i < tx_ring->ring_size; i++) {
3368 		tx_buf = &tx_ring->tx_buffer_info[i];
3369 
3370 		if (!bintime_isset(&tx_buf->timestamp))
3371 			continue;
3372 
3373 		time = curtime;
3374 		bintime_sub(&time, &tx_buf->timestamp);
3375 
3376 		/* Check again if packet is still waiting */
3377 		if (unlikely(bttosbt(time) > adapter->missing_tx_timeout)) {
3378 
3379 			if (!tx_buf->print_once)
3380 				ena_trace(ENA_WARNING, "Found a Tx that wasn't "
3381 				    "completed on time, qid %d, index %d.\n",
3382 				    tx_ring->qid, i);
3383 
3384 			tx_buf->print_once = true;
3385 			missed_tx++;
3386 
3387 			if (unlikely(missed_tx >
3388 			    adapter->missing_tx_threshold)) {
3389 				device_printf(adapter->pdev,
3390 					      "The number of lost tx completion "
3391 					      "is above the threshold (%d > %d). "
3392 					      "Reset the device\n",
3393 					      missed_tx,
3394 					      adapter->missing_tx_threshold);
3395 				adapter->reset_reason =
3396 				    ENA_REGS_RESET_MISS_TX_CMPL;
3397 				adapter->trigger_reset = true;
3398 				return (EIO);
3399 			}
3400 		}
3401 	}
3402 
3403 	return (0);
3404 }
3405 
3406 /*
3407  * Check for TX which were not completed on time.
3408  * Timeout is defined by "missing_tx_timeout".
3409  * Reset will be performed if number of incompleted
3410  * transactions exceeds "missing_tx_threshold".
3411  */
3412 static void check_for_missing_tx_completions(struct ena_adapter *adapter)
3413 {
3414 	struct ena_ring *tx_ring;
3415 	int i, budget, rc;
3416 
3417 	/* Make sure the driver doesn't turn the device in other process */
3418 	rmb();
3419 
3420 	if (!adapter->up)
3421 		return;
3422 
3423 	if (adapter->trigger_reset)
3424 		return;
3425 
3426 	if (adapter->missing_tx_timeout == 0)
3427 		return;
3428 
3429 	budget = adapter->missing_tx_max_queues;
3430 
3431 	for (i = adapter->next_monitored_tx_qid; i < adapter->num_queues; i++) {
3432 		tx_ring = &adapter->tx_ring[i];
3433 
3434 		rc = check_missing_comp_in_queue(adapter, tx_ring);
3435 		if (unlikely(rc))
3436 			return;
3437 
3438 		budget--;
3439 		if (budget == 0) {
3440 			i++;
3441 			break;
3442 		}
3443 	}
3444 
3445 	adapter->next_monitored_tx_qid = i % adapter->num_queues;
3446 }
3447 
3448 /* trigger deferred rx cleanup after 2 consecutive detections */
3449 #define EMPTY_RX_REFILL 2
3450 /* For the rare case where the device runs out of Rx descriptors and the
3451  * msix handler failed to refill new Rx descriptors (due to a lack of memory
3452  * for example).
3453  * This case will lead to a deadlock:
3454  * The device won't send interrupts since all the new Rx packets will be dropped
3455  * The msix handler won't allocate new Rx descriptors so the device won't be
3456  * able to send new packets.
3457  *
3458  * When such a situation is detected - execute rx cleanup task in another thread
3459  */
3460 static void
3461 check_for_empty_rx_ring(struct ena_adapter *adapter)
3462 {
3463 	struct ena_ring *rx_ring;
3464 	int i, refill_required;
3465 
3466 	if (!adapter->up)
3467 		return;
3468 
3469 	if (adapter->trigger_reset)
3470 		return;
3471 
3472 	for (i = 0; i < adapter->num_queues; i++) {
3473 		rx_ring = &adapter->rx_ring[i];
3474 
3475 		refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
3476 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3477 			rx_ring->empty_rx_queue++;
3478 
3479 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL)	{
3480 				counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
3481 				    1);
3482 
3483 				device_printf(adapter->pdev,
3484 				    "trigger refill for ring %d\n", i);
3485 
3486 				taskqueue_enqueue(rx_ring->cmpl_tq,
3487 				    &rx_ring->cmpl_task);
3488 				rx_ring->empty_rx_queue = 0;
3489 			}
3490 		} else {
3491 			rx_ring->empty_rx_queue = 0;
3492 		}
3493 	}
3494 }
3495 
3496 static void
3497 ena_timer_service(void *data)
3498 {
3499 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3500 	struct ena_admin_host_info *host_info =
3501 	    adapter->ena_dev->host_attr.host_info;
3502 
3503 	check_for_missing_keep_alive(adapter);
3504 
3505 	check_for_admin_com_state(adapter);
3506 
3507 	check_for_missing_tx_completions(adapter);
3508 
3509 	check_for_empty_rx_ring(adapter);
3510 
3511 	if (host_info)
3512 		ena_update_host_info(host_info, adapter->ifp);
3513 
3514 	if (unlikely(adapter->trigger_reset)) {
3515 		device_printf(adapter->pdev, "Trigger reset is on\n");
3516 		taskqueue_enqueue(adapter->reset_tq, &adapter->reset_task);
3517 		return;
3518 	}
3519 
3520 	/*
3521 	 * Schedule another timeout one second from now.
3522 	 */
3523 	callout_schedule_sbt(&adapter->timer_service, SBT_1S, SBT_1S, 0);
3524 }
3525 
3526 static void
3527 ena_reset_task(void *arg, int pending)
3528 {
3529 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3530 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3531 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3532 	bool dev_up;
3533 	int rc;
3534 
3535 	if (unlikely(!adapter->trigger_reset)) {
3536 		device_printf(adapter->pdev,
3537 		    "device reset scheduled but trigger_reset is off\n");
3538 		return;
3539 	}
3540 
3541 	sx_xlock(&adapter->ioctl_sx);
3542 
3543 	callout_drain(&adapter->timer_service);
3544 
3545 	dev_up = adapter->up;
3546 
3547 	ena_com_set_admin_running_state(ena_dev, false);
3548 	ena_down(adapter);
3549 	ena_free_mgmnt_irq(adapter);
3550 	ena_disable_msix(adapter);
3551 	ena_com_abort_admin_commands(ena_dev);
3552 	ena_com_wait_for_abort_completion(ena_dev);
3553 	ena_com_admin_destroy(ena_dev);
3554 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3555 
3556 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3557 	adapter->trigger_reset = false;
3558 
3559 	/* Finished destroy part. Restart the device */
3560 	rc = ena_device_init(adapter, adapter->pdev, &get_feat_ctx,
3561 	    &adapter->wd_active);
3562 	if (rc) {
3563 		device_printf(adapter->pdev,
3564 		    "ENA device init failed! (err: %d)\n", rc);
3565 		goto err_dev_free;
3566 	}
3567 
3568 	rc = ena_enable_msix_and_set_admin_interrupts(adapter,
3569 	    adapter->num_queues);
3570 	if (rc) {
3571 		device_printf(adapter->pdev, "Enable MSI-X failed\n");
3572 		goto err_com_free;
3573 	}
3574 
3575 	/* If the interface was up before the reset bring it up */
3576 	if (dev_up) {
3577 		rc = ena_up(adapter);
3578 		if (rc) {
3579 			device_printf(adapter->pdev,
3580 			    "Failed to create I/O queues\n");
3581 			goto err_msix_free;
3582 		}
3583 	}
3584 
3585 	callout_reset_sbt(&adapter->timer_service, SBT_1S, SBT_1S,
3586 	    ena_timer_service, (void *)adapter, 0);
3587 
3588 	sx_unlock(&adapter->ioctl_sx);
3589 
3590 	return;
3591 
3592 err_msix_free:
3593 	ena_free_mgmnt_irq(adapter);
3594 	ena_disable_msix(adapter);
3595 err_com_free:
3596 	ena_com_admin_destroy(ena_dev);
3597 err_dev_free:
3598 	device_printf(adapter->pdev, "ENA reset failed!\n");
3599 	adapter->running = false;
3600 	sx_unlock(&adapter->ioctl_sx);
3601 }
3602 
3603 /**
3604  * ena_attach - Device Initialization Routine
3605  * @pdev: device information struct
3606  *
3607  * Returns 0 on success, otherwise on failure.
3608  *
3609  * ena_attach initializes an adapter identified by a device structure.
3610  * The OS initialization, configuring of the adapter private structure,
3611  * and a hardware reset occur.
3612  **/
3613 static int
3614 ena_attach(device_t pdev)
3615 {
3616 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3617 	static int version_printed;
3618 	struct ena_adapter *adapter;
3619 	struct ena_com_dev *ena_dev = NULL;
3620 	uint16_t tx_sgl_size = 0;
3621 	uint16_t rx_sgl_size = 0;
3622 	int io_queue_num;
3623 	int queue_size;
3624 	int rc;
3625 	struct sysctl_ctx_list *ctx;
3626 	struct sysctl_oid_list *children;
3627 
3628 	adapter = device_get_softc(pdev);
3629 	adapter->pdev = pdev;
3630 	ctx = device_get_sysctl_ctx(pdev);
3631 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(pdev));
3632 
3633 	mtx_init(&adapter->global_mtx, "ENA global mtx", NULL, MTX_DEF);
3634 	sx_init(&adapter->ioctl_sx, "ENA ioctl sx");
3635 
3636 	/* Sysctl calls for Watchdog service */
3637 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "wd_active",
3638 	    CTLFLAG_RWTUN, &adapter->wd_active, 0,
3639 	    "Watchdog is active");
3640 
3641 	SYSCTL_ADD_QUAD(ctx, children, OID_AUTO, "keep_alive_timeout",
3642 	    CTLFLAG_RWTUN, &adapter->keep_alive_timeout,
3643 	    "Timeout for Keep Alive messages");
3644 
3645 	SYSCTL_ADD_QUAD(ctx, children, OID_AUTO, "missing_tx_timeout",
3646 	    CTLFLAG_RWTUN, &adapter->missing_tx_timeout,
3647 	    "Timeout for TX completion");
3648 
3649 	SYSCTL_ADD_U32(ctx, children, OID_AUTO, "missing_tx_max_queues",
3650 	    CTLFLAG_RWTUN, &adapter->missing_tx_max_queues, 0,
3651 	    "Number of TX queues to check per run");
3652 
3653 	SYSCTL_ADD_U32(ctx, children, OID_AUTO, "missing_tx_threshold",
3654 	    CTLFLAG_RWTUN, &adapter->missing_tx_threshold, 0,
3655 	    "Max number of timeouted packets");
3656 
3657 	/* Set up the timer service */
3658 	callout_init_mtx(&adapter->timer_service, &adapter->global_mtx, 0);
3659 	adapter->keep_alive_timeout = DEFAULT_KEEP_ALIVE_TO;
3660 	adapter->missing_tx_timeout = DEFAULT_TX_CMP_TO;
3661 	adapter->missing_tx_max_queues = DEFAULT_TX_MONITORED_QUEUES;
3662 	adapter->missing_tx_threshold = DEFAULT_TX_CMP_THRESHOLD;
3663 
3664 	if (version_printed++ == 0)
3665 		device_printf(pdev, "%s\n", ena_version);
3666 
3667 	rc = ena_allocate_pci_resources(adapter);
3668 	if (rc) {
3669 		device_printf(pdev, "PCI resource allocation failed!\n");
3670 		ena_free_pci_resources(adapter);
3671 		return (rc);
3672 	}
3673 
3674 	/* Allocate memory for ena_dev structure */
3675 	ena_dev = malloc(sizeof(struct ena_com_dev), M_DEVBUF,
3676 	    M_WAITOK | M_ZERO);
3677 
3678 	adapter->ena_dev = ena_dev;
3679 	ena_dev->dmadev = pdev;
3680 	ena_dev->bus = malloc(sizeof(struct ena_bus), M_DEVBUF,
3681 	    M_WAITOK | M_ZERO);
3682 
3683 	/* Store register resources */
3684 	((struct ena_bus*)(ena_dev->bus))->reg_bar_t =
3685 	    rman_get_bustag(adapter->registers);
3686 	((struct ena_bus*)(ena_dev->bus))->reg_bar_h =
3687 	    rman_get_bushandle(adapter->registers);
3688 
3689 	if (((struct ena_bus*)(ena_dev->bus))->reg_bar_h == 0) {
3690 		device_printf(pdev, "failed to pmap registers bar\n");
3691 		rc = ENXIO;
3692 		goto err_bus_free;
3693 	}
3694 
3695 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3696 
3697 	/* Device initialization */
3698 	rc = ena_device_init(adapter, pdev, &get_feat_ctx, &adapter->wd_active);
3699 	if (rc) {
3700 		device_printf(pdev, "ENA device init failed! (err: %d)\n", rc);
3701 		rc = ENXIO;
3702 		goto err_bus_free;
3703 	}
3704 
3705 	adapter->keep_alive_timestamp = getsbinuptime();
3706 
3707 	adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3708 
3709 	/* Set for sure that interface is not up */
3710 	adapter->up = false;
3711 
3712 	memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3713 	    ETHER_ADDR_LEN);
3714 
3715 	adapter->small_copy_len =
3716 	    ENA_DEFAULT_SMALL_PACKET_LEN;
3717 
3718 	/* calculate IO queue number to create */
3719 	io_queue_num = ena_calc_io_queue_num(adapter, &get_feat_ctx);
3720 
3721 	ENA_ASSERT(io_queue_num > 0, "Invalid queue number: %d\n",
3722 	    io_queue_num);
3723 	adapter->num_queues = io_queue_num;
3724 
3725 	/* calculatre ring sizes */
3726 	queue_size = ena_calc_queue_size(adapter,&tx_sgl_size,
3727 	    &rx_sgl_size, &get_feat_ctx);
3728 	if ((queue_size <= 0) || (io_queue_num <= 0)) {
3729 		rc = ENA_COM_FAULT;
3730 		goto err_com_free;
3731 	}
3732 
3733 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3734 
3735 	adapter->tx_ring_size = queue_size;
3736 	adapter->rx_ring_size = queue_size;
3737 
3738 	adapter->max_tx_sgl_size = tx_sgl_size;
3739 	adapter->max_rx_sgl_size = rx_sgl_size;
3740 
3741 	/* set up dma tags for rx and tx buffers */
3742 	rc = ena_setup_tx_dma_tag(adapter);
3743 	if (rc)
3744 		goto err_com_free;
3745 
3746 	rc = ena_setup_rx_dma_tag(adapter);
3747 	if (rc)
3748 		goto err_tx_tag_free;
3749 
3750 	/* initialize rings basic information */
3751 	device_printf(pdev, "initalize %d io queues\n", io_queue_num);
3752 	ena_init_io_rings(adapter);
3753 
3754 	/* setup network interface */
3755 	rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
3756 	if (rc) {
3757 		device_printf(pdev,"Error with network interface setup\n");
3758 		goto err_io_free;
3759 	}
3760 
3761 	rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3762 	if (rc) {
3763 		device_printf(pdev,
3764 		    "Failed to enable and set the admin interrupts\n");
3765 		goto err_ifp_free;
3766 	}
3767 
3768 	/* Initialize reset task queue */
3769 	TASK_INIT(&adapter->reset_task, 0, ena_reset_task, adapter);
3770 	adapter->reset_tq = taskqueue_create("ena_reset_enqueue",
3771 	    M_WAITOK | M_ZERO, taskqueue_thread_enqueue, &adapter->reset_tq);
3772 	taskqueue_start_threads(&adapter->reset_tq, 1, PI_NET,
3773 	    "%s rstq", device_get_nameunit(adapter->pdev));
3774 
3775 	/* Initialize statistics */
3776 	ena_alloc_counters((counter_u64_t *)&adapter->dev_stats,
3777 	    sizeof(struct ena_stats_dev));
3778 	ena_alloc_counters((counter_u64_t *)&adapter->hw_stats,
3779 	    sizeof(struct ena_hw_stats));
3780 	ena_sysctl_add_nodes(adapter);
3781 
3782 	/* Tell the stack that the interface is not active */
3783 	if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
3784 
3785 	adapter->running = true;
3786 	return (0);
3787 
3788 err_ifp_free:
3789 	if_detach(adapter->ifp);
3790 	if_free(adapter->ifp);
3791 err_io_free:
3792 	ena_free_all_io_rings_resources(adapter);
3793 	ena_free_rx_dma_tag(adapter);
3794 err_tx_tag_free:
3795 	ena_free_tx_dma_tag(adapter);
3796 err_com_free:
3797 	ena_com_admin_destroy(ena_dev);
3798 	ena_com_delete_host_info(ena_dev);
3799 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3800 err_bus_free:
3801 	free(ena_dev->bus, M_DEVBUF);
3802 	free(ena_dev, M_DEVBUF);
3803 	ena_free_pci_resources(adapter);
3804 
3805 	return (rc);
3806 }
3807 
3808 /**
3809  * ena_detach - Device Removal Routine
3810  * @pdev: device information struct
3811  *
3812  * ena_detach is called by the device subsystem to alert the driver
3813  * that it should release a PCI device.
3814  **/
3815 static int
3816 ena_detach(device_t pdev)
3817 {
3818 	struct ena_adapter *adapter = device_get_softc(pdev);
3819 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3820 	int rc;
3821 
3822 	/* Make sure VLANS are not using driver */
3823 	if (adapter->ifp->if_vlantrunk != NULL) {
3824 		device_printf(adapter->pdev ,"VLAN is in use, detach first\n");
3825 		return (EBUSY);
3826 	}
3827 
3828 	/* Free reset task and callout */
3829 	callout_drain(&adapter->timer_service);
3830 	while (taskqueue_cancel(adapter->reset_tq, &adapter->reset_task, NULL))
3831 		taskqueue_drain(adapter->reset_tq, &adapter->reset_task);
3832 	taskqueue_free(adapter->reset_tq);
3833 
3834 	sx_xlock(&adapter->ioctl_sx);
3835 	ena_down(adapter);
3836 	sx_unlock(&adapter->ioctl_sx);
3837 
3838 	if (adapter->ifp != NULL) {
3839 		ether_ifdetach(adapter->ifp);
3840 		if_free(adapter->ifp);
3841 	}
3842 
3843 	ena_free_all_io_rings_resources(adapter);
3844 
3845 	ena_free_counters((counter_u64_t *)&adapter->hw_stats,
3846 	    sizeof(struct ena_hw_stats));
3847 	ena_free_counters((counter_u64_t *)&adapter->dev_stats,
3848 	    sizeof(struct ena_stats_dev));
3849 
3850 	if (adapter->rss_support)
3851 		ena_com_rss_destroy(ena_dev);
3852 
3853 	rc = ena_free_rx_dma_tag(adapter);
3854 	if (rc)
3855 		device_printf(adapter->pdev,
3856 		    "Unmapped RX DMA tag associations\n");
3857 
3858 	rc = ena_free_tx_dma_tag(adapter);
3859 	if (rc)
3860 		device_printf(adapter->pdev,
3861 		    "Unmapped TX DMA tag associations\n");
3862 
3863 	/* Reset the device only if the device is running. */
3864 	if (adapter->running)
3865 		ena_com_dev_reset(ena_dev, adapter->reset_reason);
3866 
3867 	ena_com_delete_host_info(ena_dev);
3868 
3869 	ena_free_irqs(adapter);
3870 
3871 	ena_com_abort_admin_commands(ena_dev);
3872 
3873 	ena_com_wait_for_abort_completion(ena_dev);
3874 
3875 	ena_com_admin_destroy(ena_dev);
3876 
3877 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3878 
3879 	ena_free_pci_resources(adapter);
3880 
3881 	mtx_destroy(&adapter->global_mtx);
3882 	sx_destroy(&adapter->ioctl_sx);
3883 
3884 	if (ena_dev->bus != NULL)
3885 		free(ena_dev->bus, M_DEVBUF);
3886 
3887 	if (ena_dev != NULL)
3888 		free(ena_dev, M_DEVBUF);
3889 
3890 	return (bus_generic_detach(pdev));
3891 }
3892 
3893 /******************************************************************************
3894  ******************************** AENQ Handlers *******************************
3895  *****************************************************************************/
3896 /**
3897  * ena_update_on_link_change:
3898  * Notify the network interface about the change in link status
3899  **/
3900 static void
3901 ena_update_on_link_change(void *adapter_data,
3902     struct ena_admin_aenq_entry *aenq_e)
3903 {
3904 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3905 	struct ena_admin_aenq_link_change_desc *aenq_desc;
3906 	int status;
3907 	if_t ifp;
3908 
3909 	aenq_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
3910 	ifp = adapter->ifp;
3911 	status = aenq_desc->flags &
3912 	    ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3913 
3914 	if (status != 0) {
3915 		device_printf(adapter->pdev, "link is UP\n");
3916 		if_link_state_change(ifp, LINK_STATE_UP);
3917 	} else if (status == 0) {
3918 		device_printf(adapter->pdev, "link is DOWN\n");
3919 		if_link_state_change(ifp, LINK_STATE_DOWN);
3920 	} else {
3921 		device_printf(adapter->pdev, "invalid value recvd\n");
3922 		BUG();
3923 	}
3924 
3925 	adapter->link_status = status;
3926 
3927 	return;
3928 }
3929 
3930 /**
3931  * This handler will called for unknown event group or unimplemented handlers
3932  **/
3933 static void
3934 unimplemented_aenq_handler(void *data,
3935     struct ena_admin_aenq_entry *aenq_e)
3936 {
3937 	return;
3938 }
3939 
3940 static struct ena_aenq_handlers aenq_handlers = {
3941     .handlers = {
3942 	    [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3943 	    [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3944     },
3945     .unimplemented_handler = unimplemented_aenq_handler
3946 };
3947 
3948 /*********************************************************************
3949  *  FreeBSD Device Interface Entry Points
3950  *********************************************************************/
3951 
3952 static device_method_t ena_methods[] = {
3953     /* Device interface */
3954     DEVMETHOD(device_probe, ena_probe),
3955     DEVMETHOD(device_attach, ena_attach),
3956     DEVMETHOD(device_detach, ena_detach),
3957     DEVMETHOD_END
3958 };
3959 
3960 static driver_t ena_driver = {
3961     "ena", ena_methods, sizeof(struct ena_adapter),
3962 };
3963 
3964 devclass_t ena_devclass;
3965 DRIVER_MODULE(ena, pci, ena_driver, ena_devclass, 0, 0);
3966 MODULE_DEPEND(ena, pci, 1, 1, 1);
3967 MODULE_DEPEND(ena, ether, 1, 1, 1);
3968 
3969 /*********************************************************************/
3970