xref: /netbsd-src/sys/dev/pci/if_ena.c (revision f3cfa6f6ce31685c6c4a758bc430e69eb99f50a4)
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 #if 0
32 __FBSDID("$FreeBSD: head/sys/dev/ena/ena.c 333456 2018-05-10 09:37:54Z mw $");
33 #endif
34 __KERNEL_RCSID(0, "$NetBSD: if_ena.c,v 1.15 2019/05/29 10:07:29 msaitoh Exp $");
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/bus.h>
39 #include <sys/endian.h>
40 #include <sys/kernel.h>
41 #include <sys/kthread.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/module.h>
45 #include <sys/socket.h>
46 #include <sys/sockio.h>
47 #include <sys/sysctl.h>
48 #include <sys/time.h>
49 #include <sys/workqueue.h>
50 #include <sys/callout.h>
51 #include <sys/interrupt.h>
52 #include <sys/cpu.h>
53 
54 #include <net/if_ether.h>
55 #include <net/if_vlanvar.h>
56 
57 #include <dev/pci/if_enavar.h>
58 
59 /*********************************************************
60  *  Function prototypes
61  *********************************************************/
62 static int	ena_probe(device_t, cfdata_t, void *);
63 static int	ena_intr_msix_mgmnt(void *);
64 static int	ena_allocate_pci_resources(struct pci_attach_args *,
65 		    struct ena_adapter *);
66 static void	ena_free_pci_resources(struct ena_adapter *);
67 static int	ena_change_mtu(struct ifnet *, int);
68 static void	ena_init_io_rings_common(struct ena_adapter *,
69     struct ena_ring *, uint16_t);
70 static void	ena_init_io_rings(struct ena_adapter *);
71 static void	ena_free_io_ring_resources(struct ena_adapter *, unsigned int);
72 static void	ena_free_all_io_rings_resources(struct ena_adapter *);
73 #if 0
74 static int	ena_setup_tx_dma_tag(struct ena_adapter *);
75 static int	ena_free_tx_dma_tag(struct ena_adapter *);
76 static int	ena_setup_rx_dma_tag(struct ena_adapter *);
77 static int	ena_free_rx_dma_tag(struct ena_adapter *);
78 #endif
79 static int	ena_setup_tx_resources(struct ena_adapter *, int);
80 static void	ena_free_tx_resources(struct ena_adapter *, int);
81 static int	ena_setup_all_tx_resources(struct ena_adapter *);
82 static void	ena_free_all_tx_resources(struct ena_adapter *);
83 static inline int validate_rx_req_id(struct ena_ring *, uint16_t);
84 static int	ena_setup_rx_resources(struct ena_adapter *, unsigned int);
85 static void	ena_free_rx_resources(struct ena_adapter *, unsigned int);
86 static int	ena_setup_all_rx_resources(struct ena_adapter *);
87 static void	ena_free_all_rx_resources(struct ena_adapter *);
88 static inline int ena_alloc_rx_mbuf(struct ena_adapter *, struct ena_ring *,
89     struct ena_rx_buffer *);
90 static void	ena_free_rx_mbuf(struct ena_adapter *, struct ena_ring *,
91     struct ena_rx_buffer *);
92 static int	ena_refill_rx_bufs(struct ena_ring *, uint32_t);
93 static void	ena_free_rx_bufs(struct ena_adapter *, unsigned int);
94 static void	ena_refill_all_rx_bufs(struct ena_adapter *);
95 static void	ena_free_all_rx_bufs(struct ena_adapter *);
96 static void	ena_free_tx_bufs(struct ena_adapter *, unsigned int);
97 static void	ena_free_all_tx_bufs(struct ena_adapter *);
98 static void	ena_destroy_all_tx_queues(struct ena_adapter *);
99 static void	ena_destroy_all_rx_queues(struct ena_adapter *);
100 static void	ena_destroy_all_io_queues(struct ena_adapter *);
101 static int	ena_create_io_queues(struct ena_adapter *);
102 static int	ena_tx_cleanup(struct ena_ring *);
103 static void	ena_deferred_rx_cleanup(struct work *, void *);
104 static int	ena_rx_cleanup(struct ena_ring *);
105 static inline int validate_tx_req_id(struct ena_ring *, uint16_t);
106 #if 0
107 static void	ena_rx_hash_mbuf(struct ena_ring *, struct ena_com_rx_ctx *,
108     struct mbuf *);
109 #endif
110 static struct mbuf* ena_rx_mbuf(struct ena_ring *, struct ena_com_rx_buf_info *,
111     struct ena_com_rx_ctx *, uint16_t *);
112 static inline void ena_rx_checksum(struct ena_ring *, struct ena_com_rx_ctx *,
113     struct mbuf *);
114 static int	ena_handle_msix(void *);
115 static int	ena_enable_msix(struct ena_adapter *);
116 static int	ena_request_mgmnt_irq(struct ena_adapter *);
117 static int	ena_request_io_irq(struct ena_adapter *);
118 static void	ena_free_mgmnt_irq(struct ena_adapter *);
119 static void	ena_free_io_irq(struct ena_adapter *);
120 static void	ena_free_irqs(struct ena_adapter*);
121 static void	ena_disable_msix(struct ena_adapter *);
122 static void	ena_unmask_all_io_irqs(struct ena_adapter *);
123 static int	ena_rss_configure(struct ena_adapter *);
124 static int	ena_up_complete(struct ena_adapter *);
125 static int	ena_up(struct ena_adapter *);
126 static void	ena_down(struct ena_adapter *);
127 #if 0
128 static uint64_t	ena_get_counter(struct ifnet *, ift_counter);
129 #endif
130 static int	ena_media_change(struct ifnet *);
131 static void	ena_media_status(struct ifnet *, struct ifmediareq *);
132 static int	ena_init(struct ifnet *);
133 static int	ena_ioctl(struct ifnet *, u_long, void *);
134 static int	ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *);
135 static void	ena_update_host_info(struct ena_admin_host_info *, struct ifnet *);
136 static void	ena_update_hwassist(struct ena_adapter *);
137 static int	ena_setup_ifnet(device_t, struct ena_adapter *,
138     struct ena_com_dev_get_features_ctx *);
139 static void	ena_tx_csum(struct ena_com_tx_ctx *, struct mbuf *);
140 static int	ena_check_and_collapse_mbuf(struct ena_ring *tx_ring,
141     struct mbuf **mbuf);
142 static int	ena_xmit_mbuf(struct ena_ring *, struct mbuf **);
143 static void	ena_start_xmit(struct ena_ring *);
144 static int	ena_mq_start(struct ifnet *, struct mbuf *);
145 static void	ena_deferred_mq_start(struct work *, void *);
146 #if 0
147 static void	ena_qflush(struct ifnet *);
148 #endif
149 static int	ena_calc_io_queue_num(struct pci_attach_args *,
150     struct ena_adapter *, struct ena_com_dev_get_features_ctx *);
151 static int	ena_calc_queue_size(struct ena_adapter *, uint16_t *,
152     uint16_t *, struct ena_com_dev_get_features_ctx *);
153 #if 0
154 static int	ena_rss_init_default(struct ena_adapter *);
155 static void	ena_rss_init_default_deferred(void *);
156 #endif
157 static void	ena_config_host_info(struct ena_com_dev *);
158 static void	ena_attach(device_t, device_t, void *);
159 static int	ena_detach(device_t, int);
160 static int	ena_device_init(struct ena_adapter *, device_t,
161     struct ena_com_dev_get_features_ctx *, int *);
162 static int	ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *,
163     int);
164 static void ena_update_on_link_change(void *, struct ena_admin_aenq_entry *);
165 static void	unimplemented_aenq_handler(void *,
166     struct ena_admin_aenq_entry *);
167 static void	ena_timer_service(void *);
168 
169 static const char ena_version[] =
170     DEVICE_NAME DRV_MODULE_NAME " v" DRV_MODULE_VERSION;
171 
172 #if 0
173 static SYSCTL_NODE(_hw, OID_AUTO, ena, CTLFLAG_RD, 0, "ENA driver parameters");
174 #endif
175 
176 /*
177  * Tuneable number of buffers in the buf-ring (drbr)
178  */
179 static int ena_buf_ring_size = 4096;
180 #if 0
181 SYSCTL_INT(_hw_ena, OID_AUTO, buf_ring_size, CTLFLAG_RWTUN,
182     &ena_buf_ring_size, 0, "Size of the bufring");
183 #endif
184 
185 /*
186  * Logging level for changing verbosity of the output
187  */
188 int ena_log_level = ENA_ALERT | ENA_WARNING;
189 #if 0
190 SYSCTL_INT(_hw_ena, OID_AUTO, log_level, CTLFLAG_RWTUN,
191     &ena_log_level, 0, "Logging level indicating verbosity of the logs");
192 #endif
193 
194 static const ena_vendor_info_t ena_vendor_info_array[] = {
195     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_PF, 0},
196     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_PF, 0},
197     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_VF, 0},
198     { PCI_VENDOR_ID_AMAZON, PCI_DEV_ID_ENA_LLQ_VF, 0},
199     /* Last entry */
200     { 0, 0, 0 }
201 };
202 
203 /*
204  * Contains pointers to event handlers, e.g. link state chage.
205  */
206 static struct ena_aenq_handlers aenq_handlers;
207 
208 int
209 ena_dma_alloc(device_t dmadev, bus_size_t size,
210     ena_mem_handle_t *dma , int mapflags)
211 {
212 	struct ena_adapter *adapter = device_private(dmadev);
213 	uint32_t maxsize;
214 	bus_dma_segment_t seg;
215 	int error, nsegs;
216 
217 	maxsize = ((size - 1) / PAGE_SIZE + 1) * PAGE_SIZE;
218 
219 #if 0
220 	/* XXX what is this needed for ? */
221 	dma_space_addr = ENA_DMA_BIT_MASK(adapter->dma_width);
222 	if (unlikely(dma_space_addr == 0))
223 		dma_space_addr = BUS_SPACE_MAXADDR;
224 #endif
225 
226 	dma->tag = adapter->sc_dmat;
227 
228         if ((error = bus_dmamap_create(dma->tag, maxsize, 1, maxsize, 0,
229             BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, &dma->map)) != 0) {
230 		ena_trace(ENA_ALERT, "bus_dmamap_create(%ju) failed: %d\n",
231 		    (uintmax_t)maxsize, error);
232                 goto fail_create;
233 	}
234 
235 	error = bus_dmamem_alloc(dma->tag, maxsize, 8, 0, &seg, 1, &nsegs,
236 	    BUS_DMA_ALLOCNOW);
237 	if (error) {
238 		ena_trace(ENA_ALERT, "bus_dmamem_alloc(%ju) failed: %d\n",
239 		    (uintmax_t)maxsize, error);
240 		goto fail_alloc;
241 	}
242 
243 	error = bus_dmamem_map(dma->tag, &seg, nsegs, maxsize,
244 	    &dma->vaddr, BUS_DMA_COHERENT);
245 	if (error) {
246 		ena_trace(ENA_ALERT, "bus_dmamem_map(%ju) failed: %d\n",
247 		    (uintmax_t)maxsize, error);
248 		goto fail_map;
249 	}
250 	memset(dma->vaddr, 0, maxsize);
251 
252 	error = bus_dmamap_load(dma->tag, dma->map, dma->vaddr,
253 	    maxsize, NULL, mapflags);
254 	if (error) {
255 		ena_trace(ENA_ALERT, ": bus_dmamap_load failed: %d\n", error);
256 		goto fail_load;
257 	}
258 	dma->paddr = dma->map->dm_segs[0].ds_addr;
259 
260 	return (0);
261 
262 fail_load:
263 	bus_dmamem_unmap(dma->tag, dma->vaddr, maxsize);
264 fail_map:
265 	bus_dmamem_free(dma->tag, &seg, nsegs);
266 fail_alloc:
267 	bus_dmamap_destroy(adapter->sc_dmat, dma->map);
268 fail_create:
269 	return (error);
270 }
271 
272 static int
273 ena_allocate_pci_resources(struct pci_attach_args *pa,
274     struct ena_adapter *adapter)
275 {
276 	bus_size_t size;
277 
278 	/*
279 	 * Map control/status registers.
280 	*/
281 	pcireg_t memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, ENA_REG_BAR);
282 	if (pci_mapreg_map(pa, ENA_REG_BAR, memtype, 0, &adapter->sc_btag,
283 	    &adapter->sc_bhandle, NULL, &size)) {
284 		aprint_error(": can't map mem space\n");
285 		return ENXIO;
286 	}
287 
288 	return (0);
289 }
290 
291 static void
292 ena_free_pci_resources(struct ena_adapter *adapter)
293 {
294 	/* Nothing to do */
295 }
296 
297 static int
298 ena_probe(device_t parent, cfdata_t match, void *aux)
299 {
300 	struct pci_attach_args *pa = aux;
301 	const ena_vendor_info_t *ent;
302 
303 	for (int i = 0; i < __arraycount(ena_vendor_info_array); i++) {
304 		ent = &ena_vendor_info_array[i];
305 
306 		if ((PCI_VENDOR(pa->pa_id) == ent->vendor_id) &&
307 		    (PCI_PRODUCT(pa->pa_id) == ent->device_id)) {
308 			return 1;
309 		}
310 	}
311 
312 	return 0;
313 }
314 
315 static int
316 ena_change_mtu(struct ifnet *ifp, int new_mtu)
317 {
318 	struct ena_adapter *adapter = if_getsoftc(ifp);
319 	int rc;
320 
321 	if ((new_mtu > adapter->max_mtu) || (new_mtu < ENA_MIN_MTU)) {
322 		device_printf(adapter->pdev, "Invalid MTU setting. "
323 		    "new_mtu: %d max mtu: %d min mtu: %d\n",
324 		    new_mtu, adapter->max_mtu, ENA_MIN_MTU);
325 		return (EINVAL);
326 	}
327 
328 	rc = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
329 	if (likely(rc == 0)) {
330 		ena_trace(ENA_DBG, "set MTU to %d\n", new_mtu);
331 		if_setmtu(ifp, new_mtu);
332 	} else {
333 		device_printf(adapter->pdev, "Failed to set MTU to %d\n",
334 		    new_mtu);
335 	}
336 
337 	return (rc);
338 }
339 
340 #define EVCNT_INIT(st, f) \
341 	do {								\
342 		evcnt_attach_dynamic(&st->f, EVCNT_TYPE_MISC, NULL,	\
343 		    st->name, #f);					\
344 	} while (0)
345 
346 static inline void
347 ena_alloc_counters_rx(struct ena_stats_rx *st, int queue)
348 {
349 	snprintf(st->name, sizeof(st->name), "ena rxq%d", queue);
350 
351 	EVCNT_INIT(st, cnt);
352 	EVCNT_INIT(st, bytes);
353 	EVCNT_INIT(st, refil_partial);
354 	EVCNT_INIT(st, bad_csum);
355 	EVCNT_INIT(st, mjum_alloc_fail);
356 	EVCNT_INIT(st, mbuf_alloc_fail);
357 	EVCNT_INIT(st, dma_mapping_err);
358 	EVCNT_INIT(st, bad_desc_num);
359 	EVCNT_INIT(st, bad_req_id);
360 	EVCNT_INIT(st, empty_rx_ring);
361 
362 	/* Make sure all code is updated when new fields added */
363 	CTASSERT(offsetof(struct ena_stats_rx, empty_rx_ring)
364 	    + sizeof(st->empty_rx_ring) == sizeof(*st));
365 }
366 
367 static inline void
368 ena_alloc_counters_tx(struct ena_stats_tx *st, int queue)
369 {
370 	snprintf(st->name, sizeof(st->name), "ena txq%d", queue);
371 
372 	EVCNT_INIT(st, cnt);
373 	EVCNT_INIT(st, bytes);
374 	EVCNT_INIT(st, prepare_ctx_err);
375 	EVCNT_INIT(st, dma_mapping_err);
376 	EVCNT_INIT(st, doorbells);
377 	EVCNT_INIT(st, missing_tx_comp);
378 	EVCNT_INIT(st, bad_req_id);
379 	EVCNT_INIT(st, collapse);
380 	EVCNT_INIT(st, collapse_err);
381 
382 	/* Make sure all code is updated when new fields added */
383 	CTASSERT(offsetof(struct ena_stats_tx, collapse_err)
384 	    + sizeof(st->collapse_err) == sizeof(*st));
385 }
386 
387 static inline void
388 ena_alloc_counters_dev(struct ena_stats_dev *st, int queue)
389 {
390 	snprintf(st->name, sizeof(st->name), "ena dev ioq%d", queue);
391 
392 	EVCNT_INIT(st, wd_expired);
393 	EVCNT_INIT(st, interface_up);
394 	EVCNT_INIT(st, interface_down);
395 	EVCNT_INIT(st, admin_q_pause);
396 
397 	/* Make sure all code is updated when new fields added */
398 	CTASSERT(offsetof(struct ena_stats_dev, admin_q_pause)
399 	    + sizeof(st->admin_q_pause) == sizeof(*st));
400 }
401 
402 static inline void
403 ena_alloc_counters_hwstats(struct ena_hw_stats *st, int queue)
404 {
405 	snprintf(st->name, sizeof(st->name), "ena hw ioq%d", queue);
406 
407 	EVCNT_INIT(st, rx_packets);
408 	EVCNT_INIT(st, tx_packets);
409 	EVCNT_INIT(st, rx_bytes);
410 	EVCNT_INIT(st, tx_bytes);
411 	EVCNT_INIT(st, rx_drops);
412 
413 	/* Make sure all code is updated when new fields added */
414 	CTASSERT(offsetof(struct ena_hw_stats, rx_drops)
415 	    + sizeof(st->rx_drops) == sizeof(*st));
416 }
417 static inline void
418 ena_free_counters(struct evcnt *begin, int size)
419 {
420 	struct evcnt *end = (struct evcnt *)((char *)begin + size);
421 
422 	for (; begin < end; ++begin)
423 		counter_u64_free(*begin);
424 }
425 
426 static inline void
427 ena_reset_counters(struct evcnt *begin, int size)
428 {
429 	struct evcnt *end = (struct evcnt *)((char *)begin + size);
430 
431 	for (; begin < end; ++begin)
432 		counter_u64_zero(*begin);
433 }
434 
435 static void
436 ena_init_io_rings_common(struct ena_adapter *adapter, struct ena_ring *ring,
437     uint16_t qid)
438 {
439 
440 	ring->qid = qid;
441 	ring->adapter = adapter;
442 	ring->ena_dev = adapter->ena_dev;
443 }
444 
445 static void
446 ena_init_io_rings(struct ena_adapter *adapter)
447 {
448 	struct ena_com_dev *ena_dev;
449 	struct ena_ring *txr, *rxr;
450 	struct ena_que *que;
451 	int i;
452 
453 	ena_dev = adapter->ena_dev;
454 
455 	for (i = 0; i < adapter->num_queues; i++) {
456 		txr = &adapter->tx_ring[i];
457 		rxr = &adapter->rx_ring[i];
458 
459 		/* TX/RX common ring state */
460 		ena_init_io_rings_common(adapter, txr, i);
461 		ena_init_io_rings_common(adapter, rxr, i);
462 
463 		/* TX specific ring state */
464 		txr->ring_size = adapter->tx_ring_size;
465 		txr->tx_max_header_size = ena_dev->tx_max_header_size;
466 		txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
467 		txr->smoothed_interval =
468 		    ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
469 
470 		/* Allocate a buf ring */
471 		txr->br = buf_ring_alloc(ena_buf_ring_size, M_DEVBUF,
472 		    M_WAITOK, &txr->ring_mtx);
473 
474 		/* Alloc TX statistics. */
475 		ena_alloc_counters_tx(&txr->tx_stats, i);
476 
477 		/* RX specific ring state */
478 		rxr->ring_size = adapter->rx_ring_size;
479 		rxr->smoothed_interval =
480 		    ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
481 
482 		/* Alloc RX statistics. */
483 		ena_alloc_counters_rx(&rxr->rx_stats, i);
484 
485 		/* Initialize locks */
486 		snprintf(txr->mtx_name, sizeof(txr->mtx_name), "%s:tx(%d)",
487 		    device_xname(adapter->pdev), i);
488 		snprintf(rxr->mtx_name, sizeof(rxr->mtx_name), "%s:rx(%d)",
489 		    device_xname(adapter->pdev), i);
490 
491 		mutex_init(&txr->ring_mtx, MUTEX_DEFAULT, IPL_NET);
492 		mutex_init(&rxr->ring_mtx, MUTEX_DEFAULT, IPL_NET);
493 
494 		que = &adapter->que[i];
495 		que->adapter = adapter;
496 		que->id = i;
497 		que->tx_ring = txr;
498 		que->rx_ring = rxr;
499 
500 		txr->que = que;
501 		rxr->que = que;
502 
503 		rxr->empty_rx_queue = 0;
504 	}
505 }
506 
507 static void
508 ena_free_io_ring_resources(struct ena_adapter *adapter, unsigned int qid)
509 {
510 	struct ena_ring *txr = &adapter->tx_ring[qid];
511 	struct ena_ring *rxr = &adapter->rx_ring[qid];
512 
513 	ena_free_counters((struct evcnt *)&txr->tx_stats,
514 	    sizeof(txr->tx_stats));
515 	ena_free_counters((struct evcnt *)&rxr->rx_stats,
516 	    sizeof(rxr->rx_stats));
517 
518 	ENA_RING_MTX_LOCK(txr);
519 	drbr_free(txr->br, M_DEVBUF);
520 	ENA_RING_MTX_UNLOCK(txr);
521 
522 	mutex_destroy(&txr->ring_mtx);
523 	mutex_destroy(&rxr->ring_mtx);
524 }
525 
526 static void
527 ena_free_all_io_rings_resources(struct ena_adapter *adapter)
528 {
529 	int i;
530 
531 	for (i = 0; i < adapter->num_queues; i++)
532 		ena_free_io_ring_resources(adapter, i);
533 
534 }
535 
536 #if 0
537 static int
538 ena_setup_tx_dma_tag(struct ena_adapter *adapter)
539 {
540 	int ret;
541 
542 	/* Create DMA tag for Tx buffers */
543 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev),
544 	    1, 0,				  /* alignment, bounds 	     */
545 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
546 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
547 	    NULL, NULL,				  /* filter, filterarg 	     */
548 	    ENA_TSO_MAXSIZE,			  /* maxsize 		     */
549 	    adapter->max_tx_sgl_size - 1,	  /* nsegments 		     */
550 	    ENA_TSO_MAXSIZE,			  /* maxsegsize 	     */
551 	    0,					  /* flags 		     */
552 	    NULL,				  /* lockfunc 		     */
553 	    NULL,				  /* lockfuncarg 	     */
554 	    &adapter->tx_buf_tag);
555 
556 	return (ret);
557 }
558 #endif
559 
560 #if 0
561 static int
562 ena_setup_rx_dma_tag(struct ena_adapter *adapter)
563 {
564 	int ret;
565 
566 	/* Create DMA tag for Rx buffers*/
567 	ret = bus_dma_tag_create(bus_get_dma_tag(adapter->pdev), /* parent   */
568 	    1, 0,				  /* alignment, bounds 	     */
569 	    ENA_DMA_BIT_MASK(adapter->dma_width), /* lowaddr of excl window  */
570 	    BUS_SPACE_MAXADDR, 			  /* highaddr of excl window */
571 	    NULL, NULL,				  /* filter, filterarg 	     */
572 	    MJUM16BYTES,			  /* maxsize 		     */
573 	    adapter->max_rx_sgl_size,		  /* nsegments 		     */
574 	    MJUM16BYTES,			  /* maxsegsize 	     */
575 	    0,					  /* flags 		     */
576 	    NULL,				  /* lockfunc 		     */
577 	    NULL,				  /* lockarg 		     */
578 	    &adapter->rx_buf_tag);
579 
580 	return (ret);
581 }
582 #endif
583 
584 /**
585  * ena_setup_tx_resources - allocate Tx resources (Descriptors)
586  * @adapter: network interface device structure
587  * @qid: queue index
588  *
589  * Returns 0 on success, otherwise on failure.
590  **/
591 static int
592 ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
593 {
594 	struct ena_que *que = &adapter->que[qid];
595 	struct ena_ring *tx_ring = que->tx_ring;
596 	int size, i, err;
597 #ifdef	RSS
598 	cpuset_t cpu_mask;
599 #endif
600 
601 	size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
602 
603 	tx_ring->tx_buffer_info = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
604 	if (unlikely(tx_ring->tx_buffer_info == NULL))
605 		return (ENOMEM);
606 
607 	size = sizeof(uint16_t) * tx_ring->ring_size;
608 	tx_ring->free_tx_ids = malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO);
609 	if (unlikely(tx_ring->free_tx_ids == NULL))
610 		goto err_buf_info_free;
611 
612 	/* Req id stack for TX OOO completions */
613 	for (i = 0; i < tx_ring->ring_size; i++)
614 		tx_ring->free_tx_ids[i] = i;
615 
616 	/* Reset TX statistics. */
617 	ena_reset_counters((struct evcnt *)&tx_ring->tx_stats,
618 	    sizeof(tx_ring->tx_stats));
619 
620 	tx_ring->next_to_use = 0;
621 	tx_ring->next_to_clean = 0;
622 
623 	/* Make sure that drbr is empty */
624 	ENA_RING_MTX_LOCK(tx_ring);
625 	drbr_flush(adapter->ifp, tx_ring->br);
626 	ENA_RING_MTX_UNLOCK(tx_ring);
627 
628 	/* ... and create the buffer DMA maps */
629 	for (i = 0; i < tx_ring->ring_size; i++) {
630 		err = bus_dmamap_create(adapter->sc_dmat,
631 		    ENA_TSO_MAXSIZE, adapter->max_tx_sgl_size - 1,
632 		    ENA_TSO_MAXSIZE, 0, 0,
633 		    &tx_ring->tx_buffer_info[i].map);
634 		if (unlikely(err != 0)) {
635 			ena_trace(ENA_ALERT,
636 			     "Unable to create Tx DMA map for buffer %d\n", i);
637 			goto err_buf_info_unmap;
638 		}
639 	}
640 
641 	/* Allocate workqueues */
642 	int rc = workqueue_create(&tx_ring->enqueue_tq, "ena_tx_enq",
643 	    ena_deferred_mq_start, tx_ring, 0, IPL_NET, WQ_PERCPU | WQ_MPSAFE);
644 	if (unlikely(rc != 0)) {
645 		ena_trace(ENA_ALERT,
646 		    "Unable to create workqueue for enqueue task\n");
647 		i = tx_ring->ring_size;
648 		goto err_buf_info_unmap;
649 	}
650 
651 #if 0
652 	/* RSS set cpu for thread */
653 #ifdef RSS
654 	CPU_SETOF(que->cpu, &cpu_mask);
655 	taskqueue_start_threads_cpuset(&tx_ring->enqueue_tq, 1, IPL_NET,
656 	    &cpu_mask, "%s tx_ring enq (bucket %d)",
657 	    device_xname(adapter->pdev), que->cpu);
658 #else /* RSS */
659 	taskqueue_start_threads(&tx_ring->enqueue_tq, 1, IPL_NET,
660 	    "%s txeq %d", device_xname(adapter->pdev), que->cpu);
661 #endif /* RSS */
662 #endif
663 
664 	return (0);
665 
666 err_buf_info_unmap:
667 	while (i--) {
668 		bus_dmamap_destroy(adapter->sc_dmat,
669 		    tx_ring->tx_buffer_info[i].map);
670 	}
671 	free(tx_ring->free_tx_ids, M_DEVBUF);
672 	tx_ring->free_tx_ids = NULL;
673 err_buf_info_free:
674 	free(tx_ring->tx_buffer_info, M_DEVBUF);
675 	tx_ring->tx_buffer_info = NULL;
676 
677 	return (ENOMEM);
678 }
679 
680 /**
681  * ena_free_tx_resources - Free Tx Resources per Queue
682  * @adapter: network interface device structure
683  * @qid: queue index
684  *
685  * Free all transmit software resources
686  **/
687 static void
688 ena_free_tx_resources(struct ena_adapter *adapter, int qid)
689 {
690 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
691 
692 	workqueue_wait(tx_ring->enqueue_tq, &tx_ring->enqueue_task);
693 	workqueue_destroy(tx_ring->enqueue_tq);
694 	tx_ring->enqueue_tq = NULL;
695 
696 	ENA_RING_MTX_LOCK(tx_ring);
697 	/* Flush buffer ring, */
698 	drbr_flush(adapter->ifp, tx_ring->br);
699 
700 	/* Free buffer DMA maps, */
701 	for (int i = 0; i < tx_ring->ring_size; i++) {
702 		m_freem(tx_ring->tx_buffer_info[i].mbuf);
703 		tx_ring->tx_buffer_info[i].mbuf = NULL;
704 		bus_dmamap_unload(adapter->sc_dmat,
705 		    tx_ring->tx_buffer_info[i].map);
706 		bus_dmamap_destroy(adapter->sc_dmat,
707 		    tx_ring->tx_buffer_info[i].map);
708 	}
709 	ENA_RING_MTX_UNLOCK(tx_ring);
710 
711 	/* And free allocated memory. */
712 	free(tx_ring->tx_buffer_info, M_DEVBUF);
713 	tx_ring->tx_buffer_info = NULL;
714 
715 	free(tx_ring->free_tx_ids, M_DEVBUF);
716 	tx_ring->free_tx_ids = NULL;
717 }
718 
719 /**
720  * ena_setup_all_tx_resources - allocate all queues Tx resources
721  * @adapter: network interface device structure
722  *
723  * Returns 0 on success, otherwise on failure.
724  **/
725 static int
726 ena_setup_all_tx_resources(struct ena_adapter *adapter)
727 {
728 	int i, rc;
729 
730 	for (i = 0; i < adapter->num_queues; i++) {
731 		rc = ena_setup_tx_resources(adapter, i);
732 		if (rc != 0) {
733 			device_printf(adapter->pdev,
734 			    "Allocation for Tx Queue %u failed\n", i);
735 			goto err_setup_tx;
736 		}
737 	}
738 
739 	return (0);
740 
741 err_setup_tx:
742 	/* Rewind the index freeing the rings as we go */
743 	while (i--)
744 		ena_free_tx_resources(adapter, i);
745 	return (rc);
746 }
747 
748 /**
749  * ena_free_all_tx_resources - Free Tx Resources for All Queues
750  * @adapter: network interface device structure
751  *
752  * Free all transmit software resources
753  **/
754 static void
755 ena_free_all_tx_resources(struct ena_adapter *adapter)
756 {
757 	int i;
758 
759 	for (i = 0; i < adapter->num_queues; i++)
760 		ena_free_tx_resources(adapter, i);
761 }
762 
763 static inline int
764 validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
765 {
766 	if (likely(req_id < rx_ring->ring_size))
767 		return (0);
768 
769 	device_printf(rx_ring->adapter->pdev, "Invalid rx req_id: %hu\n",
770 	    req_id);
771 	counter_u64_add(rx_ring->rx_stats.bad_req_id, 1);
772 
773 	/* Trigger device reset */
774 	rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
775 	rx_ring->adapter->trigger_reset = true;
776 
777 	return (EFAULT);
778 }
779 
780 /**
781  * ena_setup_rx_resources - allocate Rx resources (Descriptors)
782  * @adapter: network interface device structure
783  * @qid: queue index
784  *
785  * Returns 0 on success, otherwise on failure.
786  **/
787 static int
788 ena_setup_rx_resources(struct ena_adapter *adapter, unsigned int qid)
789 {
790 	struct ena_que *que = &adapter->que[qid];
791 	struct ena_ring *rx_ring = que->rx_ring;
792 	int size, err, i;
793 #ifdef	RSS
794 	cpuset_t cpu_mask;
795 #endif
796 
797 	size = sizeof(struct ena_rx_buffer) * rx_ring->ring_size;
798 
799 	/*
800 	 * Alloc extra element so in rx path
801 	 * we can always prefetch rx_info + 1
802 	 */
803 	size += sizeof(struct ena_rx_buffer);
804 
805 	rx_ring->rx_buffer_info = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
806 
807 	size = sizeof(uint16_t) * rx_ring->ring_size;
808 	rx_ring->free_rx_ids = malloc(size, M_DEVBUF, M_WAITOK);
809 
810 	for (i = 0; i < rx_ring->ring_size; i++)
811 		rx_ring->free_rx_ids[i] = i;
812 
813 	/* Reset RX statistics. */
814 	ena_reset_counters((struct evcnt *)&rx_ring->rx_stats,
815 	    sizeof(rx_ring->rx_stats));
816 
817 	rx_ring->next_to_clean = 0;
818 	rx_ring->next_to_use = 0;
819 
820 	/* ... and create the buffer DMA maps */
821 	for (i = 0; i < rx_ring->ring_size; i++) {
822 		err = bus_dmamap_create(adapter->sc_dmat,
823 		    MJUM16BYTES, adapter->max_rx_sgl_size, MJUM16BYTES,
824 		    0, 0,
825 		    &(rx_ring->rx_buffer_info[i].map));
826 		if (err != 0) {
827 			ena_trace(ENA_ALERT,
828 			    "Unable to create Rx DMA map for buffer %d\n", i);
829 			goto err_buf_info_unmap;
830 		}
831 	}
832 
833 #ifdef LRO
834 	/* Create LRO for the ring */
835 	if ((adapter->ifp->if_capenable & IFCAP_LRO) != 0) {
836 		int err = tcp_lro_init(&rx_ring->lro);
837 		if (err != 0) {
838 			device_printf(adapter->pdev,
839 			    "LRO[%d] Initialization failed!\n", qid);
840 		} else {
841 			ena_trace(ENA_INFO,
842 			    "RX Soft LRO[%d] Initialized\n", qid);
843 			rx_ring->lro.ifp = adapter->ifp;
844 		}
845 	}
846 #endif
847 
848 	/* Allocate workqueues */
849 	int rc = workqueue_create(&rx_ring->cmpl_tq, "ena_rx_comp",
850 	    ena_deferred_rx_cleanup, rx_ring, 0, IPL_NET, WQ_PERCPU | WQ_MPSAFE);
851 	if (unlikely(rc != 0)) {
852 		ena_trace(ENA_ALERT,
853 		    "Unable to create workqueue for RX completion task\n");
854 		goto err_buf_info_unmap;
855 	}
856 
857 #if 0
858 	/* RSS set cpu for thread */
859 #ifdef RSS
860 	CPU_SETOF(que->cpu, &cpu_mask);
861 	taskqueue_start_threads_cpuset(&rx_ring->cmpl_tq, 1, IPL_NET, &cpu_mask,
862 	    "%s rx_ring cmpl (bucket %d)",
863 	    device_xname(adapter->pdev), que->cpu);
864 #else
865 	taskqueue_start_threads(&rx_ring->cmpl_tq, 1, IPL_NET,
866 	    "%s rx_ring cmpl %d", device_xname(adapter->pdev), que->cpu);
867 #endif
868 #endif
869 
870 	return (0);
871 
872 err_buf_info_unmap:
873 	while (i--) {
874 		bus_dmamap_destroy(adapter->sc_dmat,
875 		    rx_ring->rx_buffer_info[i].map);
876 	}
877 
878 	free(rx_ring->free_rx_ids, M_DEVBUF);
879 	rx_ring->free_rx_ids = NULL;
880 	free(rx_ring->rx_buffer_info, M_DEVBUF);
881 	rx_ring->rx_buffer_info = NULL;
882 	return (ENOMEM);
883 }
884 
885 /**
886  * ena_free_rx_resources - Free Rx Resources
887  * @adapter: network interface device structure
888  * @qid: queue index
889  *
890  * Free all receive software resources
891  **/
892 static void
893 ena_free_rx_resources(struct ena_adapter *adapter, unsigned int qid)
894 {
895 	struct ena_ring *rx_ring = &adapter->rx_ring[qid];
896 
897 	workqueue_wait(rx_ring->cmpl_tq, &rx_ring->cmpl_task);
898 	workqueue_destroy(rx_ring->cmpl_tq);
899 	rx_ring->cmpl_tq = NULL;
900 
901 	/* Free buffer DMA maps, */
902 	for (int i = 0; i < rx_ring->ring_size; i++) {
903 		m_freem(rx_ring->rx_buffer_info[i].mbuf);
904 		rx_ring->rx_buffer_info[i].mbuf = NULL;
905 		bus_dmamap_unload(adapter->sc_dmat,
906 		    rx_ring->rx_buffer_info[i].map);
907 		bus_dmamap_destroy(adapter->sc_dmat,
908 		    rx_ring->rx_buffer_info[i].map);
909 	}
910 
911 #ifdef LRO
912 	/* free LRO resources, */
913 	tcp_lro_free(&rx_ring->lro);
914 #endif
915 
916 	/* free allocated memory */
917 	free(rx_ring->rx_buffer_info, M_DEVBUF);
918 	rx_ring->rx_buffer_info = NULL;
919 
920 	free(rx_ring->free_rx_ids, M_DEVBUF);
921 	rx_ring->free_rx_ids = NULL;
922 }
923 
924 /**
925  * ena_setup_all_rx_resources - allocate all queues Rx resources
926  * @adapter: network interface device structure
927  *
928  * Returns 0 on success, otherwise on failure.
929  **/
930 static int
931 ena_setup_all_rx_resources(struct ena_adapter *adapter)
932 {
933 	int i, rc = 0;
934 
935 	for (i = 0; i < adapter->num_queues; i++) {
936 		rc = ena_setup_rx_resources(adapter, i);
937 		if (rc != 0) {
938 			device_printf(adapter->pdev,
939 			    "Allocation for Rx Queue %u failed\n", i);
940 			goto err_setup_rx;
941 		}
942 	}
943 	return (0);
944 
945 err_setup_rx:
946 	/* rewind the index freeing the rings as we go */
947 	while (i--)
948 		ena_free_rx_resources(adapter, i);
949 	return (rc);
950 }
951 
952 /**
953  * ena_free_all_rx_resources - Free Rx resources for all queues
954  * @adapter: network interface device structure
955  *
956  * Free all receive software resources
957  **/
958 static void
959 ena_free_all_rx_resources(struct ena_adapter *adapter)
960 {
961 	int i;
962 
963 	for (i = 0; i < adapter->num_queues; i++)
964 		ena_free_rx_resources(adapter, i);
965 }
966 
967 static inline int
968 ena_alloc_rx_mbuf(struct ena_adapter *adapter,
969     struct ena_ring *rx_ring, struct ena_rx_buffer *rx_info)
970 {
971 	struct ena_com_buf *ena_buf;
972 	int error;
973 	int mlen;
974 
975 	/* if previous allocated frag is not used */
976 	if (unlikely(rx_info->mbuf != NULL))
977 		return (0);
978 
979 	/* Get mbuf using UMA allocator */
980 	rx_info->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM16BYTES);
981 
982 	if (unlikely(rx_info->mbuf == NULL)) {
983 		counter_u64_add(rx_ring->rx_stats.mjum_alloc_fail, 1);
984 		rx_info->mbuf = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
985 		if (unlikely(rx_info->mbuf == NULL)) {
986 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
987 			return (ENOMEM);
988 		}
989 		mlen = MCLBYTES;
990 	} else {
991 		mlen = MJUM16BYTES;
992 	}
993 	/* Set mbuf length*/
994 	rx_info->mbuf->m_pkthdr.len = rx_info->mbuf->m_len = mlen;
995 
996 	/* Map packets for DMA */
997 	ena_trace(ENA_DBG | ENA_RSC | ENA_RXPTH,
998 	    "Using tag %p for buffers' DMA mapping, mbuf %p len: %d",
999 	    adapter->sc_dmat,rx_info->mbuf, rx_info->mbuf->m_len);
1000 	error = bus_dmamap_load_mbuf(adapter->sc_dmat, rx_info->map,
1001 	    rx_info->mbuf, BUS_DMA_NOWAIT);
1002 	if (unlikely((error != 0) || (rx_info->map->dm_nsegs != 1))) {
1003 		ena_trace(ENA_WARNING, "failed to map mbuf, error: %d, "
1004 		    "nsegs: %d\n", error, rx_info->map->dm_nsegs);
1005 		counter_u64_add(rx_ring->rx_stats.dma_mapping_err, 1);
1006 		goto exit;
1007 
1008 	}
1009 
1010 	bus_dmamap_sync(adapter->sc_dmat, rx_info->map, 0,
1011 	    rx_info->map->dm_mapsize, BUS_DMASYNC_PREREAD);
1012 
1013 	ena_buf = &rx_info->ena_buf;
1014 	ena_buf->paddr = rx_info->map->dm_segs[0].ds_addr;
1015 	ena_buf->len = mlen;
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 		ena_trace(ENA_WARNING, "Trying to free unallocated buffer\n");
1036 		return;
1037 	}
1038 
1039 	bus_dmamap_unload(adapter->sc_dmat, rx_info->map);
1040 	m_freem(rx_info->mbuf);
1041 	rx_info->mbuf = NULL;
1042 }
1043 
1044 /**
1045  * ena_refill_rx_bufs - Refills ring with descriptors
1046  * @rx_ring: the ring which we want to feed with free descriptors
1047  * @num: number of descriptors to refill
1048  * Refills the ring with newly allocated DMA-mapped mbufs for receiving
1049  **/
1050 static int
1051 ena_refill_rx_bufs(struct ena_ring *rx_ring, uint32_t num)
1052 {
1053 	struct ena_adapter *adapter = rx_ring->adapter;
1054 	uint16_t next_to_use, req_id;
1055 	uint32_t i;
1056 	int rc;
1057 
1058 	ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC, "refill qid: %d",
1059 	    rx_ring->qid);
1060 
1061 	next_to_use = rx_ring->next_to_use;
1062 
1063 	for (i = 0; i < num; i++) {
1064 		struct ena_rx_buffer *rx_info;
1065 
1066 		ena_trace(ENA_DBG | ENA_RXPTH | ENA_RSC,
1067 		    "RX buffer - next to use: %d", next_to_use);
1068 
1069 		req_id = rx_ring->free_rx_ids[next_to_use];
1070 		rc = validate_rx_req_id(rx_ring, req_id);
1071 		if (unlikely(rc != 0))
1072 			break;
1073 
1074 		rx_info = &rx_ring->rx_buffer_info[req_id];
1075 
1076 		rc = ena_alloc_rx_mbuf(adapter, rx_ring, rx_info);
1077 		if (unlikely(rc != 0)) {
1078 			ena_trace(ENA_WARNING,
1079 			    "failed to alloc buffer for rx queue %d\n",
1080 			    rx_ring->qid);
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 != 0)) {
1086 			ena_trace(ENA_WARNING,
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 (unlikely(i < num)) {
1096 		counter_u64_add(rx_ring->rx_stats.refil_partial, 1);
1097 		ena_trace(ENA_WARNING,
1098 		     "refilled rx qid %d with only %d mbufs (from %d)\n",
1099 		     rx_ring->qid, i, num);
1100 	}
1101 
1102 	if (likely(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 != NULL)
1120 			ena_free_rx_mbuf(adapter, rx_ring, rx_info);
1121 	}
1122 }
1123 
1124 /**
1125  * ena_refill_all_rx_bufs - allocate all queues Rx buffers
1126  * @adapter: network interface device structure
1127  *
1128  */
1129 static void
1130 ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1131 {
1132 	struct ena_ring *rx_ring;
1133 	int i, rc, bufs_num;
1134 
1135 	for (i = 0; i < adapter->num_queues; i++) {
1136 		rx_ring = &adapter->rx_ring[i];
1137 		bufs_num = rx_ring->ring_size - 1;
1138 		rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1139 
1140 		if (unlikely(rc != bufs_num))
1141 			ena_trace(ENA_WARNING, "refilling Queue %d failed. "
1142 			    "Allocated %d buffers from: %d\n", i, rc, bufs_num);
1143 	}
1144 }
1145 
1146 static void
1147 ena_free_all_rx_bufs(struct ena_adapter *adapter)
1148 {
1149 	int i;
1150 
1151 	for (i = 0; i < adapter->num_queues; i++)
1152 		ena_free_rx_bufs(adapter, i);
1153 }
1154 
1155 /**
1156  * ena_free_tx_bufs - Free Tx Buffers per Queue
1157  * @adapter: network interface device structure
1158  * @qid: queue index
1159  **/
1160 static void
1161 ena_free_tx_bufs(struct ena_adapter *adapter, unsigned int qid)
1162 {
1163 	bool print_once = true;
1164 	struct ena_ring *tx_ring = &adapter->tx_ring[qid];
1165 
1166 	ENA_RING_MTX_LOCK(tx_ring);
1167 	for (int i = 0; i < tx_ring->ring_size; i++) {
1168 		struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1169 
1170 		if (tx_info->mbuf == NULL)
1171 			continue;
1172 
1173 		if (print_once) {
1174 			device_printf(adapter->pdev,
1175 			    "free uncompleted tx mbuf qid %d idx 0x%x",
1176 			    qid, i);
1177 			print_once = false;
1178 		} else {
1179 			ena_trace(ENA_DBG,
1180 			    "free uncompleted tx mbuf qid %d idx 0x%x",
1181 			     qid, i);
1182 		}
1183 
1184 		bus_dmamap_unload(adapter->sc_dmat, tx_info->map);
1185 		m_free(tx_info->mbuf);
1186 		tx_info->mbuf = NULL;
1187 	}
1188 	ENA_RING_MTX_UNLOCK(tx_ring);
1189 }
1190 
1191 static void
1192 ena_free_all_tx_bufs(struct ena_adapter *adapter)
1193 {
1194 
1195 	for (int i = 0; i < adapter->num_queues; i++)
1196 		ena_free_tx_bufs(adapter, i);
1197 }
1198 
1199 static void
1200 ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1201 {
1202 	uint16_t ena_qid;
1203 	int i;
1204 
1205 	for (i = 0; i < adapter->num_queues; i++) {
1206 		ena_qid = ENA_IO_TXQ_IDX(i);
1207 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1208 	}
1209 }
1210 
1211 static void
1212 ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1213 {
1214 	uint16_t ena_qid;
1215 	int i;
1216 
1217 	for (i = 0; i < adapter->num_queues; i++) {
1218 		ena_qid = ENA_IO_RXQ_IDX(i);
1219 		ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1220 	}
1221 }
1222 
1223 static void
1224 ena_destroy_all_io_queues(struct ena_adapter *adapter)
1225 {
1226 	ena_destroy_all_tx_queues(adapter);
1227 	ena_destroy_all_rx_queues(adapter);
1228 }
1229 
1230 static inline int
1231 validate_tx_req_id(struct ena_ring *tx_ring, uint16_t req_id)
1232 {
1233 	struct ena_adapter *adapter = tx_ring->adapter;
1234 	struct ena_tx_buffer *tx_info = NULL;
1235 
1236 	if (likely(req_id < tx_ring->ring_size)) {
1237 		tx_info = &tx_ring->tx_buffer_info[req_id];
1238 		if (tx_info->mbuf != NULL)
1239 			return (0);
1240 	}
1241 
1242 	if (tx_info->mbuf == NULL)
1243 		device_printf(adapter->pdev,
1244 		    "tx_info doesn't have valid mbuf\n");
1245 	else
1246 		device_printf(adapter->pdev, "Invalid req_id: %hu\n", req_id);
1247 
1248 	counter_u64_add(tx_ring->tx_stats.bad_req_id, 1);
1249 
1250 	return (EFAULT);
1251 }
1252 
1253 static int
1254 ena_create_io_queues(struct ena_adapter *adapter)
1255 {
1256 	struct ena_com_dev *ena_dev = adapter->ena_dev;
1257 	struct ena_com_create_io_ctx ctx;
1258 	struct ena_ring *ring;
1259 	uint16_t ena_qid;
1260 	uint32_t msix_vector;
1261 	int rc, i;
1262 
1263 	/* Create TX queues */
1264 	for (i = 0; i < adapter->num_queues; i++) {
1265 		msix_vector = ENA_IO_IRQ_IDX(i);
1266 		ena_qid = ENA_IO_TXQ_IDX(i);
1267 		ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1268 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1269 		ctx.queue_size = adapter->tx_ring_size;
1270 		ctx.msix_vector = msix_vector;
1271 		ctx.qid = ena_qid;
1272 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1273 		if (rc != 0) {
1274 			device_printf(adapter->pdev,
1275 			    "Failed to create io TX queue #%d rc: %d\n", i, rc);
1276 			goto err_tx;
1277 		}
1278 		ring = &adapter->tx_ring[i];
1279 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1280 		    &ring->ena_com_io_sq,
1281 		    &ring->ena_com_io_cq);
1282 		if (rc != 0) {
1283 			device_printf(adapter->pdev,
1284 			    "Failed to get TX queue handlers. TX queue num"
1285 			    " %d rc: %d\n", i, rc);
1286 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1287 			goto err_tx;
1288 		}
1289 	}
1290 
1291 	/* Create RX queues */
1292 	for (i = 0; i < adapter->num_queues; i++) {
1293 		msix_vector = ENA_IO_IRQ_IDX(i);
1294 		ena_qid = ENA_IO_RXQ_IDX(i);
1295 		ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1296 		ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1297 		ctx.queue_size = adapter->rx_ring_size;
1298 		ctx.msix_vector = msix_vector;
1299 		ctx.qid = ena_qid;
1300 		rc = ena_com_create_io_queue(ena_dev, &ctx);
1301 		if (unlikely(rc != 0)) {
1302 			device_printf(adapter->pdev,
1303 			    "Failed to create io RX queue[%d] rc: %d\n", i, rc);
1304 			goto err_rx;
1305 		}
1306 
1307 		ring = &adapter->rx_ring[i];
1308 		rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1309 		    &ring->ena_com_io_sq,
1310 		    &ring->ena_com_io_cq);
1311 		if (unlikely(rc != 0)) {
1312 			device_printf(adapter->pdev,
1313 			    "Failed to get RX queue handlers. RX queue num"
1314 			    " %d rc: %d\n", i, rc);
1315 			ena_com_destroy_io_queue(ena_dev, ena_qid);
1316 			goto err_rx;
1317 		}
1318 	}
1319 
1320 	return (0);
1321 
1322 err_rx:
1323 	while (i--)
1324 		ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1325 	i = adapter->num_queues;
1326 err_tx:
1327 	while (i--)
1328 		ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1329 
1330 	return (ENXIO);
1331 }
1332 
1333 /**
1334  * ena_tx_cleanup - clear sent packets and corresponding descriptors
1335  * @tx_ring: ring for which we want to clean packets
1336  *
1337  * Once packets are sent, we ask the device in a loop for no longer used
1338  * descriptors. We find the related mbuf chain in a map (index in an array)
1339  * and free it, then update ring state.
1340  * This is performed in "endless" loop, updating ring pointers every
1341  * TX_COMMIT. The first check of free descriptor is performed before the actual
1342  * loop, then repeated at the loop end.
1343  **/
1344 static int
1345 ena_tx_cleanup(struct ena_ring *tx_ring)
1346 {
1347 	struct ena_adapter *adapter;
1348 	struct ena_com_io_cq* io_cq;
1349 	uint16_t next_to_clean;
1350 	uint16_t req_id;
1351 	uint16_t ena_qid;
1352 	unsigned int total_done = 0;
1353 	int rc;
1354 	int commit = TX_COMMIT;
1355 	int budget = TX_BUDGET;
1356 	int work_done;
1357 
1358 	adapter = tx_ring->que->adapter;
1359 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
1360 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1361 	next_to_clean = tx_ring->next_to_clean;
1362 
1363 	do {
1364 		struct ena_tx_buffer *tx_info;
1365 		struct mbuf *mbuf;
1366 
1367 		rc = ena_com_tx_comp_req_id_get(io_cq, &req_id);
1368 		if (unlikely(rc != 0))
1369 			break;
1370 
1371 		rc = validate_tx_req_id(tx_ring, req_id);
1372 		if (unlikely(rc != 0))
1373 			break;
1374 
1375 		tx_info = &tx_ring->tx_buffer_info[req_id];
1376 
1377 		mbuf = tx_info->mbuf;
1378 
1379 		tx_info->mbuf = NULL;
1380 		bintime_clear(&tx_info->timestamp);
1381 
1382 		if (likely(tx_info->num_of_bufs != 0)) {
1383 			/* Map is no longer required */
1384 			bus_dmamap_unload(adapter->sc_dmat, tx_info->map);
1385 		}
1386 
1387 		ena_trace(ENA_DBG | ENA_TXPTH, "tx: q %d mbuf %p completed",
1388 		    tx_ring->qid, mbuf);
1389 
1390 		m_freem(mbuf);
1391 
1392 		total_done += tx_info->tx_descs;
1393 
1394 		tx_ring->free_tx_ids[next_to_clean] = req_id;
1395 		next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1396 		    tx_ring->ring_size);
1397 
1398 		if (unlikely(--commit == 0)) {
1399 			commit = TX_COMMIT;
1400 			/* update ring state every TX_COMMIT descriptor */
1401 			tx_ring->next_to_clean = next_to_clean;
1402 			ena_com_comp_ack(
1403 			    &adapter->ena_dev->io_sq_queues[ena_qid],
1404 			    total_done);
1405 			ena_com_update_dev_comp_head(io_cq);
1406 			total_done = 0;
1407 		}
1408 	} while (likely(--budget));
1409 
1410 	work_done = TX_BUDGET - budget;
1411 
1412 	ena_trace(ENA_DBG | ENA_TXPTH, "tx: q %d done. total pkts: %d",
1413 	tx_ring->qid, work_done);
1414 
1415 	/* If there is still something to commit update ring state */
1416 	if (likely(commit != TX_COMMIT)) {
1417 		tx_ring->next_to_clean = next_to_clean;
1418 		ena_com_comp_ack(&adapter->ena_dev->io_sq_queues[ena_qid],
1419 		    total_done);
1420 		ena_com_update_dev_comp_head(io_cq);
1421 	}
1422 
1423 	if (atomic_cas_uint(&tx_ring->task_pending, 0, 1) == 0)
1424 		workqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task, NULL);
1425 
1426 	return (work_done);
1427 }
1428 
1429 #if 0
1430 static void
1431 ena_rx_hash_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1432     struct mbuf *mbuf)
1433 {
1434 	struct ena_adapter *adapter = rx_ring->adapter;
1435 
1436 	if (likely(adapter->rss_support)) {
1437 		mbuf->m_pkthdr.flowid = ena_rx_ctx->hash;
1438 
1439 		if (ena_rx_ctx->frag &&
1440 		    (ena_rx_ctx->l3_proto != ENA_ETH_IO_L3_PROTO_UNKNOWN)) {
1441 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1442 			return;
1443 		}
1444 
1445 		switch (ena_rx_ctx->l3_proto) {
1446 		case ENA_ETH_IO_L3_PROTO_IPV4:
1447 			switch (ena_rx_ctx->l4_proto) {
1448 			case ENA_ETH_IO_L4_PROTO_TCP:
1449 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV4);
1450 				break;
1451 			case ENA_ETH_IO_L4_PROTO_UDP:
1452 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV4);
1453 				break;
1454 			default:
1455 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV4);
1456 			}
1457 			break;
1458 		case ENA_ETH_IO_L3_PROTO_IPV6:
1459 			switch (ena_rx_ctx->l4_proto) {
1460 			case ENA_ETH_IO_L4_PROTO_TCP:
1461 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_TCP_IPV6);
1462 				break;
1463 			case ENA_ETH_IO_L4_PROTO_UDP:
1464 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_UDP_IPV6);
1465 				break;
1466 			default:
1467 				M_HASHTYPE_SET(mbuf, M_HASHTYPE_RSS_IPV6);
1468 			}
1469 			break;
1470 		case ENA_ETH_IO_L3_PROTO_UNKNOWN:
1471 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1472 			break;
1473 		default:
1474 			M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE_HASH);
1475 		}
1476 	} else {
1477 		mbuf->m_pkthdr.flowid = rx_ring->qid;
1478 		M_HASHTYPE_SET(mbuf, M_HASHTYPE_NONE);
1479 	}
1480 }
1481 #endif
1482 
1483 /**
1484  * ena_rx_mbuf - assemble mbuf from descriptors
1485  * @rx_ring: ring for which we want to clean packets
1486  * @ena_bufs: buffer info
1487  * @ena_rx_ctx: metadata for this packet(s)
1488  * @next_to_clean: ring pointer, will be updated only upon success
1489  *
1490  **/
1491 static struct mbuf*
1492 ena_rx_mbuf(struct ena_ring *rx_ring, struct ena_com_rx_buf_info *ena_bufs,
1493     struct ena_com_rx_ctx *ena_rx_ctx, uint16_t *next_to_clean)
1494 {
1495 	struct mbuf *mbuf;
1496 	struct ena_rx_buffer *rx_info;
1497 	struct ena_adapter *adapter;
1498 	unsigned int descs = ena_rx_ctx->descs;
1499 	uint16_t ntc, len, req_id, buf = 0;
1500 
1501 	ntc = *next_to_clean;
1502 	adapter = rx_ring->adapter;
1503 	rx_info = &rx_ring->rx_buffer_info[ntc];
1504 
1505 	if (unlikely(rx_info->mbuf == NULL)) {
1506 		device_printf(adapter->pdev, "NULL mbuf in rx_info");
1507 		return (NULL);
1508 	}
1509 
1510 	len = ena_bufs[buf].len;
1511 	req_id = ena_bufs[buf].req_id;
1512 	rx_info = &rx_ring->rx_buffer_info[req_id];
1513 
1514 	ena_trace(ENA_DBG | ENA_RXPTH, "rx_info %p, mbuf %p, paddr %jx",
1515 	    rx_info, rx_info->mbuf, (uintmax_t)rx_info->ena_buf.paddr);
1516 
1517 	mbuf = rx_info->mbuf;
1518 	KASSERT(mbuf->m_flags & M_PKTHDR);
1519 	mbuf->m_pkthdr.len = len;
1520 	mbuf->m_len = len;
1521 	m_set_rcvif(mbuf, rx_ring->que->adapter->ifp);
1522 
1523 	/* Fill mbuf with hash key and it's interpretation for optimization */
1524 #if 0
1525 	ena_rx_hash_mbuf(rx_ring, ena_rx_ctx, mbuf);
1526 #endif
1527 
1528 	ena_trace(ENA_DBG | ENA_RXPTH, "rx mbuf 0x%p, flags=0x%x, len: %d",
1529 	    mbuf, mbuf->m_flags, mbuf->m_pkthdr.len);
1530 
1531 	/* DMA address is not needed anymore, unmap it */
1532 	bus_dmamap_unload(rx_ring->adapter->sc_dmat, rx_info->map);
1533 
1534 	rx_info->mbuf = NULL;
1535 	rx_ring->free_rx_ids[ntc] = req_id;
1536 	ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1537 
1538 	/*
1539 	 * While we have more than 1 descriptors for one rcvd packet, append
1540 	 * other mbufs to the main one
1541 	 */
1542 	while (--descs) {
1543 		++buf;
1544 		len = ena_bufs[buf].len;
1545 		req_id = ena_bufs[buf].req_id;
1546 		rx_info = &rx_ring->rx_buffer_info[req_id];
1547 
1548 		if (unlikely(rx_info->mbuf == NULL)) {
1549 			device_printf(adapter->pdev, "NULL mbuf in rx_info");
1550 			/*
1551 			 * If one of the required mbufs was not allocated yet,
1552 			 * we can break there.
1553 			 * All earlier used descriptors will be reallocated
1554 			 * later and not used mbufs can be reused.
1555 			 * The next_to_clean pointer will not be updated in case
1556 			 * of an error, so caller should advance it manually
1557 			 * in error handling routine to keep it up to date
1558 			 * with hw ring.
1559 			 */
1560 			m_freem(mbuf);
1561 			return (NULL);
1562 		}
1563 
1564 		if (unlikely(m_append(mbuf, len, rx_info->mbuf->m_data) == 0)) {
1565 			counter_u64_add(rx_ring->rx_stats.mbuf_alloc_fail, 1);
1566 			ena_trace(ENA_WARNING, "Failed to append Rx mbuf %p",
1567 			    mbuf);
1568 		}
1569 
1570 		ena_trace(ENA_DBG | ENA_RXPTH,
1571 		    "rx mbuf updated. len %d", mbuf->m_pkthdr.len);
1572 
1573 		/* Free already appended mbuf, it won't be useful anymore */
1574 		bus_dmamap_unload(rx_ring->adapter->sc_dmat, rx_info->map);
1575 		m_freem(rx_info->mbuf);
1576 		rx_info->mbuf = NULL;
1577 
1578 		rx_ring->free_rx_ids[ntc] = req_id;
1579 		ntc = ENA_RX_RING_IDX_NEXT(ntc, rx_ring->ring_size);
1580 	}
1581 
1582 	*next_to_clean = ntc;
1583 
1584 	return (mbuf);
1585 }
1586 
1587 /**
1588  * ena_rx_checksum - indicate in mbuf if hw indicated a good cksum
1589  **/
1590 static inline void
1591 ena_rx_checksum(struct ena_ring *rx_ring, struct ena_com_rx_ctx *ena_rx_ctx,
1592     struct mbuf *mbuf)
1593 {
1594 
1595 	/* IPv4 */
1596 	if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) {
1597 		mbuf->m_pkthdr.csum_flags |= M_CSUM_IPv4;
1598 		if (ena_rx_ctx->l3_csum_err) {
1599 			/* ipv4 checksum error */
1600 			mbuf->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1601 			counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1602 			ena_trace(ENA_DBG, "RX IPv4 header checksum error");
1603 			return;
1604 		}
1605 
1606 		/*  TCP/UDP */
1607 		if ((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1608 		    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)) {
1609 			mbuf->m_pkthdr.csum_flags |= (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ? M_CSUM_TCPv4 : M_CSUM_UDPv4;
1610 			if (ena_rx_ctx->l4_csum_err) {
1611 				/* TCP/UDP checksum error */
1612 				mbuf->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1613 				counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1614 				ena_trace(ENA_DBG, "RX L4 checksum error");
1615 			}
1616 		}
1617 	}
1618 	/* IPv6 */
1619 	else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6) {
1620 		/*  TCP/UDP */
1621 		if ((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1622 		    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)) {
1623 			mbuf->m_pkthdr.csum_flags |= (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ? M_CSUM_TCPv6 : M_CSUM_UDPv6;
1624 			if (ena_rx_ctx->l4_csum_err) {
1625 				/* TCP/UDP checksum error */
1626 				mbuf->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1627 				counter_u64_add(rx_ring->rx_stats.bad_csum, 1);
1628 				ena_trace(ENA_DBG, "RX L4 checksum error");
1629 			}
1630 		}
1631 	}
1632 }
1633 
1634 static void
1635 ena_deferred_rx_cleanup(struct work *wk, void *arg)
1636 {
1637 	struct ena_ring *rx_ring = arg;
1638 	int budget = CLEAN_BUDGET;
1639 
1640 	atomic_swap_uint(&rx_ring->task_pending, 0);
1641 
1642 	ENA_RING_MTX_LOCK(rx_ring);
1643 	/*
1644 	 * If deferred task was executed, perform cleanup of all awaiting
1645 	 * descs (or until given budget is depleted to avoid infinite loop).
1646 	 */
1647 	while (likely(budget--)) {
1648 		if (ena_rx_cleanup(rx_ring) == 0)
1649 			break;
1650 	}
1651 	ENA_RING_MTX_UNLOCK(rx_ring);
1652 }
1653 
1654 /**
1655  * ena_rx_cleanup - handle rx irq
1656  * @arg: ring for which irq is being handled
1657  **/
1658 static int
1659 ena_rx_cleanup(struct ena_ring *rx_ring)
1660 {
1661 	struct ena_adapter *adapter;
1662 	struct mbuf *mbuf;
1663 	struct ena_com_rx_ctx ena_rx_ctx;
1664 	struct ena_com_io_cq* io_cq;
1665 	struct ena_com_io_sq* io_sq;
1666 	struct ifnet *ifp;
1667 	uint16_t ena_qid;
1668 	uint16_t next_to_clean;
1669 	uint32_t refill_required;
1670 	uint32_t refill_threshold;
1671 	uint32_t do_if_input = 0;
1672 	unsigned int qid;
1673 	int rc, i;
1674 	int budget = RX_BUDGET;
1675 
1676 	adapter = rx_ring->que->adapter;
1677 	ifp = adapter->ifp;
1678 	qid = rx_ring->que->id;
1679 	ena_qid = ENA_IO_RXQ_IDX(qid);
1680 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1681 	io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
1682 	next_to_clean = rx_ring->next_to_clean;
1683 
1684 	ena_trace(ENA_DBG, "rx: qid %d", qid);
1685 
1686 	do {
1687 		ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1688 		ena_rx_ctx.max_bufs = adapter->max_rx_sgl_size;
1689 		ena_rx_ctx.descs = 0;
1690 		rc = ena_com_rx_pkt(io_cq, io_sq, &ena_rx_ctx);
1691 
1692 		if (unlikely(rc != 0))
1693 			goto error;
1694 
1695 		if (unlikely(ena_rx_ctx.descs == 0))
1696 			break;
1697 
1698 		ena_trace(ENA_DBG | ENA_RXPTH, "rx: q %d got packet from ena. "
1699 		    "descs #: %d l3 proto %d l4 proto %d hash: %x",
1700 		    rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1701 		    ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1702 
1703 		/* Receive mbuf from the ring */
1704 		mbuf = ena_rx_mbuf(rx_ring, rx_ring->ena_bufs,
1705 		    &ena_rx_ctx, &next_to_clean);
1706 
1707 		/* Exit if we failed to retrieve a buffer */
1708 		if (unlikely(mbuf == NULL)) {
1709 			for (i = 0; i < ena_rx_ctx.descs; ++i) {
1710 				rx_ring->free_rx_ids[next_to_clean] =
1711 				    rx_ring->ena_bufs[i].req_id;
1712 				next_to_clean =
1713 				    ENA_RX_RING_IDX_NEXT(next_to_clean,
1714 				    rx_ring->ring_size);
1715 
1716 			}
1717 			break;
1718 		}
1719 
1720 		if (((ifp->if_capenable & IFCAP_CSUM_IPv4_Rx) != 0) ||
1721 		    ((ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx) != 0) ||
1722 		    ((ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx) != 0) ||
1723 		    ((ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx) != 0) ||
1724 		    ((ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx) != 0)) {
1725 			ena_rx_checksum(rx_ring, &ena_rx_ctx, mbuf);
1726 		}
1727 
1728 		counter_enter();
1729 		counter_u64_add_protected(rx_ring->rx_stats.bytes,
1730 		    mbuf->m_pkthdr.len);
1731 		counter_u64_add_protected(adapter->hw_stats.rx_bytes,
1732 		    mbuf->m_pkthdr.len);
1733 		counter_exit();
1734 		/*
1735 		 * LRO is only for IP/TCP packets and TCP checksum of the packet
1736 		 * should be computed by hardware.
1737 		 */
1738 		do_if_input = 1;
1739 #ifdef LRO
1740 		if (((ifp->if_capenable & IFCAP_LRO) != 0)  &&
1741 		    ((mbuf->m_pkthdr.csum_flags & CSUM_IP_VALID) != 0) &&
1742 		    (ena_rx_ctx.l4_proto == ENA_ETH_IO_L4_PROTO_TCP)) {
1743 			/*
1744 			 * Send to the stack if:
1745 			 *  - LRO not enabled, or
1746 			 *  - no LRO resources, or
1747 			 *  - lro enqueue fails
1748 			 */
1749 			if ((rx_ring->lro.lro_cnt != 0) &&
1750 			    (tcp_lro_rx(&rx_ring->lro, mbuf, 0) == 0))
1751 					do_if_input = 0;
1752 		}
1753 #endif
1754 		if (do_if_input != 0) {
1755 			ena_trace(ENA_DBG | ENA_RXPTH,
1756 			    "calling if_input() with mbuf %p", mbuf);
1757 			if_percpuq_enqueue(ifp->if_percpuq, mbuf);
1758 		}
1759 
1760 		counter_enter();
1761 		counter_u64_add_protected(rx_ring->rx_stats.cnt, 1);
1762 		counter_u64_add_protected(adapter->hw_stats.rx_packets, 1);
1763 		counter_exit();
1764 	} while (--budget);
1765 
1766 	rx_ring->next_to_clean = next_to_clean;
1767 
1768 	refill_required = ena_com_free_desc(io_sq);
1769 	refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
1770 
1771 	if (refill_required > refill_threshold) {
1772 		ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1773 		ena_refill_rx_bufs(rx_ring, refill_required);
1774 	}
1775 
1776 #ifdef LRO
1777 	tcp_lro_flush_all(&rx_ring->lro);
1778 #endif
1779 
1780 	return (RX_BUDGET - budget);
1781 
1782 error:
1783 	counter_u64_add(rx_ring->rx_stats.bad_desc_num, 1);
1784 	return (RX_BUDGET - budget);
1785 }
1786 
1787 /*********************************************************************
1788  *
1789  *  MSIX & Interrupt Service routine
1790  *
1791  **********************************************************************/
1792 
1793 /**
1794  * ena_handle_msix - MSIX Interrupt Handler for admin/async queue
1795  * @arg: interrupt number
1796  **/
1797 static int
1798 ena_intr_msix_mgmnt(void *arg)
1799 {
1800 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
1801 
1802 	ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1803 	if (likely(adapter->running))
1804 		ena_com_aenq_intr_handler(adapter->ena_dev, arg);
1805 
1806 	return 1;
1807 }
1808 
1809 /**
1810  * ena_handle_msix - MSIX Interrupt Handler for Tx/Rx
1811  * @arg: interrupt number
1812  **/
1813 static int
1814 ena_handle_msix(void *arg)
1815 {
1816 	struct ena_que	*que = arg;
1817 	struct ena_adapter *adapter = que->adapter;
1818 	struct ifnet *ifp = adapter->ifp;
1819 	struct ena_ring *tx_ring;
1820 	struct ena_ring *rx_ring;
1821 	struct ena_com_io_cq* io_cq;
1822 	struct ena_eth_io_intr_reg intr_reg;
1823 	int qid, ena_qid;
1824 	int txc, rxc, i;
1825 
1826 	if (unlikely((if_getdrvflags(ifp) & IFF_RUNNING) == 0))
1827 		return 0;
1828 
1829 	ena_trace(ENA_DBG, "MSI-X TX/RX routine");
1830 
1831 	tx_ring = que->tx_ring;
1832 	rx_ring = que->rx_ring;
1833 	qid = que->id;
1834 	ena_qid = ENA_IO_TXQ_IDX(qid);
1835 	io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
1836 
1837 	for (i = 0; i < CLEAN_BUDGET; ++i) {
1838 		/*
1839 		 * If lock cannot be acquired, then deferred cleanup task was
1840 		 * being executed and rx ring is being cleaned up in
1841 		 * another thread.
1842 		 */
1843 		if (likely(ENA_RING_MTX_TRYLOCK(rx_ring) != 0)) {
1844 			rxc = ena_rx_cleanup(rx_ring);
1845 			ENA_RING_MTX_UNLOCK(rx_ring);
1846 		} else {
1847 			rxc = 0;
1848 		}
1849 
1850 		/* Protection from calling ena_tx_cleanup from ena_start_xmit */
1851 		ENA_RING_MTX_LOCK(tx_ring);
1852 		txc = ena_tx_cleanup(tx_ring);
1853 		ENA_RING_MTX_UNLOCK(tx_ring);
1854 
1855 		if (unlikely((if_getdrvflags(ifp) & IFF_RUNNING) == 0))
1856 			return 0;
1857 
1858 		if ((txc != TX_BUDGET) && (rxc != RX_BUDGET))
1859 		       break;
1860 	}
1861 
1862 	/* Signal that work is done and unmask interrupt */
1863 	ena_com_update_intr_reg(&intr_reg,
1864 	    RX_IRQ_INTERVAL,
1865 	    TX_IRQ_INTERVAL,
1866 	    true);
1867 	ena_com_unmask_intr(io_cq, &intr_reg);
1868 
1869 	return 1;
1870 }
1871 
1872 static int
1873 ena_enable_msix(struct ena_adapter *adapter)
1874 {
1875 	int msix_req;
1876 	int counts[PCI_INTR_TYPE_SIZE];
1877 	int max_type;
1878 
1879 	/* Reserved the max msix vectors we might need */
1880 	msix_req = ENA_MAX_MSIX_VEC(adapter->num_queues);
1881 
1882 	counts[PCI_INTR_TYPE_INTX] = 0;
1883 	counts[PCI_INTR_TYPE_MSI] = 0;
1884 	counts[PCI_INTR_TYPE_MSIX] = msix_req;
1885 	max_type = PCI_INTR_TYPE_MSIX;
1886 
1887 	if (pci_intr_alloc(&adapter->sc_pa, &adapter->sc_intrs, counts,
1888 	    max_type) != 0) {
1889 		aprint_error_dev(adapter->pdev,
1890 		    "failed to allocate interrupt\n");
1891 		return ENOSPC;
1892 	}
1893 
1894 	adapter->sc_nintrs = counts[PCI_INTR_TYPE_MSIX];
1895 
1896 	if (counts[PCI_INTR_TYPE_MSIX] != msix_req) {
1897 		device_printf(adapter->pdev,
1898 		    "Enable only %d MSI-x (out of %d), reduce "
1899 		    "the number of queues\n", adapter->sc_nintrs, msix_req);
1900 		adapter->num_queues = adapter->sc_nintrs - ENA_ADMIN_MSIX_VEC;
1901 	}
1902 
1903 	return 0;
1904 }
1905 
1906 #if 0
1907 static void
1908 ena_setup_io_intr(struct ena_adapter *adapter)
1909 {
1910 	static int last_bind_cpu = -1;
1911 	int irq_idx;
1912 
1913 	for (int i = 0; i < adapter->num_queues; i++) {
1914 		irq_idx = ENA_IO_IRQ_IDX(i);
1915 
1916 		snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1917 		    "%s-TxRx-%d", device_xname(adapter->pdev), i);
1918 		adapter->irq_tbl[irq_idx].handler = ena_handle_msix;
1919 		adapter->irq_tbl[irq_idx].data = &adapter->que[i];
1920 		adapter->irq_tbl[irq_idx].vector =
1921 		    adapter->msix_entries[irq_idx].vector;
1922 		ena_trace(ENA_INFO | ENA_IOQ, "ena_setup_io_intr vector: %d\n",
1923 		    adapter->msix_entries[irq_idx].vector);
1924 #ifdef	RSS
1925 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1926 		    rss_getcpu(i % rss_getnumbuckets());
1927 #else
1928 		/*
1929 		 * We still want to bind rings to the corresponding cpu
1930 		 * using something similar to the RSS round-robin technique.
1931 		 */
1932 		if (unlikely(last_bind_cpu < 0))
1933 			last_bind_cpu = CPU_FIRST();
1934 		adapter->que[i].cpu = adapter->irq_tbl[irq_idx].cpu =
1935 		    last_bind_cpu;
1936 		last_bind_cpu = CPU_NEXT(last_bind_cpu);
1937 #endif
1938 	}
1939 }
1940 #endif
1941 
1942 static int
1943 ena_request_mgmnt_irq(struct ena_adapter *adapter)
1944 {
1945 	const char *intrstr;
1946 	char intrbuf[PCI_INTRSTR_LEN];
1947 	char intr_xname[INTRDEVNAMEBUF];
1948 	pci_chipset_tag_t pc = adapter->sc_pa.pa_pc;
1949 	const int irq_slot = ENA_MGMNT_IRQ_IDX;
1950 
1951 	KASSERT(adapter->sc_intrs != NULL);
1952 	KASSERT(adapter->sc_ihs[irq_slot] == NULL);
1953 
1954 	snprintf(intr_xname, sizeof(intr_xname), "%s mgmnt",
1955 	    device_xname(adapter->pdev));
1956 	intrstr = pci_intr_string(pc, adapter->sc_intrs[irq_slot],
1957 	    intrbuf, sizeof(intrbuf));
1958 
1959 	adapter->sc_ihs[irq_slot] = pci_intr_establish_xname(
1960 	    pc, adapter->sc_intrs[irq_slot],
1961 	    IPL_NET, ena_intr_msix_mgmnt, adapter, intr_xname);
1962 
1963 	if (adapter->sc_ihs[irq_slot] == NULL) {
1964 		device_printf(adapter->pdev, "failed to register "
1965 		    "interrupt handler for MGMNT irq %s\n",
1966 		    intrstr);
1967 		return ENOMEM;
1968 	}
1969 
1970 	aprint_normal_dev(adapter->pdev,
1971 	    "for MGMNT interrupting at %s\n", intrstr);
1972 
1973 	return 0;
1974 }
1975 
1976 static int
1977 ena_request_io_irq(struct ena_adapter *adapter)
1978 {
1979 	const char *intrstr;
1980 	char intrbuf[PCI_INTRSTR_LEN];
1981 	char intr_xname[INTRDEVNAMEBUF];
1982 	pci_chipset_tag_t pc = adapter->sc_pa.pa_pc;
1983 	const int irq_off = ENA_IO_IRQ_FIRST_IDX;
1984 	void *vih;
1985 	kcpuset_t *affinity;
1986 	int i;
1987 
1988 	KASSERT(adapter->sc_intrs != NULL);
1989 
1990 	kcpuset_create(&affinity, false);
1991 
1992 	for (i = 0; i < adapter->num_queues; i++) {
1993 		int irq_slot = i + irq_off;
1994 		int affinity_to = (irq_slot) % ncpu;
1995 
1996 		KASSERT((void *)adapter->sc_intrs[irq_slot] != NULL);
1997 		KASSERT(adapter->sc_ihs[irq_slot] == NULL);
1998 
1999 		snprintf(intr_xname, sizeof(intr_xname), "%s ioq%d",
2000 		    device_xname(adapter->pdev), i);
2001 		intrstr = pci_intr_string(pc, adapter->sc_intrs[irq_slot],
2002 		    intrbuf, sizeof(intrbuf));
2003 
2004 		vih = pci_intr_establish_xname(adapter->sc_pa.pa_pc,
2005 		    adapter->sc_intrs[irq_slot], IPL_NET,
2006 		    ena_handle_msix, &adapter->que[i], intr_xname);
2007 
2008 		if (adapter->sc_ihs[ENA_MGMNT_IRQ_IDX] == NULL) {
2009 			device_printf(adapter->pdev, "failed to register "
2010 			    "interrupt handler for IO queue %d irq %s\n",
2011 			    i, intrstr);
2012 			goto err;
2013 		}
2014 
2015 		kcpuset_zero(affinity);
2016 		/* Round-robin affinity */
2017 		kcpuset_set(affinity, affinity_to);
2018 		int error = interrupt_distribute(vih, affinity, NULL);
2019 		if (error == 0) {
2020 			aprint_normal_dev(adapter->pdev,
2021 			    "for IO queue %d interrupting at %s"
2022 			    " affinity to %u\n", i, intrstr, affinity_to);
2023 		} else {
2024 			aprint_normal_dev(adapter->pdev,
2025 			    "for IO queue %d interrupting at %s\n", i, intrstr);
2026 		}
2027 
2028 		adapter->sc_ihs[irq_slot] = vih;
2029 
2030 #ifdef	RSS
2031 		ena_trace(ENA_INFO, "queue %d - RSS bucket %d\n",
2032 		    i - ENA_IO_IRQ_FIRST_IDX, irq->cpu);
2033 #else
2034 		ena_trace(ENA_INFO, "queue %d - cpu %d\n",
2035 		    i - ENA_IO_IRQ_FIRST_IDX, affinity_to);
2036 #endif
2037 	}
2038 
2039 	kcpuset_destroy(affinity);
2040 	return 0;
2041 
2042 err:
2043 	kcpuset_destroy(affinity);
2044 
2045 	for (i--; i >= 0; i--) {
2046 #if defined(DEBUG) || defined(DIAGNOSTIC)
2047 		int irq_slot = i + irq_off;
2048 #endif
2049 		KASSERT(adapter->sc_ihs[irq_slot] != NULL);
2050 		pci_intr_disestablish(adapter->sc_pa.pa_pc, adapter->sc_ihs[i]);
2051 		adapter->sc_ihs[i] = NULL;
2052 	}
2053 
2054 	return ENOSPC;
2055 }
2056 
2057 static void
2058 ena_free_mgmnt_irq(struct ena_adapter *adapter)
2059 {
2060 	const int irq_slot = ENA_MGMNT_IRQ_IDX;
2061 
2062 	if (adapter->sc_ihs[irq_slot]) {
2063 		pci_intr_disestablish(adapter->sc_pa.pa_pc,
2064 		    adapter->sc_ihs[irq_slot]);
2065 		adapter->sc_ihs[irq_slot] = NULL;
2066 	}
2067 }
2068 
2069 static void
2070 ena_free_io_irq(struct ena_adapter *adapter)
2071 {
2072 	const int irq_off = ENA_IO_IRQ_FIRST_IDX;
2073 
2074 	for (int i = 0; i < adapter->num_queues; i++) {
2075 		int irq_slot = i + irq_off;
2076 
2077 		if (adapter->sc_ihs[irq_slot]) {
2078 			pci_intr_disestablish(adapter->sc_pa.pa_pc,
2079 			    adapter->sc_ihs[i]);
2080 			adapter->sc_ihs[i] = NULL;
2081 		}
2082 	}
2083 }
2084 
2085 static void
2086 ena_free_irqs(struct ena_adapter* adapter)
2087 {
2088 
2089 	ena_free_io_irq(adapter);
2090 	ena_free_mgmnt_irq(adapter);
2091 	ena_disable_msix(adapter);
2092 }
2093 
2094 static void
2095 ena_disable_msix(struct ena_adapter *adapter)
2096 {
2097 	pci_intr_release(adapter->sc_pa.pa_pc, adapter->sc_intrs,
2098 	    adapter->sc_nintrs);
2099 }
2100 
2101 static void
2102 ena_unmask_all_io_irqs(struct ena_adapter *adapter)
2103 {
2104 	struct ena_com_io_cq* io_cq;
2105 	struct ena_eth_io_intr_reg intr_reg;
2106 	uint16_t ena_qid;
2107 	int i;
2108 
2109 	/* Unmask interrupts for all queues */
2110 	for (i = 0; i < adapter->num_queues; i++) {
2111 		ena_qid = ENA_IO_TXQ_IDX(i);
2112 		io_cq = &adapter->ena_dev->io_cq_queues[ena_qid];
2113 		ena_com_update_intr_reg(&intr_reg, 0, 0, true);
2114 		ena_com_unmask_intr(io_cq, &intr_reg);
2115 	}
2116 }
2117 
2118 /* Configure the Rx forwarding */
2119 static int
2120 ena_rss_configure(struct ena_adapter *adapter)
2121 {
2122 	struct ena_com_dev *ena_dev = adapter->ena_dev;
2123 	int rc;
2124 
2125 	/* Set indirect table */
2126 	rc = ena_com_indirect_table_set(ena_dev);
2127 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2128 		return (rc);
2129 
2130 	/* Configure hash function (if supported) */
2131 	rc = ena_com_set_hash_function(ena_dev);
2132 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2133 		return (rc);
2134 
2135 	/* Configure hash inputs (if supported) */
2136 	rc = ena_com_set_hash_ctrl(ena_dev);
2137 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP)))
2138 		return (rc);
2139 
2140 	return (0);
2141 }
2142 
2143 static int
2144 ena_up_complete(struct ena_adapter *adapter)
2145 {
2146 	int rc;
2147 
2148 	if (likely(adapter->rss_support)) {
2149 		rc = ena_rss_configure(adapter);
2150 		if (rc != 0)
2151 			return (rc);
2152 	}
2153 
2154 	rc = ena_change_mtu(adapter->ifp, adapter->ifp->if_mtu);
2155 	if (unlikely(rc != 0))
2156 		return (rc);
2157 
2158 	ena_refill_all_rx_bufs(adapter);
2159 	ena_reset_counters((struct evcnt *)&adapter->hw_stats,
2160 	    sizeof(adapter->hw_stats));
2161 
2162 	return (0);
2163 }
2164 
2165 static int
2166 ena_up(struct ena_adapter *adapter)
2167 {
2168 	int rc = 0;
2169 
2170 #if 0
2171 	if (unlikely(device_is_attached(adapter->pdev) == 0)) {
2172 		device_printf(adapter->pdev, "device is not attached!\n");
2173 		return (ENXIO);
2174 	}
2175 #endif
2176 
2177 	if (unlikely(!adapter->running)) {
2178 		device_printf(adapter->pdev, "device is not running!\n");
2179 		return (ENXIO);
2180 	}
2181 
2182 	if (!adapter->up) {
2183 		device_printf(adapter->pdev, "device is going UP\n");
2184 
2185 		/* setup interrupts for IO queues */
2186 		rc = ena_request_io_irq(adapter);
2187 		if (unlikely(rc != 0)) {
2188 			ena_trace(ENA_ALERT, "err_req_irq");
2189 			goto err_req_irq;
2190 		}
2191 
2192 		/* allocate transmit descriptors */
2193 		rc = ena_setup_all_tx_resources(adapter);
2194 		if (unlikely(rc != 0)) {
2195 			ena_trace(ENA_ALERT, "err_setup_tx");
2196 			goto err_setup_tx;
2197 		}
2198 
2199 		/* allocate receive descriptors */
2200 		rc = ena_setup_all_rx_resources(adapter);
2201 		if (unlikely(rc != 0)) {
2202 			ena_trace(ENA_ALERT, "err_setup_rx");
2203 			goto err_setup_rx;
2204 		}
2205 
2206 		/* create IO queues for Rx & Tx */
2207 		rc = ena_create_io_queues(adapter);
2208 		if (unlikely(rc != 0)) {
2209 			ena_trace(ENA_ALERT,
2210 			    "create IO queues failed");
2211 			goto err_io_que;
2212 		}
2213 
2214 		if (unlikely(adapter->link_status))
2215 			if_link_state_change(adapter->ifp, LINK_STATE_UP);
2216 
2217 		rc = ena_up_complete(adapter);
2218 		if (unlikely(rc != 0))
2219 			goto err_up_complete;
2220 
2221 		counter_u64_add(adapter->dev_stats.interface_up, 1);
2222 
2223 		ena_update_hwassist(adapter);
2224 
2225 		if_setdrvflagbits(adapter->ifp, IFF_RUNNING,
2226 		    IFF_OACTIVE);
2227 
2228 		callout_reset(&adapter->timer_service, hz,
2229 		    ena_timer_service, (void *)adapter);
2230 
2231 		adapter->up = true;
2232 
2233 		ena_unmask_all_io_irqs(adapter);
2234 	}
2235 
2236 	return (0);
2237 
2238 err_up_complete:
2239 	ena_destroy_all_io_queues(adapter);
2240 err_io_que:
2241 	ena_free_all_rx_resources(adapter);
2242 err_setup_rx:
2243 	ena_free_all_tx_resources(adapter);
2244 err_setup_tx:
2245 	ena_free_io_irq(adapter);
2246 err_req_irq:
2247 	return (rc);
2248 }
2249 
2250 #if 0
2251 static uint64_t
2252 ena_get_counter(struct ifnet *ifp, ift_counter cnt)
2253 {
2254 	struct ena_adapter *adapter;
2255 	struct ena_hw_stats *stats;
2256 
2257 	adapter = if_getsoftc(ifp);
2258 	stats = &adapter->hw_stats;
2259 
2260 	switch (cnt) {
2261 	case IFCOUNTER_IPACKETS:
2262 		return (counter_u64_fetch(stats->rx_packets));
2263 	case IFCOUNTER_OPACKETS:
2264 		return (counter_u64_fetch(stats->tx_packets));
2265 	case IFCOUNTER_IBYTES:
2266 		return (counter_u64_fetch(stats->rx_bytes));
2267 	case IFCOUNTER_OBYTES:
2268 		return (counter_u64_fetch(stats->tx_bytes));
2269 	case IFCOUNTER_IQDROPS:
2270 		return (counter_u64_fetch(stats->rx_drops));
2271 	default:
2272 		return (if_get_counter_default(ifp, cnt));
2273 	}
2274 }
2275 #endif
2276 
2277 static int
2278 ena_media_change(struct ifnet *ifp)
2279 {
2280 	/* Media Change is not supported by firmware */
2281 	return (0);
2282 }
2283 
2284 static void
2285 ena_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2286 {
2287 	struct ena_adapter *adapter = if_getsoftc(ifp);
2288 	ena_trace(ENA_DBG, "enter");
2289 
2290 	mutex_enter(&adapter->global_mtx);
2291 
2292 	ifmr->ifm_status = IFM_AVALID;
2293 	ifmr->ifm_active = IFM_ETHER;
2294 
2295 	if (!adapter->link_status) {
2296 		mutex_exit(&adapter->global_mtx);
2297 		ena_trace(ENA_INFO, "link_status = false");
2298 		return;
2299 	}
2300 
2301 	ifmr->ifm_status |= IFM_ACTIVE;
2302 	ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
2303 
2304 	mutex_exit(&adapter->global_mtx);
2305 }
2306 
2307 static int
2308 ena_init(struct ifnet *ifp)
2309 {
2310 	struct ena_adapter *adapter = if_getsoftc(ifp);
2311 
2312 	if (!adapter->up) {
2313 		rw_enter(&adapter->ioctl_sx, RW_WRITER);
2314 		ena_up(adapter);
2315 		rw_exit(&adapter->ioctl_sx);
2316 	}
2317 
2318 	return 0;
2319 }
2320 
2321 static int
2322 ena_ioctl(struct ifnet *ifp, u_long command, void *data)
2323 {
2324 	struct ena_adapter *adapter;
2325 	struct ifreq *ifr;
2326 	int rc;
2327 
2328 	adapter = ifp->if_softc;
2329 	ifr = (struct ifreq *)data;
2330 
2331 	/*
2332 	 * Acquiring lock to prevent from running up and down routines parallel.
2333 	 */
2334 	rc = 0;
2335 	switch (command) {
2336 	case SIOCSIFMTU:
2337 		if (ifp->if_mtu == ifr->ifr_mtu)
2338 			break;
2339 		rw_enter(&adapter->ioctl_sx, RW_WRITER);
2340 		ena_down(adapter);
2341 
2342 		ena_change_mtu(ifp, ifr->ifr_mtu);
2343 
2344 		rc = ena_up(adapter);
2345 		rw_exit(&adapter->ioctl_sx);
2346 		break;
2347 
2348 	case SIOCSIFFLAGS:
2349 		if ((ifp->if_flags & IFF_UP) != 0) {
2350 			if ((if_getdrvflags(ifp) & IFF_RUNNING) != 0) {
2351 				if ((ifp->if_flags & (IFF_PROMISC |
2352 				    IFF_ALLMULTI)) != 0) {
2353 					device_printf(adapter->pdev,
2354 					    "ioctl promisc/allmulti\n");
2355 				}
2356 			} else {
2357 				rw_enter(&adapter->ioctl_sx, RW_WRITER);
2358 				rc = ena_up(adapter);
2359 				rw_exit(&adapter->ioctl_sx);
2360 			}
2361 		} else {
2362 			if ((if_getdrvflags(ifp) & IFF_RUNNING) != 0) {
2363 				rw_enter(&adapter->ioctl_sx, RW_WRITER);
2364 				ena_down(adapter);
2365 				rw_exit(&adapter->ioctl_sx);
2366 			}
2367 		}
2368 		break;
2369 
2370 	case SIOCADDMULTI:
2371 	case SIOCDELMULTI:
2372 		break;
2373 
2374 	case SIOCSIFCAP:
2375 		{
2376 			struct ifcapreq *ifcr = data;
2377 			int reinit = 0;
2378 
2379 			if (ifcr->ifcr_capenable != ifp->if_capenable) {
2380 				ifp->if_capenable = ifcr->ifcr_capenable;
2381 				reinit = 1;
2382 			}
2383 
2384 			if ((reinit != 0) &&
2385 			    ((if_getdrvflags(ifp) & IFF_RUNNING) != 0)) {
2386 				rw_enter(&adapter->ioctl_sx, RW_WRITER);
2387 				ena_down(adapter);
2388 				rc = ena_up(adapter);
2389 				rw_exit(&adapter->ioctl_sx);
2390 			}
2391 		}
2392 
2393 		break;
2394 	default:
2395 		rc = ether_ioctl(ifp, command, data);
2396 		break;
2397 	}
2398 
2399 	return (rc);
2400 }
2401 
2402 static int
2403 ena_get_dev_offloads(struct ena_com_dev_get_features_ctx *feat)
2404 {
2405 	int caps = 0;
2406 
2407 	if ((feat->offload.tx &
2408 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2409 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK |
2410 		ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK)) != 0)
2411 		caps |= IFCAP_CSUM_IPv4_Tx;
2412 
2413 	if ((feat->offload.tx &
2414 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_FULL_MASK |
2415 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)) != 0)
2416 		caps |= IFCAP_CSUM_TCPv6_Tx | IFCAP_CSUM_UDPv6_Tx;
2417 
2418 	if ((feat->offload.tx &
2419 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0)
2420 		caps |= IFCAP_TSOv4;
2421 
2422 	if ((feat->offload.tx &
2423 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK) != 0)
2424 		caps |= IFCAP_TSOv6;
2425 
2426 	if ((feat->offload.rx_supported &
2427 	    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK |
2428 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L3_CSUM_IPV4_MASK)) != 0)
2429 		caps |= IFCAP_CSUM_IPv4_Rx;
2430 
2431 	if ((feat->offload.rx_supported &
2432 	    ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK) != 0)
2433 		caps |= IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_UDPv6_Rx;
2434 
2435 	caps |= IFCAP_LRO;
2436 
2437 	return (caps);
2438 }
2439 
2440 static void
2441 ena_update_host_info(struct ena_admin_host_info *host_info, struct ifnet *ifp)
2442 {
2443 
2444 	host_info->supported_network_features[0] =
2445 	    (uint32_t)if_getcapabilities(ifp);
2446 }
2447 
2448 static void
2449 ena_update_hwassist(struct ena_adapter *adapter)
2450 {
2451 	struct ifnet *ifp = adapter->ifp;
2452 	uint32_t feat = adapter->tx_offload_cap;
2453 	int cap = if_getcapenable(ifp);
2454 	int flags = 0;
2455 
2456 	if_clearhwassist(ifp);
2457 
2458 	if ((cap & (IFCAP_CSUM_IPv4_Tx|IFCAP_CSUM_TCPv4_Tx|IFCAP_CSUM_UDPv4_Tx))
2459 	    != 0) {
2460 		if ((feat &
2461 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L3_CSUM_IPV4_MASK) != 0)
2462 			flags |= M_CSUM_IPv4;
2463 		if ((feat &
2464 		    (ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_FULL_MASK |
2465 		    ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)) != 0)
2466 			flags |= M_CSUM_TCPv4 | M_CSUM_UDPv4;
2467 	}
2468 
2469 	if ((cap & IFCAP_CSUM_TCPv6_Tx) != 0)
2470 		flags |= M_CSUM_TCPv6;
2471 
2472 	if ((cap & IFCAP_CSUM_UDPv6_Tx) != 0)
2473 		flags |= M_CSUM_UDPv6;
2474 
2475 	if ((cap & IFCAP_TSOv4) != 0)
2476 		flags |= M_CSUM_TSOv4;
2477 
2478 	if ((cap & IFCAP_TSOv6) != 0)
2479 		flags |= M_CSUM_TSOv6;
2480 
2481 	if_sethwassistbits(ifp, flags, 0);
2482 }
2483 
2484 static int
2485 ena_setup_ifnet(device_t pdev, struct ena_adapter *adapter,
2486     struct ena_com_dev_get_features_ctx *feat)
2487 {
2488 	struct ifnet *ifp;
2489 	int caps = 0;
2490 
2491 	ifp = adapter->ifp = &adapter->sc_ec.ec_if;
2492 	if (unlikely(ifp == NULL)) {
2493 		ena_trace(ENA_ALERT, "can not allocate ifnet structure\n");
2494 		return (ENXIO);
2495 	}
2496 	if_initname(ifp, "ena", device_unit(pdev));
2497 	if_setdev(ifp, pdev);
2498 	if_setsoftc(ifp, adapter);
2499 
2500 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
2501 	if_setinitfn(ifp, ena_init);
2502 	if_settransmitfn(ifp, ena_mq_start);
2503 #if 0
2504 	if_setqflushfn(ifp, ena_qflush);
2505 #endif
2506 	if_setioctlfn(ifp, ena_ioctl);
2507 #if 0
2508 	if_setgetcounterfn(ifp, ena_get_counter);
2509 #endif
2510 
2511 	if_setsendqlen(ifp, adapter->tx_ring_size);
2512 	if_setsendqready(ifp);
2513 	if_setmtu(ifp, ETHERMTU);
2514 	if_setbaudrate(ifp, 0);
2515 	/* Zeroize capabilities... */
2516 	if_setcapabilities(ifp, 0);
2517 	if_setcapenable(ifp, 0);
2518 	/* check hardware support */
2519 	caps = ena_get_dev_offloads(feat);
2520 	/* ... and set them */
2521 	if_setcapabilitiesbit(ifp, caps, 0);
2522 	adapter->sc_ec.ec_capabilities |= ETHERCAP_JUMBO_MTU;
2523 
2524 #if 0
2525 	/* TSO parameters */
2526 	/* XXX no limits on NetBSD, guarded by virtue of dmamap load failing */
2527 	ifp->if_hw_tsomax = ENA_TSO_MAXSIZE -
2528 	    (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2529 	ifp->if_hw_tsomaxsegcount = adapter->max_tx_sgl_size - 1;
2530 	ifp->if_hw_tsomaxsegsize = ENA_TSO_MAXSIZE;
2531 #endif
2532 
2533 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
2534 	if_setcapenable(ifp, if_getcapabilities(ifp));
2535 
2536 	/*
2537 	 * Specify the media types supported by this adapter and register
2538 	 * callbacks to update media and link information
2539 	 */
2540 	adapter->sc_ec.ec_ifmedia = &adapter->media;
2541 	ifmedia_init(&adapter->media, IFM_IMASK,
2542 	    ena_media_change, ena_media_status);
2543 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2544 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2545 
2546 	if_attach(ifp);
2547 	if_deferred_start_init(ifp, NULL);
2548 
2549 	ether_ifattach(ifp, adapter->mac_addr);
2550 
2551 	return (0);
2552 }
2553 
2554 static void
2555 ena_down(struct ena_adapter *adapter)
2556 {
2557 	int rc;
2558 
2559 	if (adapter->up) {
2560 		device_printf(adapter->pdev, "device is going DOWN\n");
2561 
2562 		callout_halt(&adapter->timer_service, &adapter->global_mtx);
2563 
2564 		adapter->up = false;
2565 		if_setdrvflagbits(adapter->ifp, IFF_OACTIVE,
2566 		    IFF_RUNNING);
2567 
2568 		ena_free_io_irq(adapter);
2569 
2570 		if (adapter->trigger_reset) {
2571 			rc = ena_com_dev_reset(adapter->ena_dev,
2572 			    adapter->reset_reason);
2573 			if (unlikely(rc != 0))
2574 				device_printf(adapter->pdev,
2575 				    "Device reset failed\n");
2576 		}
2577 
2578 		ena_destroy_all_io_queues(adapter);
2579 
2580 		ena_free_all_tx_bufs(adapter);
2581 		ena_free_all_rx_bufs(adapter);
2582 		ena_free_all_tx_resources(adapter);
2583 		ena_free_all_rx_resources(adapter);
2584 
2585 		counter_u64_add(adapter->dev_stats.interface_down, 1);
2586 	}
2587 }
2588 
2589 static void
2590 ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct mbuf *mbuf)
2591 {
2592 	struct ena_com_tx_meta *ena_meta;
2593 	struct ether_vlan_header *eh;
2594 	u32 mss;
2595 	bool offload;
2596 	uint16_t etype;
2597 	int ehdrlen;
2598 	struct ip *ip;
2599 	int iphlen;
2600 	struct tcphdr *th;
2601 
2602 	offload = false;
2603 	ena_meta = &ena_tx_ctx->ena_meta;
2604 
2605 #if 0
2606 	u32 mss = mbuf->m_pkthdr.tso_segsz;
2607 
2608 	if (mss != 0)
2609 		offload = true;
2610 #else
2611 	mss = mbuf->m_pkthdr.len;	/* XXX don't have tso_segsz */
2612 #endif
2613 
2614 	if ((mbuf->m_pkthdr.csum_flags & (M_CSUM_TSOv4 | M_CSUM_TSOv6)) != 0)
2615 		offload = true;
2616 
2617 	if ((mbuf->m_pkthdr.csum_flags & CSUM_OFFLOAD) != 0)
2618 		offload = true;
2619 
2620 	if (!offload) {
2621 		ena_tx_ctx->meta_valid = 0;
2622 		return;
2623 	}
2624 
2625 	/* Determine where frame payload starts. */
2626 	eh = mtod(mbuf, struct ether_vlan_header *);
2627 	if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2628 		etype = ntohs(eh->evl_proto);
2629 		ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
2630 	} else {
2631 		etype = htons(eh->evl_encap_proto);
2632 		ehdrlen = ETHER_HDR_LEN;
2633 	}
2634 
2635 	ip = (struct ip *)(mbuf->m_data + ehdrlen);
2636 	iphlen = ip->ip_hl << 2;
2637 	th = (struct tcphdr *)((vaddr_t)ip + iphlen);
2638 
2639 	if ((mbuf->m_pkthdr.csum_flags & M_CSUM_IPv4) != 0) {
2640 		ena_tx_ctx->l3_csum_enable = 1;
2641 	}
2642 	if ((mbuf->m_pkthdr.csum_flags & (M_CSUM_TSOv4 | M_CSUM_TSOv6)) != 0) {
2643 		ena_tx_ctx->tso_enable = 1;
2644 		ena_meta->l4_hdr_len = (th->th_off);
2645 	}
2646 
2647 	switch (etype) {
2648 	case ETHERTYPE_IP:
2649 		ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2650 		if ((ip->ip_off & htons(IP_DF)) != 0)
2651 			ena_tx_ctx->df = 1;
2652 		break;
2653 	case ETHERTYPE_IPV6:
2654 		ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2655 
2656 	default:
2657 		break;
2658 	}
2659 
2660 	if (ip->ip_p == IPPROTO_TCP) {
2661 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2662 		if ((mbuf->m_pkthdr.csum_flags &
2663 		    (M_CSUM_TCPv4 | M_CSUM_TCPv6)) != 0)
2664 			ena_tx_ctx->l4_csum_enable = 1;
2665 		else
2666 			ena_tx_ctx->l4_csum_enable = 0;
2667 	} else if (ip->ip_p == IPPROTO_UDP) {
2668 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2669 		if ((mbuf->m_pkthdr.csum_flags &
2670 		    (M_CSUM_UDPv4 | M_CSUM_UDPv6)) != 0)
2671 			ena_tx_ctx->l4_csum_enable = 1;
2672 		else
2673 			ena_tx_ctx->l4_csum_enable = 0;
2674 	} else {
2675 		ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
2676 		ena_tx_ctx->l4_csum_enable = 0;
2677 	}
2678 
2679 	ena_meta->mss = mss;
2680 	ena_meta->l3_hdr_len = iphlen;
2681 	ena_meta->l3_hdr_offset = ehdrlen;
2682 	ena_tx_ctx->meta_valid = 1;
2683 }
2684 
2685 static int
2686 ena_check_and_collapse_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2687 {
2688 	struct ena_adapter *adapter;
2689 	struct mbuf *collapsed_mbuf;
2690 	int num_frags;
2691 
2692 	adapter = tx_ring->adapter;
2693 	num_frags = ena_mbuf_count(*mbuf);
2694 
2695 	/* One segment must be reserved for configuration descriptor. */
2696 	if (num_frags < adapter->max_tx_sgl_size)
2697 		return (0);
2698 	counter_u64_add(tx_ring->tx_stats.collapse, 1);
2699 
2700 	collapsed_mbuf = m_collapse(*mbuf, M_NOWAIT,
2701 	    adapter->max_tx_sgl_size - 1);
2702 	if (unlikely(collapsed_mbuf == NULL)) {
2703 		counter_u64_add(tx_ring->tx_stats.collapse_err, 1);
2704 		return (ENOMEM);
2705 	}
2706 
2707 	/* If mbuf was collapsed succesfully, original mbuf is released. */
2708 	*mbuf = collapsed_mbuf;
2709 
2710 	return (0);
2711 }
2712 
2713 static int
2714 ena_xmit_mbuf(struct ena_ring *tx_ring, struct mbuf **mbuf)
2715 {
2716 	struct ena_adapter *adapter;
2717 	struct ena_tx_buffer *tx_info;
2718 	struct ena_com_tx_ctx ena_tx_ctx;
2719 	struct ena_com_dev *ena_dev;
2720 	struct ena_com_buf *ena_buf;
2721 	struct ena_com_io_sq* io_sq;
2722 	void *push_hdr;
2723 	uint16_t next_to_use;
2724 	uint16_t req_id;
2725 	uint16_t ena_qid;
2726 	uint32_t header_len;
2727 	int i, rc;
2728 	int nb_hw_desc;
2729 
2730 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2731 	adapter = tx_ring->que->adapter;
2732 	ena_dev = adapter->ena_dev;
2733 	io_sq = &ena_dev->io_sq_queues[ena_qid];
2734 
2735 	rc = ena_check_and_collapse_mbuf(tx_ring, mbuf);
2736 	if (unlikely(rc != 0)) {
2737 		ena_trace(ENA_WARNING,
2738 		    "Failed to collapse mbuf! err: %d", rc);
2739 		return (rc);
2740 	}
2741 
2742 	next_to_use = tx_ring->next_to_use;
2743 	req_id = tx_ring->free_tx_ids[next_to_use];
2744 	tx_info = &tx_ring->tx_buffer_info[req_id];
2745 
2746 	tx_info->mbuf = *mbuf;
2747 	tx_info->num_of_bufs = 0;
2748 
2749 	ena_buf = tx_info->bufs;
2750 
2751 	ena_trace(ENA_DBG | ENA_TXPTH, "Tx: %d bytes", (*mbuf)->m_pkthdr.len);
2752 
2753 	/*
2754 	 * header_len is just a hint for the device. Because FreeBSD is not
2755 	 * giving us information about packet header length and it is not
2756 	 * guaranteed that all packet headers will be in the 1st mbuf, setting
2757 	 * header_len to 0 is making the device ignore this value and resolve
2758 	 * header on it's own.
2759 	 */
2760 	header_len = 0;
2761 	push_hdr = NULL;
2762 
2763 	rc = bus_dmamap_load_mbuf(adapter->sc_dmat, tx_info->map,
2764 	    *mbuf, BUS_DMA_NOWAIT);
2765 
2766 	if (unlikely((rc != 0) || (tx_info->map->dm_nsegs == 0))) {
2767 		ena_trace(ENA_WARNING,
2768 		    "dmamap load failed! err: %d nsegs: %d", rc,
2769 		    tx_info->map->dm_nsegs);
2770 		counter_u64_add(tx_ring->tx_stats.dma_mapping_err, 1);
2771 		tx_info->mbuf = NULL;
2772 		if (rc == ENOMEM)
2773 			return (ENA_COM_NO_MEM);
2774 		else
2775 			return (ENA_COM_INVAL);
2776 	}
2777 
2778 	for (i = 0; i < tx_info->map->dm_nsegs; i++) {
2779 		ena_buf->len = tx_info->map->dm_segs[i].ds_len;
2780 		ena_buf->paddr = tx_info->map->dm_segs[i].ds_addr;
2781 		ena_buf++;
2782 	}
2783 	tx_info->num_of_bufs = tx_info->map->dm_nsegs;
2784 
2785 	memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2786 	ena_tx_ctx.ena_bufs = tx_info->bufs;
2787 	ena_tx_ctx.push_header = push_hdr;
2788 	ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2789 	ena_tx_ctx.req_id = req_id;
2790 	ena_tx_ctx.header_len = header_len;
2791 
2792 	/* Set flags and meta data */
2793 	ena_tx_csum(&ena_tx_ctx, *mbuf);
2794 	/* Prepare the packet's descriptors and send them to device */
2795 	rc = ena_com_prepare_tx(io_sq, &ena_tx_ctx, &nb_hw_desc);
2796 	if (unlikely(rc != 0)) {
2797 		device_printf(adapter->pdev, "failed to prepare tx bufs\n");
2798 		counter_u64_add(tx_ring->tx_stats.prepare_ctx_err, 1);
2799 		goto dma_error;
2800 	}
2801 
2802 	counter_enter();
2803 	counter_u64_add_protected(tx_ring->tx_stats.cnt, 1);
2804 	counter_u64_add_protected(tx_ring->tx_stats.bytes,
2805 	    (*mbuf)->m_pkthdr.len);
2806 
2807 	counter_u64_add_protected(adapter->hw_stats.tx_packets, 1);
2808 	counter_u64_add_protected(adapter->hw_stats.tx_bytes,
2809 	    (*mbuf)->m_pkthdr.len);
2810 	counter_exit();
2811 
2812 	tx_info->tx_descs = nb_hw_desc;
2813 	getbinuptime(&tx_info->timestamp);
2814 	tx_info->print_once = true;
2815 
2816 	tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2817 	    tx_ring->ring_size);
2818 
2819 	bus_dmamap_sync(adapter->sc_dmat, tx_info->map, 0,
2820 	    tx_info->map->dm_mapsize, BUS_DMASYNC_PREWRITE);
2821 
2822 	return (0);
2823 
2824 dma_error:
2825 	tx_info->mbuf = NULL;
2826 	bus_dmamap_unload(adapter->sc_dmat, tx_info->map);
2827 
2828 	return (rc);
2829 }
2830 
2831 static void
2832 ena_start_xmit(struct ena_ring *tx_ring)
2833 {
2834 	struct mbuf *mbuf;
2835 	struct ena_adapter *adapter = tx_ring->adapter;
2836 	struct ena_com_io_sq* io_sq;
2837 	int ena_qid;
2838 	int acum_pkts = 0;
2839 	int ret = 0;
2840 
2841 	if (unlikely((if_getdrvflags(adapter->ifp) & IFF_RUNNING) == 0))
2842 		return;
2843 
2844 	if (unlikely(!adapter->link_status))
2845 		return;
2846 
2847 	ena_qid = ENA_IO_TXQ_IDX(tx_ring->que->id);
2848 	io_sq = &adapter->ena_dev->io_sq_queues[ena_qid];
2849 
2850 	while ((mbuf = drbr_peek(adapter->ifp, tx_ring->br)) != NULL) {
2851 		ena_trace(ENA_DBG | ENA_TXPTH, "\ndequeued mbuf %p with flags %#x and"
2852 		    " header csum flags %#jx",
2853 		    mbuf, mbuf->m_flags, (uint64_t)mbuf->m_pkthdr.csum_flags);
2854 
2855 		if (unlikely(!ena_com_sq_have_enough_space(io_sq,
2856 		    ENA_TX_CLEANUP_THRESHOLD)))
2857 			ena_tx_cleanup(tx_ring);
2858 
2859 		if (unlikely((ret = ena_xmit_mbuf(tx_ring, &mbuf)) != 0)) {
2860 			if (ret == ENA_COM_NO_MEM) {
2861 				drbr_putback(adapter->ifp, tx_ring->br, mbuf);
2862 			} else if (ret == ENA_COM_NO_SPACE) {
2863 				drbr_putback(adapter->ifp, tx_ring->br, mbuf);
2864 			} else {
2865 				m_freem(mbuf);
2866 				drbr_advance(adapter->ifp, tx_ring->br);
2867 			}
2868 
2869 			break;
2870 		}
2871 
2872 		drbr_advance(adapter->ifp, tx_ring->br);
2873 
2874 		if (unlikely((if_getdrvflags(adapter->ifp) &
2875 		    IFF_RUNNING) == 0))
2876 			return;
2877 
2878 		acum_pkts++;
2879 
2880 		/*
2881 		 * If there's a BPF listener, bounce a copy of this frame
2882 		 * to him.
2883 		 */
2884 		bpf_mtap(adapter->ifp, mbuf, BPF_D_OUT);
2885 
2886 		if (unlikely(acum_pkts == DB_THRESHOLD)) {
2887 			acum_pkts = 0;
2888 			wmb();
2889 			/* Trigger the dma engine */
2890 			ena_com_write_sq_doorbell(io_sq);
2891 			counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2892 		}
2893 
2894 	}
2895 
2896 	if (likely(acum_pkts != 0)) {
2897 		wmb();
2898 		/* Trigger the dma engine */
2899 		ena_com_write_sq_doorbell(io_sq);
2900 		counter_u64_add(tx_ring->tx_stats.doorbells, 1);
2901 	}
2902 
2903 	if (!ena_com_sq_have_enough_space(io_sq, ENA_TX_CLEANUP_THRESHOLD))
2904 		ena_tx_cleanup(tx_ring);
2905 }
2906 
2907 static void
2908 ena_deferred_mq_start(struct work *wk, void *arg)
2909 {
2910 	struct ena_ring *tx_ring = (struct ena_ring *)arg;
2911 	struct ifnet *ifp = tx_ring->adapter->ifp;
2912 
2913 	atomic_swap_uint(&tx_ring->task_pending, 0);
2914 
2915 	while (!drbr_empty(ifp, tx_ring->br) &&
2916 	    (if_getdrvflags(ifp) & IFF_RUNNING) != 0) {
2917 		ENA_RING_MTX_LOCK(tx_ring);
2918 		ena_start_xmit(tx_ring);
2919 		ENA_RING_MTX_UNLOCK(tx_ring);
2920 	}
2921 }
2922 
2923 static int
2924 ena_mq_start(struct ifnet *ifp, struct mbuf *m)
2925 {
2926 	struct ena_adapter *adapter = ifp->if_softc;
2927 	struct ena_ring *tx_ring;
2928 	int ret, is_drbr_empty;
2929 	uint32_t i;
2930 
2931 	if (unlikely((if_getdrvflags(adapter->ifp) & IFF_RUNNING) == 0))
2932 		return (ENODEV);
2933 
2934 	/* Which queue to use */
2935 	/*
2936 	 * If everything is setup correctly, it should be the
2937 	 * same bucket that the current CPU we're on is.
2938 	 * It should improve performance.
2939 	 */
2940 #if 0
2941 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) {
2942 #ifdef	RSS
2943 		if (rss_hash2bucket(m->m_pkthdr.flowid,
2944 		    M_HASHTYPE_GET(m), &i) == 0) {
2945 			i = i % adapter->num_queues;
2946 
2947 		} else
2948 #endif
2949 		{
2950 			i = m->m_pkthdr.flowid % adapter->num_queues;
2951 		}
2952 	} else {
2953 #endif
2954 		i = cpu_index(curcpu()) % adapter->num_queues;
2955 #if 0
2956 	}
2957 #endif
2958 	tx_ring = &adapter->tx_ring[i];
2959 
2960 	/* Check if drbr is empty before putting packet */
2961 	is_drbr_empty = drbr_empty(ifp, tx_ring->br);
2962 	ret = drbr_enqueue(ifp, tx_ring->br, m);
2963 	if (unlikely(ret != 0)) {
2964 		if (atomic_cas_uint(&tx_ring->task_pending, 0, 1) == 0)
2965 			workqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task,
2966 			    curcpu());
2967 		return (ret);
2968 	}
2969 
2970 	if ((is_drbr_empty != 0) && (ENA_RING_MTX_TRYLOCK(tx_ring) != 0)) {
2971 		ena_start_xmit(tx_ring);
2972 		ENA_RING_MTX_UNLOCK(tx_ring);
2973 	} else {
2974 		if (atomic_cas_uint(&tx_ring->task_pending, 0, 1) == 0)
2975 			workqueue_enqueue(tx_ring->enqueue_tq, &tx_ring->enqueue_task,
2976 			    curcpu());
2977 	}
2978 
2979 	return (0);
2980 }
2981 
2982 #if 0
2983 static void
2984 ena_qflush(struct ifnet *ifp)
2985 {
2986 	struct ena_adapter *adapter = ifp->if_softc;
2987 	struct ena_ring *tx_ring = adapter->tx_ring;
2988 	int i;
2989 
2990 	for(i = 0; i < adapter->num_queues; ++i, ++tx_ring)
2991 		if (!drbr_empty(ifp, tx_ring->br)) {
2992 			ENA_RING_MTX_LOCK(tx_ring);
2993 			drbr_flush(ifp, tx_ring->br);
2994 			ENA_RING_MTX_UNLOCK(tx_ring);
2995 		}
2996 
2997 	if_qflush(ifp);
2998 }
2999 #endif
3000 
3001 static int
3002 ena_calc_io_queue_num(struct pci_attach_args *pa,
3003     struct ena_adapter *adapter,
3004     struct ena_com_dev_get_features_ctx *get_feat_ctx)
3005 {
3006 	int io_sq_num, io_cq_num, io_queue_num;
3007 
3008 	io_sq_num = get_feat_ctx->max_queues.max_sq_num;
3009 	io_cq_num = get_feat_ctx->max_queues.max_cq_num;
3010 
3011 	io_queue_num = min_t(int, mp_ncpus, ENA_MAX_NUM_IO_QUEUES);
3012 	io_queue_num = min_t(int, io_queue_num, io_sq_num);
3013 	io_queue_num = min_t(int, io_queue_num, io_cq_num);
3014 	/* 1 IRQ for for mgmnt and 1 IRQ for each TX/RX pair */
3015 	io_queue_num = min_t(int, io_queue_num,
3016 	    pci_msix_count(pa->pa_pc, pa->pa_tag) - 1);
3017 #ifdef	RSS
3018 	io_queue_num = min_t(int, io_queue_num, rss_getnumbuckets());
3019 #endif
3020 
3021 	return (io_queue_num);
3022 }
3023 
3024 static int
3025 ena_calc_queue_size(struct ena_adapter *adapter, uint16_t *max_tx_sgl_size,
3026     uint16_t *max_rx_sgl_size, struct ena_com_dev_get_features_ctx *feat)
3027 {
3028 	uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
3029 	uint32_t v;
3030 	uint32_t q;
3031 
3032 	queue_size = min_t(uint32_t, queue_size,
3033 	    feat->max_queues.max_cq_depth);
3034 	queue_size = min_t(uint32_t, queue_size,
3035 	    feat->max_queues.max_sq_depth);
3036 
3037 	/* round down to the nearest power of 2 */
3038 	v = queue_size;
3039 	while (v != 0) {
3040 		if (powerof2(queue_size) != 0)
3041 			break;
3042 		v /= 2;
3043 		q = rounddown2(queue_size, v);
3044 		if (q != 0) {
3045 			queue_size = q;
3046 			break;
3047 		}
3048 	}
3049 
3050 	if (unlikely(queue_size == 0)) {
3051 		device_printf(adapter->pdev, "Invalid queue size\n");
3052 		return (ENA_COM_FAULT);
3053 	}
3054 
3055 	*max_tx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3056 	    feat->max_queues.max_packet_tx_descs);
3057 	*max_rx_sgl_size = min_t(uint16_t, ENA_PKT_MAX_BUFS,
3058 	    feat->max_queues.max_packet_rx_descs);
3059 
3060 	return (queue_size);
3061 }
3062 
3063 #if 0
3064 static int
3065 ena_rss_init_default(struct ena_adapter *adapter)
3066 {
3067 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3068 	device_t dev = adapter->pdev;
3069 	int qid, rc, i;
3070 
3071 	rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3072 	if (unlikely(rc != 0)) {
3073 		device_printf(dev, "Cannot init indirect table\n");
3074 		return (rc);
3075 	}
3076 
3077 	for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3078 #ifdef	RSS
3079 		qid = rss_get_indirection_to_bucket(i);
3080 		qid = qid % adapter->num_queues;
3081 #else
3082 		qid = i % adapter->num_queues;
3083 #endif
3084 		rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3085 		    ENA_IO_RXQ_IDX(qid));
3086 		if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3087 			device_printf(dev, "Cannot fill indirect table\n");
3088 			goto err_rss_destroy;
3089 		}
3090 	}
3091 
3092 	rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3093 	    ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3094 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3095 		device_printf(dev, "Cannot fill hash function\n");
3096 		goto err_rss_destroy;
3097 	}
3098 
3099 	rc = ena_com_set_default_hash_ctrl(ena_dev);
3100 	if (unlikely((rc != 0) && (rc != EOPNOTSUPP))) {
3101 		device_printf(dev, "Cannot fill hash control\n");
3102 		goto err_rss_destroy;
3103 	}
3104 
3105 	return (0);
3106 
3107 err_rss_destroy:
3108 	ena_com_rss_destroy(ena_dev);
3109 	return (rc);
3110 }
3111 
3112 static void
3113 ena_rss_init_default_deferred(void *arg)
3114 {
3115 	struct ena_adapter *adapter;
3116 	devclass_t dc;
3117 	int max;
3118 	int rc;
3119 
3120 	dc = devclass_find("ena");
3121 	if (unlikely(dc == NULL)) {
3122 		ena_trace(ENA_ALERT, "No devclass ena\n");
3123 		return;
3124 	}
3125 
3126 	max = devclass_get_maxunit(dc);
3127 	while (max-- >= 0) {
3128 		adapter = devclass_get_softc(dc, max);
3129 		if (adapter != NULL) {
3130 			rc = ena_rss_init_default(adapter);
3131 			adapter->rss_support = true;
3132 			if (unlikely(rc != 0)) {
3133 				device_printf(adapter->pdev,
3134 				    "WARNING: RSS was not properly initialized,"
3135 				    " it will affect bandwidth\n");
3136 				adapter->rss_support = false;
3137 			}
3138 		}
3139 	}
3140 }
3141 SYSINIT(ena_rss_init, SI_SUB_KICK_SCHEDULER, SI_ORDER_SECOND, ena_rss_init_default_deferred, NULL);
3142 #endif
3143 
3144 static void
3145 ena_config_host_info(struct ena_com_dev *ena_dev)
3146 {
3147 	struct ena_admin_host_info *host_info;
3148 	int rc;
3149 
3150 	/* Allocate only the host info */
3151 	rc = ena_com_allocate_host_info(ena_dev);
3152 	if (unlikely(rc != 0)) {
3153 		ena_trace(ENA_ALERT, "Cannot allocate host info\n");
3154 		return;
3155 	}
3156 
3157 	host_info = ena_dev->host_attr.host_info;
3158 
3159 	host_info->os_type = ENA_ADMIN_OS_FREEBSD;
3160 	host_info->kernel_ver = osreldate;
3161 
3162 	snprintf(host_info->kernel_ver_str, sizeof(host_info->kernel_ver_str),
3163 	    "%d", osreldate);
3164 	host_info->os_dist = 0;
3165 	strncpy(host_info->os_dist_str, osrelease,
3166 	    sizeof(host_info->os_dist_str) - 1);
3167 
3168 	host_info->driver_version =
3169 		(DRV_MODULE_VER_MAJOR) |
3170 		(DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3171 		(DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
3172 
3173 	rc = ena_com_set_host_attributes(ena_dev);
3174 	if (unlikely(rc != 0)) {
3175 		if (rc == EOPNOTSUPP)
3176 			ena_trace(ENA_WARNING, "Cannot set host attributes\n");
3177 		else
3178 			ena_trace(ENA_ALERT, "Cannot set host attributes\n");
3179 
3180 		goto err;
3181 	}
3182 
3183 	return;
3184 
3185 err:
3186 	ena_com_delete_host_info(ena_dev);
3187 }
3188 
3189 static int
3190 ena_device_init(struct ena_adapter *adapter, device_t pdev,
3191     struct ena_com_dev_get_features_ctx *get_feat_ctx, int *wd_active)
3192 {
3193 	struct ena_com_dev* ena_dev = adapter->ena_dev;
3194 	bool readless_supported;
3195 	uint32_t aenq_groups;
3196 	int dma_width;
3197 	int rc;
3198 
3199 	rc = ena_com_mmio_reg_read_request_init(ena_dev);
3200 	if (unlikely(rc != 0)) {
3201 		device_printf(pdev, "failed to init mmio read less\n");
3202 		return (rc);
3203 	}
3204 
3205 	/*
3206 	 * The PCIe configuration space revision id indicate if mmio reg
3207 	 * read is disabled
3208 	 */
3209 	const int rev = PCI_REVISION(adapter->sc_pa.pa_class);
3210 	readless_supported = ((rev & ENA_MMIO_DISABLE_REG_READ) == 0);
3211 	ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3212 
3213 	rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3214 	if (unlikely(rc != 0)) {
3215 		device_printf(pdev, "Can not reset device\n");
3216 		goto err_mmio_read_less;
3217 	}
3218 
3219 	rc = ena_com_validate_version(ena_dev);
3220 	if (unlikely(rc != 0)) {
3221 		device_printf(pdev, "device version is too low\n");
3222 		goto err_mmio_read_less;
3223 	}
3224 
3225 	dma_width = ena_com_get_dma_width(ena_dev);
3226 	if (unlikely(dma_width < 0)) {
3227 		device_printf(pdev, "Invalid dma width value %d", dma_width);
3228 		rc = dma_width;
3229 		goto err_mmio_read_less;
3230 	}
3231 	adapter->dma_width = dma_width;
3232 
3233 	/* ENA admin level init */
3234 	rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
3235 	if (unlikely(rc != 0)) {
3236 		device_printf(pdev,
3237 		    "Can not initialize ena admin queue with device\n");
3238 		goto err_mmio_read_less;
3239 	}
3240 
3241 	/*
3242 	 * To enable the msix interrupts the driver needs to know the number
3243 	 * of queues. So the driver uses polling mode to retrieve this
3244 	 * information
3245 	 */
3246 	ena_com_set_admin_polling_mode(ena_dev, true);
3247 
3248 	ena_config_host_info(ena_dev);
3249 
3250 	/* Get Device Attributes */
3251 	rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3252 	if (unlikely(rc != 0)) {
3253 		device_printf(pdev,
3254 		    "Cannot get attribute for ena device rc: %d\n", rc);
3255 		goto err_admin_init;
3256 	}
3257 
3258 	aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) | BIT(ENA_ADMIN_KEEP_ALIVE);
3259 
3260 	aenq_groups &= get_feat_ctx->aenq.supported_groups;
3261 	rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3262 	if (unlikely(rc != 0)) {
3263 		device_printf(pdev, "Cannot configure aenq groups rc: %d\n", rc);
3264 		goto err_admin_init;
3265 	}
3266 
3267 	*wd_active = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3268 
3269 	return (0);
3270 
3271 err_admin_init:
3272 	ena_com_delete_host_info(ena_dev);
3273 	ena_com_admin_destroy(ena_dev);
3274 err_mmio_read_less:
3275 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3276 
3277 	return (rc);
3278 }
3279 
3280 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
3281     int io_vectors)
3282 {
3283 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3284 	int rc;
3285 
3286 	rc = ena_enable_msix(adapter);
3287 	if (unlikely(rc != 0)) {
3288 		device_printf(adapter->pdev, "Error with MSI-X enablement\n");
3289 		return (rc);
3290 	}
3291 
3292 	rc = ena_request_mgmnt_irq(adapter);
3293 	if (unlikely(rc != 0)) {
3294 		device_printf(adapter->pdev, "Cannot setup mgmnt queue intr\n");
3295 		goto err_disable_msix;
3296 	}
3297 
3298 	ena_com_set_admin_polling_mode(ena_dev, false);
3299 
3300 	ena_com_admin_aenq_enable(ena_dev);
3301 
3302 	return (0);
3303 
3304 err_disable_msix:
3305 	ena_disable_msix(adapter);
3306 
3307 	return (rc);
3308 }
3309 
3310 /* Function called on ENA_ADMIN_KEEP_ALIVE event */
3311 static void ena_keep_alive_wd(void *adapter_data,
3312     struct ena_admin_aenq_entry *aenq_e)
3313 {
3314 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3315 	struct ena_admin_aenq_keep_alive_desc *desc;
3316 	sbintime_t stime;
3317 	uint64_t rx_drops;
3318 
3319 	desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3320 
3321 	rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
3322 	counter_u64_zero(adapter->hw_stats.rx_drops);
3323 	counter_u64_add(adapter->hw_stats.rx_drops, rx_drops);
3324 
3325 	stime = getsbinuptime();
3326 	(void) atomic_swap_64(&adapter->keep_alive_timestamp, stime);
3327 }
3328 
3329 /* Check for keep alive expiration */
3330 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3331 {
3332 	sbintime_t timestamp, time;
3333 
3334 	if (adapter->wd_active == 0)
3335 		return;
3336 
3337 	if (likely(adapter->keep_alive_timeout == 0))
3338 		return;
3339 
3340 	/* FreeBSD uses atomic_load_acq_64() in place of the membar + read */
3341 	membar_sync();
3342 	timestamp = adapter->keep_alive_timestamp;
3343 
3344 	time = getsbinuptime() - timestamp;
3345 	if (unlikely(time > adapter->keep_alive_timeout)) {
3346 		device_printf(adapter->pdev,
3347 		    "Keep alive watchdog timeout.\n");
3348 		counter_u64_add(adapter->dev_stats.wd_expired, 1);
3349 		adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3350 		adapter->trigger_reset = true;
3351 	}
3352 }
3353 
3354 /* Check if admin queue is enabled */
3355 static void check_for_admin_com_state(struct ena_adapter *adapter)
3356 {
3357 	if (unlikely(ena_com_get_admin_running_state(adapter->ena_dev) ==
3358 	    false)) {
3359 		device_printf(adapter->pdev,
3360 		    "ENA admin queue is not in running state!\n");
3361 		counter_u64_add(adapter->dev_stats.admin_q_pause, 1);
3362 		adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3363 		adapter->trigger_reset = true;
3364 	}
3365 }
3366 
3367 static int
3368 check_missing_comp_in_queue(struct ena_adapter *adapter,
3369     struct ena_ring *tx_ring)
3370 {
3371 	struct bintime curtime, time;
3372 	struct ena_tx_buffer *tx_buf;
3373 	uint32_t missed_tx = 0;
3374 	int i;
3375 
3376 	getbinuptime(&curtime);
3377 
3378 	for (i = 0; i < tx_ring->ring_size; i++) {
3379 		tx_buf = &tx_ring->tx_buffer_info[i];
3380 
3381 		if (bintime_isset(&tx_buf->timestamp) == 0)
3382 			continue;
3383 
3384 		time = curtime;
3385 		bintime_sub(&time, &tx_buf->timestamp);
3386 
3387 		/* Check again if packet is still waiting */
3388 		if (unlikely(bttosbt(time) > adapter->missing_tx_timeout)) {
3389 
3390 			if (!tx_buf->print_once)
3391 				ena_trace(ENA_WARNING, "Found a Tx that wasn't "
3392 				    "completed on time, qid %d, index %d.\n",
3393 				    tx_ring->qid, i);
3394 
3395 			tx_buf->print_once = true;
3396 			missed_tx++;
3397 			counter_u64_add(tx_ring->tx_stats.missing_tx_comp, 1);
3398 
3399 			if (unlikely(missed_tx >
3400 			    adapter->missing_tx_threshold)) {
3401 				device_printf(adapter->pdev,
3402 				    "The number of lost tx completion "
3403 				    "is above the threshold (%d > %d). "
3404 				    "Reset the device\n",
3405 				    missed_tx, adapter->missing_tx_threshold);
3406 				adapter->reset_reason =
3407 				    ENA_REGS_RESET_MISS_TX_CMPL;
3408 				adapter->trigger_reset = true;
3409 				return (EIO);
3410 			}
3411 		}
3412 	}
3413 
3414 	return (0);
3415 }
3416 
3417 /*
3418  * Check for TX which were not completed on time.
3419  * Timeout is defined by "missing_tx_timeout".
3420  * Reset will be performed if number of incompleted
3421  * transactions exceeds "missing_tx_threshold".
3422  */
3423 static void
3424 check_for_missing_tx_completions(struct ena_adapter *adapter)
3425 {
3426 	struct ena_ring *tx_ring;
3427 	int i, budget, rc;
3428 
3429 	/* Make sure the driver doesn't turn the device in other process */
3430 	rmb();
3431 
3432 	if (!adapter->up)
3433 		return;
3434 
3435 	if (adapter->trigger_reset)
3436 		return;
3437 
3438 	if (adapter->missing_tx_timeout == 0)
3439 		return;
3440 
3441 	budget = adapter->missing_tx_max_queues;
3442 
3443 	for (i = adapter->next_monitored_tx_qid; i < adapter->num_queues; i++) {
3444 		tx_ring = &adapter->tx_ring[i];
3445 
3446 		rc = check_missing_comp_in_queue(adapter, tx_ring);
3447 		if (unlikely(rc != 0))
3448 			return;
3449 
3450 		budget--;
3451 		if (budget == 0) {
3452 			i++;
3453 			break;
3454 		}
3455 	}
3456 
3457 	adapter->next_monitored_tx_qid = i % adapter->num_queues;
3458 }
3459 
3460 /* trigger deferred rx cleanup after 2 consecutive detections */
3461 #define EMPTY_RX_REFILL 2
3462 /* For the rare case where the device runs out of Rx descriptors and the
3463  * msix handler failed to refill new Rx descriptors (due to a lack of memory
3464  * for example).
3465  * This case will lead to a deadlock:
3466  * The device won't send interrupts since all the new Rx packets will be dropped
3467  * The msix handler won't allocate new Rx descriptors so the device won't be
3468  * able to send new packets.
3469  *
3470  * When such a situation is detected - execute rx cleanup task in another thread
3471  */
3472 static void
3473 check_for_empty_rx_ring(struct ena_adapter *adapter)
3474 {
3475 	struct ena_ring *rx_ring;
3476 	int i, refill_required;
3477 
3478 	if (!adapter->up)
3479 		return;
3480 
3481 	if (adapter->trigger_reset)
3482 		return;
3483 
3484 	for (i = 0; i < adapter->num_queues; i++) {
3485 		rx_ring = &adapter->rx_ring[i];
3486 
3487 		refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
3488 		if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3489 			rx_ring->empty_rx_queue++;
3490 
3491 			if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL)	{
3492 				counter_u64_add(rx_ring->rx_stats.empty_rx_ring,
3493 				    1);
3494 
3495 				device_printf(adapter->pdev,
3496 				    "trigger refill for ring %d\n", i);
3497 
3498 				if (atomic_cas_uint(&rx_ring->task_pending, 0, 1) == 0)
3499 					workqueue_enqueue(rx_ring->cmpl_tq,
3500 					    &rx_ring->cmpl_task, curcpu());
3501 				rx_ring->empty_rx_queue = 0;
3502 			}
3503 		} else {
3504 			rx_ring->empty_rx_queue = 0;
3505 		}
3506 	}
3507 }
3508 
3509 static void
3510 ena_timer_service(void *data)
3511 {
3512 	struct ena_adapter *adapter = (struct ena_adapter *)data;
3513 	struct ena_admin_host_info *host_info =
3514 	    adapter->ena_dev->host_attr.host_info;
3515 
3516 	check_for_missing_keep_alive(adapter);
3517 
3518 	check_for_admin_com_state(adapter);
3519 
3520 	check_for_missing_tx_completions(adapter);
3521 
3522 	check_for_empty_rx_ring(adapter);
3523 
3524 	if (host_info != NULL)
3525 		ena_update_host_info(host_info, adapter->ifp);
3526 
3527 	if (unlikely(adapter->trigger_reset)) {
3528 		device_printf(adapter->pdev, "Trigger reset is on\n");
3529 		workqueue_enqueue(adapter->reset_tq, &adapter->reset_task,
3530 		    curcpu());
3531 		return;
3532 	}
3533 
3534 	/*
3535 	 * Schedule another timeout one second from now.
3536 	 */
3537 	callout_schedule(&adapter->timer_service, hz);
3538 }
3539 
3540 static void
3541 ena_reset_task(struct work *wk, void *arg)
3542 {
3543 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3544 	struct ena_adapter *adapter = (struct ena_adapter *)arg;
3545 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3546 	bool dev_up;
3547 	int rc;
3548 
3549 	if (unlikely(!adapter->trigger_reset)) {
3550 		device_printf(adapter->pdev,
3551 		    "device reset scheduled but trigger_reset is off\n");
3552 		return;
3553 	}
3554 
3555 	rw_enter(&adapter->ioctl_sx, RW_WRITER);
3556 
3557 	callout_halt(&adapter->timer_service, &adapter->global_mtx);
3558 
3559 	dev_up = adapter->up;
3560 
3561 	ena_com_set_admin_running_state(ena_dev, false);
3562 	ena_down(adapter);
3563 	ena_free_mgmnt_irq(adapter);
3564 	ena_disable_msix(adapter);
3565 	ena_com_abort_admin_commands(ena_dev);
3566 	ena_com_wait_for_abort_completion(ena_dev);
3567 	ena_com_admin_destroy(ena_dev);
3568 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3569 
3570 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3571 	adapter->trigger_reset = false;
3572 
3573 	/* Finished destroy part. Restart the device */
3574 	rc = ena_device_init(adapter, adapter->pdev, &get_feat_ctx,
3575 	    &adapter->wd_active);
3576 	if (unlikely(rc != 0)) {
3577 		device_printf(adapter->pdev,
3578 		    "ENA device init failed! (err: %d)\n", rc);
3579 		goto err_dev_free;
3580 	}
3581 
3582 	/* XXX dealloc and realloc MSI-X, probably a waste */
3583 	rc = ena_enable_msix_and_set_admin_interrupts(adapter,
3584 	    adapter->num_queues);
3585 	if (unlikely(rc != 0)) {
3586 		device_printf(adapter->pdev, "Enable MSI-X failed\n");
3587 		goto err_com_free;
3588 	}
3589 
3590 	/* If the interface was up before the reset bring it up */
3591 	if (dev_up) {
3592 		rc = ena_up(adapter);
3593 		if (unlikely(rc != 0)) {
3594 			device_printf(adapter->pdev,
3595 			    "Failed to create I/O queues\n");
3596 			goto err_msix_free;
3597 		}
3598 	}
3599 
3600 	callout_reset(&adapter->timer_service, hz,
3601 	    ena_timer_service, (void *)adapter);
3602 
3603 	rw_exit(&adapter->ioctl_sx);
3604 
3605 	return;
3606 
3607 err_msix_free:
3608 	ena_free_mgmnt_irq(adapter);
3609 	ena_disable_msix(adapter);
3610 err_com_free:
3611 	ena_com_admin_destroy(ena_dev);
3612 err_dev_free:
3613 	device_printf(adapter->pdev, "ENA reset failed!\n");
3614 	adapter->running = false;
3615 	rw_exit(&adapter->ioctl_sx);
3616 }
3617 
3618 /**
3619  * ena_attach - Device Initialization Routine
3620  * @pdev: device information struct
3621  *
3622  * Returns 0 on success, otherwise on failure.
3623  *
3624  * ena_attach initializes an adapter identified by a device structure.
3625  * The OS initialization, configuring of the adapter private structure,
3626  * and a hardware reset occur.
3627  **/
3628 static void
3629 ena_attach(device_t parent, device_t self, void *aux)
3630 {
3631 	struct pci_attach_args *pa = aux;
3632 	struct ena_com_dev_get_features_ctx get_feat_ctx;
3633 	static int version_printed;
3634 	struct ena_adapter *adapter = device_private(self);
3635 	struct ena_com_dev *ena_dev = NULL;
3636 	uint16_t tx_sgl_size = 0;
3637 	uint16_t rx_sgl_size = 0;
3638 	pcireg_t reg;
3639 	int io_queue_num;
3640 	int queue_size;
3641 	int rc;
3642 
3643 	adapter->pdev = self;
3644 	adapter->ifp = &adapter->sc_ec.ec_if;
3645 	adapter->sc_pa = *pa;	/* used after attach for adapter reset too */
3646 
3647 	if (pci_dma64_available(pa))
3648 		adapter->sc_dmat = pa->pa_dmat64;
3649 	else
3650 		adapter->sc_dmat = pa->pa_dmat;
3651 
3652 	pci_aprint_devinfo(pa, NULL);
3653 
3654 	reg = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
3655 	if ((reg & PCI_COMMAND_MASTER_ENABLE) == 0) {
3656 		reg |= PCI_COMMAND_MASTER_ENABLE;
3657         	pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, reg);
3658 	}
3659 
3660 	mutex_init(&adapter->global_mtx, MUTEX_DEFAULT, IPL_NET);
3661 	rw_init(&adapter->ioctl_sx);
3662 
3663 	/* Set up the timer service */
3664 	adapter->keep_alive_timeout = DEFAULT_KEEP_ALIVE_TO;
3665 	adapter->missing_tx_timeout = DEFAULT_TX_CMP_TO;
3666 	adapter->missing_tx_max_queues = DEFAULT_TX_MONITORED_QUEUES;
3667 	adapter->missing_tx_threshold = DEFAULT_TX_CMP_THRESHOLD;
3668 
3669 	if (version_printed++ == 0)
3670 		device_printf(parent, "%s\n", ena_version);
3671 
3672 	rc = ena_allocate_pci_resources(pa, adapter);
3673 	if (unlikely(rc != 0)) {
3674 		device_printf(parent, "PCI resource allocation failed!\n");
3675 		ena_free_pci_resources(adapter);
3676 		return;
3677 	}
3678 
3679 	/* Allocate memory for ena_dev structure */
3680 	ena_dev = malloc(sizeof(struct ena_com_dev), M_DEVBUF,
3681 	    M_WAITOK | M_ZERO);
3682 
3683 	adapter->ena_dev = ena_dev;
3684 	ena_dev->dmadev = self;
3685 	ena_dev->bus = malloc(sizeof(struct ena_bus), M_DEVBUF,
3686 	    M_WAITOK | M_ZERO);
3687 
3688 	/* Store register resources */
3689 	((struct ena_bus*)(ena_dev->bus))->reg_bar_t = adapter->sc_btag;
3690 	((struct ena_bus*)(ena_dev->bus))->reg_bar_h = adapter->sc_bhandle;
3691 
3692 	ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3693 
3694 	/* Device initialization */
3695 	rc = ena_device_init(adapter, self, &get_feat_ctx, &adapter->wd_active);
3696 	if (unlikely(rc != 0)) {
3697 		device_printf(self, "ENA device init failed! (err: %d)\n", rc);
3698 		rc = ENXIO;
3699 		goto err_bus_free;
3700 	}
3701 
3702 	adapter->keep_alive_timestamp = getsbinuptime();
3703 
3704 	adapter->tx_offload_cap = get_feat_ctx.offload.tx;
3705 
3706 	/* Set for sure that interface is not up */
3707 	adapter->up = false;
3708 
3709 	memcpy(adapter->mac_addr, get_feat_ctx.dev_attr.mac_addr,
3710 	    ETHER_ADDR_LEN);
3711 
3712 	/* calculate IO queue number to create */
3713 	io_queue_num = ena_calc_io_queue_num(pa, adapter, &get_feat_ctx);
3714 
3715 	ENA_ASSERT(io_queue_num > 0, "Invalid queue number: %d\n",
3716 	    io_queue_num);
3717 	adapter->num_queues = io_queue_num;
3718 
3719 	adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
3720 
3721 	/* calculatre ring sizes */
3722 	queue_size = ena_calc_queue_size(adapter,&tx_sgl_size,
3723 	    &rx_sgl_size, &get_feat_ctx);
3724 	if (unlikely((queue_size <= 0) || (io_queue_num <= 0))) {
3725 		rc = ENA_COM_FAULT;
3726 		goto err_com_free;
3727 	}
3728 
3729 	adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3730 
3731 	adapter->tx_ring_size = queue_size;
3732 	adapter->rx_ring_size = queue_size;
3733 
3734 	adapter->max_tx_sgl_size = tx_sgl_size;
3735 	adapter->max_rx_sgl_size = rx_sgl_size;
3736 
3737 #if 0
3738 	/* set up dma tags for rx and tx buffers */
3739 	rc = ena_setup_tx_dma_tag(adapter);
3740 	if (unlikely(rc != 0)) {
3741 		device_printf(self, "Failed to create TX DMA tag\n");
3742 		goto err_com_free;
3743 	}
3744 
3745 	rc = ena_setup_rx_dma_tag(adapter);
3746 	if (unlikely(rc != 0)) {
3747 		device_printf(self, "Failed to create RX DMA tag\n");
3748 		goto err_tx_tag_free;
3749 	}
3750 #endif
3751 
3752 	/* initialize rings basic information */
3753 	device_printf(self, "initalize %d io queues\n", io_queue_num);
3754 	ena_init_io_rings(adapter);
3755 
3756 	/* setup network interface */
3757 	rc = ena_setup_ifnet(self, adapter, &get_feat_ctx);
3758 	if (unlikely(rc != 0)) {
3759 		device_printf(self, "Error with network interface setup\n");
3760 		goto err_io_free;
3761 	}
3762 
3763 	rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3764 	if (unlikely(rc != 0)) {
3765 		device_printf(self,
3766 		    "Failed to enable and set the admin interrupts\n");
3767 		goto err_ifp_free;
3768 	}
3769 
3770 	callout_init(&adapter->timer_service, CALLOUT_MPSAFE);
3771 
3772 	/* Initialize reset task queue */
3773 	rc = workqueue_create(&adapter->reset_tq, "ena_reset_enq",
3774 	    ena_reset_task, adapter, 0, IPL_NET, WQ_PERCPU | WQ_MPSAFE);
3775 	if (unlikely(rc != 0)) {
3776 		ena_trace(ENA_ALERT,
3777 		    "Unable to create workqueue for reset task\n");
3778 		goto err_ifp_free;
3779 	}
3780 
3781 	/* Initialize statistics */
3782 	ena_alloc_counters_dev(&adapter->dev_stats, io_queue_num);
3783 	ena_alloc_counters_hwstats(&adapter->hw_stats, io_queue_num);
3784 #if 0
3785 	ena_sysctl_add_nodes(adapter);
3786 #endif
3787 
3788 	/* Tell the stack that the interface is not active */
3789 	if_setdrvflagbits(adapter->ifp, IFF_OACTIVE, IFF_RUNNING);
3790 
3791 	adapter->running = true;
3792 	return;
3793 
3794 err_ifp_free:
3795 	if_detach(adapter->ifp);
3796 	if_free(adapter->ifp);
3797 err_io_free:
3798 	ena_free_all_io_rings_resources(adapter);
3799 #if 0
3800 	ena_free_rx_dma_tag(adapter);
3801 err_tx_tag_free:
3802 	ena_free_tx_dma_tag(adapter);
3803 #endif
3804 err_com_free:
3805 	ena_com_admin_destroy(ena_dev);
3806 	ena_com_delete_host_info(ena_dev);
3807 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3808 err_bus_free:
3809 	free(ena_dev->bus, M_DEVBUF);
3810 	free(ena_dev, M_DEVBUF);
3811 	ena_free_pci_resources(adapter);
3812 }
3813 
3814 /**
3815  * ena_detach - Device Removal Routine
3816  * @pdev: device information struct
3817  *
3818  * ena_detach is called by the device subsystem to alert the driver
3819  * that it should release a PCI device.
3820  **/
3821 static int
3822 ena_detach(device_t pdev, int flags)
3823 {
3824 	struct ena_adapter *adapter = device_private(pdev);
3825 	struct ena_com_dev *ena_dev = adapter->ena_dev;
3826 #if 0
3827 	int rc;
3828 #endif
3829 
3830 	/* Make sure VLANS are not using driver */
3831 	if (VLAN_ATTACHED(&adapter->sc_ec)) {
3832 		device_printf(adapter->pdev ,"VLAN is in use, detach first\n");
3833 		return (EBUSY);
3834 	}
3835 
3836 	/* Free reset task and callout */
3837 	callout_halt(&adapter->timer_service, &adapter->global_mtx);
3838 	callout_destroy(&adapter->timer_service);
3839 	workqueue_wait(adapter->reset_tq, &adapter->reset_task);
3840 	workqueue_destroy(adapter->reset_tq);
3841 	adapter->reset_tq = NULL;
3842 
3843 	rw_enter(&adapter->ioctl_sx, RW_WRITER);
3844 	ena_down(adapter);
3845 	rw_exit(&adapter->ioctl_sx);
3846 
3847 	if (adapter->ifp != NULL) {
3848 		ether_ifdetach(adapter->ifp);
3849 		if_free(adapter->ifp);
3850 	}
3851 
3852 	ena_free_all_io_rings_resources(adapter);
3853 
3854 	ena_free_counters((struct evcnt *)&adapter->hw_stats,
3855 	    sizeof(struct ena_hw_stats));
3856 	ena_free_counters((struct evcnt *)&adapter->dev_stats,
3857 	    sizeof(struct ena_stats_dev));
3858 
3859 	if (likely(adapter->rss_support))
3860 		ena_com_rss_destroy(ena_dev);
3861 
3862 #if 0
3863 	rc = ena_free_rx_dma_tag(adapter);
3864 	if (unlikely(rc != 0))
3865 		device_printf(adapter->pdev,
3866 		    "Unmapped RX DMA tag associations\n");
3867 
3868 	rc = ena_free_tx_dma_tag(adapter);
3869 	if (unlikely(rc != 0))
3870 		device_printf(adapter->pdev,
3871 		    "Unmapped TX DMA tag associations\n");
3872 #endif
3873 
3874 	/* Reset the device only if the device is running. */
3875 	if (adapter->running)
3876 		ena_com_dev_reset(ena_dev, adapter->reset_reason);
3877 
3878 	ena_com_delete_host_info(ena_dev);
3879 
3880 	ena_free_irqs(adapter);
3881 
3882 	ena_com_abort_admin_commands(ena_dev);
3883 
3884 	ena_com_wait_for_abort_completion(ena_dev);
3885 
3886 	ena_com_admin_destroy(ena_dev);
3887 
3888 	ena_com_mmio_reg_read_request_destroy(ena_dev);
3889 
3890 	ena_free_pci_resources(adapter);
3891 
3892 	mutex_destroy(&adapter->global_mtx);
3893 	rw_destroy(&adapter->ioctl_sx);
3894 
3895 	if (ena_dev->bus != NULL)
3896 		free(ena_dev->bus, M_DEVBUF);
3897 
3898 	if (ena_dev != NULL)
3899 		free(ena_dev, M_DEVBUF);
3900 
3901 	return 0;
3902 }
3903 
3904 /******************************************************************************
3905  ******************************** AENQ Handlers *******************************
3906  *****************************************************************************/
3907 /**
3908  * ena_update_on_link_change:
3909  * Notify the network interface about the change in link status
3910  **/
3911 static void
3912 ena_update_on_link_change(void *adapter_data,
3913     struct ena_admin_aenq_entry *aenq_e)
3914 {
3915 	struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3916 	struct ena_admin_aenq_link_change_desc *aenq_desc;
3917 	int status;
3918 	struct ifnet *ifp;
3919 
3920 	aenq_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
3921 	ifp = adapter->ifp;
3922 	status = aenq_desc->flags &
3923 	    ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3924 
3925 	if (status != 0) {
3926 		device_printf(adapter->pdev, "link is UP\n");
3927 		if_link_state_change(ifp, LINK_STATE_UP);
3928 	} else if (status == 0) {
3929 		device_printf(adapter->pdev, "link is DOWN\n");
3930 		if_link_state_change(ifp, LINK_STATE_DOWN);
3931 	} else {
3932 		device_printf(adapter->pdev, "invalid value recvd\n");
3933 		BUG();
3934 	}
3935 
3936 	adapter->link_status = status;
3937 }
3938 
3939 /**
3940  * This handler will called for unknown event group or unimplemented handlers
3941  **/
3942 static void
3943 unimplemented_aenq_handler(void *data,
3944     struct ena_admin_aenq_entry *aenq_e)
3945 {
3946 	return;
3947 }
3948 
3949 static struct ena_aenq_handlers aenq_handlers = {
3950     .handlers = {
3951 	    [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3952 	    [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3953     },
3954     .unimplemented_handler = unimplemented_aenq_handler
3955 };
3956 
3957 #ifdef __FreeBSD__
3958 /*********************************************************************
3959  *  FreeBSD Device Interface Entry Points
3960  *********************************************************************/
3961 
3962 static device_method_t ena_methods[] = {
3963     /* Device interface */
3964     DEVMETHOD(device_probe, ena_probe),
3965     DEVMETHOD(device_attach, ena_attach),
3966     DEVMETHOD(device_detach, ena_detach),
3967     DEVMETHOD_END
3968 };
3969 
3970 static driver_t ena_driver = {
3971     "ena", ena_methods, sizeof(struct ena_adapter),
3972 };
3973 
3974 devclass_t ena_devclass;
3975 DRIVER_MODULE(ena, pci, ena_driver, ena_devclass, 0, 0);
3976 MODULE_DEPEND(ena, pci, 1, 1, 1);
3977 MODULE_DEPEND(ena, ether, 1, 1, 1);
3978 
3979 /*********************************************************************/
3980 #endif /* __FreeBSD__ */
3981 
3982 #ifdef __NetBSD__
3983 CFATTACH_DECL_NEW(ena, sizeof(struct ena_adapter), ena_probe, ena_attach,
3984 	ena_detach, NULL);
3985 #endif /* __NetBSD */
3986