xref: /freebsd/sys/dev/vnic/nicvf_queues.c (revision e17f5b1d307b7b8910d67883e57a9604305906d5)
1 /*
2  * Copyright (C) 2015 Cavium Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  *
28  */
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/bitset.h>
38 #include <sys/bitstring.h>
39 #include <sys/buf_ring.h>
40 #include <sys/bus.h>
41 #include <sys/endian.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/module.h>
45 #include <sys/rman.h>
46 #include <sys/pciio.h>
47 #include <sys/pcpu.h>
48 #include <sys/proc.h>
49 #include <sys/sockio.h>
50 #include <sys/socket.h>
51 #include <sys/stdatomic.h>
52 #include <sys/cpuset.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/smp.h>
56 #include <sys/taskqueue.h>
57 
58 #include <vm/vm.h>
59 #include <vm/pmap.h>
60 
61 #include <machine/bus.h>
62 #include <machine/vmparam.h>
63 
64 #include <net/if.h>
65 #include <net/if_var.h>
66 #include <net/if_media.h>
67 #include <net/ifq.h>
68 #include <net/bpf.h>
69 #include <net/ethernet.h>
70 
71 #include <netinet/in_systm.h>
72 #include <netinet/in.h>
73 #include <netinet/if_ether.h>
74 #include <netinet/ip.h>
75 #include <netinet/ip6.h>
76 #include <netinet/sctp.h>
77 #include <netinet/tcp.h>
78 #include <netinet/tcp_lro.h>
79 #include <netinet/udp.h>
80 
81 #include <netinet6/ip6_var.h>
82 
83 #include <dev/pci/pcireg.h>
84 #include <dev/pci/pcivar.h>
85 
86 #include "thunder_bgx.h"
87 #include "nic_reg.h"
88 #include "nic.h"
89 #include "q_struct.h"
90 #include "nicvf_queues.h"
91 
92 #define	DEBUG
93 #undef DEBUG
94 
95 #ifdef DEBUG
96 #define	dprintf(dev, fmt, ...)	device_printf(dev, fmt, ##__VA_ARGS__)
97 #else
98 #define	dprintf(dev, fmt, ...)
99 #endif
100 
101 MALLOC_DECLARE(M_NICVF);
102 
103 static void nicvf_free_snd_queue(struct nicvf *, struct snd_queue *);
104 static struct mbuf * nicvf_get_rcv_mbuf(struct nicvf *, struct cqe_rx_t *);
105 static void nicvf_sq_disable(struct nicvf *, int);
106 static void nicvf_sq_enable(struct nicvf *, struct snd_queue *, int);
107 static void nicvf_put_sq_desc(struct snd_queue *, int);
108 static void nicvf_cmp_queue_config(struct nicvf *, struct queue_set *, int,
109     boolean_t);
110 static void nicvf_sq_free_used_descs(struct nicvf *, struct snd_queue *, int);
111 
112 static int nicvf_tx_mbuf_locked(struct snd_queue *, struct mbuf **);
113 
114 static void nicvf_rbdr_task(void *, int);
115 static void nicvf_rbdr_task_nowait(void *, int);
116 
117 struct rbuf_info {
118 	bus_dma_tag_t	dmat;
119 	bus_dmamap_t	dmap;
120 	struct mbuf *	mbuf;
121 };
122 
123 #define GET_RBUF_INFO(x) ((struct rbuf_info *)((x) - NICVF_RCV_BUF_ALIGN_BYTES))
124 
125 /* Poll a register for a specific value */
126 static int nicvf_poll_reg(struct nicvf *nic, int qidx,
127 			  uint64_t reg, int bit_pos, int bits, int val)
128 {
129 	uint64_t bit_mask;
130 	uint64_t reg_val;
131 	int timeout = 10;
132 
133 	bit_mask = (1UL << bits) - 1;
134 	bit_mask = (bit_mask << bit_pos);
135 
136 	while (timeout) {
137 		reg_val = nicvf_queue_reg_read(nic, reg, qidx);
138 		if (((reg_val & bit_mask) >> bit_pos) == val)
139 			return (0);
140 
141 		DELAY(1000);
142 		timeout--;
143 	}
144 	device_printf(nic->dev, "Poll on reg 0x%lx failed\n", reg);
145 	return (ETIMEDOUT);
146 }
147 
148 /* Callback for bus_dmamap_load() */
149 static void
150 nicvf_dmamap_q_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
151 {
152 	bus_addr_t *paddr;
153 
154 	KASSERT(nseg == 1, ("wrong number of segments, should be 1"));
155 	paddr = arg;
156 	*paddr = segs->ds_addr;
157 }
158 
159 /* Allocate memory for a queue's descriptors */
160 static int
161 nicvf_alloc_q_desc_mem(struct nicvf *nic, struct q_desc_mem *dmem,
162     int q_len, int desc_size, int align_bytes)
163 {
164 	int err, err_dmat;
165 
166 	/* Create DMA tag first */
167 	err = bus_dma_tag_create(
168 	    bus_get_dma_tag(nic->dev),		/* parent tag */
169 	    align_bytes,			/* alignment */
170 	    0,					/* boundary */
171 	    BUS_SPACE_MAXADDR,			/* lowaddr */
172 	    BUS_SPACE_MAXADDR,			/* highaddr */
173 	    NULL, NULL,				/* filtfunc, filtfuncarg */
174 	    (q_len * desc_size),		/* maxsize */
175 	    1,					/* nsegments */
176 	    (q_len * desc_size),		/* maxsegsize */
177 	    0,					/* flags */
178 	    NULL, NULL,				/* lockfunc, lockfuncarg */
179 	    &dmem->dmat);			/* dmat */
180 
181 	if (err != 0) {
182 		device_printf(nic->dev,
183 		    "Failed to create busdma tag for descriptors ring\n");
184 		return (err);
185 	}
186 
187 	/* Allocate segment of continuous DMA safe memory */
188 	err = bus_dmamem_alloc(
189 	    dmem->dmat,				/* DMA tag */
190 	    &dmem->base,			/* virtual address */
191 	    (BUS_DMA_NOWAIT | BUS_DMA_ZERO),	/* flags */
192 	    &dmem->dmap);			/* DMA map */
193 	if (err != 0) {
194 		device_printf(nic->dev, "Failed to allocate DMA safe memory for"
195 		    "descriptors ring\n");
196 		goto dmamem_fail;
197 	}
198 
199 	err = bus_dmamap_load(
200 	    dmem->dmat,
201 	    dmem->dmap,
202 	    dmem->base,
203 	    (q_len * desc_size),		/* allocation size */
204 	    nicvf_dmamap_q_cb,			/* map to DMA address cb. */
205 	    &dmem->phys_base,			/* physical address */
206 	    BUS_DMA_NOWAIT);
207 	if (err != 0) {
208 		device_printf(nic->dev,
209 		    "Cannot load DMA map of descriptors ring\n");
210 		goto dmamap_fail;
211 	}
212 
213 	dmem->q_len = q_len;
214 	dmem->size = (desc_size * q_len);
215 
216 	return (0);
217 
218 dmamap_fail:
219 	bus_dmamem_free(dmem->dmat, dmem->base, dmem->dmap);
220 	dmem->phys_base = 0;
221 dmamem_fail:
222 	err_dmat = bus_dma_tag_destroy(dmem->dmat);
223 	dmem->base = NULL;
224 	KASSERT(err_dmat == 0,
225 	    ("%s: Trying to destroy BUSY DMA tag", __func__));
226 
227 	return (err);
228 }
229 
230 /* Free queue's descriptor memory */
231 static void
232 nicvf_free_q_desc_mem(struct nicvf *nic, struct q_desc_mem *dmem)
233 {
234 	int err;
235 
236 	if ((dmem == NULL) || (dmem->base == NULL))
237 		return;
238 
239 	/* Unload a map */
240 	bus_dmamap_sync(dmem->dmat, dmem->dmap, BUS_DMASYNC_POSTREAD);
241 	bus_dmamap_unload(dmem->dmat, dmem->dmap);
242 	/* Free DMA memory */
243 	bus_dmamem_free(dmem->dmat, dmem->base, dmem->dmap);
244 	/* Destroy DMA tag */
245 	err = bus_dma_tag_destroy(dmem->dmat);
246 
247 	KASSERT(err == 0,
248 	    ("%s: Trying to destroy BUSY DMA tag", __func__));
249 
250 	dmem->phys_base = 0;
251 	dmem->base = NULL;
252 }
253 
254 /*
255  * Allocate buffer for packet reception
256  * HW returns memory address where packet is DMA'ed but not a pointer
257  * into RBDR ring, so save buffer address at the start of fragment and
258  * align the start address to a cache aligned address
259  */
260 static __inline int
261 nicvf_alloc_rcv_buffer(struct nicvf *nic, struct rbdr *rbdr,
262     bus_dmamap_t dmap, int mflags, uint32_t buf_len, bus_addr_t *rbuf)
263 {
264 	struct mbuf *mbuf;
265 	struct rbuf_info *rinfo;
266 	bus_dma_segment_t segs[1];
267 	int nsegs;
268 	int err;
269 
270 	mbuf = m_getjcl(mflags, MT_DATA, M_PKTHDR, MCLBYTES);
271 	if (mbuf == NULL)
272 		return (ENOMEM);
273 
274 	/*
275 	 * The length is equal to the actual length + one 128b line
276 	 * used as a room for rbuf_info structure.
277 	 */
278 	mbuf->m_len = mbuf->m_pkthdr.len = buf_len;
279 
280 	err = bus_dmamap_load_mbuf_sg(rbdr->rbdr_buff_dmat, dmap, mbuf, segs,
281 	    &nsegs, BUS_DMA_NOWAIT);
282 	if (err != 0) {
283 		device_printf(nic->dev,
284 		    "Failed to map mbuf into DMA visible memory, err: %d\n",
285 		    err);
286 		m_freem(mbuf);
287 		bus_dmamap_destroy(rbdr->rbdr_buff_dmat, dmap);
288 		return (err);
289 	}
290 	if (nsegs != 1)
291 		panic("Unexpected number of DMA segments for RB: %d", nsegs);
292 	/*
293 	 * Now use the room for rbuf_info structure
294 	 * and adjust mbuf data and length.
295 	 */
296 	rinfo = (struct rbuf_info *)mbuf->m_data;
297 	m_adj(mbuf, NICVF_RCV_BUF_ALIGN_BYTES);
298 
299 	rinfo->dmat = rbdr->rbdr_buff_dmat;
300 	rinfo->dmap = dmap;
301 	rinfo->mbuf = mbuf;
302 
303 	*rbuf = segs[0].ds_addr + NICVF_RCV_BUF_ALIGN_BYTES;
304 
305 	return (0);
306 }
307 
308 /* Retrieve mbuf for received packet */
309 static struct mbuf *
310 nicvf_rb_ptr_to_mbuf(struct nicvf *nic, bus_addr_t rb_ptr)
311 {
312 	struct mbuf *mbuf;
313 	struct rbuf_info *rinfo;
314 
315 	/* Get buffer start address and alignment offset */
316 	rinfo = GET_RBUF_INFO(PHYS_TO_DMAP(rb_ptr));
317 
318 	/* Now retrieve mbuf to give to stack */
319 	mbuf = rinfo->mbuf;
320 	if (__predict_false(mbuf == NULL)) {
321 		panic("%s: Received packet fragment with NULL mbuf",
322 		    device_get_nameunit(nic->dev));
323 	}
324 	/*
325 	 * Clear the mbuf in the descriptor to indicate
326 	 * that this slot is processed and free to use.
327 	 */
328 	rinfo->mbuf = NULL;
329 
330 	bus_dmamap_sync(rinfo->dmat, rinfo->dmap, BUS_DMASYNC_POSTREAD);
331 	bus_dmamap_unload(rinfo->dmat, rinfo->dmap);
332 
333 	return (mbuf);
334 }
335 
336 /* Allocate RBDR ring and populate receive buffers */
337 static int
338 nicvf_init_rbdr(struct nicvf *nic, struct rbdr *rbdr, int ring_len,
339     int buf_size, int qidx)
340 {
341 	bus_dmamap_t dmap;
342 	bus_addr_t rbuf;
343 	struct rbdr_entry_t *desc;
344 	int idx;
345 	int err;
346 
347 	/* Allocate rbdr descriptors ring */
348 	err = nicvf_alloc_q_desc_mem(nic, &rbdr->dmem, ring_len,
349 	    sizeof(struct rbdr_entry_t), NICVF_RCV_BUF_ALIGN_BYTES);
350 	if (err != 0) {
351 		device_printf(nic->dev,
352 		    "Failed to create RBDR descriptors ring\n");
353 		return (err);
354 	}
355 
356 	rbdr->desc = rbdr->dmem.base;
357 	/*
358 	 * Buffer size has to be in multiples of 128 bytes.
359 	 * Make room for metadata of size of one line (128 bytes).
360 	 */
361 	rbdr->dma_size = buf_size - NICVF_RCV_BUF_ALIGN_BYTES;
362 	rbdr->enable = TRUE;
363 	rbdr->thresh = RBDR_THRESH;
364 	rbdr->nic = nic;
365 	rbdr->idx = qidx;
366 
367 	/*
368 	 * Create DMA tag for Rx buffers.
369 	 * Each map created using this tag is intended to store Rx payload for
370 	 * one fragment and one header structure containing rbuf_info (thus
371 	 * additional 128 byte line since RB must be a multiple of 128 byte
372 	 * cache line).
373 	 */
374 	if (buf_size > MCLBYTES) {
375 		device_printf(nic->dev,
376 		    "Buffer size to large for mbuf cluster\n");
377 		return (EINVAL);
378 	}
379 	err = bus_dma_tag_create(
380 	    bus_get_dma_tag(nic->dev),		/* parent tag */
381 	    NICVF_RCV_BUF_ALIGN_BYTES,		/* alignment */
382 	    0,					/* boundary */
383 	    DMAP_MAX_PHYSADDR,			/* lowaddr */
384 	    DMAP_MIN_PHYSADDR,			/* highaddr */
385 	    NULL, NULL,				/* filtfunc, filtfuncarg */
386 	    roundup2(buf_size, MCLBYTES),	/* maxsize */
387 	    1,					/* nsegments */
388 	    roundup2(buf_size, MCLBYTES),	/* maxsegsize */
389 	    0,					/* flags */
390 	    NULL, NULL,				/* lockfunc, lockfuncarg */
391 	    &rbdr->rbdr_buff_dmat);		/* dmat */
392 
393 	if (err != 0) {
394 		device_printf(nic->dev,
395 		    "Failed to create busdma tag for RBDR buffers\n");
396 		return (err);
397 	}
398 
399 	rbdr->rbdr_buff_dmaps = malloc(sizeof(*rbdr->rbdr_buff_dmaps) *
400 	    ring_len, M_NICVF, (M_WAITOK | M_ZERO));
401 
402 	for (idx = 0; idx < ring_len; idx++) {
403 		err = bus_dmamap_create(rbdr->rbdr_buff_dmat, 0, &dmap);
404 		if (err != 0) {
405 			device_printf(nic->dev,
406 			    "Failed to create DMA map for RB\n");
407 			return (err);
408 		}
409 		rbdr->rbdr_buff_dmaps[idx] = dmap;
410 
411 		err = nicvf_alloc_rcv_buffer(nic, rbdr, dmap, M_WAITOK,
412 		    DMA_BUFFER_LEN, &rbuf);
413 		if (err != 0)
414 			return (err);
415 
416 		desc = GET_RBDR_DESC(rbdr, idx);
417 		desc->buf_addr = (rbuf >> NICVF_RCV_BUF_ALIGN);
418 	}
419 
420 	/* Allocate taskqueue */
421 	TASK_INIT(&rbdr->rbdr_task, 0, nicvf_rbdr_task, rbdr);
422 	TASK_INIT(&rbdr->rbdr_task_nowait, 0, nicvf_rbdr_task_nowait, rbdr);
423 	rbdr->rbdr_taskq = taskqueue_create_fast("nicvf_rbdr_taskq", M_WAITOK,
424 	    taskqueue_thread_enqueue, &rbdr->rbdr_taskq);
425 	taskqueue_start_threads(&rbdr->rbdr_taskq, 1, PI_NET, "%s: rbdr_taskq",
426 	    device_get_nameunit(nic->dev));
427 
428 	return (0);
429 }
430 
431 /* Free RBDR ring and its receive buffers */
432 static void
433 nicvf_free_rbdr(struct nicvf *nic, struct rbdr *rbdr)
434 {
435 	struct mbuf *mbuf;
436 	struct queue_set *qs;
437 	struct rbdr_entry_t *desc;
438 	struct rbuf_info *rinfo;
439 	bus_addr_t buf_addr;
440 	int head, tail, idx;
441 	int err;
442 
443 	qs = nic->qs;
444 
445 	if ((qs == NULL) || (rbdr == NULL))
446 		return;
447 
448 	rbdr->enable = FALSE;
449 	if (rbdr->rbdr_taskq != NULL) {
450 		/* Remove tasks */
451 		while (taskqueue_cancel(rbdr->rbdr_taskq,
452 		    &rbdr->rbdr_task_nowait, NULL) != 0) {
453 			/* Finish the nowait task first */
454 			taskqueue_drain(rbdr->rbdr_taskq,
455 			    &rbdr->rbdr_task_nowait);
456 		}
457 		taskqueue_free(rbdr->rbdr_taskq);
458 		rbdr->rbdr_taskq = NULL;
459 
460 		while (taskqueue_cancel(taskqueue_thread,
461 		    &rbdr->rbdr_task, NULL) != 0) {
462 			/* Now finish the sleepable task */
463 			taskqueue_drain(taskqueue_thread, &rbdr->rbdr_task);
464 		}
465 	}
466 
467 	/*
468 	 * Free all of the memory under the RB descriptors.
469 	 * There are assumptions here:
470 	 * 1. Corresponding RBDR is disabled
471 	 *    - it is safe to operate using head and tail indexes
472 	 * 2. All bffers that were received are properly freed by
473 	 *    the receive handler
474 	 *    - there is no need to unload DMA map and free MBUF for other
475 	 *      descriptors than unused ones
476 	 */
477 	if (rbdr->rbdr_buff_dmat != NULL) {
478 		head = rbdr->head;
479 		tail = rbdr->tail;
480 		while (head != tail) {
481 			desc = GET_RBDR_DESC(rbdr, head);
482 			buf_addr = desc->buf_addr << NICVF_RCV_BUF_ALIGN;
483 			rinfo = GET_RBUF_INFO(PHYS_TO_DMAP(buf_addr));
484 			bus_dmamap_unload(rbdr->rbdr_buff_dmat, rinfo->dmap);
485 			mbuf = rinfo->mbuf;
486 			/* This will destroy everything including rinfo! */
487 			m_freem(mbuf);
488 			head++;
489 			head &= (rbdr->dmem.q_len - 1);
490 		}
491 		/* Free tail descriptor */
492 		desc = GET_RBDR_DESC(rbdr, tail);
493 		buf_addr = desc->buf_addr << NICVF_RCV_BUF_ALIGN;
494 		rinfo = GET_RBUF_INFO(PHYS_TO_DMAP(buf_addr));
495 		bus_dmamap_unload(rbdr->rbdr_buff_dmat, rinfo->dmap);
496 		mbuf = rinfo->mbuf;
497 		/* This will destroy everything including rinfo! */
498 		m_freem(mbuf);
499 
500 		/* Destroy DMA maps */
501 		for (idx = 0; idx < qs->rbdr_len; idx++) {
502 			if (rbdr->rbdr_buff_dmaps[idx] == NULL)
503 				continue;
504 			err = bus_dmamap_destroy(rbdr->rbdr_buff_dmat,
505 			    rbdr->rbdr_buff_dmaps[idx]);
506 			KASSERT(err == 0,
507 			    ("%s: Could not destroy DMA map for RB, desc: %d",
508 			    __func__, idx));
509 			rbdr->rbdr_buff_dmaps[idx] = NULL;
510 		}
511 
512 		/* Now destroy the tag */
513 		err = bus_dma_tag_destroy(rbdr->rbdr_buff_dmat);
514 		KASSERT(err == 0,
515 		    ("%s: Trying to destroy BUSY DMA tag", __func__));
516 
517 		rbdr->head = 0;
518 		rbdr->tail = 0;
519 	}
520 
521 	/* Free RBDR ring */
522 	nicvf_free_q_desc_mem(nic, &rbdr->dmem);
523 }
524 
525 /*
526  * Refill receive buffer descriptors with new buffers.
527  */
528 static int
529 nicvf_refill_rbdr(struct rbdr *rbdr, int mflags)
530 {
531 	struct nicvf *nic;
532 	struct queue_set *qs;
533 	int rbdr_idx;
534 	int tail, qcount;
535 	int refill_rb_cnt;
536 	struct rbdr_entry_t *desc;
537 	bus_dmamap_t dmap;
538 	bus_addr_t rbuf;
539 	boolean_t rb_alloc_fail;
540 	int new_rb;
541 
542 	rb_alloc_fail = TRUE;
543 	new_rb = 0;
544 	nic = rbdr->nic;
545 	qs = nic->qs;
546 	rbdr_idx = rbdr->idx;
547 
548 	/* Check if it's enabled */
549 	if (!rbdr->enable)
550 		return (0);
551 
552 	/* Get no of desc's to be refilled */
553 	qcount = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_STATUS0, rbdr_idx);
554 	qcount &= 0x7FFFF;
555 	/* Doorbell can be ringed with a max of ring size minus 1 */
556 	if (qcount >= (qs->rbdr_len - 1)) {
557 		rb_alloc_fail = FALSE;
558 		goto out;
559 	} else
560 		refill_rb_cnt = qs->rbdr_len - qcount - 1;
561 
562 	/* Start filling descs from tail */
563 	tail = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_TAIL, rbdr_idx) >> 3;
564 	while (refill_rb_cnt) {
565 		tail++;
566 		tail &= (rbdr->dmem.q_len - 1);
567 
568 		dmap = rbdr->rbdr_buff_dmaps[tail];
569 		if (nicvf_alloc_rcv_buffer(nic, rbdr, dmap, mflags,
570 		    DMA_BUFFER_LEN, &rbuf)) {
571 			/* Something went wrong. Resign */
572 			break;
573 		}
574 		desc = GET_RBDR_DESC(rbdr, tail);
575 		desc->buf_addr = (rbuf >> NICVF_RCV_BUF_ALIGN);
576 		refill_rb_cnt--;
577 		new_rb++;
578 	}
579 
580 	/* make sure all memory stores are done before ringing doorbell */
581 	wmb();
582 
583 	/* Check if buffer allocation failed */
584 	if (refill_rb_cnt == 0)
585 		rb_alloc_fail = FALSE;
586 
587 	/* Notify HW */
588 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_DOOR,
589 			      rbdr_idx, new_rb);
590 out:
591 	if (!rb_alloc_fail) {
592 		/*
593 		 * Re-enable RBDR interrupts only
594 		 * if buffer allocation is success.
595 		 */
596 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, rbdr_idx);
597 
598 		return (0);
599 	}
600 
601 	return (ENOMEM);
602 }
603 
604 /* Refill RBs even if sleep is needed to reclaim memory */
605 static void
606 nicvf_rbdr_task(void *arg, int pending)
607 {
608 	struct rbdr *rbdr;
609 	int err;
610 
611 	rbdr = (struct rbdr *)arg;
612 
613 	err = nicvf_refill_rbdr(rbdr, M_WAITOK);
614 	if (__predict_false(err != 0)) {
615 		panic("%s: Failed to refill RBs even when sleep enabled",
616 		    __func__);
617 	}
618 }
619 
620 /* Refill RBs as soon as possible without waiting */
621 static void
622 nicvf_rbdr_task_nowait(void *arg, int pending)
623 {
624 	struct rbdr *rbdr;
625 	int err;
626 
627 	rbdr = (struct rbdr *)arg;
628 
629 	err = nicvf_refill_rbdr(rbdr, M_NOWAIT);
630 	if (err != 0) {
631 		/*
632 		 * Schedule another, sleepable kernel thread
633 		 * that will for sure refill the buffers.
634 		 */
635 		taskqueue_enqueue(taskqueue_thread, &rbdr->rbdr_task);
636 	}
637 }
638 
639 static int
640 nicvf_rcv_pkt_handler(struct nicvf *nic, struct cmp_queue *cq,
641     struct cqe_rx_t *cqe_rx, int cqe_type)
642 {
643 	struct mbuf *mbuf;
644 	struct rcv_queue *rq;
645 	int rq_idx;
646 	int err = 0;
647 
648 	rq_idx = cqe_rx->rq_idx;
649 	rq = &nic->qs->rq[rq_idx];
650 
651 	/* Check for errors */
652 	err = nicvf_check_cqe_rx_errs(nic, cq, cqe_rx);
653 	if (err && !cqe_rx->rb_cnt)
654 		return (0);
655 
656 	mbuf = nicvf_get_rcv_mbuf(nic, cqe_rx);
657 	if (mbuf == NULL) {
658 		dprintf(nic->dev, "Packet not received\n");
659 		return (0);
660 	}
661 
662 	/* If error packet */
663 	if (err != 0) {
664 		m_freem(mbuf);
665 		return (0);
666 	}
667 
668 	if (rq->lro_enabled &&
669 	    ((cqe_rx->l3_type == L3TYPE_IPV4) && (cqe_rx->l4_type == L4TYPE_TCP)) &&
670 	    (mbuf->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) ==
671             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) {
672 		/*
673 		 * At this point it is known that there are no errors in the
674 		 * packet. Attempt to LRO enqueue. Send to stack if no resources
675 		 * or enqueue error.
676 		 */
677 		if ((rq->lro.lro_cnt != 0) &&
678 		    (tcp_lro_rx(&rq->lro, mbuf, 0) == 0))
679 			return (0);
680 	}
681 	/*
682 	 * Push this packet to the stack later to avoid
683 	 * unlocking completion task in the middle of work.
684 	 */
685 	err = buf_ring_enqueue(cq->rx_br, mbuf);
686 	if (err != 0) {
687 		/*
688 		 * Failed to enqueue this mbuf.
689 		 * We don't drop it, just schedule another task.
690 		 */
691 		return (err);
692 	}
693 
694 	return (0);
695 }
696 
697 static void
698 nicvf_snd_pkt_handler(struct nicvf *nic, struct cmp_queue *cq,
699     struct cqe_send_t *cqe_tx, int cqe_type)
700 {
701 	bus_dmamap_t dmap;
702 	struct mbuf *mbuf;
703 	struct snd_queue *sq;
704 	struct sq_hdr_subdesc *hdr;
705 
706 	mbuf = NULL;
707 	sq = &nic->qs->sq[cqe_tx->sq_idx];
708 
709 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr);
710 	if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER)
711 		return;
712 
713 	dprintf(nic->dev,
714 	    "%s Qset #%d SQ #%d SQ ptr #%d subdesc count %d\n",
715 	    __func__, cqe_tx->sq_qs, cqe_tx->sq_idx,
716 	    cqe_tx->sqe_ptr, hdr->subdesc_cnt);
717 
718 	dmap = (bus_dmamap_t)sq->snd_buff[cqe_tx->sqe_ptr].dmap;
719 	bus_dmamap_unload(sq->snd_buff_dmat, dmap);
720 
721 	mbuf = (struct mbuf *)sq->snd_buff[cqe_tx->sqe_ptr].mbuf;
722 	if (mbuf != NULL) {
723 		m_freem(mbuf);
724 		sq->snd_buff[cqe_tx->sqe_ptr].mbuf = NULL;
725 		nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1);
726 	}
727 
728 	nicvf_check_cqe_tx_errs(nic, cq, cqe_tx);
729 }
730 
731 static int
732 nicvf_cq_intr_handler(struct nicvf *nic, uint8_t cq_idx)
733 {
734 	struct mbuf *mbuf;
735 	struct ifnet *ifp;
736 	int processed_cqe, work_done = 0, tx_done = 0;
737 	int cqe_count, cqe_head;
738 	struct queue_set *qs = nic->qs;
739 	struct cmp_queue *cq = &qs->cq[cq_idx];
740 	struct snd_queue *sq = &qs->sq[cq_idx];
741 	struct rcv_queue *rq;
742 	struct cqe_rx_t *cq_desc;
743 	struct lro_ctrl	*lro;
744 	int rq_idx;
745 	int cmp_err;
746 
747 	NICVF_CMP_LOCK(cq);
748 	cmp_err = 0;
749 	processed_cqe = 0;
750 	/* Get no of valid CQ entries to process */
751 	cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx);
752 	cqe_count &= CQ_CQE_COUNT;
753 	if (cqe_count == 0)
754 		goto out;
755 
756 	/* Get head of the valid CQ entries */
757 	cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9;
758 	cqe_head &= 0xFFFF;
759 
760 	dprintf(nic->dev, "%s CQ%d cqe_count %d cqe_head %d\n",
761 	    __func__, cq_idx, cqe_count, cqe_head);
762 	while (processed_cqe < cqe_count) {
763 		/* Get the CQ descriptor */
764 		cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head);
765 		cqe_head++;
766 		cqe_head &= (cq->dmem.q_len - 1);
767 		/* Prefetch next CQ descriptor */
768 		__builtin_prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head));
769 
770 		dprintf(nic->dev, "CQ%d cq_desc->cqe_type %d\n", cq_idx,
771 		    cq_desc->cqe_type);
772 		switch (cq_desc->cqe_type) {
773 		case CQE_TYPE_RX:
774 			cmp_err = nicvf_rcv_pkt_handler(nic, cq, cq_desc,
775 			    CQE_TYPE_RX);
776 			if (__predict_false(cmp_err != 0)) {
777 				/*
778 				 * Ups. Cannot finish now.
779 				 * Let's try again later.
780 				 */
781 				goto done;
782 			}
783 			work_done++;
784 			break;
785 		case CQE_TYPE_SEND:
786 			nicvf_snd_pkt_handler(nic, cq, (void *)cq_desc,
787 			    CQE_TYPE_SEND);
788 			tx_done++;
789 			break;
790 		case CQE_TYPE_INVALID:
791 		case CQE_TYPE_RX_SPLIT:
792 		case CQE_TYPE_RX_TCP:
793 		case CQE_TYPE_SEND_PTP:
794 			/* Ignore for now */
795 			break;
796 		}
797 		processed_cqe++;
798 	}
799 done:
800 	dprintf(nic->dev,
801 	    "%s CQ%d processed_cqe %d work_done %d\n",
802 	    __func__, cq_idx, processed_cqe, work_done);
803 
804 	/* Ring doorbell to inform H/W to reuse processed CQEs */
805 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR, cq_idx, processed_cqe);
806 
807 	if ((tx_done > 0) &&
808 	    ((if_getdrvflags(nic->ifp) & IFF_DRV_RUNNING) != 0)) {
809 		/* Reenable TXQ if its stopped earlier due to SQ full */
810 		if_setdrvflagbits(nic->ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
811 		taskqueue_enqueue(sq->snd_taskq, &sq->snd_task);
812 	}
813 out:
814 	/*
815 	 * Flush any outstanding LRO work
816 	 */
817 	rq_idx = cq_idx;
818 	rq = &nic->qs->rq[rq_idx];
819 	lro = &rq->lro;
820 	tcp_lro_flush_all(lro);
821 
822 	NICVF_CMP_UNLOCK(cq);
823 
824 	ifp = nic->ifp;
825 	/* Push received MBUFs to the stack */
826 	while (!buf_ring_empty(cq->rx_br)) {
827 		mbuf = buf_ring_dequeue_mc(cq->rx_br);
828 		if (__predict_true(mbuf != NULL))
829 			(*ifp->if_input)(ifp, mbuf);
830 	}
831 
832 	return (cmp_err);
833 }
834 
835 /*
836  * Qset error interrupt handler
837  *
838  * As of now only CQ errors are handled
839  */
840 static void
841 nicvf_qs_err_task(void *arg, int pending)
842 {
843 	struct nicvf *nic;
844 	struct queue_set *qs;
845 	int qidx;
846 	uint64_t status;
847 	boolean_t enable = TRUE;
848 
849 	nic = (struct nicvf *)arg;
850 	qs = nic->qs;
851 
852 	/* Deactivate network interface */
853 	if_setdrvflagbits(nic->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
854 
855 	/* Check if it is CQ err */
856 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
857 		status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS,
858 		    qidx);
859 		if ((status & CQ_ERR_MASK) == 0)
860 			continue;
861 		/* Process already queued CQEs and reconfig CQ */
862 		nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
863 		nicvf_sq_disable(nic, qidx);
864 		(void)nicvf_cq_intr_handler(nic, qidx);
865 		nicvf_cmp_queue_config(nic, qs, qidx, enable);
866 		nicvf_sq_free_used_descs(nic, &qs->sq[qidx], qidx);
867 		nicvf_sq_enable(nic, &qs->sq[qidx], qidx);
868 		nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx);
869 	}
870 
871 	if_setdrvflagbits(nic->ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
872 	/* Re-enable Qset error interrupt */
873 	nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0);
874 }
875 
876 static void
877 nicvf_cmp_task(void *arg, int pending)
878 {
879 	struct cmp_queue *cq;
880 	struct nicvf *nic;
881 	int cmp_err;
882 
883 	cq = (struct cmp_queue *)arg;
884 	nic = cq->nic;
885 
886 	/* Handle CQ descriptors */
887 	cmp_err = nicvf_cq_intr_handler(nic, cq->idx);
888 	if (__predict_false(cmp_err != 0)) {
889 		/*
890 		 * Schedule another thread here since we did not
891 		 * process the entire CQ due to Tx or Rx CQ parse error.
892 		 */
893 		taskqueue_enqueue(cq->cmp_taskq, &cq->cmp_task);
894 
895 	}
896 
897 	nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->idx);
898 	/* Reenable interrupt (previously disabled in nicvf_intr_handler() */
899 	nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->idx);
900 
901 }
902 
903 /* Initialize completion queue */
904 static int
905 nicvf_init_cmp_queue(struct nicvf *nic, struct cmp_queue *cq, int q_len,
906     int qidx)
907 {
908 	int err;
909 
910 	/* Initizalize lock */
911 	snprintf(cq->mtx_name, sizeof(cq->mtx_name), "%s: CQ(%d) lock",
912 	    device_get_nameunit(nic->dev), qidx);
913 	mtx_init(&cq->mtx, cq->mtx_name, NULL, MTX_DEF);
914 
915 	err = nicvf_alloc_q_desc_mem(nic, &cq->dmem, q_len, CMP_QUEUE_DESC_SIZE,
916 				     NICVF_CQ_BASE_ALIGN_BYTES);
917 
918 	if (err != 0) {
919 		device_printf(nic->dev,
920 		    "Could not allocate DMA memory for CQ\n");
921 		return (err);
922 	}
923 
924 	cq->desc = cq->dmem.base;
925 	cq->thresh = pass1_silicon(nic->dev) ? 0 : CMP_QUEUE_CQE_THRESH;
926 	cq->nic = nic;
927 	cq->idx = qidx;
928 	nic->cq_coalesce_usecs = (CMP_QUEUE_TIMER_THRESH * 0.05) - 1;
929 
930 	cq->rx_br = buf_ring_alloc(CMP_QUEUE_LEN * 8, M_DEVBUF, M_WAITOK,
931 	    &cq->mtx);
932 
933 	/* Allocate taskqueue */
934 	NET_TASK_INIT(&cq->cmp_task, 0, nicvf_cmp_task, cq);
935 	cq->cmp_taskq = taskqueue_create_fast("nicvf_cmp_taskq", M_WAITOK,
936 	    taskqueue_thread_enqueue, &cq->cmp_taskq);
937 	taskqueue_start_threads(&cq->cmp_taskq, 1, PI_NET, "%s: cmp_taskq(%d)",
938 	    device_get_nameunit(nic->dev), qidx);
939 
940 	return (0);
941 }
942 
943 static void
944 nicvf_free_cmp_queue(struct nicvf *nic, struct cmp_queue *cq)
945 {
946 
947 	if (cq == NULL)
948 		return;
949 	/*
950 	 * The completion queue itself should be disabled by now
951 	 * (ref. nicvf_snd_queue_config()).
952 	 * Ensure that it is safe to disable it or panic.
953 	 */
954 	if (cq->enable)
955 		panic("%s: Trying to free working CQ(%d)", __func__, cq->idx);
956 
957 	if (cq->cmp_taskq != NULL) {
958 		/* Remove task */
959 		while (taskqueue_cancel(cq->cmp_taskq, &cq->cmp_task, NULL) != 0)
960 			taskqueue_drain(cq->cmp_taskq, &cq->cmp_task);
961 
962 		taskqueue_free(cq->cmp_taskq);
963 		cq->cmp_taskq = NULL;
964 	}
965 	/*
966 	 * Completion interrupt will possibly enable interrupts again
967 	 * so disable interrupting now after we finished processing
968 	 * completion task. It is safe to do so since the corresponding CQ
969 	 * was already disabled.
970 	 */
971 	nicvf_disable_intr(nic, NICVF_INTR_CQ, cq->idx);
972 	nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->idx);
973 
974 	NICVF_CMP_LOCK(cq);
975 	nicvf_free_q_desc_mem(nic, &cq->dmem);
976 	drbr_free(cq->rx_br, M_DEVBUF);
977 	NICVF_CMP_UNLOCK(cq);
978 	mtx_destroy(&cq->mtx);
979 	memset(cq->mtx_name, 0, sizeof(cq->mtx_name));
980 }
981 
982 int
983 nicvf_xmit_locked(struct snd_queue *sq)
984 {
985 	struct nicvf *nic;
986 	struct ifnet *ifp;
987 	struct mbuf *next;
988 	int err;
989 
990 	NICVF_TX_LOCK_ASSERT(sq);
991 
992 	nic = sq->nic;
993 	ifp = nic->ifp;
994 	err = 0;
995 
996 	while ((next = drbr_peek(ifp, sq->br)) != NULL) {
997 		/* Send a copy of the frame to the BPF listener */
998 		ETHER_BPF_MTAP(ifp, next);
999 
1000 		err = nicvf_tx_mbuf_locked(sq, &next);
1001 		if (err != 0) {
1002 			if (next == NULL)
1003 				drbr_advance(ifp, sq->br);
1004 			else
1005 				drbr_putback(ifp, sq->br, next);
1006 
1007 			break;
1008 		}
1009 		drbr_advance(ifp, sq->br);
1010 	}
1011 	return (err);
1012 }
1013 
1014 static void
1015 nicvf_snd_task(void *arg, int pending)
1016 {
1017 	struct snd_queue *sq = (struct snd_queue *)arg;
1018 	struct nicvf *nic;
1019 	struct ifnet *ifp;
1020 	int err;
1021 
1022 	nic = sq->nic;
1023 	ifp = nic->ifp;
1024 
1025 	/*
1026 	 * Skip sending anything if the driver is not running,
1027 	 * SQ full or link is down.
1028 	 */
1029 	if (((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1030 	    IFF_DRV_RUNNING) || !nic->link_up)
1031 		return;
1032 
1033 	NICVF_TX_LOCK(sq);
1034 	err = nicvf_xmit_locked(sq);
1035 	NICVF_TX_UNLOCK(sq);
1036 	/* Try again */
1037 	if (err != 0)
1038 		taskqueue_enqueue(sq->snd_taskq, &sq->snd_task);
1039 }
1040 
1041 /* Initialize transmit queue */
1042 static int
1043 nicvf_init_snd_queue(struct nicvf *nic, struct snd_queue *sq, int q_len,
1044     int qidx)
1045 {
1046 	size_t i;
1047 	int err;
1048 
1049 	/* Initizalize TX lock for this queue */
1050 	snprintf(sq->mtx_name, sizeof(sq->mtx_name), "%s: SQ(%d) lock",
1051 	    device_get_nameunit(nic->dev), qidx);
1052 	mtx_init(&sq->mtx, sq->mtx_name, NULL, MTX_DEF);
1053 
1054 	NICVF_TX_LOCK(sq);
1055 	/* Allocate buffer ring */
1056 	sq->br = buf_ring_alloc(q_len / MIN_SQ_DESC_PER_PKT_XMIT, M_DEVBUF,
1057 	    M_NOWAIT, &sq->mtx);
1058 	if (sq->br == NULL) {
1059 		device_printf(nic->dev,
1060 		    "ERROR: Could not set up buf ring for SQ(%d)\n", qidx);
1061 		err = ENOMEM;
1062 		goto error;
1063 	}
1064 
1065 	/* Allocate DMA memory for Tx descriptors */
1066 	err = nicvf_alloc_q_desc_mem(nic, &sq->dmem, q_len, SND_QUEUE_DESC_SIZE,
1067 				     NICVF_SQ_BASE_ALIGN_BYTES);
1068 	if (err != 0) {
1069 		device_printf(nic->dev,
1070 		    "Could not allocate DMA memory for SQ\n");
1071 		goto error;
1072 	}
1073 
1074 	sq->desc = sq->dmem.base;
1075 	sq->head = sq->tail = 0;
1076 	atomic_store_rel_int(&sq->free_cnt, q_len - 1);
1077 	sq->thresh = SND_QUEUE_THRESH;
1078 	sq->idx = qidx;
1079 	sq->nic = nic;
1080 
1081 	/*
1082 	 * Allocate DMA maps for Tx buffers
1083 	 */
1084 
1085 	/* Create DMA tag first */
1086 	err = bus_dma_tag_create(
1087 	    bus_get_dma_tag(nic->dev),		/* parent tag */
1088 	    1,					/* alignment */
1089 	    0,					/* boundary */
1090 	    BUS_SPACE_MAXADDR,			/* lowaddr */
1091 	    BUS_SPACE_MAXADDR,			/* highaddr */
1092 	    NULL, NULL,				/* filtfunc, filtfuncarg */
1093 	    NICVF_TSO_MAXSIZE,			/* maxsize */
1094 	    NICVF_TSO_NSEGS,			/* nsegments */
1095 	    MCLBYTES,				/* maxsegsize */
1096 	    0,					/* flags */
1097 	    NULL, NULL,				/* lockfunc, lockfuncarg */
1098 	    &sq->snd_buff_dmat);		/* dmat */
1099 
1100 	if (err != 0) {
1101 		device_printf(nic->dev,
1102 		    "Failed to create busdma tag for Tx buffers\n");
1103 		goto error;
1104 	}
1105 
1106 	/* Allocate send buffers array */
1107 	sq->snd_buff = malloc(sizeof(*sq->snd_buff) * q_len, M_NICVF,
1108 	    (M_NOWAIT | M_ZERO));
1109 	if (sq->snd_buff == NULL) {
1110 		device_printf(nic->dev,
1111 		    "Could not allocate memory for Tx buffers array\n");
1112 		err = ENOMEM;
1113 		goto error;
1114 	}
1115 
1116 	/* Now populate maps */
1117 	for (i = 0; i < q_len; i++) {
1118 		err = bus_dmamap_create(sq->snd_buff_dmat, 0,
1119 		    &sq->snd_buff[i].dmap);
1120 		if (err != 0) {
1121 			device_printf(nic->dev,
1122 			    "Failed to create DMA maps for Tx buffers\n");
1123 			goto error;
1124 		}
1125 	}
1126 	NICVF_TX_UNLOCK(sq);
1127 
1128 	/* Allocate taskqueue */
1129 	TASK_INIT(&sq->snd_task, 0, nicvf_snd_task, sq);
1130 	sq->snd_taskq = taskqueue_create_fast("nicvf_snd_taskq", M_WAITOK,
1131 	    taskqueue_thread_enqueue, &sq->snd_taskq);
1132 	taskqueue_start_threads(&sq->snd_taskq, 1, PI_NET, "%s: snd_taskq(%d)",
1133 	    device_get_nameunit(nic->dev), qidx);
1134 
1135 	return (0);
1136 error:
1137 	NICVF_TX_UNLOCK(sq);
1138 	return (err);
1139 }
1140 
1141 static void
1142 nicvf_free_snd_queue(struct nicvf *nic, struct snd_queue *sq)
1143 {
1144 	struct queue_set *qs = nic->qs;
1145 	size_t i;
1146 	int err;
1147 
1148 	if (sq == NULL)
1149 		return;
1150 
1151 	if (sq->snd_taskq != NULL) {
1152 		/* Remove task */
1153 		while (taskqueue_cancel(sq->snd_taskq, &sq->snd_task, NULL) != 0)
1154 			taskqueue_drain(sq->snd_taskq, &sq->snd_task);
1155 
1156 		taskqueue_free(sq->snd_taskq);
1157 		sq->snd_taskq = NULL;
1158 	}
1159 
1160 	NICVF_TX_LOCK(sq);
1161 	if (sq->snd_buff_dmat != NULL) {
1162 		if (sq->snd_buff != NULL) {
1163 			for (i = 0; i < qs->sq_len; i++) {
1164 				m_freem(sq->snd_buff[i].mbuf);
1165 				sq->snd_buff[i].mbuf = NULL;
1166 
1167 				bus_dmamap_unload(sq->snd_buff_dmat,
1168 				    sq->snd_buff[i].dmap);
1169 				err = bus_dmamap_destroy(sq->snd_buff_dmat,
1170 				    sq->snd_buff[i].dmap);
1171 				/*
1172 				 * If bus_dmamap_destroy fails it can cause
1173 				 * random panic later if the tag is also
1174 				 * destroyed in the process.
1175 				 */
1176 				KASSERT(err == 0,
1177 				    ("%s: Could not destroy DMA map for SQ",
1178 				    __func__));
1179 			}
1180 		}
1181 
1182 		free(sq->snd_buff, M_NICVF);
1183 
1184 		err = bus_dma_tag_destroy(sq->snd_buff_dmat);
1185 		KASSERT(err == 0,
1186 		    ("%s: Trying to destroy BUSY DMA tag", __func__));
1187 	}
1188 
1189 	/* Free private driver ring for this send queue */
1190 	if (sq->br != NULL)
1191 		drbr_free(sq->br, M_DEVBUF);
1192 
1193 	if (sq->dmem.base != NULL)
1194 		nicvf_free_q_desc_mem(nic, &sq->dmem);
1195 
1196 	NICVF_TX_UNLOCK(sq);
1197 	/* Destroy Tx lock */
1198 	mtx_destroy(&sq->mtx);
1199 	memset(sq->mtx_name, 0, sizeof(sq->mtx_name));
1200 }
1201 
1202 static void
1203 nicvf_reclaim_snd_queue(struct nicvf *nic, struct queue_set *qs, int qidx)
1204 {
1205 
1206 	/* Disable send queue */
1207 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, 0);
1208 	/* Check if SQ is stopped */
1209 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_SQ_0_7_STATUS, 21, 1, 0x01))
1210 		return;
1211 	/* Reset send queue */
1212 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, NICVF_SQ_RESET);
1213 }
1214 
1215 static void
1216 nicvf_reclaim_rcv_queue(struct nicvf *nic, struct queue_set *qs, int qidx)
1217 {
1218 	union nic_mbx mbx = {};
1219 
1220 	/* Make sure all packets in the pipeline are written back into mem */
1221 	mbx.msg.msg = NIC_MBOX_MSG_RQ_SW_SYNC;
1222 	nicvf_send_msg_to_pf(nic, &mbx);
1223 }
1224 
1225 static void
1226 nicvf_reclaim_cmp_queue(struct nicvf *nic, struct queue_set *qs, int qidx)
1227 {
1228 
1229 	/* Disable timer threshold (doesn't get reset upon CQ reset */
1230 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG2, qidx, 0);
1231 	/* Disable completion queue */
1232 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, 0);
1233 	/* Reset completion queue */
1234 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, NICVF_CQ_RESET);
1235 }
1236 
1237 static void
1238 nicvf_reclaim_rbdr(struct nicvf *nic, struct rbdr *rbdr, int qidx)
1239 {
1240 	uint64_t tmp, fifo_state;
1241 	int timeout = 10;
1242 
1243 	/* Save head and tail pointers for feeing up buffers */
1244 	rbdr->head =
1245 	    nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_HEAD, qidx) >> 3;
1246 	rbdr->tail =
1247 	    nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_TAIL, qidx) >> 3;
1248 
1249 	/*
1250 	 * If RBDR FIFO is in 'FAIL' state then do a reset first
1251 	 * before relaiming.
1252 	 */
1253 	fifo_state = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_STATUS0, qidx);
1254 	if (((fifo_state >> 62) & 0x03) == 0x3) {
1255 		nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG,
1256 		    qidx, NICVF_RBDR_RESET);
1257 	}
1258 
1259 	/* Disable RBDR */
1260 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx, 0);
1261 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x00))
1262 		return;
1263 	while (1) {
1264 		tmp = nicvf_queue_reg_read(nic,
1265 		    NIC_QSET_RBDR_0_1_PREFETCH_STATUS, qidx);
1266 		if ((tmp & 0xFFFFFFFF) == ((tmp >> 32) & 0xFFFFFFFF))
1267 			break;
1268 
1269 		DELAY(1000);
1270 		timeout--;
1271 		if (!timeout) {
1272 			device_printf(nic->dev,
1273 			    "Failed polling on prefetch status\n");
1274 			return;
1275 		}
1276 	}
1277 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx,
1278 	    NICVF_RBDR_RESET);
1279 
1280 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x02))
1281 		return;
1282 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx, 0x00);
1283 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x00))
1284 		return;
1285 }
1286 
1287 /* Configures receive queue */
1288 static void
1289 nicvf_rcv_queue_config(struct nicvf *nic, struct queue_set *qs,
1290     int qidx, bool enable)
1291 {
1292 	union nic_mbx mbx = {};
1293 	struct rcv_queue *rq;
1294 	struct rq_cfg rq_cfg;
1295 	struct ifnet *ifp;
1296 	struct lro_ctrl	*lro;
1297 
1298 	ifp = nic->ifp;
1299 
1300 	rq = &qs->rq[qidx];
1301 	rq->enable = enable;
1302 
1303 	lro = &rq->lro;
1304 
1305 	/* Disable receive queue */
1306 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_0_7_CFG, qidx, 0);
1307 
1308 	if (!rq->enable) {
1309 		nicvf_reclaim_rcv_queue(nic, qs, qidx);
1310 		/* Free LRO memory */
1311 		tcp_lro_free(lro);
1312 		rq->lro_enabled = FALSE;
1313 		return;
1314 	}
1315 
1316 	/* Configure LRO if enabled */
1317 	rq->lro_enabled = FALSE;
1318 	if ((if_getcapenable(ifp) & IFCAP_LRO) != 0) {
1319 		if (tcp_lro_init(lro) != 0) {
1320 			device_printf(nic->dev,
1321 			    "Failed to initialize LRO for RXQ%d\n", qidx);
1322 		} else {
1323 			rq->lro_enabled = TRUE;
1324 			lro->ifp = nic->ifp;
1325 		}
1326 	}
1327 
1328 	rq->cq_qs = qs->vnic_id;
1329 	rq->cq_idx = qidx;
1330 	rq->start_rbdr_qs = qs->vnic_id;
1331 	rq->start_qs_rbdr_idx = qs->rbdr_cnt - 1;
1332 	rq->cont_rbdr_qs = qs->vnic_id;
1333 	rq->cont_qs_rbdr_idx = qs->rbdr_cnt - 1;
1334 	/* all writes of RBDR data to be loaded into L2 Cache as well*/
1335 	rq->caching = 1;
1336 
1337 	/* Send a mailbox msg to PF to config RQ */
1338 	mbx.rq.msg = NIC_MBOX_MSG_RQ_CFG;
1339 	mbx.rq.qs_num = qs->vnic_id;
1340 	mbx.rq.rq_num = qidx;
1341 	mbx.rq.cfg = (rq->caching << 26) | (rq->cq_qs << 19) |
1342 	    (rq->cq_idx << 16) | (rq->cont_rbdr_qs << 9) |
1343 	    (rq->cont_qs_rbdr_idx << 8) | (rq->start_rbdr_qs << 1) |
1344 	    (rq->start_qs_rbdr_idx);
1345 	nicvf_send_msg_to_pf(nic, &mbx);
1346 
1347 	mbx.rq.msg = NIC_MBOX_MSG_RQ_BP_CFG;
1348 	mbx.rq.cfg = (1UL << 63) | (1UL << 62) | (qs->vnic_id << 0);
1349 	nicvf_send_msg_to_pf(nic, &mbx);
1350 
1351 	/*
1352 	 * RQ drop config
1353 	 * Enable CQ drop to reserve sufficient CQEs for all tx packets
1354 	 */
1355 	mbx.rq.msg = NIC_MBOX_MSG_RQ_DROP_CFG;
1356 	mbx.rq.cfg = (1UL << 62) | (RQ_CQ_DROP << 8);
1357 	nicvf_send_msg_to_pf(nic, &mbx);
1358 
1359 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_GEN_CFG, 0, 0x00);
1360 
1361 	/* Enable Receive queue */
1362 	rq_cfg.ena = 1;
1363 	rq_cfg.tcp_ena = 0;
1364 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_0_7_CFG, qidx,
1365 	    *(uint64_t *)&rq_cfg);
1366 }
1367 
1368 /* Configures completion queue */
1369 static void
1370 nicvf_cmp_queue_config(struct nicvf *nic, struct queue_set *qs,
1371     int qidx, boolean_t enable)
1372 {
1373 	struct cmp_queue *cq;
1374 	struct cq_cfg cq_cfg;
1375 
1376 	cq = &qs->cq[qidx];
1377 	cq->enable = enable;
1378 
1379 	if (!cq->enable) {
1380 		nicvf_reclaim_cmp_queue(nic, qs, qidx);
1381 		return;
1382 	}
1383 
1384 	/* Reset completion queue */
1385 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, NICVF_CQ_RESET);
1386 
1387 	/* Set completion queue base address */
1388 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_BASE, qidx,
1389 	    (uint64_t)(cq->dmem.phys_base));
1390 
1391 	/* Enable Completion queue */
1392 	cq_cfg.ena = 1;
1393 	cq_cfg.reset = 0;
1394 	cq_cfg.caching = 0;
1395 	cq_cfg.qsize = CMP_QSIZE;
1396 	cq_cfg.avg_con = 0;
1397 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, *(uint64_t *)&cq_cfg);
1398 
1399 	/* Set threshold value for interrupt generation */
1400 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_THRESH, qidx, cq->thresh);
1401 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG2, qidx,
1402 	    nic->cq_coalesce_usecs);
1403 }
1404 
1405 /* Configures transmit queue */
1406 static void
1407 nicvf_snd_queue_config(struct nicvf *nic, struct queue_set *qs, int qidx,
1408     boolean_t enable)
1409 {
1410 	union nic_mbx mbx = {};
1411 	struct snd_queue *sq;
1412 	struct sq_cfg sq_cfg;
1413 
1414 	sq = &qs->sq[qidx];
1415 	sq->enable = enable;
1416 
1417 	if (!sq->enable) {
1418 		nicvf_reclaim_snd_queue(nic, qs, qidx);
1419 		return;
1420 	}
1421 
1422 	/* Reset send queue */
1423 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, NICVF_SQ_RESET);
1424 
1425 	sq->cq_qs = qs->vnic_id;
1426 	sq->cq_idx = qidx;
1427 
1428 	/* Send a mailbox msg to PF to config SQ */
1429 	mbx.sq.msg = NIC_MBOX_MSG_SQ_CFG;
1430 	mbx.sq.qs_num = qs->vnic_id;
1431 	mbx.sq.sq_num = qidx;
1432 	mbx.sq.sqs_mode = nic->sqs_mode;
1433 	mbx.sq.cfg = (sq->cq_qs << 3) | sq->cq_idx;
1434 	nicvf_send_msg_to_pf(nic, &mbx);
1435 
1436 	/* Set queue base address */
1437 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_BASE, qidx,
1438 	    (uint64_t)(sq->dmem.phys_base));
1439 
1440 	/* Enable send queue  & set queue size */
1441 	sq_cfg.ena = 1;
1442 	sq_cfg.reset = 0;
1443 	sq_cfg.ldwb = 0;
1444 	sq_cfg.qsize = SND_QSIZE;
1445 	sq_cfg.tstmp_bgx_intf = 0;
1446 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, *(uint64_t *)&sq_cfg);
1447 
1448 	/* Set threshold value for interrupt generation */
1449 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_THRESH, qidx, sq->thresh);
1450 }
1451 
1452 /* Configures receive buffer descriptor ring */
1453 static void
1454 nicvf_rbdr_config(struct nicvf *nic, struct queue_set *qs, int qidx,
1455     boolean_t enable)
1456 {
1457 	struct rbdr *rbdr;
1458 	struct rbdr_cfg rbdr_cfg;
1459 
1460 	rbdr = &qs->rbdr[qidx];
1461 	nicvf_reclaim_rbdr(nic, rbdr, qidx);
1462 	if (!enable)
1463 		return;
1464 
1465 	/* Set descriptor base address */
1466 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_BASE, qidx,
1467 	    (uint64_t)(rbdr->dmem.phys_base));
1468 
1469 	/* Enable RBDR  & set queue size */
1470 	/* Buffer size should be in multiples of 128 bytes */
1471 	rbdr_cfg.ena = 1;
1472 	rbdr_cfg.reset = 0;
1473 	rbdr_cfg.ldwb = 0;
1474 	rbdr_cfg.qsize = RBDR_SIZE;
1475 	rbdr_cfg.avg_con = 0;
1476 	rbdr_cfg.lines = rbdr->dma_size / 128;
1477 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx,
1478 	    *(uint64_t *)&rbdr_cfg);
1479 
1480 	/* Notify HW */
1481 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_DOOR, qidx,
1482 	    qs->rbdr_len - 1);
1483 
1484 	/* Set threshold value for interrupt generation */
1485 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_THRESH, qidx,
1486 	    rbdr->thresh - 1);
1487 }
1488 
1489 /* Requests PF to assign and enable Qset */
1490 void
1491 nicvf_qset_config(struct nicvf *nic, boolean_t enable)
1492 {
1493 	union nic_mbx mbx = {};
1494 	struct queue_set *qs;
1495 	struct qs_cfg *qs_cfg;
1496 
1497 	qs = nic->qs;
1498 	if (qs == NULL) {
1499 		device_printf(nic->dev,
1500 		    "Qset is still not allocated, don't init queues\n");
1501 		return;
1502 	}
1503 
1504 	qs->enable = enable;
1505 	qs->vnic_id = nic->vf_id;
1506 
1507 	/* Send a mailbox msg to PF to config Qset */
1508 	mbx.qs.msg = NIC_MBOX_MSG_QS_CFG;
1509 	mbx.qs.num = qs->vnic_id;
1510 
1511 	mbx.qs.cfg = 0;
1512 	qs_cfg = (struct qs_cfg *)&mbx.qs.cfg;
1513 	if (qs->enable) {
1514 		qs_cfg->ena = 1;
1515 		qs_cfg->vnic = qs->vnic_id;
1516 	}
1517 	nicvf_send_msg_to_pf(nic, &mbx);
1518 }
1519 
1520 static void
1521 nicvf_free_resources(struct nicvf *nic)
1522 {
1523 	int qidx;
1524 	struct queue_set *qs;
1525 
1526 	qs = nic->qs;
1527 	/*
1528 	 * Remove QS error task first since it has to be dead
1529 	 * to safely free completion queue tasks.
1530 	 */
1531 	if (qs->qs_err_taskq != NULL) {
1532 		/* Shut down QS error tasks */
1533 		while (taskqueue_cancel(qs->qs_err_taskq,
1534 		    &qs->qs_err_task,  NULL) != 0) {
1535 			taskqueue_drain(qs->qs_err_taskq, &qs->qs_err_task);
1536 
1537 		}
1538 		taskqueue_free(qs->qs_err_taskq);
1539 		qs->qs_err_taskq = NULL;
1540 	}
1541 	/* Free receive buffer descriptor ring */
1542 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1543 		nicvf_free_rbdr(nic, &qs->rbdr[qidx]);
1544 
1545 	/* Free completion queue */
1546 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1547 		nicvf_free_cmp_queue(nic, &qs->cq[qidx]);
1548 
1549 	/* Free send queue */
1550 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1551 		nicvf_free_snd_queue(nic, &qs->sq[qidx]);
1552 }
1553 
1554 static int
1555 nicvf_alloc_resources(struct nicvf *nic)
1556 {
1557 	struct queue_set *qs = nic->qs;
1558 	int qidx;
1559 
1560 	/* Alloc receive buffer descriptor ring */
1561 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
1562 		if (nicvf_init_rbdr(nic, &qs->rbdr[qidx], qs->rbdr_len,
1563 				    DMA_BUFFER_LEN, qidx))
1564 			goto alloc_fail;
1565 	}
1566 
1567 	/* Alloc send queue */
1568 	for (qidx = 0; qidx < qs->sq_cnt; qidx++) {
1569 		if (nicvf_init_snd_queue(nic, &qs->sq[qidx], qs->sq_len, qidx))
1570 			goto alloc_fail;
1571 	}
1572 
1573 	/* Alloc completion queue */
1574 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
1575 		if (nicvf_init_cmp_queue(nic, &qs->cq[qidx], qs->cq_len, qidx))
1576 			goto alloc_fail;
1577 	}
1578 
1579 	/* Allocate QS error taskqueue */
1580 	NET_TASK_INIT(&qs->qs_err_task, 0, nicvf_qs_err_task, nic);
1581 	qs->qs_err_taskq = taskqueue_create_fast("nicvf_qs_err_taskq", M_WAITOK,
1582 	    taskqueue_thread_enqueue, &qs->qs_err_taskq);
1583 	taskqueue_start_threads(&qs->qs_err_taskq, 1, PI_NET, "%s: qs_taskq",
1584 	    device_get_nameunit(nic->dev));
1585 
1586 	return (0);
1587 alloc_fail:
1588 	nicvf_free_resources(nic);
1589 	return (ENOMEM);
1590 }
1591 
1592 int
1593 nicvf_set_qset_resources(struct nicvf *nic)
1594 {
1595 	struct queue_set *qs;
1596 
1597 	qs = malloc(sizeof(*qs), M_NICVF, (M_ZERO | M_WAITOK));
1598 	nic->qs = qs;
1599 
1600 	/* Set count of each queue */
1601 	qs->rbdr_cnt = RBDR_CNT;
1602 	qs->rq_cnt = RCV_QUEUE_CNT;
1603 
1604 	qs->sq_cnt = SND_QUEUE_CNT;
1605 	qs->cq_cnt = CMP_QUEUE_CNT;
1606 
1607 	/* Set queue lengths */
1608 	qs->rbdr_len = RCV_BUF_COUNT;
1609 	qs->sq_len = SND_QUEUE_LEN;
1610 	qs->cq_len = CMP_QUEUE_LEN;
1611 
1612 	nic->rx_queues = qs->rq_cnt;
1613 	nic->tx_queues = qs->sq_cnt;
1614 
1615 	return (0);
1616 }
1617 
1618 int
1619 nicvf_config_data_transfer(struct nicvf *nic, boolean_t enable)
1620 {
1621 	boolean_t disable = FALSE;
1622 	struct queue_set *qs;
1623 	int qidx;
1624 
1625 	qs = nic->qs;
1626 	if (qs == NULL)
1627 		return (0);
1628 
1629 	if (enable) {
1630 		if (nicvf_alloc_resources(nic) != 0)
1631 			return (ENOMEM);
1632 
1633 		for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1634 			nicvf_snd_queue_config(nic, qs, qidx, enable);
1635 		for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1636 			nicvf_cmp_queue_config(nic, qs, qidx, enable);
1637 		for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1638 			nicvf_rbdr_config(nic, qs, qidx, enable);
1639 		for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1640 			nicvf_rcv_queue_config(nic, qs, qidx, enable);
1641 	} else {
1642 		for (qidx = 0; qidx < qs->rq_cnt; qidx++)
1643 			nicvf_rcv_queue_config(nic, qs, qidx, disable);
1644 		for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
1645 			nicvf_rbdr_config(nic, qs, qidx, disable);
1646 		for (qidx = 0; qidx < qs->sq_cnt; qidx++)
1647 			nicvf_snd_queue_config(nic, qs, qidx, disable);
1648 		for (qidx = 0; qidx < qs->cq_cnt; qidx++)
1649 			nicvf_cmp_queue_config(nic, qs, qidx, disable);
1650 
1651 		nicvf_free_resources(nic);
1652 	}
1653 
1654 	return (0);
1655 }
1656 
1657 /*
1658  * Get a free desc from SQ
1659  * returns descriptor ponter & descriptor number
1660  */
1661 static __inline int
1662 nicvf_get_sq_desc(struct snd_queue *sq, int desc_cnt)
1663 {
1664 	int qentry;
1665 
1666 	qentry = sq->tail;
1667 	atomic_subtract_int(&sq->free_cnt, desc_cnt);
1668 	sq->tail += desc_cnt;
1669 	sq->tail &= (sq->dmem.q_len - 1);
1670 
1671 	return (qentry);
1672 }
1673 
1674 /* Free descriptor back to SQ for future use */
1675 static void
1676 nicvf_put_sq_desc(struct snd_queue *sq, int desc_cnt)
1677 {
1678 
1679 	atomic_add_int(&sq->free_cnt, desc_cnt);
1680 	sq->head += desc_cnt;
1681 	sq->head &= (sq->dmem.q_len - 1);
1682 }
1683 
1684 static __inline int
1685 nicvf_get_nxt_sqentry(struct snd_queue *sq, int qentry)
1686 {
1687 	qentry++;
1688 	qentry &= (sq->dmem.q_len - 1);
1689 	return (qentry);
1690 }
1691 
1692 static void
1693 nicvf_sq_enable(struct nicvf *nic, struct snd_queue *sq, int qidx)
1694 {
1695 	uint64_t sq_cfg;
1696 
1697 	sq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_CFG, qidx);
1698 	sq_cfg |= NICVF_SQ_EN;
1699 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, sq_cfg);
1700 	/* Ring doorbell so that H/W restarts processing SQEs */
1701 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_DOOR, qidx, 0);
1702 }
1703 
1704 static void
1705 nicvf_sq_disable(struct nicvf *nic, int qidx)
1706 {
1707 	uint64_t sq_cfg;
1708 
1709 	sq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_CFG, qidx);
1710 	sq_cfg &= ~NICVF_SQ_EN;
1711 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, sq_cfg);
1712 }
1713 
1714 static void
1715 nicvf_sq_free_used_descs(struct nicvf *nic, struct snd_queue *sq, int qidx)
1716 {
1717 	uint64_t head;
1718 	struct snd_buff *snd_buff;
1719 	struct sq_hdr_subdesc *hdr;
1720 
1721 	NICVF_TX_LOCK(sq);
1722 	head = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_HEAD, qidx) >> 4;
1723 	while (sq->head != head) {
1724 		hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, sq->head);
1725 		if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER) {
1726 			nicvf_put_sq_desc(sq, 1);
1727 			continue;
1728 		}
1729 		snd_buff = &sq->snd_buff[sq->head];
1730 		if (snd_buff->mbuf != NULL) {
1731 			bus_dmamap_unload(sq->snd_buff_dmat, snd_buff->dmap);
1732 			m_freem(snd_buff->mbuf);
1733 			sq->snd_buff[sq->head].mbuf = NULL;
1734 		}
1735 		nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1);
1736 	}
1737 	NICVF_TX_UNLOCK(sq);
1738 }
1739 
1740 /*
1741  * Add SQ HEADER subdescriptor.
1742  * First subdescriptor for every send descriptor.
1743  */
1744 static __inline int
1745 nicvf_sq_add_hdr_subdesc(struct snd_queue *sq, int qentry,
1746 			 int subdesc_cnt, struct mbuf *mbuf, int len)
1747 {
1748 	struct nicvf *nic;
1749 	struct sq_hdr_subdesc *hdr;
1750 	struct ether_vlan_header *eh;
1751 #ifdef INET
1752 	struct ip *ip;
1753 	struct tcphdr *th;
1754 #endif
1755 	uint16_t etype;
1756 	int ehdrlen, iphlen, poff, proto;
1757 
1758 	nic = sq->nic;
1759 
1760 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, qentry);
1761 	sq->snd_buff[qentry].mbuf = mbuf;
1762 
1763 	memset(hdr, 0, SND_QUEUE_DESC_SIZE);
1764 	hdr->subdesc_type = SQ_DESC_TYPE_HEADER;
1765 	/* Enable notification via CQE after processing SQE */
1766 	hdr->post_cqe = 1;
1767 	/* No of subdescriptors following this */
1768 	hdr->subdesc_cnt = subdesc_cnt;
1769 	hdr->tot_len = len;
1770 
1771 	eh = mtod(mbuf, struct ether_vlan_header *);
1772 	if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1773 		ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1774 		etype = ntohs(eh->evl_proto);
1775 	} else {
1776 		ehdrlen = ETHER_HDR_LEN;
1777 		etype = ntohs(eh->evl_encap_proto);
1778 	}
1779 
1780 	poff = proto = -1;
1781 	switch (etype) {
1782 #ifdef INET6
1783 	case ETHERTYPE_IPV6:
1784 		if (mbuf->m_len < ehdrlen + sizeof(struct ip6_hdr)) {
1785 			mbuf = m_pullup(mbuf, ehdrlen +sizeof(struct ip6_hdr));
1786 			sq->snd_buff[qentry].mbuf = NULL;
1787 			if (mbuf == NULL)
1788 				return (ENOBUFS);
1789 		}
1790 		poff = ip6_lasthdr(mbuf, ehdrlen, IPPROTO_IPV6, &proto);
1791 		if (poff < 0)
1792 			return (ENOBUFS);
1793 		poff += ehdrlen;
1794 		break;
1795 #endif
1796 #ifdef INET
1797 	case ETHERTYPE_IP:
1798 		if (mbuf->m_len < ehdrlen + sizeof(struct ip)) {
1799 			mbuf = m_pullup(mbuf, ehdrlen + sizeof(struct ip));
1800 			sq->snd_buff[qentry].mbuf = mbuf;
1801 			if (mbuf == NULL)
1802 				return (ENOBUFS);
1803 		}
1804 		if (mbuf->m_pkthdr.csum_flags & CSUM_IP)
1805 			hdr->csum_l3 = 1; /* Enable IP csum calculation */
1806 
1807 		ip = (struct ip *)(mbuf->m_data + ehdrlen);
1808 		iphlen = ip->ip_hl << 2;
1809 		poff = ehdrlen + iphlen;
1810 		proto = ip->ip_p;
1811 		break;
1812 #endif
1813 	}
1814 
1815 #if defined(INET6) || defined(INET)
1816 	if (poff > 0 && mbuf->m_pkthdr.csum_flags != 0) {
1817 		switch (proto) {
1818 		case IPPROTO_TCP:
1819 			if ((mbuf->m_pkthdr.csum_flags & CSUM_TCP) == 0)
1820 				break;
1821 
1822 			if (mbuf->m_len < (poff + sizeof(struct tcphdr))) {
1823 				mbuf = m_pullup(mbuf, poff + sizeof(struct tcphdr));
1824 				sq->snd_buff[qentry].mbuf = mbuf;
1825 				if (mbuf == NULL)
1826 					return (ENOBUFS);
1827 			}
1828 			hdr->csum_l4 = SEND_L4_CSUM_TCP;
1829 			break;
1830 		case IPPROTO_UDP:
1831 			if ((mbuf->m_pkthdr.csum_flags & CSUM_UDP) == 0)
1832 				break;
1833 
1834 			if (mbuf->m_len < (poff + sizeof(struct udphdr))) {
1835 				mbuf = m_pullup(mbuf, poff + sizeof(struct udphdr));
1836 				sq->snd_buff[qentry].mbuf = mbuf;
1837 				if (mbuf == NULL)
1838 					return (ENOBUFS);
1839 			}
1840 			hdr->csum_l4 = SEND_L4_CSUM_UDP;
1841 			break;
1842 		case IPPROTO_SCTP:
1843 			if ((mbuf->m_pkthdr.csum_flags & CSUM_SCTP) == 0)
1844 				break;
1845 
1846 			if (mbuf->m_len < (poff + sizeof(struct sctphdr))) {
1847 				mbuf = m_pullup(mbuf, poff + sizeof(struct sctphdr));
1848 				sq->snd_buff[qentry].mbuf = mbuf;
1849 				if (mbuf == NULL)
1850 					return (ENOBUFS);
1851 			}
1852 			hdr->csum_l4 = SEND_L4_CSUM_SCTP;
1853 			break;
1854 		default:
1855 			break;
1856 		}
1857 		hdr->l3_offset = ehdrlen;
1858 		hdr->l4_offset = poff;
1859 	}
1860 
1861 	if ((mbuf->m_pkthdr.tso_segsz != 0) && nic->hw_tso) {
1862 		th = (struct tcphdr *)((caddr_t)(mbuf->m_data + poff));
1863 
1864 		hdr->tso = 1;
1865 		hdr->tso_start = poff + (th->th_off * 4);
1866 		hdr->tso_max_paysize = mbuf->m_pkthdr.tso_segsz;
1867 		hdr->inner_l3_offset = ehdrlen - 2;
1868 		nic->drv_stats.tx_tso++;
1869 	}
1870 #endif
1871 
1872 	return (0);
1873 }
1874 
1875 /*
1876  * SQ GATHER subdescriptor
1877  * Must follow HDR descriptor
1878  */
1879 static inline void nicvf_sq_add_gather_subdesc(struct snd_queue *sq, int qentry,
1880 					       int size, uint64_t data)
1881 {
1882 	struct sq_gather_subdesc *gather;
1883 
1884 	qentry &= (sq->dmem.q_len - 1);
1885 	gather = (struct sq_gather_subdesc *)GET_SQ_DESC(sq, qentry);
1886 
1887 	memset(gather, 0, SND_QUEUE_DESC_SIZE);
1888 	gather->subdesc_type = SQ_DESC_TYPE_GATHER;
1889 	gather->ld_type = NIC_SEND_LD_TYPE_E_LDD;
1890 	gather->size = size;
1891 	gather->addr = data;
1892 }
1893 
1894 /* Put an mbuf to a SQ for packet transfer. */
1895 static int
1896 nicvf_tx_mbuf_locked(struct snd_queue *sq, struct mbuf **mbufp)
1897 {
1898 	bus_dma_segment_t segs[256];
1899 	struct snd_buff *snd_buff;
1900 	size_t seg;
1901 	int nsegs, qentry;
1902 	int subdesc_cnt;
1903 	int err;
1904 
1905 	NICVF_TX_LOCK_ASSERT(sq);
1906 
1907 	if (sq->free_cnt == 0)
1908 		return (ENOBUFS);
1909 
1910 	snd_buff = &sq->snd_buff[sq->tail];
1911 
1912 	err = bus_dmamap_load_mbuf_sg(sq->snd_buff_dmat, snd_buff->dmap,
1913 	    *mbufp, segs, &nsegs, BUS_DMA_NOWAIT);
1914 	if (__predict_false(err != 0)) {
1915 		/* ARM64TODO: Add mbuf defragmenting if we lack maps */
1916 		m_freem(*mbufp);
1917 		*mbufp = NULL;
1918 		return (err);
1919 	}
1920 
1921 	/* Set how many subdescriptors is required */
1922 	subdesc_cnt = MIN_SQ_DESC_PER_PKT_XMIT + nsegs - 1;
1923 	if (subdesc_cnt > sq->free_cnt) {
1924 		/* ARM64TODO: Add mbuf defragmentation if we lack descriptors */
1925 		bus_dmamap_unload(sq->snd_buff_dmat, snd_buff->dmap);
1926 		return (ENOBUFS);
1927 	}
1928 
1929 	qentry = nicvf_get_sq_desc(sq, subdesc_cnt);
1930 
1931 	/* Add SQ header subdesc */
1932 	err = nicvf_sq_add_hdr_subdesc(sq, qentry, subdesc_cnt - 1, *mbufp,
1933 	    (*mbufp)->m_pkthdr.len);
1934 	if (err != 0) {
1935 		nicvf_put_sq_desc(sq, subdesc_cnt);
1936 		bus_dmamap_unload(sq->snd_buff_dmat, snd_buff->dmap);
1937 		if (err == ENOBUFS) {
1938 			m_freem(*mbufp);
1939 			*mbufp = NULL;
1940 		}
1941 		return (err);
1942 	}
1943 
1944 	/* Add SQ gather subdescs */
1945 	for (seg = 0; seg < nsegs; seg++) {
1946 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1947 		nicvf_sq_add_gather_subdesc(sq, qentry, segs[seg].ds_len,
1948 		    segs[seg].ds_addr);
1949 	}
1950 
1951 	/* make sure all memory stores are done before ringing doorbell */
1952 	bus_dmamap_sync(sq->dmem.dmat, sq->dmem.dmap, BUS_DMASYNC_PREWRITE);
1953 
1954 	dprintf(sq->nic->dev, "%s: sq->idx: %d, subdesc_cnt: %d\n",
1955 	    __func__, sq->idx, subdesc_cnt);
1956 	/* Inform HW to xmit new packet */
1957 	nicvf_queue_reg_write(sq->nic, NIC_QSET_SQ_0_7_DOOR,
1958 	    sq->idx, subdesc_cnt);
1959 	return (0);
1960 }
1961 
1962 static __inline u_int
1963 frag_num(u_int i)
1964 {
1965 #if BYTE_ORDER == BIG_ENDIAN
1966 	return ((i & ~3) + 3 - (i & 3));
1967 #else
1968 	return (i);
1969 #endif
1970 }
1971 
1972 /* Returns MBUF for a received packet */
1973 struct mbuf *
1974 nicvf_get_rcv_mbuf(struct nicvf *nic, struct cqe_rx_t *cqe_rx)
1975 {
1976 	int frag;
1977 	int payload_len = 0;
1978 	struct mbuf *mbuf;
1979 	struct mbuf *mbuf_frag;
1980 	uint16_t *rb_lens = NULL;
1981 	uint64_t *rb_ptrs = NULL;
1982 
1983 	mbuf = NULL;
1984 	rb_lens = (uint16_t *)((uint8_t *)cqe_rx + (3 * sizeof(uint64_t)));
1985 	rb_ptrs = (uint64_t *)((uint8_t *)cqe_rx + (6 * sizeof(uint64_t)));
1986 
1987 	dprintf(nic->dev, "%s rb_cnt %d rb0_ptr %lx rb0_sz %d\n",
1988 	    __func__, cqe_rx->rb_cnt, cqe_rx->rb0_ptr, cqe_rx->rb0_sz);
1989 
1990 	for (frag = 0; frag < cqe_rx->rb_cnt; frag++) {
1991 		payload_len = rb_lens[frag_num(frag)];
1992 		if (frag == 0) {
1993 			/* First fragment */
1994 			mbuf = nicvf_rb_ptr_to_mbuf(nic,
1995 			    (*rb_ptrs - cqe_rx->align_pad));
1996 			mbuf->m_len = payload_len;
1997 			mbuf->m_data += cqe_rx->align_pad;
1998 			if_setrcvif(mbuf, nic->ifp);
1999 		} else {
2000 			/* Add fragments */
2001 			mbuf_frag = nicvf_rb_ptr_to_mbuf(nic, *rb_ptrs);
2002 			m_append(mbuf, payload_len, mbuf_frag->m_data);
2003 			m_freem(mbuf_frag);
2004 		}
2005 		/* Next buffer pointer */
2006 		rb_ptrs++;
2007 	}
2008 
2009 	if (__predict_true(mbuf != NULL)) {
2010 		m_fixhdr(mbuf);
2011 		mbuf->m_pkthdr.flowid = cqe_rx->rq_idx;
2012 		M_HASHTYPE_SET(mbuf, M_HASHTYPE_OPAQUE);
2013 		if (__predict_true((if_getcapenable(nic->ifp) & IFCAP_RXCSUM) != 0)) {
2014 			/*
2015 			 * HW by default verifies IP & TCP/UDP/SCTP checksums
2016 			 */
2017 			if (__predict_true(cqe_rx->l3_type == L3TYPE_IPV4)) {
2018 				mbuf->m_pkthdr.csum_flags =
2019 				    (CSUM_IP_CHECKED | CSUM_IP_VALID);
2020 			}
2021 
2022 			switch (cqe_rx->l4_type) {
2023 			case L4TYPE_UDP:
2024 			case L4TYPE_TCP: /* fall through */
2025 				mbuf->m_pkthdr.csum_flags |=
2026 				    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
2027 				mbuf->m_pkthdr.csum_data = 0xffff;
2028 				break;
2029 			case L4TYPE_SCTP:
2030 				mbuf->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
2031 				break;
2032 			default:
2033 				break;
2034 			}
2035 		}
2036 	}
2037 
2038 	return (mbuf);
2039 }
2040 
2041 /* Enable interrupt */
2042 void
2043 nicvf_enable_intr(struct nicvf *nic, int int_type, int q_idx)
2044 {
2045 	uint64_t reg_val;
2046 
2047 	reg_val = nicvf_reg_read(nic, NIC_VF_ENA_W1S);
2048 
2049 	switch (int_type) {
2050 	case NICVF_INTR_CQ:
2051 		reg_val |= ((1UL << q_idx) << NICVF_INTR_CQ_SHIFT);
2052 		break;
2053 	case NICVF_INTR_SQ:
2054 		reg_val |= ((1UL << q_idx) << NICVF_INTR_SQ_SHIFT);
2055 		break;
2056 	case NICVF_INTR_RBDR:
2057 		reg_val |= ((1UL << q_idx) << NICVF_INTR_RBDR_SHIFT);
2058 		break;
2059 	case NICVF_INTR_PKT_DROP:
2060 		reg_val |= (1UL << NICVF_INTR_PKT_DROP_SHIFT);
2061 		break;
2062 	case NICVF_INTR_TCP_TIMER:
2063 		reg_val |= (1UL << NICVF_INTR_TCP_TIMER_SHIFT);
2064 		break;
2065 	case NICVF_INTR_MBOX:
2066 		reg_val |= (1UL << NICVF_INTR_MBOX_SHIFT);
2067 		break;
2068 	case NICVF_INTR_QS_ERR:
2069 		reg_val |= (1UL << NICVF_INTR_QS_ERR_SHIFT);
2070 		break;
2071 	default:
2072 		device_printf(nic->dev,
2073 			   "Failed to enable interrupt: unknown type\n");
2074 		break;
2075 	}
2076 
2077 	nicvf_reg_write(nic, NIC_VF_ENA_W1S, reg_val);
2078 }
2079 
2080 /* Disable interrupt */
2081 void
2082 nicvf_disable_intr(struct nicvf *nic, int int_type, int q_idx)
2083 {
2084 	uint64_t reg_val = 0;
2085 
2086 	switch (int_type) {
2087 	case NICVF_INTR_CQ:
2088 		reg_val |= ((1UL << q_idx) << NICVF_INTR_CQ_SHIFT);
2089 		break;
2090 	case NICVF_INTR_SQ:
2091 		reg_val |= ((1UL << q_idx) << NICVF_INTR_SQ_SHIFT);
2092 		break;
2093 	case NICVF_INTR_RBDR:
2094 		reg_val |= ((1UL << q_idx) << NICVF_INTR_RBDR_SHIFT);
2095 		break;
2096 	case NICVF_INTR_PKT_DROP:
2097 		reg_val |= (1UL << NICVF_INTR_PKT_DROP_SHIFT);
2098 		break;
2099 	case NICVF_INTR_TCP_TIMER:
2100 		reg_val |= (1UL << NICVF_INTR_TCP_TIMER_SHIFT);
2101 		break;
2102 	case NICVF_INTR_MBOX:
2103 		reg_val |= (1UL << NICVF_INTR_MBOX_SHIFT);
2104 		break;
2105 	case NICVF_INTR_QS_ERR:
2106 		reg_val |= (1UL << NICVF_INTR_QS_ERR_SHIFT);
2107 		break;
2108 	default:
2109 		device_printf(nic->dev,
2110 			   "Failed to disable interrupt: unknown type\n");
2111 		break;
2112 	}
2113 
2114 	nicvf_reg_write(nic, NIC_VF_ENA_W1C, reg_val);
2115 }
2116 
2117 /* Clear interrupt */
2118 void
2119 nicvf_clear_intr(struct nicvf *nic, int int_type, int q_idx)
2120 {
2121 	uint64_t reg_val = 0;
2122 
2123 	switch (int_type) {
2124 	case NICVF_INTR_CQ:
2125 		reg_val = ((1UL << q_idx) << NICVF_INTR_CQ_SHIFT);
2126 		break;
2127 	case NICVF_INTR_SQ:
2128 		reg_val = ((1UL << q_idx) << NICVF_INTR_SQ_SHIFT);
2129 		break;
2130 	case NICVF_INTR_RBDR:
2131 		reg_val = ((1UL << q_idx) << NICVF_INTR_RBDR_SHIFT);
2132 		break;
2133 	case NICVF_INTR_PKT_DROP:
2134 		reg_val = (1UL << NICVF_INTR_PKT_DROP_SHIFT);
2135 		break;
2136 	case NICVF_INTR_TCP_TIMER:
2137 		reg_val = (1UL << NICVF_INTR_TCP_TIMER_SHIFT);
2138 		break;
2139 	case NICVF_INTR_MBOX:
2140 		reg_val = (1UL << NICVF_INTR_MBOX_SHIFT);
2141 		break;
2142 	case NICVF_INTR_QS_ERR:
2143 		reg_val |= (1UL << NICVF_INTR_QS_ERR_SHIFT);
2144 		break;
2145 	default:
2146 		device_printf(nic->dev,
2147 			   "Failed to clear interrupt: unknown type\n");
2148 		break;
2149 	}
2150 
2151 	nicvf_reg_write(nic, NIC_VF_INT, reg_val);
2152 }
2153 
2154 /* Check if interrupt is enabled */
2155 int
2156 nicvf_is_intr_enabled(struct nicvf *nic, int int_type, int q_idx)
2157 {
2158 	uint64_t reg_val;
2159 	uint64_t mask = 0xff;
2160 
2161 	reg_val = nicvf_reg_read(nic, NIC_VF_ENA_W1S);
2162 
2163 	switch (int_type) {
2164 	case NICVF_INTR_CQ:
2165 		mask = ((1UL << q_idx) << NICVF_INTR_CQ_SHIFT);
2166 		break;
2167 	case NICVF_INTR_SQ:
2168 		mask = ((1UL << q_idx) << NICVF_INTR_SQ_SHIFT);
2169 		break;
2170 	case NICVF_INTR_RBDR:
2171 		mask = ((1UL << q_idx) << NICVF_INTR_RBDR_SHIFT);
2172 		break;
2173 	case NICVF_INTR_PKT_DROP:
2174 		mask = NICVF_INTR_PKT_DROP_MASK;
2175 		break;
2176 	case NICVF_INTR_TCP_TIMER:
2177 		mask = NICVF_INTR_TCP_TIMER_MASK;
2178 		break;
2179 	case NICVF_INTR_MBOX:
2180 		mask = NICVF_INTR_MBOX_MASK;
2181 		break;
2182 	case NICVF_INTR_QS_ERR:
2183 		mask = NICVF_INTR_QS_ERR_MASK;
2184 		break;
2185 	default:
2186 		device_printf(nic->dev,
2187 			   "Failed to check interrupt enable: unknown type\n");
2188 		break;
2189 	}
2190 
2191 	return (reg_val & mask);
2192 }
2193 
2194 void
2195 nicvf_update_rq_stats(struct nicvf *nic, int rq_idx)
2196 {
2197 	struct rcv_queue *rq;
2198 
2199 #define GET_RQ_STATS(reg) \
2200 	nicvf_reg_read(nic, NIC_QSET_RQ_0_7_STAT_0_1 |\
2201 			    (rq_idx << NIC_Q_NUM_SHIFT) | (reg << 3))
2202 
2203 	rq = &nic->qs->rq[rq_idx];
2204 	rq->stats.bytes = GET_RQ_STATS(RQ_SQ_STATS_OCTS);
2205 	rq->stats.pkts = GET_RQ_STATS(RQ_SQ_STATS_PKTS);
2206 }
2207 
2208 void
2209 nicvf_update_sq_stats(struct nicvf *nic, int sq_idx)
2210 {
2211 	struct snd_queue *sq;
2212 
2213 #define GET_SQ_STATS(reg) \
2214 	nicvf_reg_read(nic, NIC_QSET_SQ_0_7_STAT_0_1 |\
2215 			    (sq_idx << NIC_Q_NUM_SHIFT) | (reg << 3))
2216 
2217 	sq = &nic->qs->sq[sq_idx];
2218 	sq->stats.bytes = GET_SQ_STATS(RQ_SQ_STATS_OCTS);
2219 	sq->stats.pkts = GET_SQ_STATS(RQ_SQ_STATS_PKTS);
2220 }
2221 
2222 /* Check for errors in the receive cmp.queue entry */
2223 int
2224 nicvf_check_cqe_rx_errs(struct nicvf *nic, struct cmp_queue *cq,
2225     struct cqe_rx_t *cqe_rx)
2226 {
2227 	struct nicvf_hw_stats *stats = &nic->hw_stats;
2228 	struct nicvf_drv_stats *drv_stats = &nic->drv_stats;
2229 
2230 	if (!cqe_rx->err_level && !cqe_rx->err_opcode) {
2231 		drv_stats->rx_frames_ok++;
2232 		return (0);
2233 	}
2234 
2235 	switch (cqe_rx->err_opcode) {
2236 	case CQ_RX_ERROP_RE_PARTIAL:
2237 		stats->rx_bgx_truncated_pkts++;
2238 		break;
2239 	case CQ_RX_ERROP_RE_JABBER:
2240 		stats->rx_jabber_errs++;
2241 		break;
2242 	case CQ_RX_ERROP_RE_FCS:
2243 		stats->rx_fcs_errs++;
2244 		break;
2245 	case CQ_RX_ERROP_RE_RX_CTL:
2246 		stats->rx_bgx_errs++;
2247 		break;
2248 	case CQ_RX_ERROP_PREL2_ERR:
2249 		stats->rx_prel2_errs++;
2250 		break;
2251 	case CQ_RX_ERROP_L2_MAL:
2252 		stats->rx_l2_hdr_malformed++;
2253 		break;
2254 	case CQ_RX_ERROP_L2_OVERSIZE:
2255 		stats->rx_oversize++;
2256 		break;
2257 	case CQ_RX_ERROP_L2_UNDERSIZE:
2258 		stats->rx_undersize++;
2259 		break;
2260 	case CQ_RX_ERROP_L2_LENMISM:
2261 		stats->rx_l2_len_mismatch++;
2262 		break;
2263 	case CQ_RX_ERROP_L2_PCLP:
2264 		stats->rx_l2_pclp++;
2265 		break;
2266 	case CQ_RX_ERROP_IP_NOT:
2267 		stats->rx_ip_ver_errs++;
2268 		break;
2269 	case CQ_RX_ERROP_IP_CSUM_ERR:
2270 		stats->rx_ip_csum_errs++;
2271 		break;
2272 	case CQ_RX_ERROP_IP_MAL:
2273 		stats->rx_ip_hdr_malformed++;
2274 		break;
2275 	case CQ_RX_ERROP_IP_MALD:
2276 		stats->rx_ip_payload_malformed++;
2277 		break;
2278 	case CQ_RX_ERROP_IP_HOP:
2279 		stats->rx_ip_ttl_errs++;
2280 		break;
2281 	case CQ_RX_ERROP_L3_PCLP:
2282 		stats->rx_l3_pclp++;
2283 		break;
2284 	case CQ_RX_ERROP_L4_MAL:
2285 		stats->rx_l4_malformed++;
2286 		break;
2287 	case CQ_RX_ERROP_L4_CHK:
2288 		stats->rx_l4_csum_errs++;
2289 		break;
2290 	case CQ_RX_ERROP_UDP_LEN:
2291 		stats->rx_udp_len_errs++;
2292 		break;
2293 	case CQ_RX_ERROP_L4_PORT:
2294 		stats->rx_l4_port_errs++;
2295 		break;
2296 	case CQ_RX_ERROP_TCP_FLAG:
2297 		stats->rx_tcp_flag_errs++;
2298 		break;
2299 	case CQ_RX_ERROP_TCP_OFFSET:
2300 		stats->rx_tcp_offset_errs++;
2301 		break;
2302 	case CQ_RX_ERROP_L4_PCLP:
2303 		stats->rx_l4_pclp++;
2304 		break;
2305 	case CQ_RX_ERROP_RBDR_TRUNC:
2306 		stats->rx_truncated_pkts++;
2307 		break;
2308 	}
2309 
2310 	return (1);
2311 }
2312 
2313 /* Check for errors in the send cmp.queue entry */
2314 int
2315 nicvf_check_cqe_tx_errs(struct nicvf *nic, struct cmp_queue *cq,
2316     struct cqe_send_t *cqe_tx)
2317 {
2318 	struct cmp_queue_stats *stats = &cq->stats;
2319 
2320 	switch (cqe_tx->send_status) {
2321 	case CQ_TX_ERROP_GOOD:
2322 		stats->tx.good++;
2323 		return (0);
2324 	case CQ_TX_ERROP_DESC_FAULT:
2325 		stats->tx.desc_fault++;
2326 		break;
2327 	case CQ_TX_ERROP_HDR_CONS_ERR:
2328 		stats->tx.hdr_cons_err++;
2329 		break;
2330 	case CQ_TX_ERROP_SUBDC_ERR:
2331 		stats->tx.subdesc_err++;
2332 		break;
2333 	case CQ_TX_ERROP_IMM_SIZE_OFLOW:
2334 		stats->tx.imm_size_oflow++;
2335 		break;
2336 	case CQ_TX_ERROP_DATA_SEQUENCE_ERR:
2337 		stats->tx.data_seq_err++;
2338 		break;
2339 	case CQ_TX_ERROP_MEM_SEQUENCE_ERR:
2340 		stats->tx.mem_seq_err++;
2341 		break;
2342 	case CQ_TX_ERROP_LOCK_VIOL:
2343 		stats->tx.lock_viol++;
2344 		break;
2345 	case CQ_TX_ERROP_DATA_FAULT:
2346 		stats->tx.data_fault++;
2347 		break;
2348 	case CQ_TX_ERROP_TSTMP_CONFLICT:
2349 		stats->tx.tstmp_conflict++;
2350 		break;
2351 	case CQ_TX_ERROP_TSTMP_TIMEOUT:
2352 		stats->tx.tstmp_timeout++;
2353 		break;
2354 	case CQ_TX_ERROP_MEM_FAULT:
2355 		stats->tx.mem_fault++;
2356 		break;
2357 	case CQ_TX_ERROP_CK_OVERLAP:
2358 		stats->tx.csum_overlap++;
2359 		break;
2360 	case CQ_TX_ERROP_CK_OFLOW:
2361 		stats->tx.csum_overflow++;
2362 		break;
2363 	}
2364 
2365 	return (1);
2366 }
2367