1 /*
2  * Copyright (C) 2015 Cavium, Inc.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License
6  * as published by the Free Software Foundation.
7  */
8 
9 #include <linux/pci.h>
10 #include <linux/netdevice.h>
11 #include <linux/ip.h>
12 #include <linux/etherdevice.h>
13 #include <net/ip.h>
14 #include <net/tso.h>
15 
16 #include "nic_reg.h"
17 #include "nic.h"
18 #include "q_struct.h"
19 #include "nicvf_queues.h"
20 
21 static void nicvf_get_page(struct nicvf *nic)
22 {
23 	if (!nic->rb_pageref || !nic->rb_page)
24 		return;
25 
26 	page_ref_add(nic->rb_page, nic->rb_pageref);
27 	nic->rb_pageref = 0;
28 }
29 
30 /* Poll a register for a specific value */
31 static int nicvf_poll_reg(struct nicvf *nic, int qidx,
32 			  u64 reg, int bit_pos, int bits, int val)
33 {
34 	u64 bit_mask;
35 	u64 reg_val;
36 	int timeout = 10;
37 
38 	bit_mask = (1ULL << bits) - 1;
39 	bit_mask = (bit_mask << bit_pos);
40 
41 	while (timeout) {
42 		reg_val = nicvf_queue_reg_read(nic, reg, qidx);
43 		if (((reg_val & bit_mask) >> bit_pos) == val)
44 			return 0;
45 		usleep_range(1000, 2000);
46 		timeout--;
47 	}
48 	netdev_err(nic->netdev, "Poll on reg 0x%llx failed\n", reg);
49 	return 1;
50 }
51 
52 /* Allocate memory for a queue's descriptors */
53 static int nicvf_alloc_q_desc_mem(struct nicvf *nic, struct q_desc_mem *dmem,
54 				  int q_len, int desc_size, int align_bytes)
55 {
56 	dmem->q_len = q_len;
57 	dmem->size = (desc_size * q_len) + align_bytes;
58 	/* Save address, need it while freeing */
59 	dmem->unalign_base = dma_zalloc_coherent(&nic->pdev->dev, dmem->size,
60 						&dmem->dma, GFP_KERNEL);
61 	if (!dmem->unalign_base)
62 		return -ENOMEM;
63 
64 	/* Align memory address for 'align_bytes' */
65 	dmem->phys_base = NICVF_ALIGNED_ADDR((u64)dmem->dma, align_bytes);
66 	dmem->base = dmem->unalign_base + (dmem->phys_base - dmem->dma);
67 	return 0;
68 }
69 
70 /* Free queue's descriptor memory */
71 static void nicvf_free_q_desc_mem(struct nicvf *nic, struct q_desc_mem *dmem)
72 {
73 	if (!dmem)
74 		return;
75 
76 	dma_free_coherent(&nic->pdev->dev, dmem->size,
77 			  dmem->unalign_base, dmem->dma);
78 	dmem->unalign_base = NULL;
79 	dmem->base = NULL;
80 }
81 
82 /* Allocate buffer for packet reception
83  * HW returns memory address where packet is DMA'ed but not a pointer
84  * into RBDR ring, so save buffer address at the start of fragment and
85  * align the start address to a cache aligned address
86  */
87 static inline int nicvf_alloc_rcv_buffer(struct nicvf *nic, gfp_t gfp,
88 					 u32 buf_len, u64 **rbuf)
89 {
90 	int order = (PAGE_SIZE <= 4096) ?  PAGE_ALLOC_COSTLY_ORDER : 0;
91 
92 	/* Check if request can be accomodated in previous allocated page */
93 	if (nic->rb_page &&
94 	    ((nic->rb_page_offset + buf_len) < (PAGE_SIZE << order))) {
95 		nic->rb_pageref++;
96 		goto ret;
97 	}
98 
99 	nicvf_get_page(nic);
100 	nic->rb_page = NULL;
101 
102 	/* Allocate a new page */
103 	if (!nic->rb_page) {
104 		nic->rb_page = alloc_pages(gfp | __GFP_COMP | __GFP_NOWARN,
105 					   order);
106 		if (!nic->rb_page) {
107 			nic->drv_stats.rcv_buffer_alloc_failures++;
108 			return -ENOMEM;
109 		}
110 		nic->rb_page_offset = 0;
111 	}
112 
113 ret:
114 	*rbuf = (u64 *)((u64)page_address(nic->rb_page) + nic->rb_page_offset);
115 	nic->rb_page_offset += buf_len;
116 
117 	return 0;
118 }
119 
120 /* Build skb around receive buffer */
121 static struct sk_buff *nicvf_rb_ptr_to_skb(struct nicvf *nic,
122 					   u64 rb_ptr, int len)
123 {
124 	void *data;
125 	struct sk_buff *skb;
126 
127 	data = phys_to_virt(rb_ptr);
128 
129 	/* Now build an skb to give to stack */
130 	skb = build_skb(data, RCV_FRAG_LEN);
131 	if (!skb) {
132 		put_page(virt_to_page(data));
133 		return NULL;
134 	}
135 
136 	prefetch(skb->data);
137 	return skb;
138 }
139 
140 /* Allocate RBDR ring and populate receive buffers */
141 static int  nicvf_init_rbdr(struct nicvf *nic, struct rbdr *rbdr,
142 			    int ring_len, int buf_size)
143 {
144 	int idx;
145 	u64 *rbuf;
146 	struct rbdr_entry_t *desc;
147 	int err;
148 
149 	err = nicvf_alloc_q_desc_mem(nic, &rbdr->dmem, ring_len,
150 				     sizeof(struct rbdr_entry_t),
151 				     NICVF_RCV_BUF_ALIGN_BYTES);
152 	if (err)
153 		return err;
154 
155 	rbdr->desc = rbdr->dmem.base;
156 	/* Buffer size has to be in multiples of 128 bytes */
157 	rbdr->dma_size = buf_size;
158 	rbdr->enable = true;
159 	rbdr->thresh = RBDR_THRESH;
160 
161 	nic->rb_page = NULL;
162 	for (idx = 0; idx < ring_len; idx++) {
163 		err = nicvf_alloc_rcv_buffer(nic, GFP_KERNEL, RCV_FRAG_LEN,
164 					     &rbuf);
165 		if (err)
166 			return err;
167 
168 		desc = GET_RBDR_DESC(rbdr, idx);
169 		desc->buf_addr = virt_to_phys(rbuf) >> NICVF_RCV_BUF_ALIGN;
170 	}
171 
172 	nicvf_get_page(nic);
173 
174 	return 0;
175 }
176 
177 /* Free RBDR ring and its receive buffers */
178 static void nicvf_free_rbdr(struct nicvf *nic, struct rbdr *rbdr)
179 {
180 	int head, tail;
181 	u64 buf_addr;
182 	struct rbdr_entry_t *desc;
183 
184 	if (!rbdr)
185 		return;
186 
187 	rbdr->enable = false;
188 	if (!rbdr->dmem.base)
189 		return;
190 
191 	head = rbdr->head;
192 	tail = rbdr->tail;
193 
194 	/* Free SKBs */
195 	while (head != tail) {
196 		desc = GET_RBDR_DESC(rbdr, head);
197 		buf_addr = desc->buf_addr << NICVF_RCV_BUF_ALIGN;
198 		put_page(virt_to_page(phys_to_virt(buf_addr)));
199 		head++;
200 		head &= (rbdr->dmem.q_len - 1);
201 	}
202 	/* Free SKB of tail desc */
203 	desc = GET_RBDR_DESC(rbdr, tail);
204 	buf_addr = desc->buf_addr << NICVF_RCV_BUF_ALIGN;
205 	put_page(virt_to_page(phys_to_virt(buf_addr)));
206 
207 	/* Free RBDR ring */
208 	nicvf_free_q_desc_mem(nic, &rbdr->dmem);
209 }
210 
211 /* Refill receive buffer descriptors with new buffers.
212  */
213 static void nicvf_refill_rbdr(struct nicvf *nic, gfp_t gfp)
214 {
215 	struct queue_set *qs = nic->qs;
216 	int rbdr_idx = qs->rbdr_cnt;
217 	int tail, qcount;
218 	int refill_rb_cnt;
219 	struct rbdr *rbdr;
220 	struct rbdr_entry_t *desc;
221 	u64 *rbuf;
222 	int new_rb = 0;
223 
224 refill:
225 	if (!rbdr_idx)
226 		return;
227 	rbdr_idx--;
228 	rbdr = &qs->rbdr[rbdr_idx];
229 	/* Check if it's enabled */
230 	if (!rbdr->enable)
231 		goto next_rbdr;
232 
233 	/* Get no of desc's to be refilled */
234 	qcount = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_STATUS0, rbdr_idx);
235 	qcount &= 0x7FFFF;
236 	/* Doorbell can be ringed with a max of ring size minus 1 */
237 	if (qcount >= (qs->rbdr_len - 1))
238 		goto next_rbdr;
239 	else
240 		refill_rb_cnt = qs->rbdr_len - qcount - 1;
241 
242 	/* Start filling descs from tail */
243 	tail = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_TAIL, rbdr_idx) >> 3;
244 	while (refill_rb_cnt) {
245 		tail++;
246 		tail &= (rbdr->dmem.q_len - 1);
247 
248 		if (nicvf_alloc_rcv_buffer(nic, gfp, RCV_FRAG_LEN, &rbuf))
249 			break;
250 
251 		desc = GET_RBDR_DESC(rbdr, tail);
252 		desc->buf_addr = virt_to_phys(rbuf) >> NICVF_RCV_BUF_ALIGN;
253 		refill_rb_cnt--;
254 		new_rb++;
255 	}
256 
257 	nicvf_get_page(nic);
258 
259 	/* make sure all memory stores are done before ringing doorbell */
260 	smp_wmb();
261 
262 	/* Check if buffer allocation failed */
263 	if (refill_rb_cnt)
264 		nic->rb_alloc_fail = true;
265 	else
266 		nic->rb_alloc_fail = false;
267 
268 	/* Notify HW */
269 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_DOOR,
270 			      rbdr_idx, new_rb);
271 next_rbdr:
272 	/* Re-enable RBDR interrupts only if buffer allocation is success */
273 	if (!nic->rb_alloc_fail && rbdr->enable)
274 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, rbdr_idx);
275 
276 	if (rbdr_idx)
277 		goto refill;
278 }
279 
280 /* Alloc rcv buffers in non-atomic mode for better success */
281 void nicvf_rbdr_work(struct work_struct *work)
282 {
283 	struct nicvf *nic = container_of(work, struct nicvf, rbdr_work.work);
284 
285 	nicvf_refill_rbdr(nic, GFP_KERNEL);
286 	if (nic->rb_alloc_fail)
287 		schedule_delayed_work(&nic->rbdr_work, msecs_to_jiffies(10));
288 	else
289 		nic->rb_work_scheduled = false;
290 }
291 
292 /* In Softirq context, alloc rcv buffers in atomic mode */
293 void nicvf_rbdr_task(unsigned long data)
294 {
295 	struct nicvf *nic = (struct nicvf *)data;
296 
297 	nicvf_refill_rbdr(nic, GFP_ATOMIC);
298 	if (nic->rb_alloc_fail) {
299 		nic->rb_work_scheduled = true;
300 		schedule_delayed_work(&nic->rbdr_work, msecs_to_jiffies(10));
301 	}
302 }
303 
304 /* Initialize completion queue */
305 static int nicvf_init_cmp_queue(struct nicvf *nic,
306 				struct cmp_queue *cq, int q_len)
307 {
308 	int err;
309 
310 	err = nicvf_alloc_q_desc_mem(nic, &cq->dmem, q_len, CMP_QUEUE_DESC_SIZE,
311 				     NICVF_CQ_BASE_ALIGN_BYTES);
312 	if (err)
313 		return err;
314 
315 	cq->desc = cq->dmem.base;
316 	cq->thresh = pass1_silicon(nic->pdev) ? 0 : CMP_QUEUE_CQE_THRESH;
317 	nic->cq_coalesce_usecs = (CMP_QUEUE_TIMER_THRESH * 0.05) - 1;
318 
319 	return 0;
320 }
321 
322 static void nicvf_free_cmp_queue(struct nicvf *nic, struct cmp_queue *cq)
323 {
324 	if (!cq)
325 		return;
326 	if (!cq->dmem.base)
327 		return;
328 
329 	nicvf_free_q_desc_mem(nic, &cq->dmem);
330 }
331 
332 /* Initialize transmit queue */
333 static int nicvf_init_snd_queue(struct nicvf *nic,
334 				struct snd_queue *sq, int q_len)
335 {
336 	int err;
337 
338 	err = nicvf_alloc_q_desc_mem(nic, &sq->dmem, q_len, SND_QUEUE_DESC_SIZE,
339 				     NICVF_SQ_BASE_ALIGN_BYTES);
340 	if (err)
341 		return err;
342 
343 	sq->desc = sq->dmem.base;
344 	sq->skbuff = kcalloc(q_len, sizeof(u64), GFP_KERNEL);
345 	if (!sq->skbuff)
346 		return -ENOMEM;
347 	sq->head = 0;
348 	sq->tail = 0;
349 	atomic_set(&sq->free_cnt, q_len - 1);
350 	sq->thresh = SND_QUEUE_THRESH;
351 
352 	/* Preallocate memory for TSO segment's header */
353 	sq->tso_hdrs = dma_alloc_coherent(&nic->pdev->dev,
354 					  q_len * TSO_HEADER_SIZE,
355 					  &sq->tso_hdrs_phys, GFP_KERNEL);
356 	if (!sq->tso_hdrs)
357 		return -ENOMEM;
358 
359 	return 0;
360 }
361 
362 static void nicvf_free_snd_queue(struct nicvf *nic, struct snd_queue *sq)
363 {
364 	if (!sq)
365 		return;
366 	if (!sq->dmem.base)
367 		return;
368 
369 	if (sq->tso_hdrs)
370 		dma_free_coherent(&nic->pdev->dev,
371 				  sq->dmem.q_len * TSO_HEADER_SIZE,
372 				  sq->tso_hdrs, sq->tso_hdrs_phys);
373 
374 	kfree(sq->skbuff);
375 	nicvf_free_q_desc_mem(nic, &sq->dmem);
376 }
377 
378 static void nicvf_reclaim_snd_queue(struct nicvf *nic,
379 				    struct queue_set *qs, int qidx)
380 {
381 	/* Disable send queue */
382 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, 0);
383 	/* Check if SQ is stopped */
384 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_SQ_0_7_STATUS, 21, 1, 0x01))
385 		return;
386 	/* Reset send queue */
387 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, NICVF_SQ_RESET);
388 }
389 
390 static void nicvf_reclaim_rcv_queue(struct nicvf *nic,
391 				    struct queue_set *qs, int qidx)
392 {
393 	union nic_mbx mbx = {};
394 
395 	/* Make sure all packets in the pipeline are written back into mem */
396 	mbx.msg.msg = NIC_MBOX_MSG_RQ_SW_SYNC;
397 	nicvf_send_msg_to_pf(nic, &mbx);
398 }
399 
400 static void nicvf_reclaim_cmp_queue(struct nicvf *nic,
401 				    struct queue_set *qs, int qidx)
402 {
403 	/* Disable timer threshold (doesn't get reset upon CQ reset */
404 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG2, qidx, 0);
405 	/* Disable completion queue */
406 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, 0);
407 	/* Reset completion queue */
408 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, NICVF_CQ_RESET);
409 }
410 
411 static void nicvf_reclaim_rbdr(struct nicvf *nic,
412 			       struct rbdr *rbdr, int qidx)
413 {
414 	u64 tmp, fifo_state;
415 	int timeout = 10;
416 
417 	/* Save head and tail pointers for feeing up buffers */
418 	rbdr->head = nicvf_queue_reg_read(nic,
419 					  NIC_QSET_RBDR_0_1_HEAD,
420 					  qidx) >> 3;
421 	rbdr->tail = nicvf_queue_reg_read(nic,
422 					  NIC_QSET_RBDR_0_1_TAIL,
423 					  qidx) >> 3;
424 
425 	/* If RBDR FIFO is in 'FAIL' state then do a reset first
426 	 * before relaiming.
427 	 */
428 	fifo_state = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_STATUS0, qidx);
429 	if (((fifo_state >> 62) & 0x03) == 0x3)
430 		nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG,
431 				      qidx, NICVF_RBDR_RESET);
432 
433 	/* Disable RBDR */
434 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx, 0);
435 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x00))
436 		return;
437 	while (1) {
438 		tmp = nicvf_queue_reg_read(nic,
439 					   NIC_QSET_RBDR_0_1_PREFETCH_STATUS,
440 					   qidx);
441 		if ((tmp & 0xFFFFFFFF) == ((tmp >> 32) & 0xFFFFFFFF))
442 			break;
443 		usleep_range(1000, 2000);
444 		timeout--;
445 		if (!timeout) {
446 			netdev_err(nic->netdev,
447 				   "Failed polling on prefetch status\n");
448 			return;
449 		}
450 	}
451 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG,
452 			      qidx, NICVF_RBDR_RESET);
453 
454 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x02))
455 		return;
456 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx, 0x00);
457 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x00))
458 		return;
459 }
460 
461 void nicvf_config_vlan_stripping(struct nicvf *nic, netdev_features_t features)
462 {
463 	u64 rq_cfg;
464 	int sqs;
465 
466 	rq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_RQ_GEN_CFG, 0);
467 
468 	/* Enable first VLAN stripping */
469 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
470 		rq_cfg |= (1ULL << 25);
471 	else
472 		rq_cfg &= ~(1ULL << 25);
473 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_GEN_CFG, 0, rq_cfg);
474 
475 	/* Configure Secondary Qsets, if any */
476 	for (sqs = 0; sqs < nic->sqs_count; sqs++)
477 		if (nic->snicvf[sqs])
478 			nicvf_queue_reg_write(nic->snicvf[sqs],
479 					      NIC_QSET_RQ_GEN_CFG, 0, rq_cfg);
480 }
481 
482 static void nicvf_reset_rcv_queue_stats(struct nicvf *nic)
483 {
484 	union nic_mbx mbx = {};
485 
486 	/* Reset all RXQ's stats */
487 	mbx.reset_stat.msg = NIC_MBOX_MSG_RESET_STAT_COUNTER;
488 	mbx.reset_stat.rq_stat_mask = 0xFFFF;
489 	nicvf_send_msg_to_pf(nic, &mbx);
490 }
491 
492 /* Configures receive queue */
493 static void nicvf_rcv_queue_config(struct nicvf *nic, struct queue_set *qs,
494 				   int qidx, bool enable)
495 {
496 	union nic_mbx mbx = {};
497 	struct rcv_queue *rq;
498 	struct rq_cfg rq_cfg;
499 
500 	rq = &qs->rq[qidx];
501 	rq->enable = enable;
502 
503 	/* Disable receive queue */
504 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_0_7_CFG, qidx, 0);
505 
506 	if (!rq->enable) {
507 		nicvf_reclaim_rcv_queue(nic, qs, qidx);
508 		return;
509 	}
510 
511 	rq->cq_qs = qs->vnic_id;
512 	rq->cq_idx = qidx;
513 	rq->start_rbdr_qs = qs->vnic_id;
514 	rq->start_qs_rbdr_idx = qs->rbdr_cnt - 1;
515 	rq->cont_rbdr_qs = qs->vnic_id;
516 	rq->cont_qs_rbdr_idx = qs->rbdr_cnt - 1;
517 	/* all writes of RBDR data to be loaded into L2 Cache as well*/
518 	rq->caching = 1;
519 
520 	/* Send a mailbox msg to PF to config RQ */
521 	mbx.rq.msg = NIC_MBOX_MSG_RQ_CFG;
522 	mbx.rq.qs_num = qs->vnic_id;
523 	mbx.rq.rq_num = qidx;
524 	mbx.rq.cfg = (rq->caching << 26) | (rq->cq_qs << 19) |
525 			  (rq->cq_idx << 16) | (rq->cont_rbdr_qs << 9) |
526 			  (rq->cont_qs_rbdr_idx << 8) |
527 			  (rq->start_rbdr_qs << 1) | (rq->start_qs_rbdr_idx);
528 	nicvf_send_msg_to_pf(nic, &mbx);
529 
530 	mbx.rq.msg = NIC_MBOX_MSG_RQ_BP_CFG;
531 	mbx.rq.cfg = (1ULL << 63) | (1ULL << 62) | (qs->vnic_id << 0);
532 	nicvf_send_msg_to_pf(nic, &mbx);
533 
534 	/* RQ drop config
535 	 * Enable CQ drop to reserve sufficient CQEs for all tx packets
536 	 */
537 	mbx.rq.msg = NIC_MBOX_MSG_RQ_DROP_CFG;
538 	mbx.rq.cfg = (1ULL << 62) | (RQ_CQ_DROP << 8);
539 	nicvf_send_msg_to_pf(nic, &mbx);
540 
541 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_GEN_CFG, 0, 0x00);
542 	if (!nic->sqs_mode)
543 		nicvf_config_vlan_stripping(nic, nic->netdev->features);
544 
545 	/* Enable Receive queue */
546 	memset(&rq_cfg, 0, sizeof(struct rq_cfg));
547 	rq_cfg.ena = 1;
548 	rq_cfg.tcp_ena = 0;
549 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_0_7_CFG, qidx, *(u64 *)&rq_cfg);
550 }
551 
552 /* Configures completion queue */
553 void nicvf_cmp_queue_config(struct nicvf *nic, struct queue_set *qs,
554 			    int qidx, bool enable)
555 {
556 	struct cmp_queue *cq;
557 	struct cq_cfg cq_cfg;
558 
559 	cq = &qs->cq[qidx];
560 	cq->enable = enable;
561 
562 	if (!cq->enable) {
563 		nicvf_reclaim_cmp_queue(nic, qs, qidx);
564 		return;
565 	}
566 
567 	/* Reset completion queue */
568 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, NICVF_CQ_RESET);
569 
570 	if (!cq->enable)
571 		return;
572 
573 	spin_lock_init(&cq->lock);
574 	/* Set completion queue base address */
575 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_BASE,
576 			      qidx, (u64)(cq->dmem.phys_base));
577 
578 	/* Enable Completion queue */
579 	memset(&cq_cfg, 0, sizeof(struct cq_cfg));
580 	cq_cfg.ena = 1;
581 	cq_cfg.reset = 0;
582 	cq_cfg.caching = 0;
583 	cq_cfg.qsize = CMP_QSIZE;
584 	cq_cfg.avg_con = 0;
585 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, *(u64 *)&cq_cfg);
586 
587 	/* Set threshold value for interrupt generation */
588 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_THRESH, qidx, cq->thresh);
589 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG2,
590 			      qidx, CMP_QUEUE_TIMER_THRESH);
591 }
592 
593 /* Configures transmit queue */
594 static void nicvf_snd_queue_config(struct nicvf *nic, struct queue_set *qs,
595 				   int qidx, bool enable)
596 {
597 	union nic_mbx mbx = {};
598 	struct snd_queue *sq;
599 	struct sq_cfg sq_cfg;
600 
601 	sq = &qs->sq[qidx];
602 	sq->enable = enable;
603 
604 	if (!sq->enable) {
605 		nicvf_reclaim_snd_queue(nic, qs, qidx);
606 		return;
607 	}
608 
609 	/* Reset send queue */
610 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, NICVF_SQ_RESET);
611 
612 	sq->cq_qs = qs->vnic_id;
613 	sq->cq_idx = qidx;
614 
615 	/* Send a mailbox msg to PF to config SQ */
616 	mbx.sq.msg = NIC_MBOX_MSG_SQ_CFG;
617 	mbx.sq.qs_num = qs->vnic_id;
618 	mbx.sq.sq_num = qidx;
619 	mbx.sq.sqs_mode = nic->sqs_mode;
620 	mbx.sq.cfg = (sq->cq_qs << 3) | sq->cq_idx;
621 	nicvf_send_msg_to_pf(nic, &mbx);
622 
623 	/* Set queue base address */
624 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_BASE,
625 			      qidx, (u64)(sq->dmem.phys_base));
626 
627 	/* Enable send queue  & set queue size */
628 	memset(&sq_cfg, 0, sizeof(struct sq_cfg));
629 	sq_cfg.ena = 1;
630 	sq_cfg.reset = 0;
631 	sq_cfg.ldwb = 0;
632 	sq_cfg.qsize = SND_QSIZE;
633 	sq_cfg.tstmp_bgx_intf = 0;
634 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, *(u64 *)&sq_cfg);
635 
636 	/* Set threshold value for interrupt generation */
637 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_THRESH, qidx, sq->thresh);
638 
639 	/* Set queue:cpu affinity for better load distribution */
640 	if (cpu_online(qidx)) {
641 		cpumask_set_cpu(qidx, &sq->affinity_mask);
642 		netif_set_xps_queue(nic->netdev,
643 				    &sq->affinity_mask, qidx);
644 	}
645 }
646 
647 /* Configures receive buffer descriptor ring */
648 static void nicvf_rbdr_config(struct nicvf *nic, struct queue_set *qs,
649 			      int qidx, bool enable)
650 {
651 	struct rbdr *rbdr;
652 	struct rbdr_cfg rbdr_cfg;
653 
654 	rbdr = &qs->rbdr[qidx];
655 	nicvf_reclaim_rbdr(nic, rbdr, qidx);
656 	if (!enable)
657 		return;
658 
659 	/* Set descriptor base address */
660 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_BASE,
661 			      qidx, (u64)(rbdr->dmem.phys_base));
662 
663 	/* Enable RBDR  & set queue size */
664 	/* Buffer size should be in multiples of 128 bytes */
665 	memset(&rbdr_cfg, 0, sizeof(struct rbdr_cfg));
666 	rbdr_cfg.ena = 1;
667 	rbdr_cfg.reset = 0;
668 	rbdr_cfg.ldwb = 0;
669 	rbdr_cfg.qsize = RBDR_SIZE;
670 	rbdr_cfg.avg_con = 0;
671 	rbdr_cfg.lines = rbdr->dma_size / 128;
672 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG,
673 			      qidx, *(u64 *)&rbdr_cfg);
674 
675 	/* Notify HW */
676 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_DOOR,
677 			      qidx, qs->rbdr_len - 1);
678 
679 	/* Set threshold value for interrupt generation */
680 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_THRESH,
681 			      qidx, rbdr->thresh - 1);
682 }
683 
684 /* Requests PF to assign and enable Qset */
685 void nicvf_qset_config(struct nicvf *nic, bool enable)
686 {
687 	union nic_mbx mbx = {};
688 	struct queue_set *qs = nic->qs;
689 	struct qs_cfg *qs_cfg;
690 
691 	if (!qs) {
692 		netdev_warn(nic->netdev,
693 			    "Qset is still not allocated, don't init queues\n");
694 		return;
695 	}
696 
697 	qs->enable = enable;
698 	qs->vnic_id = nic->vf_id;
699 
700 	/* Send a mailbox msg to PF to config Qset */
701 	mbx.qs.msg = NIC_MBOX_MSG_QS_CFG;
702 	mbx.qs.num = qs->vnic_id;
703 	mbx.qs.sqs_count = nic->sqs_count;
704 
705 	mbx.qs.cfg = 0;
706 	qs_cfg = (struct qs_cfg *)&mbx.qs.cfg;
707 	if (qs->enable) {
708 		qs_cfg->ena = 1;
709 #ifdef __BIG_ENDIAN
710 		qs_cfg->be = 1;
711 #endif
712 		qs_cfg->vnic = qs->vnic_id;
713 	}
714 	nicvf_send_msg_to_pf(nic, &mbx);
715 }
716 
717 static void nicvf_free_resources(struct nicvf *nic)
718 {
719 	int qidx;
720 	struct queue_set *qs = nic->qs;
721 
722 	/* Free receive buffer descriptor ring */
723 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
724 		nicvf_free_rbdr(nic, &qs->rbdr[qidx]);
725 
726 	/* Free completion queue */
727 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
728 		nicvf_free_cmp_queue(nic, &qs->cq[qidx]);
729 
730 	/* Free send queue */
731 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
732 		nicvf_free_snd_queue(nic, &qs->sq[qidx]);
733 }
734 
735 static int nicvf_alloc_resources(struct nicvf *nic)
736 {
737 	int qidx;
738 	struct queue_set *qs = nic->qs;
739 
740 	/* Alloc receive buffer descriptor ring */
741 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
742 		if (nicvf_init_rbdr(nic, &qs->rbdr[qidx], qs->rbdr_len,
743 				    DMA_BUFFER_LEN))
744 			goto alloc_fail;
745 	}
746 
747 	/* Alloc send queue */
748 	for (qidx = 0; qidx < qs->sq_cnt; qidx++) {
749 		if (nicvf_init_snd_queue(nic, &qs->sq[qidx], qs->sq_len))
750 			goto alloc_fail;
751 	}
752 
753 	/* Alloc completion queue */
754 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
755 		if (nicvf_init_cmp_queue(nic, &qs->cq[qidx], qs->cq_len))
756 			goto alloc_fail;
757 	}
758 
759 	return 0;
760 alloc_fail:
761 	nicvf_free_resources(nic);
762 	return -ENOMEM;
763 }
764 
765 int nicvf_set_qset_resources(struct nicvf *nic)
766 {
767 	struct queue_set *qs;
768 
769 	qs = devm_kzalloc(&nic->pdev->dev, sizeof(*qs), GFP_KERNEL);
770 	if (!qs)
771 		return -ENOMEM;
772 	nic->qs = qs;
773 
774 	/* Set count of each queue */
775 	qs->rbdr_cnt = DEFAULT_RBDR_CNT;
776 	qs->rq_cnt = min_t(u8, MAX_RCV_QUEUES_PER_QS, num_online_cpus());
777 	qs->sq_cnt = min_t(u8, MAX_SND_QUEUES_PER_QS, num_online_cpus());
778 	qs->cq_cnt = max_t(u8, qs->rq_cnt, qs->sq_cnt);
779 
780 	/* Set queue lengths */
781 	qs->rbdr_len = RCV_BUF_COUNT;
782 	qs->sq_len = SND_QUEUE_LEN;
783 	qs->cq_len = CMP_QUEUE_LEN;
784 
785 	nic->rx_queues = qs->rq_cnt;
786 	nic->tx_queues = qs->sq_cnt;
787 
788 	return 0;
789 }
790 
791 int nicvf_config_data_transfer(struct nicvf *nic, bool enable)
792 {
793 	bool disable = false;
794 	struct queue_set *qs = nic->qs;
795 	int qidx;
796 
797 	if (!qs)
798 		return 0;
799 
800 	if (enable) {
801 		if (nicvf_alloc_resources(nic))
802 			return -ENOMEM;
803 
804 		for (qidx = 0; qidx < qs->sq_cnt; qidx++)
805 			nicvf_snd_queue_config(nic, qs, qidx, enable);
806 		for (qidx = 0; qidx < qs->cq_cnt; qidx++)
807 			nicvf_cmp_queue_config(nic, qs, qidx, enable);
808 		for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
809 			nicvf_rbdr_config(nic, qs, qidx, enable);
810 		for (qidx = 0; qidx < qs->rq_cnt; qidx++)
811 			nicvf_rcv_queue_config(nic, qs, qidx, enable);
812 	} else {
813 		for (qidx = 0; qidx < qs->rq_cnt; qidx++)
814 			nicvf_rcv_queue_config(nic, qs, qidx, disable);
815 		for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
816 			nicvf_rbdr_config(nic, qs, qidx, disable);
817 		for (qidx = 0; qidx < qs->sq_cnt; qidx++)
818 			nicvf_snd_queue_config(nic, qs, qidx, disable);
819 		for (qidx = 0; qidx < qs->cq_cnt; qidx++)
820 			nicvf_cmp_queue_config(nic, qs, qidx, disable);
821 
822 		nicvf_free_resources(nic);
823 	}
824 
825 	/* Reset RXQ's stats.
826 	 * SQ's stats will get reset automatically once SQ is reset.
827 	 */
828 	nicvf_reset_rcv_queue_stats(nic);
829 
830 	return 0;
831 }
832 
833 /* Get a free desc from SQ
834  * returns descriptor ponter & descriptor number
835  */
836 static inline int nicvf_get_sq_desc(struct snd_queue *sq, int desc_cnt)
837 {
838 	int qentry;
839 
840 	qentry = sq->tail;
841 	atomic_sub(desc_cnt, &sq->free_cnt);
842 	sq->tail += desc_cnt;
843 	sq->tail &= (sq->dmem.q_len - 1);
844 
845 	return qentry;
846 }
847 
848 /* Free descriptor back to SQ for future use */
849 void nicvf_put_sq_desc(struct snd_queue *sq, int desc_cnt)
850 {
851 	atomic_add(desc_cnt, &sq->free_cnt);
852 	sq->head += desc_cnt;
853 	sq->head &= (sq->dmem.q_len - 1);
854 }
855 
856 static inline int nicvf_get_nxt_sqentry(struct snd_queue *sq, int qentry)
857 {
858 	qentry++;
859 	qentry &= (sq->dmem.q_len - 1);
860 	return qentry;
861 }
862 
863 void nicvf_sq_enable(struct nicvf *nic, struct snd_queue *sq, int qidx)
864 {
865 	u64 sq_cfg;
866 
867 	sq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_CFG, qidx);
868 	sq_cfg |= NICVF_SQ_EN;
869 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, sq_cfg);
870 	/* Ring doorbell so that H/W restarts processing SQEs */
871 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_DOOR, qidx, 0);
872 }
873 
874 void nicvf_sq_disable(struct nicvf *nic, int qidx)
875 {
876 	u64 sq_cfg;
877 
878 	sq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_CFG, qidx);
879 	sq_cfg &= ~NICVF_SQ_EN;
880 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, sq_cfg);
881 }
882 
883 void nicvf_sq_free_used_descs(struct net_device *netdev, struct snd_queue *sq,
884 			      int qidx)
885 {
886 	u64 head, tail;
887 	struct sk_buff *skb;
888 	struct nicvf *nic = netdev_priv(netdev);
889 	struct sq_hdr_subdesc *hdr;
890 
891 	head = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_HEAD, qidx) >> 4;
892 	tail = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_TAIL, qidx) >> 4;
893 	while (sq->head != head) {
894 		hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, sq->head);
895 		if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER) {
896 			nicvf_put_sq_desc(sq, 1);
897 			continue;
898 		}
899 		skb = (struct sk_buff *)sq->skbuff[sq->head];
900 		if (skb)
901 			dev_kfree_skb_any(skb);
902 		atomic64_add(1, (atomic64_t *)&netdev->stats.tx_packets);
903 		atomic64_add(hdr->tot_len,
904 			     (atomic64_t *)&netdev->stats.tx_bytes);
905 		nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1);
906 	}
907 }
908 
909 /* Calculate no of SQ subdescriptors needed to transmit all
910  * segments of this TSO packet.
911  * Taken from 'Tilera network driver' with a minor modification.
912  */
913 static int nicvf_tso_count_subdescs(struct sk_buff *skb)
914 {
915 	struct skb_shared_info *sh = skb_shinfo(skb);
916 	unsigned int sh_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
917 	unsigned int data_len = skb->len - sh_len;
918 	unsigned int p_len = sh->gso_size;
919 	long f_id = -1;    /* id of the current fragment */
920 	long f_size = skb_headlen(skb) - sh_len;  /* current fragment size */
921 	long f_used = 0;  /* bytes used from the current fragment */
922 	long n;            /* size of the current piece of payload */
923 	int num_edescs = 0;
924 	int segment;
925 
926 	for (segment = 0; segment < sh->gso_segs; segment++) {
927 		unsigned int p_used = 0;
928 
929 		/* One edesc for header and for each piece of the payload. */
930 		for (num_edescs++; p_used < p_len; num_edescs++) {
931 			/* Advance as needed. */
932 			while (f_used >= f_size) {
933 				f_id++;
934 				f_size = skb_frag_size(&sh->frags[f_id]);
935 				f_used = 0;
936 			}
937 
938 			/* Use bytes from the current fragment. */
939 			n = p_len - p_used;
940 			if (n > f_size - f_used)
941 				n = f_size - f_used;
942 			f_used += n;
943 			p_used += n;
944 		}
945 
946 		/* The last segment may be less than gso_size. */
947 		data_len -= p_len;
948 		if (data_len < p_len)
949 			p_len = data_len;
950 	}
951 
952 	/* '+ gso_segs' for SQ_HDR_SUDESCs for each segment */
953 	return num_edescs + sh->gso_segs;
954 }
955 
956 #define POST_CQE_DESC_COUNT 2
957 
958 /* Get the number of SQ descriptors needed to xmit this skb */
959 static int nicvf_sq_subdesc_required(struct nicvf *nic, struct sk_buff *skb)
960 {
961 	int subdesc_cnt = MIN_SQ_DESC_PER_PKT_XMIT;
962 
963 	if (skb_shinfo(skb)->gso_size && !nic->hw_tso) {
964 		subdesc_cnt = nicvf_tso_count_subdescs(skb);
965 		return subdesc_cnt;
966 	}
967 
968 	/* Dummy descriptors to get TSO pkt completion notification */
969 	if (nic->t88 && nic->hw_tso && skb_shinfo(skb)->gso_size)
970 		subdesc_cnt += POST_CQE_DESC_COUNT;
971 
972 	if (skb_shinfo(skb)->nr_frags)
973 		subdesc_cnt += skb_shinfo(skb)->nr_frags;
974 
975 	return subdesc_cnt;
976 }
977 
978 /* Add SQ HEADER subdescriptor.
979  * First subdescriptor for every send descriptor.
980  */
981 static inline void
982 nicvf_sq_add_hdr_subdesc(struct nicvf *nic, struct snd_queue *sq, int qentry,
983 			 int subdesc_cnt, struct sk_buff *skb, int len)
984 {
985 	int proto;
986 	struct sq_hdr_subdesc *hdr;
987 
988 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, qentry);
989 	memset(hdr, 0, SND_QUEUE_DESC_SIZE);
990 	hdr->subdesc_type = SQ_DESC_TYPE_HEADER;
991 
992 	if (nic->t88 && nic->hw_tso && skb_shinfo(skb)->gso_size) {
993 		/* post_cqe = 0, to avoid HW posting a CQE for every TSO
994 		 * segment transmitted on 88xx.
995 		 */
996 		hdr->subdesc_cnt = subdesc_cnt - POST_CQE_DESC_COUNT;
997 	} else {
998 		sq->skbuff[qentry] = (u64)skb;
999 		/* Enable notification via CQE after processing SQE */
1000 		hdr->post_cqe = 1;
1001 		/* No of subdescriptors following this */
1002 		hdr->subdesc_cnt = subdesc_cnt;
1003 	}
1004 	hdr->tot_len = len;
1005 
1006 	/* Offload checksum calculation to HW */
1007 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1008 		hdr->csum_l3 = 1; /* Enable IP csum calculation */
1009 		hdr->l3_offset = skb_network_offset(skb);
1010 		hdr->l4_offset = skb_transport_offset(skb);
1011 
1012 		proto = ip_hdr(skb)->protocol;
1013 		switch (proto) {
1014 		case IPPROTO_TCP:
1015 			hdr->csum_l4 = SEND_L4_CSUM_TCP;
1016 			break;
1017 		case IPPROTO_UDP:
1018 			hdr->csum_l4 = SEND_L4_CSUM_UDP;
1019 			break;
1020 		case IPPROTO_SCTP:
1021 			hdr->csum_l4 = SEND_L4_CSUM_SCTP;
1022 			break;
1023 		}
1024 	}
1025 
1026 	if (nic->hw_tso && skb_shinfo(skb)->gso_size) {
1027 		hdr->tso = 1;
1028 		hdr->tso_start = skb_transport_offset(skb) + tcp_hdrlen(skb);
1029 		hdr->tso_max_paysize = skb_shinfo(skb)->gso_size;
1030 		/* For non-tunneled pkts, point this to L2 ethertype */
1031 		hdr->inner_l3_offset = skb_network_offset(skb) - 2;
1032 		nic->drv_stats.tx_tso++;
1033 	}
1034 }
1035 
1036 /* SQ GATHER subdescriptor
1037  * Must follow HDR descriptor
1038  */
1039 static inline void nicvf_sq_add_gather_subdesc(struct snd_queue *sq, int qentry,
1040 					       int size, u64 data)
1041 {
1042 	struct sq_gather_subdesc *gather;
1043 
1044 	qentry &= (sq->dmem.q_len - 1);
1045 	gather = (struct sq_gather_subdesc *)GET_SQ_DESC(sq, qentry);
1046 
1047 	memset(gather, 0, SND_QUEUE_DESC_SIZE);
1048 	gather->subdesc_type = SQ_DESC_TYPE_GATHER;
1049 	gather->ld_type = NIC_SEND_LD_TYPE_E_LDD;
1050 	gather->size = size;
1051 	gather->addr = data;
1052 }
1053 
1054 /* Add HDR + IMMEDIATE subdescriptors right after descriptors of a TSO
1055  * packet so that a CQE is posted as a notifation for transmission of
1056  * TSO packet.
1057  */
1058 static inline void nicvf_sq_add_cqe_subdesc(struct snd_queue *sq, int qentry,
1059 					    int tso_sqe, struct sk_buff *skb)
1060 {
1061 	struct sq_imm_subdesc *imm;
1062 	struct sq_hdr_subdesc *hdr;
1063 
1064 	sq->skbuff[qentry] = (u64)skb;
1065 
1066 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, qentry);
1067 	memset(hdr, 0, SND_QUEUE_DESC_SIZE);
1068 	hdr->subdesc_type = SQ_DESC_TYPE_HEADER;
1069 	/* Enable notification via CQE after processing SQE */
1070 	hdr->post_cqe = 1;
1071 	/* There is no packet to transmit here */
1072 	hdr->dont_send = 1;
1073 	hdr->subdesc_cnt = POST_CQE_DESC_COUNT - 1;
1074 	hdr->tot_len = 1;
1075 	/* Actual TSO header SQE index, needed for cleanup */
1076 	hdr->rsvd2 = tso_sqe;
1077 
1078 	qentry = nicvf_get_nxt_sqentry(sq, qentry);
1079 	imm = (struct sq_imm_subdesc *)GET_SQ_DESC(sq, qentry);
1080 	memset(imm, 0, SND_QUEUE_DESC_SIZE);
1081 	imm->subdesc_type = SQ_DESC_TYPE_IMMEDIATE;
1082 	imm->len = 1;
1083 }
1084 
1085 static inline void nicvf_sq_doorbell(struct nicvf *nic, struct sk_buff *skb,
1086 				     int sq_num, int desc_cnt)
1087 {
1088 	struct netdev_queue *txq;
1089 
1090 	txq = netdev_get_tx_queue(nic->pnicvf->netdev,
1091 				  skb_get_queue_mapping(skb));
1092 
1093 	netdev_tx_sent_queue(txq, skb->len);
1094 
1095 	/* make sure all memory stores are done before ringing doorbell */
1096 	smp_wmb();
1097 
1098 	/* Inform HW to xmit all TSO segments */
1099 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_DOOR,
1100 			      sq_num, desc_cnt);
1101 }
1102 
1103 /* Segment a TSO packet into 'gso_size' segments and append
1104  * them to SQ for transfer
1105  */
1106 static int nicvf_sq_append_tso(struct nicvf *nic, struct snd_queue *sq,
1107 			       int sq_num, int qentry, struct sk_buff *skb)
1108 {
1109 	struct tso_t tso;
1110 	int seg_subdescs = 0, desc_cnt = 0;
1111 	int seg_len, total_len, data_left;
1112 	int hdr_qentry = qentry;
1113 	int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1114 
1115 	tso_start(skb, &tso);
1116 	total_len = skb->len - hdr_len;
1117 	while (total_len > 0) {
1118 		char *hdr;
1119 
1120 		/* Save Qentry for adding HDR_SUBDESC at the end */
1121 		hdr_qentry = qentry;
1122 
1123 		data_left = min_t(int, skb_shinfo(skb)->gso_size, total_len);
1124 		total_len -= data_left;
1125 
1126 		/* Add segment's header */
1127 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1128 		hdr = sq->tso_hdrs + qentry * TSO_HEADER_SIZE;
1129 		tso_build_hdr(skb, hdr, &tso, data_left, total_len == 0);
1130 		nicvf_sq_add_gather_subdesc(sq, qentry, hdr_len,
1131 					    sq->tso_hdrs_phys +
1132 					    qentry * TSO_HEADER_SIZE);
1133 		/* HDR_SUDESC + GATHER */
1134 		seg_subdescs = 2;
1135 		seg_len = hdr_len;
1136 
1137 		/* Add segment's payload fragments */
1138 		while (data_left > 0) {
1139 			int size;
1140 
1141 			size = min_t(int, tso.size, data_left);
1142 
1143 			qentry = nicvf_get_nxt_sqentry(sq, qentry);
1144 			nicvf_sq_add_gather_subdesc(sq, qentry, size,
1145 						    virt_to_phys(tso.data));
1146 			seg_subdescs++;
1147 			seg_len += size;
1148 
1149 			data_left -= size;
1150 			tso_build_data(skb, &tso, size);
1151 		}
1152 		nicvf_sq_add_hdr_subdesc(nic, sq, hdr_qentry,
1153 					 seg_subdescs - 1, skb, seg_len);
1154 		sq->skbuff[hdr_qentry] = (u64)NULL;
1155 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1156 
1157 		desc_cnt += seg_subdescs;
1158 	}
1159 	/* Save SKB in the last segment for freeing */
1160 	sq->skbuff[hdr_qentry] = (u64)skb;
1161 
1162 	nicvf_sq_doorbell(nic, skb, sq_num, desc_cnt);
1163 
1164 	nic->drv_stats.tx_tso++;
1165 	return 1;
1166 }
1167 
1168 /* Append an skb to a SQ for packet transfer. */
1169 int nicvf_sq_append_skb(struct nicvf *nic, struct sk_buff *skb)
1170 {
1171 	int i, size;
1172 	int subdesc_cnt, tso_sqe = 0;
1173 	int sq_num, qentry;
1174 	struct queue_set *qs;
1175 	struct snd_queue *sq;
1176 
1177 	sq_num = skb_get_queue_mapping(skb);
1178 	if (sq_num >= MAX_SND_QUEUES_PER_QS) {
1179 		/* Get secondary Qset's SQ structure */
1180 		i = sq_num / MAX_SND_QUEUES_PER_QS;
1181 		if (!nic->snicvf[i - 1]) {
1182 			netdev_warn(nic->netdev,
1183 				    "Secondary Qset#%d's ptr not initialized\n",
1184 				    i - 1);
1185 			return 1;
1186 		}
1187 		nic = (struct nicvf *)nic->snicvf[i - 1];
1188 		sq_num = sq_num % MAX_SND_QUEUES_PER_QS;
1189 	}
1190 
1191 	qs = nic->qs;
1192 	sq = &qs->sq[sq_num];
1193 
1194 	subdesc_cnt = nicvf_sq_subdesc_required(nic, skb);
1195 	if (subdesc_cnt > atomic_read(&sq->free_cnt))
1196 		goto append_fail;
1197 
1198 	qentry = nicvf_get_sq_desc(sq, subdesc_cnt);
1199 
1200 	/* Check if its a TSO packet */
1201 	if (skb_shinfo(skb)->gso_size && !nic->hw_tso)
1202 		return nicvf_sq_append_tso(nic, sq, sq_num, qentry, skb);
1203 
1204 	/* Add SQ header subdesc */
1205 	nicvf_sq_add_hdr_subdesc(nic, sq, qentry, subdesc_cnt - 1,
1206 				 skb, skb->len);
1207 	tso_sqe = qentry;
1208 
1209 	/* Add SQ gather subdescs */
1210 	qentry = nicvf_get_nxt_sqentry(sq, qentry);
1211 	size = skb_is_nonlinear(skb) ? skb_headlen(skb) : skb->len;
1212 	nicvf_sq_add_gather_subdesc(sq, qentry, size, virt_to_phys(skb->data));
1213 
1214 	/* Check for scattered buffer */
1215 	if (!skb_is_nonlinear(skb))
1216 		goto doorbell;
1217 
1218 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1219 		const struct skb_frag_struct *frag;
1220 
1221 		frag = &skb_shinfo(skb)->frags[i];
1222 
1223 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1224 		size = skb_frag_size(frag);
1225 		nicvf_sq_add_gather_subdesc(sq, qentry, size,
1226 					    virt_to_phys(
1227 					    skb_frag_address(frag)));
1228 	}
1229 
1230 doorbell:
1231 	if (nic->t88 && skb_shinfo(skb)->gso_size) {
1232 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1233 		nicvf_sq_add_cqe_subdesc(sq, qentry, tso_sqe, skb);
1234 	}
1235 
1236 	nicvf_sq_doorbell(nic, skb, sq_num, subdesc_cnt);
1237 
1238 	return 1;
1239 
1240 append_fail:
1241 	/* Use original PCI dev for debug log */
1242 	nic = nic->pnicvf;
1243 	netdev_dbg(nic->netdev, "Not enough SQ descriptors to xmit pkt\n");
1244 	return 0;
1245 }
1246 
1247 static inline unsigned frag_num(unsigned i)
1248 {
1249 #ifdef __BIG_ENDIAN
1250 	return (i & ~3) + 3 - (i & 3);
1251 #else
1252 	return i;
1253 #endif
1254 }
1255 
1256 /* Returns SKB for a received packet */
1257 struct sk_buff *nicvf_get_rcv_skb(struct nicvf *nic, struct cqe_rx_t *cqe_rx)
1258 {
1259 	int frag;
1260 	int payload_len = 0;
1261 	struct sk_buff *skb = NULL;
1262 	struct page *page;
1263 	int offset;
1264 	u16 *rb_lens = NULL;
1265 	u64 *rb_ptrs = NULL;
1266 
1267 	rb_lens = (void *)cqe_rx + (3 * sizeof(u64));
1268 	/* Except 88xx pass1 on all other chips CQE_RX2_S is added to
1269 	 * CQE_RX at word6, hence buffer pointers move by word
1270 	 *
1271 	 * Use existing 'hw_tso' flag which will be set for all chips
1272 	 * except 88xx pass1 instead of a additional cache line
1273 	 * access (or miss) by using pci dev's revision.
1274 	 */
1275 	if (!nic->hw_tso)
1276 		rb_ptrs = (void *)cqe_rx + (6 * sizeof(u64));
1277 	else
1278 		rb_ptrs = (void *)cqe_rx + (7 * sizeof(u64));
1279 
1280 	netdev_dbg(nic->netdev, "%s rb_cnt %d rb0_ptr %llx rb0_sz %d\n",
1281 		   __func__, cqe_rx->rb_cnt, cqe_rx->rb0_ptr, cqe_rx->rb0_sz);
1282 
1283 	for (frag = 0; frag < cqe_rx->rb_cnt; frag++) {
1284 		payload_len = rb_lens[frag_num(frag)];
1285 		if (!frag) {
1286 			/* First fragment */
1287 			skb = nicvf_rb_ptr_to_skb(nic,
1288 						  *rb_ptrs - cqe_rx->align_pad,
1289 						  payload_len);
1290 			if (!skb)
1291 				return NULL;
1292 			skb_reserve(skb, cqe_rx->align_pad);
1293 			skb_put(skb, payload_len);
1294 		} else {
1295 			/* Add fragments */
1296 			page = virt_to_page(phys_to_virt(*rb_ptrs));
1297 			offset = phys_to_virt(*rb_ptrs) - page_address(page);
1298 			skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1299 					offset, payload_len, RCV_FRAG_LEN);
1300 		}
1301 		/* Next buffer pointer */
1302 		rb_ptrs++;
1303 	}
1304 	return skb;
1305 }
1306 
1307 static u64 nicvf_int_type_to_mask(int int_type, int q_idx)
1308 {
1309 	u64 reg_val;
1310 
1311 	switch (int_type) {
1312 	case NICVF_INTR_CQ:
1313 		reg_val = ((1ULL << q_idx) << NICVF_INTR_CQ_SHIFT);
1314 		break;
1315 	case NICVF_INTR_SQ:
1316 		reg_val = ((1ULL << q_idx) << NICVF_INTR_SQ_SHIFT);
1317 		break;
1318 	case NICVF_INTR_RBDR:
1319 		reg_val = ((1ULL << q_idx) << NICVF_INTR_RBDR_SHIFT);
1320 		break;
1321 	case NICVF_INTR_PKT_DROP:
1322 		reg_val = (1ULL << NICVF_INTR_PKT_DROP_SHIFT);
1323 		break;
1324 	case NICVF_INTR_TCP_TIMER:
1325 		reg_val = (1ULL << NICVF_INTR_TCP_TIMER_SHIFT);
1326 		break;
1327 	case NICVF_INTR_MBOX:
1328 		reg_val = (1ULL << NICVF_INTR_MBOX_SHIFT);
1329 		break;
1330 	case NICVF_INTR_QS_ERR:
1331 		reg_val = (1ULL << NICVF_INTR_QS_ERR_SHIFT);
1332 		break;
1333 	default:
1334 		reg_val = 0;
1335 	}
1336 
1337 	return reg_val;
1338 }
1339 
1340 /* Enable interrupt */
1341 void nicvf_enable_intr(struct nicvf *nic, int int_type, int q_idx)
1342 {
1343 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1344 
1345 	if (!mask) {
1346 		netdev_dbg(nic->netdev,
1347 			   "Failed to enable interrupt: unknown type\n");
1348 		return;
1349 	}
1350 	nicvf_reg_write(nic, NIC_VF_ENA_W1S,
1351 			nicvf_reg_read(nic, NIC_VF_ENA_W1S) | mask);
1352 }
1353 
1354 /* Disable interrupt */
1355 void nicvf_disable_intr(struct nicvf *nic, int int_type, int q_idx)
1356 {
1357 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1358 
1359 	if (!mask) {
1360 		netdev_dbg(nic->netdev,
1361 			   "Failed to disable interrupt: unknown type\n");
1362 		return;
1363 	}
1364 
1365 	nicvf_reg_write(nic, NIC_VF_ENA_W1C, mask);
1366 }
1367 
1368 /* Clear interrupt */
1369 void nicvf_clear_intr(struct nicvf *nic, int int_type, int q_idx)
1370 {
1371 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1372 
1373 	if (!mask) {
1374 		netdev_dbg(nic->netdev,
1375 			   "Failed to clear interrupt: unknown type\n");
1376 		return;
1377 	}
1378 
1379 	nicvf_reg_write(nic, NIC_VF_INT, mask);
1380 }
1381 
1382 /* Check if interrupt is enabled */
1383 int nicvf_is_intr_enabled(struct nicvf *nic, int int_type, int q_idx)
1384 {
1385 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1386 	/* If interrupt type is unknown, we treat it disabled. */
1387 	if (!mask) {
1388 		netdev_dbg(nic->netdev,
1389 			   "Failed to check interrupt enable: unknown type\n");
1390 		return 0;
1391 	}
1392 
1393 	return mask & nicvf_reg_read(nic, NIC_VF_ENA_W1S);
1394 }
1395 
1396 void nicvf_update_rq_stats(struct nicvf *nic, int rq_idx)
1397 {
1398 	struct rcv_queue *rq;
1399 
1400 #define GET_RQ_STATS(reg) \
1401 	nicvf_reg_read(nic, NIC_QSET_RQ_0_7_STAT_0_1 |\
1402 			    (rq_idx << NIC_Q_NUM_SHIFT) | (reg << 3))
1403 
1404 	rq = &nic->qs->rq[rq_idx];
1405 	rq->stats.bytes = GET_RQ_STATS(RQ_SQ_STATS_OCTS);
1406 	rq->stats.pkts = GET_RQ_STATS(RQ_SQ_STATS_PKTS);
1407 }
1408 
1409 void nicvf_update_sq_stats(struct nicvf *nic, int sq_idx)
1410 {
1411 	struct snd_queue *sq;
1412 
1413 #define GET_SQ_STATS(reg) \
1414 	nicvf_reg_read(nic, NIC_QSET_SQ_0_7_STAT_0_1 |\
1415 			    (sq_idx << NIC_Q_NUM_SHIFT) | (reg << 3))
1416 
1417 	sq = &nic->qs->sq[sq_idx];
1418 	sq->stats.bytes = GET_SQ_STATS(RQ_SQ_STATS_OCTS);
1419 	sq->stats.pkts = GET_SQ_STATS(RQ_SQ_STATS_PKTS);
1420 }
1421 
1422 /* Check for errors in the receive cmp.queue entry */
1423 int nicvf_check_cqe_rx_errs(struct nicvf *nic, struct cqe_rx_t *cqe_rx)
1424 {
1425 	struct nicvf_hw_stats *stats = &nic->hw_stats;
1426 
1427 	if (!cqe_rx->err_level && !cqe_rx->err_opcode)
1428 		return 0;
1429 
1430 	if (netif_msg_rx_err(nic))
1431 		netdev_err(nic->netdev,
1432 			   "%s: RX error CQE err_level 0x%x err_opcode 0x%x\n",
1433 			   nic->netdev->name,
1434 			   cqe_rx->err_level, cqe_rx->err_opcode);
1435 
1436 	switch (cqe_rx->err_opcode) {
1437 	case CQ_RX_ERROP_RE_PARTIAL:
1438 		stats->rx_bgx_truncated_pkts++;
1439 		break;
1440 	case CQ_RX_ERROP_RE_JABBER:
1441 		stats->rx_jabber_errs++;
1442 		break;
1443 	case CQ_RX_ERROP_RE_FCS:
1444 		stats->rx_fcs_errs++;
1445 		break;
1446 	case CQ_RX_ERROP_RE_RX_CTL:
1447 		stats->rx_bgx_errs++;
1448 		break;
1449 	case CQ_RX_ERROP_PREL2_ERR:
1450 		stats->rx_prel2_errs++;
1451 		break;
1452 	case CQ_RX_ERROP_L2_MAL:
1453 		stats->rx_l2_hdr_malformed++;
1454 		break;
1455 	case CQ_RX_ERROP_L2_OVERSIZE:
1456 		stats->rx_oversize++;
1457 		break;
1458 	case CQ_RX_ERROP_L2_UNDERSIZE:
1459 		stats->rx_undersize++;
1460 		break;
1461 	case CQ_RX_ERROP_L2_LENMISM:
1462 		stats->rx_l2_len_mismatch++;
1463 		break;
1464 	case CQ_RX_ERROP_L2_PCLP:
1465 		stats->rx_l2_pclp++;
1466 		break;
1467 	case CQ_RX_ERROP_IP_NOT:
1468 		stats->rx_ip_ver_errs++;
1469 		break;
1470 	case CQ_RX_ERROP_IP_CSUM_ERR:
1471 		stats->rx_ip_csum_errs++;
1472 		break;
1473 	case CQ_RX_ERROP_IP_MAL:
1474 		stats->rx_ip_hdr_malformed++;
1475 		break;
1476 	case CQ_RX_ERROP_IP_MALD:
1477 		stats->rx_ip_payload_malformed++;
1478 		break;
1479 	case CQ_RX_ERROP_IP_HOP:
1480 		stats->rx_ip_ttl_errs++;
1481 		break;
1482 	case CQ_RX_ERROP_L3_PCLP:
1483 		stats->rx_l3_pclp++;
1484 		break;
1485 	case CQ_RX_ERROP_L4_MAL:
1486 		stats->rx_l4_malformed++;
1487 		break;
1488 	case CQ_RX_ERROP_L4_CHK:
1489 		stats->rx_l4_csum_errs++;
1490 		break;
1491 	case CQ_RX_ERROP_UDP_LEN:
1492 		stats->rx_udp_len_errs++;
1493 		break;
1494 	case CQ_RX_ERROP_L4_PORT:
1495 		stats->rx_l4_port_errs++;
1496 		break;
1497 	case CQ_RX_ERROP_TCP_FLAG:
1498 		stats->rx_tcp_flag_errs++;
1499 		break;
1500 	case CQ_RX_ERROP_TCP_OFFSET:
1501 		stats->rx_tcp_offset_errs++;
1502 		break;
1503 	case CQ_RX_ERROP_L4_PCLP:
1504 		stats->rx_l4_pclp++;
1505 		break;
1506 	case CQ_RX_ERROP_RBDR_TRUNC:
1507 		stats->rx_truncated_pkts++;
1508 		break;
1509 	}
1510 
1511 	return 1;
1512 }
1513 
1514 /* Check for errors in the send cmp.queue entry */
1515 int nicvf_check_cqe_tx_errs(struct nicvf *nic,
1516 			    struct cmp_queue *cq, struct cqe_send_t *cqe_tx)
1517 {
1518 	struct cmp_queue_stats *stats = &cq->stats;
1519 
1520 	switch (cqe_tx->send_status) {
1521 	case CQ_TX_ERROP_GOOD:
1522 		stats->tx.good++;
1523 		return 0;
1524 	case CQ_TX_ERROP_DESC_FAULT:
1525 		stats->tx.desc_fault++;
1526 		break;
1527 	case CQ_TX_ERROP_HDR_CONS_ERR:
1528 		stats->tx.hdr_cons_err++;
1529 		break;
1530 	case CQ_TX_ERROP_SUBDC_ERR:
1531 		stats->tx.subdesc_err++;
1532 		break;
1533 	case CQ_TX_ERROP_IMM_SIZE_OFLOW:
1534 		stats->tx.imm_size_oflow++;
1535 		break;
1536 	case CQ_TX_ERROP_DATA_SEQUENCE_ERR:
1537 		stats->tx.data_seq_err++;
1538 		break;
1539 	case CQ_TX_ERROP_MEM_SEQUENCE_ERR:
1540 		stats->tx.mem_seq_err++;
1541 		break;
1542 	case CQ_TX_ERROP_LOCK_VIOL:
1543 		stats->tx.lock_viol++;
1544 		break;
1545 	case CQ_TX_ERROP_DATA_FAULT:
1546 		stats->tx.data_fault++;
1547 		break;
1548 	case CQ_TX_ERROP_TSTMP_CONFLICT:
1549 		stats->tx.tstmp_conflict++;
1550 		break;
1551 	case CQ_TX_ERROP_TSTMP_TIMEOUT:
1552 		stats->tx.tstmp_timeout++;
1553 		break;
1554 	case CQ_TX_ERROP_MEM_FAULT:
1555 		stats->tx.mem_fault++;
1556 		break;
1557 	case CQ_TX_ERROP_CK_OVERLAP:
1558 		stats->tx.csum_overlap++;
1559 		break;
1560 	case CQ_TX_ERROP_CK_OFLOW:
1561 		stats->tx.csum_overflow++;
1562 		break;
1563 	}
1564 
1565 	return 1;
1566 }
1567