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 			this_cpu_inc(nic->pnicvf->drv_stats->
108 				     rcv_buffer_alloc_failures);
109 			return -ENOMEM;
110 		}
111 		nic->rb_page_offset = 0;
112 	}
113 
114 ret:
115 	*rbuf = (u64 *)((u64)page_address(nic->rb_page) + nic->rb_page_offset);
116 	nic->rb_page_offset += buf_len;
117 
118 	return 0;
119 }
120 
121 /* Build skb around receive buffer */
122 static struct sk_buff *nicvf_rb_ptr_to_skb(struct nicvf *nic,
123 					   u64 rb_ptr, int len)
124 {
125 	void *data;
126 	struct sk_buff *skb;
127 
128 	data = phys_to_virt(rb_ptr);
129 
130 	/* Now build an skb to give to stack */
131 	skb = build_skb(data, RCV_FRAG_LEN);
132 	if (!skb) {
133 		put_page(virt_to_page(data));
134 		return NULL;
135 	}
136 
137 	prefetch(skb->data);
138 	return skb;
139 }
140 
141 /* Allocate RBDR ring and populate receive buffers */
142 static int  nicvf_init_rbdr(struct nicvf *nic, struct rbdr *rbdr,
143 			    int ring_len, int buf_size)
144 {
145 	int idx;
146 	u64 *rbuf;
147 	struct rbdr_entry_t *desc;
148 	int err;
149 
150 	err = nicvf_alloc_q_desc_mem(nic, &rbdr->dmem, ring_len,
151 				     sizeof(struct rbdr_entry_t),
152 				     NICVF_RCV_BUF_ALIGN_BYTES);
153 	if (err)
154 		return err;
155 
156 	rbdr->desc = rbdr->dmem.base;
157 	/* Buffer size has to be in multiples of 128 bytes */
158 	rbdr->dma_size = buf_size;
159 	rbdr->enable = true;
160 	rbdr->thresh = RBDR_THRESH;
161 
162 	nic->rb_page = NULL;
163 	for (idx = 0; idx < ring_len; idx++) {
164 		err = nicvf_alloc_rcv_buffer(nic, GFP_KERNEL, RCV_FRAG_LEN,
165 					     &rbuf);
166 		if (err)
167 			return err;
168 
169 		desc = GET_RBDR_DESC(rbdr, idx);
170 		desc->buf_addr = virt_to_phys(rbuf) >> NICVF_RCV_BUF_ALIGN;
171 	}
172 
173 	nicvf_get_page(nic);
174 
175 	return 0;
176 }
177 
178 /* Free RBDR ring and its receive buffers */
179 static void nicvf_free_rbdr(struct nicvf *nic, struct rbdr *rbdr)
180 {
181 	int head, tail;
182 	u64 buf_addr;
183 	struct rbdr_entry_t *desc;
184 
185 	if (!rbdr)
186 		return;
187 
188 	rbdr->enable = false;
189 	if (!rbdr->dmem.base)
190 		return;
191 
192 	head = rbdr->head;
193 	tail = rbdr->tail;
194 
195 	/* Free SKBs */
196 	while (head != tail) {
197 		desc = GET_RBDR_DESC(rbdr, head);
198 		buf_addr = desc->buf_addr << NICVF_RCV_BUF_ALIGN;
199 		put_page(virt_to_page(phys_to_virt(buf_addr)));
200 		head++;
201 		head &= (rbdr->dmem.q_len - 1);
202 	}
203 	/* Free SKB of tail desc */
204 	desc = GET_RBDR_DESC(rbdr, tail);
205 	buf_addr = desc->buf_addr << NICVF_RCV_BUF_ALIGN;
206 	put_page(virt_to_page(phys_to_virt(buf_addr)));
207 
208 	/* Free RBDR ring */
209 	nicvf_free_q_desc_mem(nic, &rbdr->dmem);
210 }
211 
212 /* Refill receive buffer descriptors with new buffers.
213  */
214 static void nicvf_refill_rbdr(struct nicvf *nic, gfp_t gfp)
215 {
216 	struct queue_set *qs = nic->qs;
217 	int rbdr_idx = qs->rbdr_cnt;
218 	int tail, qcount;
219 	int refill_rb_cnt;
220 	struct rbdr *rbdr;
221 	struct rbdr_entry_t *desc;
222 	u64 *rbuf;
223 	int new_rb = 0;
224 
225 refill:
226 	if (!rbdr_idx)
227 		return;
228 	rbdr_idx--;
229 	rbdr = &qs->rbdr[rbdr_idx];
230 	/* Check if it's enabled */
231 	if (!rbdr->enable)
232 		goto next_rbdr;
233 
234 	/* Get no of desc's to be refilled */
235 	qcount = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_STATUS0, rbdr_idx);
236 	qcount &= 0x7FFFF;
237 	/* Doorbell can be ringed with a max of ring size minus 1 */
238 	if (qcount >= (qs->rbdr_len - 1))
239 		goto next_rbdr;
240 	else
241 		refill_rb_cnt = qs->rbdr_len - qcount - 1;
242 
243 	/* Start filling descs from tail */
244 	tail = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_TAIL, rbdr_idx) >> 3;
245 	while (refill_rb_cnt) {
246 		tail++;
247 		tail &= (rbdr->dmem.q_len - 1);
248 
249 		if (nicvf_alloc_rcv_buffer(nic, gfp, RCV_FRAG_LEN, &rbuf))
250 			break;
251 
252 		desc = GET_RBDR_DESC(rbdr, tail);
253 		desc->buf_addr = virt_to_phys(rbuf) >> NICVF_RCV_BUF_ALIGN;
254 		refill_rb_cnt--;
255 		new_rb++;
256 	}
257 
258 	nicvf_get_page(nic);
259 
260 	/* make sure all memory stores are done before ringing doorbell */
261 	smp_wmb();
262 
263 	/* Check if buffer allocation failed */
264 	if (refill_rb_cnt)
265 		nic->rb_alloc_fail = true;
266 	else
267 		nic->rb_alloc_fail = false;
268 
269 	/* Notify HW */
270 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_DOOR,
271 			      rbdr_idx, new_rb);
272 next_rbdr:
273 	/* Re-enable RBDR interrupts only if buffer allocation is success */
274 	if (!nic->rb_alloc_fail && rbdr->enable &&
275 	    netif_running(nic->pnicvf->netdev))
276 		nicvf_enable_intr(nic, NICVF_INTR_RBDR, rbdr_idx);
277 
278 	if (rbdr_idx)
279 		goto refill;
280 }
281 
282 /* Alloc rcv buffers in non-atomic mode for better success */
283 void nicvf_rbdr_work(struct work_struct *work)
284 {
285 	struct nicvf *nic = container_of(work, struct nicvf, rbdr_work.work);
286 
287 	nicvf_refill_rbdr(nic, GFP_KERNEL);
288 	if (nic->rb_alloc_fail)
289 		schedule_delayed_work(&nic->rbdr_work, msecs_to_jiffies(10));
290 	else
291 		nic->rb_work_scheduled = false;
292 }
293 
294 /* In Softirq context, alloc rcv buffers in atomic mode */
295 void nicvf_rbdr_task(unsigned long data)
296 {
297 	struct nicvf *nic = (struct nicvf *)data;
298 
299 	nicvf_refill_rbdr(nic, GFP_ATOMIC);
300 	if (nic->rb_alloc_fail) {
301 		nic->rb_work_scheduled = true;
302 		schedule_delayed_work(&nic->rbdr_work, msecs_to_jiffies(10));
303 	}
304 }
305 
306 /* Initialize completion queue */
307 static int nicvf_init_cmp_queue(struct nicvf *nic,
308 				struct cmp_queue *cq, int q_len)
309 {
310 	int err;
311 
312 	err = nicvf_alloc_q_desc_mem(nic, &cq->dmem, q_len, CMP_QUEUE_DESC_SIZE,
313 				     NICVF_CQ_BASE_ALIGN_BYTES);
314 	if (err)
315 		return err;
316 
317 	cq->desc = cq->dmem.base;
318 	cq->thresh = pass1_silicon(nic->pdev) ? 0 : CMP_QUEUE_CQE_THRESH;
319 	nic->cq_coalesce_usecs = (CMP_QUEUE_TIMER_THRESH * 0.05) - 1;
320 
321 	return 0;
322 }
323 
324 static void nicvf_free_cmp_queue(struct nicvf *nic, struct cmp_queue *cq)
325 {
326 	if (!cq)
327 		return;
328 	if (!cq->dmem.base)
329 		return;
330 
331 	nicvf_free_q_desc_mem(nic, &cq->dmem);
332 }
333 
334 /* Initialize transmit queue */
335 static int nicvf_init_snd_queue(struct nicvf *nic,
336 				struct snd_queue *sq, int q_len)
337 {
338 	int err;
339 
340 	err = nicvf_alloc_q_desc_mem(nic, &sq->dmem, q_len, SND_QUEUE_DESC_SIZE,
341 				     NICVF_SQ_BASE_ALIGN_BYTES);
342 	if (err)
343 		return err;
344 
345 	sq->desc = sq->dmem.base;
346 	sq->skbuff = kcalloc(q_len, sizeof(u64), GFP_KERNEL);
347 	if (!sq->skbuff)
348 		return -ENOMEM;
349 	sq->head = 0;
350 	sq->tail = 0;
351 	atomic_set(&sq->free_cnt, q_len - 1);
352 	sq->thresh = SND_QUEUE_THRESH;
353 
354 	/* Preallocate memory for TSO segment's header */
355 	sq->tso_hdrs = dma_alloc_coherent(&nic->pdev->dev,
356 					  q_len * TSO_HEADER_SIZE,
357 					  &sq->tso_hdrs_phys, GFP_KERNEL);
358 	if (!sq->tso_hdrs)
359 		return -ENOMEM;
360 
361 	return 0;
362 }
363 
364 static void nicvf_free_snd_queue(struct nicvf *nic, struct snd_queue *sq)
365 {
366 	struct sk_buff *skb;
367 
368 	if (!sq)
369 		return;
370 	if (!sq->dmem.base)
371 		return;
372 
373 	if (sq->tso_hdrs)
374 		dma_free_coherent(&nic->pdev->dev,
375 				  sq->dmem.q_len * TSO_HEADER_SIZE,
376 				  sq->tso_hdrs, sq->tso_hdrs_phys);
377 
378 	/* Free pending skbs in the queue */
379 	smp_rmb();
380 	while (sq->head != sq->tail) {
381 		skb = (struct sk_buff *)sq->skbuff[sq->head];
382 		if (skb)
383 			dev_kfree_skb_any(skb);
384 		sq->head++;
385 		sq->head &= (sq->dmem.q_len - 1);
386 	}
387 	kfree(sq->skbuff);
388 	nicvf_free_q_desc_mem(nic, &sq->dmem);
389 }
390 
391 static void nicvf_reclaim_snd_queue(struct nicvf *nic,
392 				    struct queue_set *qs, int qidx)
393 {
394 	/* Disable send queue */
395 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, 0);
396 	/* Check if SQ is stopped */
397 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_SQ_0_7_STATUS, 21, 1, 0x01))
398 		return;
399 	/* Reset send queue */
400 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, NICVF_SQ_RESET);
401 }
402 
403 static void nicvf_reclaim_rcv_queue(struct nicvf *nic,
404 				    struct queue_set *qs, int qidx)
405 {
406 	union nic_mbx mbx = {};
407 
408 	/* Make sure all packets in the pipeline are written back into mem */
409 	mbx.msg.msg = NIC_MBOX_MSG_RQ_SW_SYNC;
410 	nicvf_send_msg_to_pf(nic, &mbx);
411 }
412 
413 static void nicvf_reclaim_cmp_queue(struct nicvf *nic,
414 				    struct queue_set *qs, int qidx)
415 {
416 	/* Disable timer threshold (doesn't get reset upon CQ reset */
417 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG2, qidx, 0);
418 	/* Disable completion queue */
419 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, 0);
420 	/* Reset completion queue */
421 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, NICVF_CQ_RESET);
422 }
423 
424 static void nicvf_reclaim_rbdr(struct nicvf *nic,
425 			       struct rbdr *rbdr, int qidx)
426 {
427 	u64 tmp, fifo_state;
428 	int timeout = 10;
429 
430 	/* Save head and tail pointers for feeing up buffers */
431 	rbdr->head = nicvf_queue_reg_read(nic,
432 					  NIC_QSET_RBDR_0_1_HEAD,
433 					  qidx) >> 3;
434 	rbdr->tail = nicvf_queue_reg_read(nic,
435 					  NIC_QSET_RBDR_0_1_TAIL,
436 					  qidx) >> 3;
437 
438 	/* If RBDR FIFO is in 'FAIL' state then do a reset first
439 	 * before relaiming.
440 	 */
441 	fifo_state = nicvf_queue_reg_read(nic, NIC_QSET_RBDR_0_1_STATUS0, qidx);
442 	if (((fifo_state >> 62) & 0x03) == 0x3)
443 		nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG,
444 				      qidx, NICVF_RBDR_RESET);
445 
446 	/* Disable RBDR */
447 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx, 0);
448 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x00))
449 		return;
450 	while (1) {
451 		tmp = nicvf_queue_reg_read(nic,
452 					   NIC_QSET_RBDR_0_1_PREFETCH_STATUS,
453 					   qidx);
454 		if ((tmp & 0xFFFFFFFF) == ((tmp >> 32) & 0xFFFFFFFF))
455 			break;
456 		usleep_range(1000, 2000);
457 		timeout--;
458 		if (!timeout) {
459 			netdev_err(nic->netdev,
460 				   "Failed polling on prefetch status\n");
461 			return;
462 		}
463 	}
464 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG,
465 			      qidx, NICVF_RBDR_RESET);
466 
467 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x02))
468 		return;
469 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG, qidx, 0x00);
470 	if (nicvf_poll_reg(nic, qidx, NIC_QSET_RBDR_0_1_STATUS0, 62, 2, 0x00))
471 		return;
472 }
473 
474 void nicvf_config_vlan_stripping(struct nicvf *nic, netdev_features_t features)
475 {
476 	u64 rq_cfg;
477 	int sqs;
478 
479 	rq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_RQ_GEN_CFG, 0);
480 
481 	/* Enable first VLAN stripping */
482 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
483 		rq_cfg |= (1ULL << 25);
484 	else
485 		rq_cfg &= ~(1ULL << 25);
486 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_GEN_CFG, 0, rq_cfg);
487 
488 	/* Configure Secondary Qsets, if any */
489 	for (sqs = 0; sqs < nic->sqs_count; sqs++)
490 		if (nic->snicvf[sqs])
491 			nicvf_queue_reg_write(nic->snicvf[sqs],
492 					      NIC_QSET_RQ_GEN_CFG, 0, rq_cfg);
493 }
494 
495 static void nicvf_reset_rcv_queue_stats(struct nicvf *nic)
496 {
497 	union nic_mbx mbx = {};
498 
499 	/* Reset all RQ/SQ and VF stats */
500 	mbx.reset_stat.msg = NIC_MBOX_MSG_RESET_STAT_COUNTER;
501 	mbx.reset_stat.rx_stat_mask = 0x3FFF;
502 	mbx.reset_stat.tx_stat_mask = 0x1F;
503 	mbx.reset_stat.rq_stat_mask = 0xFFFF;
504 	mbx.reset_stat.sq_stat_mask = 0xFFFF;
505 	nicvf_send_msg_to_pf(nic, &mbx);
506 }
507 
508 /* Configures receive queue */
509 static void nicvf_rcv_queue_config(struct nicvf *nic, struct queue_set *qs,
510 				   int qidx, bool enable)
511 {
512 	union nic_mbx mbx = {};
513 	struct rcv_queue *rq;
514 	struct rq_cfg rq_cfg;
515 
516 	rq = &qs->rq[qidx];
517 	rq->enable = enable;
518 
519 	/* Disable receive queue */
520 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_0_7_CFG, qidx, 0);
521 
522 	if (!rq->enable) {
523 		nicvf_reclaim_rcv_queue(nic, qs, qidx);
524 		return;
525 	}
526 
527 	rq->cq_qs = qs->vnic_id;
528 	rq->cq_idx = qidx;
529 	rq->start_rbdr_qs = qs->vnic_id;
530 	rq->start_qs_rbdr_idx = qs->rbdr_cnt - 1;
531 	rq->cont_rbdr_qs = qs->vnic_id;
532 	rq->cont_qs_rbdr_idx = qs->rbdr_cnt - 1;
533 	/* all writes of RBDR data to be loaded into L2 Cache as well*/
534 	rq->caching = 1;
535 
536 	/* Send a mailbox msg to PF to config RQ */
537 	mbx.rq.msg = NIC_MBOX_MSG_RQ_CFG;
538 	mbx.rq.qs_num = qs->vnic_id;
539 	mbx.rq.rq_num = qidx;
540 	mbx.rq.cfg = (rq->caching << 26) | (rq->cq_qs << 19) |
541 			  (rq->cq_idx << 16) | (rq->cont_rbdr_qs << 9) |
542 			  (rq->cont_qs_rbdr_idx << 8) |
543 			  (rq->start_rbdr_qs << 1) | (rq->start_qs_rbdr_idx);
544 	nicvf_send_msg_to_pf(nic, &mbx);
545 
546 	mbx.rq.msg = NIC_MBOX_MSG_RQ_BP_CFG;
547 	mbx.rq.cfg = BIT_ULL(63) | BIT_ULL(62) |
548 		     (RQ_PASS_RBDR_LVL << 16) | (RQ_PASS_CQ_LVL << 8) |
549 		     (qs->vnic_id << 0);
550 	nicvf_send_msg_to_pf(nic, &mbx);
551 
552 	/* RQ drop config
553 	 * Enable CQ drop to reserve sufficient CQEs for all tx packets
554 	 */
555 	mbx.rq.msg = NIC_MBOX_MSG_RQ_DROP_CFG;
556 	mbx.rq.cfg = BIT_ULL(63) | BIT_ULL(62) |
557 		     (RQ_PASS_RBDR_LVL << 40) | (RQ_DROP_RBDR_LVL << 32) |
558 		     (RQ_PASS_CQ_LVL << 16) | (RQ_DROP_CQ_LVL << 8);
559 	nicvf_send_msg_to_pf(nic, &mbx);
560 
561 	if (!nic->sqs_mode && (qidx == 0)) {
562 		/* Enable checking L3/L4 length and TCP/UDP checksums */
563 		nicvf_queue_reg_write(nic, NIC_QSET_RQ_GEN_CFG, 0,
564 				      (BIT(24) | BIT(23) | BIT(21)));
565 		nicvf_config_vlan_stripping(nic, nic->netdev->features);
566 	}
567 
568 	/* Enable Receive queue */
569 	memset(&rq_cfg, 0, sizeof(struct rq_cfg));
570 	rq_cfg.ena = 1;
571 	rq_cfg.tcp_ena = 0;
572 	nicvf_queue_reg_write(nic, NIC_QSET_RQ_0_7_CFG, qidx, *(u64 *)&rq_cfg);
573 }
574 
575 /* Configures completion queue */
576 void nicvf_cmp_queue_config(struct nicvf *nic, struct queue_set *qs,
577 			    int qidx, bool enable)
578 {
579 	struct cmp_queue *cq;
580 	struct cq_cfg cq_cfg;
581 
582 	cq = &qs->cq[qidx];
583 	cq->enable = enable;
584 
585 	if (!cq->enable) {
586 		nicvf_reclaim_cmp_queue(nic, qs, qidx);
587 		return;
588 	}
589 
590 	/* Reset completion queue */
591 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, NICVF_CQ_RESET);
592 
593 	if (!cq->enable)
594 		return;
595 
596 	spin_lock_init(&cq->lock);
597 	/* Set completion queue base address */
598 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_BASE,
599 			      qidx, (u64)(cq->dmem.phys_base));
600 
601 	/* Enable Completion queue */
602 	memset(&cq_cfg, 0, sizeof(struct cq_cfg));
603 	cq_cfg.ena = 1;
604 	cq_cfg.reset = 0;
605 	cq_cfg.caching = 0;
606 	cq_cfg.qsize = CMP_QSIZE;
607 	cq_cfg.avg_con = 0;
608 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG, qidx, *(u64 *)&cq_cfg);
609 
610 	/* Set threshold value for interrupt generation */
611 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_THRESH, qidx, cq->thresh);
612 	nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_CFG2,
613 			      qidx, CMP_QUEUE_TIMER_THRESH);
614 }
615 
616 /* Configures transmit queue */
617 static void nicvf_snd_queue_config(struct nicvf *nic, struct queue_set *qs,
618 				   int qidx, bool enable)
619 {
620 	union nic_mbx mbx = {};
621 	struct snd_queue *sq;
622 	struct sq_cfg sq_cfg;
623 
624 	sq = &qs->sq[qidx];
625 	sq->enable = enable;
626 
627 	if (!sq->enable) {
628 		nicvf_reclaim_snd_queue(nic, qs, qidx);
629 		return;
630 	}
631 
632 	/* Reset send queue */
633 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, NICVF_SQ_RESET);
634 
635 	sq->cq_qs = qs->vnic_id;
636 	sq->cq_idx = qidx;
637 
638 	/* Send a mailbox msg to PF to config SQ */
639 	mbx.sq.msg = NIC_MBOX_MSG_SQ_CFG;
640 	mbx.sq.qs_num = qs->vnic_id;
641 	mbx.sq.sq_num = qidx;
642 	mbx.sq.sqs_mode = nic->sqs_mode;
643 	mbx.sq.cfg = (sq->cq_qs << 3) | sq->cq_idx;
644 	nicvf_send_msg_to_pf(nic, &mbx);
645 
646 	/* Set queue base address */
647 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_BASE,
648 			      qidx, (u64)(sq->dmem.phys_base));
649 
650 	/* Enable send queue  & set queue size */
651 	memset(&sq_cfg, 0, sizeof(struct sq_cfg));
652 	sq_cfg.ena = 1;
653 	sq_cfg.reset = 0;
654 	sq_cfg.ldwb = 0;
655 	sq_cfg.qsize = SND_QSIZE;
656 	sq_cfg.tstmp_bgx_intf = 0;
657 	sq_cfg.cq_limit = 0;
658 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, *(u64 *)&sq_cfg);
659 
660 	/* Set threshold value for interrupt generation */
661 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_THRESH, qidx, sq->thresh);
662 
663 	/* Set queue:cpu affinity for better load distribution */
664 	if (cpu_online(qidx)) {
665 		cpumask_set_cpu(qidx, &sq->affinity_mask);
666 		netif_set_xps_queue(nic->netdev,
667 				    &sq->affinity_mask, qidx);
668 	}
669 }
670 
671 /* Configures receive buffer descriptor ring */
672 static void nicvf_rbdr_config(struct nicvf *nic, struct queue_set *qs,
673 			      int qidx, bool enable)
674 {
675 	struct rbdr *rbdr;
676 	struct rbdr_cfg rbdr_cfg;
677 
678 	rbdr = &qs->rbdr[qidx];
679 	nicvf_reclaim_rbdr(nic, rbdr, qidx);
680 	if (!enable)
681 		return;
682 
683 	/* Set descriptor base address */
684 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_BASE,
685 			      qidx, (u64)(rbdr->dmem.phys_base));
686 
687 	/* Enable RBDR  & set queue size */
688 	/* Buffer size should be in multiples of 128 bytes */
689 	memset(&rbdr_cfg, 0, sizeof(struct rbdr_cfg));
690 	rbdr_cfg.ena = 1;
691 	rbdr_cfg.reset = 0;
692 	rbdr_cfg.ldwb = 0;
693 	rbdr_cfg.qsize = RBDR_SIZE;
694 	rbdr_cfg.avg_con = 0;
695 	rbdr_cfg.lines = rbdr->dma_size / 128;
696 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_CFG,
697 			      qidx, *(u64 *)&rbdr_cfg);
698 
699 	/* Notify HW */
700 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_DOOR,
701 			      qidx, qs->rbdr_len - 1);
702 
703 	/* Set threshold value for interrupt generation */
704 	nicvf_queue_reg_write(nic, NIC_QSET_RBDR_0_1_THRESH,
705 			      qidx, rbdr->thresh - 1);
706 }
707 
708 /* Requests PF to assign and enable Qset */
709 void nicvf_qset_config(struct nicvf *nic, bool enable)
710 {
711 	union nic_mbx mbx = {};
712 	struct queue_set *qs = nic->qs;
713 	struct qs_cfg *qs_cfg;
714 
715 	if (!qs) {
716 		netdev_warn(nic->netdev,
717 			    "Qset is still not allocated, don't init queues\n");
718 		return;
719 	}
720 
721 	qs->enable = enable;
722 	qs->vnic_id = nic->vf_id;
723 
724 	/* Send a mailbox msg to PF to config Qset */
725 	mbx.qs.msg = NIC_MBOX_MSG_QS_CFG;
726 	mbx.qs.num = qs->vnic_id;
727 	mbx.qs.sqs_count = nic->sqs_count;
728 
729 	mbx.qs.cfg = 0;
730 	qs_cfg = (struct qs_cfg *)&mbx.qs.cfg;
731 	if (qs->enable) {
732 		qs_cfg->ena = 1;
733 #ifdef __BIG_ENDIAN
734 		qs_cfg->be = 1;
735 #endif
736 		qs_cfg->vnic = qs->vnic_id;
737 	}
738 	nicvf_send_msg_to_pf(nic, &mbx);
739 }
740 
741 static void nicvf_free_resources(struct nicvf *nic)
742 {
743 	int qidx;
744 	struct queue_set *qs = nic->qs;
745 
746 	/* Free receive buffer descriptor ring */
747 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
748 		nicvf_free_rbdr(nic, &qs->rbdr[qidx]);
749 
750 	/* Free completion queue */
751 	for (qidx = 0; qidx < qs->cq_cnt; qidx++)
752 		nicvf_free_cmp_queue(nic, &qs->cq[qidx]);
753 
754 	/* Free send queue */
755 	for (qidx = 0; qidx < qs->sq_cnt; qidx++)
756 		nicvf_free_snd_queue(nic, &qs->sq[qidx]);
757 }
758 
759 static int nicvf_alloc_resources(struct nicvf *nic)
760 {
761 	int qidx;
762 	struct queue_set *qs = nic->qs;
763 
764 	/* Alloc receive buffer descriptor ring */
765 	for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) {
766 		if (nicvf_init_rbdr(nic, &qs->rbdr[qidx], qs->rbdr_len,
767 				    DMA_BUFFER_LEN))
768 			goto alloc_fail;
769 	}
770 
771 	/* Alloc send queue */
772 	for (qidx = 0; qidx < qs->sq_cnt; qidx++) {
773 		if (nicvf_init_snd_queue(nic, &qs->sq[qidx], qs->sq_len))
774 			goto alloc_fail;
775 	}
776 
777 	/* Alloc completion queue */
778 	for (qidx = 0; qidx < qs->cq_cnt; qidx++) {
779 		if (nicvf_init_cmp_queue(nic, &qs->cq[qidx], qs->cq_len))
780 			goto alloc_fail;
781 	}
782 
783 	return 0;
784 alloc_fail:
785 	nicvf_free_resources(nic);
786 	return -ENOMEM;
787 }
788 
789 int nicvf_set_qset_resources(struct nicvf *nic)
790 {
791 	struct queue_set *qs;
792 
793 	qs = devm_kzalloc(&nic->pdev->dev, sizeof(*qs), GFP_KERNEL);
794 	if (!qs)
795 		return -ENOMEM;
796 	nic->qs = qs;
797 
798 	/* Set count of each queue */
799 	qs->rbdr_cnt = DEFAULT_RBDR_CNT;
800 	qs->rq_cnt = min_t(u8, MAX_RCV_QUEUES_PER_QS, num_online_cpus());
801 	qs->sq_cnt = min_t(u8, MAX_SND_QUEUES_PER_QS, num_online_cpus());
802 	qs->cq_cnt = max_t(u8, qs->rq_cnt, qs->sq_cnt);
803 
804 	/* Set queue lengths */
805 	qs->rbdr_len = RCV_BUF_COUNT;
806 	qs->sq_len = SND_QUEUE_LEN;
807 	qs->cq_len = CMP_QUEUE_LEN;
808 
809 	nic->rx_queues = qs->rq_cnt;
810 	nic->tx_queues = qs->sq_cnt;
811 
812 	return 0;
813 }
814 
815 int nicvf_config_data_transfer(struct nicvf *nic, bool enable)
816 {
817 	bool disable = false;
818 	struct queue_set *qs = nic->qs;
819 	int qidx;
820 
821 	if (!qs)
822 		return 0;
823 
824 	if (enable) {
825 		if (nicvf_alloc_resources(nic))
826 			return -ENOMEM;
827 
828 		for (qidx = 0; qidx < qs->sq_cnt; qidx++)
829 			nicvf_snd_queue_config(nic, qs, qidx, enable);
830 		for (qidx = 0; qidx < qs->cq_cnt; qidx++)
831 			nicvf_cmp_queue_config(nic, qs, qidx, enable);
832 		for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
833 			nicvf_rbdr_config(nic, qs, qidx, enable);
834 		for (qidx = 0; qidx < qs->rq_cnt; qidx++)
835 			nicvf_rcv_queue_config(nic, qs, qidx, enable);
836 	} else {
837 		for (qidx = 0; qidx < qs->rq_cnt; qidx++)
838 			nicvf_rcv_queue_config(nic, qs, qidx, disable);
839 		for (qidx = 0; qidx < qs->rbdr_cnt; qidx++)
840 			nicvf_rbdr_config(nic, qs, qidx, disable);
841 		for (qidx = 0; qidx < qs->sq_cnt; qidx++)
842 			nicvf_snd_queue_config(nic, qs, qidx, disable);
843 		for (qidx = 0; qidx < qs->cq_cnt; qidx++)
844 			nicvf_cmp_queue_config(nic, qs, qidx, disable);
845 
846 		nicvf_free_resources(nic);
847 	}
848 
849 	/* Reset RXQ's stats.
850 	 * SQ's stats will get reset automatically once SQ is reset.
851 	 */
852 	nicvf_reset_rcv_queue_stats(nic);
853 
854 	return 0;
855 }
856 
857 /* Get a free desc from SQ
858  * returns descriptor ponter & descriptor number
859  */
860 static inline int nicvf_get_sq_desc(struct snd_queue *sq, int desc_cnt)
861 {
862 	int qentry;
863 
864 	qentry = sq->tail;
865 	atomic_sub(desc_cnt, &sq->free_cnt);
866 	sq->tail += desc_cnt;
867 	sq->tail &= (sq->dmem.q_len - 1);
868 
869 	return qentry;
870 }
871 
872 /* Free descriptor back to SQ for future use */
873 void nicvf_put_sq_desc(struct snd_queue *sq, int desc_cnt)
874 {
875 	atomic_add(desc_cnt, &sq->free_cnt);
876 	sq->head += desc_cnt;
877 	sq->head &= (sq->dmem.q_len - 1);
878 }
879 
880 static inline int nicvf_get_nxt_sqentry(struct snd_queue *sq, int qentry)
881 {
882 	qentry++;
883 	qentry &= (sq->dmem.q_len - 1);
884 	return qentry;
885 }
886 
887 void nicvf_sq_enable(struct nicvf *nic, struct snd_queue *sq, int qidx)
888 {
889 	u64 sq_cfg;
890 
891 	sq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_CFG, qidx);
892 	sq_cfg |= NICVF_SQ_EN;
893 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, sq_cfg);
894 	/* Ring doorbell so that H/W restarts processing SQEs */
895 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_DOOR, qidx, 0);
896 }
897 
898 void nicvf_sq_disable(struct nicvf *nic, int qidx)
899 {
900 	u64 sq_cfg;
901 
902 	sq_cfg = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_CFG, qidx);
903 	sq_cfg &= ~NICVF_SQ_EN;
904 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_CFG, qidx, sq_cfg);
905 }
906 
907 void nicvf_sq_free_used_descs(struct net_device *netdev, struct snd_queue *sq,
908 			      int qidx)
909 {
910 	u64 head, tail;
911 	struct sk_buff *skb;
912 	struct nicvf *nic = netdev_priv(netdev);
913 	struct sq_hdr_subdesc *hdr;
914 
915 	head = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_HEAD, qidx) >> 4;
916 	tail = nicvf_queue_reg_read(nic, NIC_QSET_SQ_0_7_TAIL, qidx) >> 4;
917 	while (sq->head != head) {
918 		hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, sq->head);
919 		if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER) {
920 			nicvf_put_sq_desc(sq, 1);
921 			continue;
922 		}
923 		skb = (struct sk_buff *)sq->skbuff[sq->head];
924 		if (skb)
925 			dev_kfree_skb_any(skb);
926 		atomic64_add(1, (atomic64_t *)&netdev->stats.tx_packets);
927 		atomic64_add(hdr->tot_len,
928 			     (atomic64_t *)&netdev->stats.tx_bytes);
929 		nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1);
930 	}
931 }
932 
933 /* Calculate no of SQ subdescriptors needed to transmit all
934  * segments of this TSO packet.
935  * Taken from 'Tilera network driver' with a minor modification.
936  */
937 static int nicvf_tso_count_subdescs(struct sk_buff *skb)
938 {
939 	struct skb_shared_info *sh = skb_shinfo(skb);
940 	unsigned int sh_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
941 	unsigned int data_len = skb->len - sh_len;
942 	unsigned int p_len = sh->gso_size;
943 	long f_id = -1;    /* id of the current fragment */
944 	long f_size = skb_headlen(skb) - sh_len;  /* current fragment size */
945 	long f_used = 0;  /* bytes used from the current fragment */
946 	long n;            /* size of the current piece of payload */
947 	int num_edescs = 0;
948 	int segment;
949 
950 	for (segment = 0; segment < sh->gso_segs; segment++) {
951 		unsigned int p_used = 0;
952 
953 		/* One edesc for header and for each piece of the payload. */
954 		for (num_edescs++; p_used < p_len; num_edescs++) {
955 			/* Advance as needed. */
956 			while (f_used >= f_size) {
957 				f_id++;
958 				f_size = skb_frag_size(&sh->frags[f_id]);
959 				f_used = 0;
960 			}
961 
962 			/* Use bytes from the current fragment. */
963 			n = p_len - p_used;
964 			if (n > f_size - f_used)
965 				n = f_size - f_used;
966 			f_used += n;
967 			p_used += n;
968 		}
969 
970 		/* The last segment may be less than gso_size. */
971 		data_len -= p_len;
972 		if (data_len < p_len)
973 			p_len = data_len;
974 	}
975 
976 	/* '+ gso_segs' for SQ_HDR_SUDESCs for each segment */
977 	return num_edescs + sh->gso_segs;
978 }
979 
980 #define POST_CQE_DESC_COUNT 2
981 
982 /* Get the number of SQ descriptors needed to xmit this skb */
983 static int nicvf_sq_subdesc_required(struct nicvf *nic, struct sk_buff *skb)
984 {
985 	int subdesc_cnt = MIN_SQ_DESC_PER_PKT_XMIT;
986 
987 	if (skb_shinfo(skb)->gso_size && !nic->hw_tso) {
988 		subdesc_cnt = nicvf_tso_count_subdescs(skb);
989 		return subdesc_cnt;
990 	}
991 
992 	/* Dummy descriptors to get TSO pkt completion notification */
993 	if (nic->t88 && nic->hw_tso && skb_shinfo(skb)->gso_size)
994 		subdesc_cnt += POST_CQE_DESC_COUNT;
995 
996 	if (skb_shinfo(skb)->nr_frags)
997 		subdesc_cnt += skb_shinfo(skb)->nr_frags;
998 
999 	return subdesc_cnt;
1000 }
1001 
1002 /* Add SQ HEADER subdescriptor.
1003  * First subdescriptor for every send descriptor.
1004  */
1005 static inline void
1006 nicvf_sq_add_hdr_subdesc(struct nicvf *nic, struct snd_queue *sq, int qentry,
1007 			 int subdesc_cnt, struct sk_buff *skb, int len)
1008 {
1009 	int proto;
1010 	struct sq_hdr_subdesc *hdr;
1011 
1012 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, qentry);
1013 	memset(hdr, 0, SND_QUEUE_DESC_SIZE);
1014 	hdr->subdesc_type = SQ_DESC_TYPE_HEADER;
1015 
1016 	if (nic->t88 && nic->hw_tso && skb_shinfo(skb)->gso_size) {
1017 		/* post_cqe = 0, to avoid HW posting a CQE for every TSO
1018 		 * segment transmitted on 88xx.
1019 		 */
1020 		hdr->subdesc_cnt = subdesc_cnt - POST_CQE_DESC_COUNT;
1021 	} else {
1022 		sq->skbuff[qentry] = (u64)skb;
1023 		/* Enable notification via CQE after processing SQE */
1024 		hdr->post_cqe = 1;
1025 		/* No of subdescriptors following this */
1026 		hdr->subdesc_cnt = subdesc_cnt;
1027 	}
1028 	hdr->tot_len = len;
1029 
1030 	/* Offload checksum calculation to HW */
1031 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1032 		hdr->csum_l3 = 1; /* Enable IP csum calculation */
1033 		hdr->l3_offset = skb_network_offset(skb);
1034 		hdr->l4_offset = skb_transport_offset(skb);
1035 
1036 		proto = ip_hdr(skb)->protocol;
1037 		switch (proto) {
1038 		case IPPROTO_TCP:
1039 			hdr->csum_l4 = SEND_L4_CSUM_TCP;
1040 			break;
1041 		case IPPROTO_UDP:
1042 			hdr->csum_l4 = SEND_L4_CSUM_UDP;
1043 			break;
1044 		case IPPROTO_SCTP:
1045 			hdr->csum_l4 = SEND_L4_CSUM_SCTP;
1046 			break;
1047 		}
1048 	}
1049 
1050 	if (nic->hw_tso && skb_shinfo(skb)->gso_size) {
1051 		hdr->tso = 1;
1052 		hdr->tso_start = skb_transport_offset(skb) + tcp_hdrlen(skb);
1053 		hdr->tso_max_paysize = skb_shinfo(skb)->gso_size;
1054 		/* For non-tunneled pkts, point this to L2 ethertype */
1055 		hdr->inner_l3_offset = skb_network_offset(skb) - 2;
1056 		this_cpu_inc(nic->pnicvf->drv_stats->tx_tso);
1057 	}
1058 }
1059 
1060 /* SQ GATHER subdescriptor
1061  * Must follow HDR descriptor
1062  */
1063 static inline void nicvf_sq_add_gather_subdesc(struct snd_queue *sq, int qentry,
1064 					       int size, u64 data)
1065 {
1066 	struct sq_gather_subdesc *gather;
1067 
1068 	qentry &= (sq->dmem.q_len - 1);
1069 	gather = (struct sq_gather_subdesc *)GET_SQ_DESC(sq, qentry);
1070 
1071 	memset(gather, 0, SND_QUEUE_DESC_SIZE);
1072 	gather->subdesc_type = SQ_DESC_TYPE_GATHER;
1073 	gather->ld_type = NIC_SEND_LD_TYPE_E_LDD;
1074 	gather->size = size;
1075 	gather->addr = data;
1076 }
1077 
1078 /* Add HDR + IMMEDIATE subdescriptors right after descriptors of a TSO
1079  * packet so that a CQE is posted as a notifation for transmission of
1080  * TSO packet.
1081  */
1082 static inline void nicvf_sq_add_cqe_subdesc(struct snd_queue *sq, int qentry,
1083 					    int tso_sqe, struct sk_buff *skb)
1084 {
1085 	struct sq_imm_subdesc *imm;
1086 	struct sq_hdr_subdesc *hdr;
1087 
1088 	sq->skbuff[qentry] = (u64)skb;
1089 
1090 	hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, qentry);
1091 	memset(hdr, 0, SND_QUEUE_DESC_SIZE);
1092 	hdr->subdesc_type = SQ_DESC_TYPE_HEADER;
1093 	/* Enable notification via CQE after processing SQE */
1094 	hdr->post_cqe = 1;
1095 	/* There is no packet to transmit here */
1096 	hdr->dont_send = 1;
1097 	hdr->subdesc_cnt = POST_CQE_DESC_COUNT - 1;
1098 	hdr->tot_len = 1;
1099 	/* Actual TSO header SQE index, needed for cleanup */
1100 	hdr->rsvd2 = tso_sqe;
1101 
1102 	qentry = nicvf_get_nxt_sqentry(sq, qentry);
1103 	imm = (struct sq_imm_subdesc *)GET_SQ_DESC(sq, qentry);
1104 	memset(imm, 0, SND_QUEUE_DESC_SIZE);
1105 	imm->subdesc_type = SQ_DESC_TYPE_IMMEDIATE;
1106 	imm->len = 1;
1107 }
1108 
1109 static inline void nicvf_sq_doorbell(struct nicvf *nic, struct sk_buff *skb,
1110 				     int sq_num, int desc_cnt)
1111 {
1112 	struct netdev_queue *txq;
1113 
1114 	txq = netdev_get_tx_queue(nic->pnicvf->netdev,
1115 				  skb_get_queue_mapping(skb));
1116 
1117 	netdev_tx_sent_queue(txq, skb->len);
1118 
1119 	/* make sure all memory stores are done before ringing doorbell */
1120 	smp_wmb();
1121 
1122 	/* Inform HW to xmit all TSO segments */
1123 	nicvf_queue_reg_write(nic, NIC_QSET_SQ_0_7_DOOR,
1124 			      sq_num, desc_cnt);
1125 }
1126 
1127 /* Segment a TSO packet into 'gso_size' segments and append
1128  * them to SQ for transfer
1129  */
1130 static int nicvf_sq_append_tso(struct nicvf *nic, struct snd_queue *sq,
1131 			       int sq_num, int qentry, struct sk_buff *skb)
1132 {
1133 	struct tso_t tso;
1134 	int seg_subdescs = 0, desc_cnt = 0;
1135 	int seg_len, total_len, data_left;
1136 	int hdr_qentry = qentry;
1137 	int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1138 
1139 	tso_start(skb, &tso);
1140 	total_len = skb->len - hdr_len;
1141 	while (total_len > 0) {
1142 		char *hdr;
1143 
1144 		/* Save Qentry for adding HDR_SUBDESC at the end */
1145 		hdr_qentry = qentry;
1146 
1147 		data_left = min_t(int, skb_shinfo(skb)->gso_size, total_len);
1148 		total_len -= data_left;
1149 
1150 		/* Add segment's header */
1151 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1152 		hdr = sq->tso_hdrs + qentry * TSO_HEADER_SIZE;
1153 		tso_build_hdr(skb, hdr, &tso, data_left, total_len == 0);
1154 		nicvf_sq_add_gather_subdesc(sq, qentry, hdr_len,
1155 					    sq->tso_hdrs_phys +
1156 					    qentry * TSO_HEADER_SIZE);
1157 		/* HDR_SUDESC + GATHER */
1158 		seg_subdescs = 2;
1159 		seg_len = hdr_len;
1160 
1161 		/* Add segment's payload fragments */
1162 		while (data_left > 0) {
1163 			int size;
1164 
1165 			size = min_t(int, tso.size, data_left);
1166 
1167 			qentry = nicvf_get_nxt_sqentry(sq, qentry);
1168 			nicvf_sq_add_gather_subdesc(sq, qentry, size,
1169 						    virt_to_phys(tso.data));
1170 			seg_subdescs++;
1171 			seg_len += size;
1172 
1173 			data_left -= size;
1174 			tso_build_data(skb, &tso, size);
1175 		}
1176 		nicvf_sq_add_hdr_subdesc(nic, sq, hdr_qentry,
1177 					 seg_subdescs - 1, skb, seg_len);
1178 		sq->skbuff[hdr_qentry] = (u64)NULL;
1179 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1180 
1181 		desc_cnt += seg_subdescs;
1182 	}
1183 	/* Save SKB in the last segment for freeing */
1184 	sq->skbuff[hdr_qentry] = (u64)skb;
1185 
1186 	nicvf_sq_doorbell(nic, skb, sq_num, desc_cnt);
1187 
1188 	this_cpu_inc(nic->pnicvf->drv_stats->tx_tso);
1189 	return 1;
1190 }
1191 
1192 /* Append an skb to a SQ for packet transfer. */
1193 int nicvf_sq_append_skb(struct nicvf *nic, struct snd_queue *sq,
1194 			struct sk_buff *skb, u8 sq_num)
1195 {
1196 	int i, size;
1197 	int subdesc_cnt, tso_sqe = 0;
1198 	int qentry;
1199 
1200 	subdesc_cnt = nicvf_sq_subdesc_required(nic, skb);
1201 	if (subdesc_cnt > atomic_read(&sq->free_cnt))
1202 		goto append_fail;
1203 
1204 	qentry = nicvf_get_sq_desc(sq, subdesc_cnt);
1205 
1206 	/* Check if its a TSO packet */
1207 	if (skb_shinfo(skb)->gso_size && !nic->hw_tso)
1208 		return nicvf_sq_append_tso(nic, sq, sq_num, qentry, skb);
1209 
1210 	/* Add SQ header subdesc */
1211 	nicvf_sq_add_hdr_subdesc(nic, sq, qentry, subdesc_cnt - 1,
1212 				 skb, skb->len);
1213 	tso_sqe = qentry;
1214 
1215 	/* Add SQ gather subdescs */
1216 	qentry = nicvf_get_nxt_sqentry(sq, qentry);
1217 	size = skb_is_nonlinear(skb) ? skb_headlen(skb) : skb->len;
1218 	nicvf_sq_add_gather_subdesc(sq, qentry, size, virt_to_phys(skb->data));
1219 
1220 	/* Check for scattered buffer */
1221 	if (!skb_is_nonlinear(skb))
1222 		goto doorbell;
1223 
1224 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1225 		const struct skb_frag_struct *frag;
1226 
1227 		frag = &skb_shinfo(skb)->frags[i];
1228 
1229 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1230 		size = skb_frag_size(frag);
1231 		nicvf_sq_add_gather_subdesc(sq, qentry, size,
1232 					    virt_to_phys(
1233 					    skb_frag_address(frag)));
1234 	}
1235 
1236 doorbell:
1237 	if (nic->t88 && skb_shinfo(skb)->gso_size) {
1238 		qentry = nicvf_get_nxt_sqentry(sq, qentry);
1239 		nicvf_sq_add_cqe_subdesc(sq, qentry, tso_sqe, skb);
1240 	}
1241 
1242 	nicvf_sq_doorbell(nic, skb, sq_num, subdesc_cnt);
1243 
1244 	return 1;
1245 
1246 append_fail:
1247 	/* Use original PCI dev for debug log */
1248 	nic = nic->pnicvf;
1249 	netdev_dbg(nic->netdev, "Not enough SQ descriptors to xmit pkt\n");
1250 	return 0;
1251 }
1252 
1253 static inline unsigned frag_num(unsigned i)
1254 {
1255 #ifdef __BIG_ENDIAN
1256 	return (i & ~3) + 3 - (i & 3);
1257 #else
1258 	return i;
1259 #endif
1260 }
1261 
1262 /* Returns SKB for a received packet */
1263 struct sk_buff *nicvf_get_rcv_skb(struct nicvf *nic, struct cqe_rx_t *cqe_rx)
1264 {
1265 	int frag;
1266 	int payload_len = 0;
1267 	struct sk_buff *skb = NULL;
1268 	struct page *page;
1269 	int offset;
1270 	u16 *rb_lens = NULL;
1271 	u64 *rb_ptrs = NULL;
1272 
1273 	rb_lens = (void *)cqe_rx + (3 * sizeof(u64));
1274 	/* Except 88xx pass1 on all other chips CQE_RX2_S is added to
1275 	 * CQE_RX at word6, hence buffer pointers move by word
1276 	 *
1277 	 * Use existing 'hw_tso' flag which will be set for all chips
1278 	 * except 88xx pass1 instead of a additional cache line
1279 	 * access (or miss) by using pci dev's revision.
1280 	 */
1281 	if (!nic->hw_tso)
1282 		rb_ptrs = (void *)cqe_rx + (6 * sizeof(u64));
1283 	else
1284 		rb_ptrs = (void *)cqe_rx + (7 * sizeof(u64));
1285 
1286 	netdev_dbg(nic->netdev, "%s rb_cnt %d rb0_ptr %llx rb0_sz %d\n",
1287 		   __func__, cqe_rx->rb_cnt, cqe_rx->rb0_ptr, cqe_rx->rb0_sz);
1288 
1289 	for (frag = 0; frag < cqe_rx->rb_cnt; frag++) {
1290 		payload_len = rb_lens[frag_num(frag)];
1291 		if (!frag) {
1292 			/* First fragment */
1293 			skb = nicvf_rb_ptr_to_skb(nic,
1294 						  *rb_ptrs - cqe_rx->align_pad,
1295 						  payload_len);
1296 			if (!skb)
1297 				return NULL;
1298 			skb_reserve(skb, cqe_rx->align_pad);
1299 			skb_put(skb, payload_len);
1300 		} else {
1301 			/* Add fragments */
1302 			page = virt_to_page(phys_to_virt(*rb_ptrs));
1303 			offset = phys_to_virt(*rb_ptrs) - page_address(page);
1304 			skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1305 					offset, payload_len, RCV_FRAG_LEN);
1306 		}
1307 		/* Next buffer pointer */
1308 		rb_ptrs++;
1309 	}
1310 	return skb;
1311 }
1312 
1313 static u64 nicvf_int_type_to_mask(int int_type, int q_idx)
1314 {
1315 	u64 reg_val;
1316 
1317 	switch (int_type) {
1318 	case NICVF_INTR_CQ:
1319 		reg_val = ((1ULL << q_idx) << NICVF_INTR_CQ_SHIFT);
1320 		break;
1321 	case NICVF_INTR_SQ:
1322 		reg_val = ((1ULL << q_idx) << NICVF_INTR_SQ_SHIFT);
1323 		break;
1324 	case NICVF_INTR_RBDR:
1325 		reg_val = ((1ULL << q_idx) << NICVF_INTR_RBDR_SHIFT);
1326 		break;
1327 	case NICVF_INTR_PKT_DROP:
1328 		reg_val = (1ULL << NICVF_INTR_PKT_DROP_SHIFT);
1329 		break;
1330 	case NICVF_INTR_TCP_TIMER:
1331 		reg_val = (1ULL << NICVF_INTR_TCP_TIMER_SHIFT);
1332 		break;
1333 	case NICVF_INTR_MBOX:
1334 		reg_val = (1ULL << NICVF_INTR_MBOX_SHIFT);
1335 		break;
1336 	case NICVF_INTR_QS_ERR:
1337 		reg_val = (1ULL << NICVF_INTR_QS_ERR_SHIFT);
1338 		break;
1339 	default:
1340 		reg_val = 0;
1341 	}
1342 
1343 	return reg_val;
1344 }
1345 
1346 /* Enable interrupt */
1347 void nicvf_enable_intr(struct nicvf *nic, int int_type, int q_idx)
1348 {
1349 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1350 
1351 	if (!mask) {
1352 		netdev_dbg(nic->netdev,
1353 			   "Failed to enable interrupt: unknown type\n");
1354 		return;
1355 	}
1356 	nicvf_reg_write(nic, NIC_VF_ENA_W1S,
1357 			nicvf_reg_read(nic, NIC_VF_ENA_W1S) | mask);
1358 }
1359 
1360 /* Disable interrupt */
1361 void nicvf_disable_intr(struct nicvf *nic, int int_type, int q_idx)
1362 {
1363 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1364 
1365 	if (!mask) {
1366 		netdev_dbg(nic->netdev,
1367 			   "Failed to disable interrupt: unknown type\n");
1368 		return;
1369 	}
1370 
1371 	nicvf_reg_write(nic, NIC_VF_ENA_W1C, mask);
1372 }
1373 
1374 /* Clear interrupt */
1375 void nicvf_clear_intr(struct nicvf *nic, int int_type, int q_idx)
1376 {
1377 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1378 
1379 	if (!mask) {
1380 		netdev_dbg(nic->netdev,
1381 			   "Failed to clear interrupt: unknown type\n");
1382 		return;
1383 	}
1384 
1385 	nicvf_reg_write(nic, NIC_VF_INT, mask);
1386 }
1387 
1388 /* Check if interrupt is enabled */
1389 int nicvf_is_intr_enabled(struct nicvf *nic, int int_type, int q_idx)
1390 {
1391 	u64 mask = nicvf_int_type_to_mask(int_type, q_idx);
1392 	/* If interrupt type is unknown, we treat it disabled. */
1393 	if (!mask) {
1394 		netdev_dbg(nic->netdev,
1395 			   "Failed to check interrupt enable: unknown type\n");
1396 		return 0;
1397 	}
1398 
1399 	return mask & nicvf_reg_read(nic, NIC_VF_ENA_W1S);
1400 }
1401 
1402 void nicvf_update_rq_stats(struct nicvf *nic, int rq_idx)
1403 {
1404 	struct rcv_queue *rq;
1405 
1406 #define GET_RQ_STATS(reg) \
1407 	nicvf_reg_read(nic, NIC_QSET_RQ_0_7_STAT_0_1 |\
1408 			    (rq_idx << NIC_Q_NUM_SHIFT) | (reg << 3))
1409 
1410 	rq = &nic->qs->rq[rq_idx];
1411 	rq->stats.bytes = GET_RQ_STATS(RQ_SQ_STATS_OCTS);
1412 	rq->stats.pkts = GET_RQ_STATS(RQ_SQ_STATS_PKTS);
1413 }
1414 
1415 void nicvf_update_sq_stats(struct nicvf *nic, int sq_idx)
1416 {
1417 	struct snd_queue *sq;
1418 
1419 #define GET_SQ_STATS(reg) \
1420 	nicvf_reg_read(nic, NIC_QSET_SQ_0_7_STAT_0_1 |\
1421 			    (sq_idx << NIC_Q_NUM_SHIFT) | (reg << 3))
1422 
1423 	sq = &nic->qs->sq[sq_idx];
1424 	sq->stats.bytes = GET_SQ_STATS(RQ_SQ_STATS_OCTS);
1425 	sq->stats.pkts = GET_SQ_STATS(RQ_SQ_STATS_PKTS);
1426 }
1427 
1428 /* Check for errors in the receive cmp.queue entry */
1429 int nicvf_check_cqe_rx_errs(struct nicvf *nic, struct cqe_rx_t *cqe_rx)
1430 {
1431 	if (!cqe_rx->err_level && !cqe_rx->err_opcode)
1432 		return 0;
1433 
1434 	if (netif_msg_rx_err(nic))
1435 		netdev_err(nic->netdev,
1436 			   "%s: RX error CQE err_level 0x%x err_opcode 0x%x\n",
1437 			   nic->netdev->name,
1438 			   cqe_rx->err_level, cqe_rx->err_opcode);
1439 
1440 	switch (cqe_rx->err_opcode) {
1441 	case CQ_RX_ERROP_RE_PARTIAL:
1442 		this_cpu_inc(nic->drv_stats->rx_bgx_truncated_pkts);
1443 		break;
1444 	case CQ_RX_ERROP_RE_JABBER:
1445 		this_cpu_inc(nic->drv_stats->rx_jabber_errs);
1446 		break;
1447 	case CQ_RX_ERROP_RE_FCS:
1448 		this_cpu_inc(nic->drv_stats->rx_fcs_errs);
1449 		break;
1450 	case CQ_RX_ERROP_RE_RX_CTL:
1451 		this_cpu_inc(nic->drv_stats->rx_bgx_errs);
1452 		break;
1453 	case CQ_RX_ERROP_PREL2_ERR:
1454 		this_cpu_inc(nic->drv_stats->rx_prel2_errs);
1455 		break;
1456 	case CQ_RX_ERROP_L2_MAL:
1457 		this_cpu_inc(nic->drv_stats->rx_l2_hdr_malformed);
1458 		break;
1459 	case CQ_RX_ERROP_L2_OVERSIZE:
1460 		this_cpu_inc(nic->drv_stats->rx_oversize);
1461 		break;
1462 	case CQ_RX_ERROP_L2_UNDERSIZE:
1463 		this_cpu_inc(nic->drv_stats->rx_undersize);
1464 		break;
1465 	case CQ_RX_ERROP_L2_LENMISM:
1466 		this_cpu_inc(nic->drv_stats->rx_l2_len_mismatch);
1467 		break;
1468 	case CQ_RX_ERROP_L2_PCLP:
1469 		this_cpu_inc(nic->drv_stats->rx_l2_pclp);
1470 		break;
1471 	case CQ_RX_ERROP_IP_NOT:
1472 		this_cpu_inc(nic->drv_stats->rx_ip_ver_errs);
1473 		break;
1474 	case CQ_RX_ERROP_IP_CSUM_ERR:
1475 		this_cpu_inc(nic->drv_stats->rx_ip_csum_errs);
1476 		break;
1477 	case CQ_RX_ERROP_IP_MAL:
1478 		this_cpu_inc(nic->drv_stats->rx_ip_hdr_malformed);
1479 		break;
1480 	case CQ_RX_ERROP_IP_MALD:
1481 		this_cpu_inc(nic->drv_stats->rx_ip_payload_malformed);
1482 		break;
1483 	case CQ_RX_ERROP_IP_HOP:
1484 		this_cpu_inc(nic->drv_stats->rx_ip_ttl_errs);
1485 		break;
1486 	case CQ_RX_ERROP_L3_PCLP:
1487 		this_cpu_inc(nic->drv_stats->rx_l3_pclp);
1488 		break;
1489 	case CQ_RX_ERROP_L4_MAL:
1490 		this_cpu_inc(nic->drv_stats->rx_l4_malformed);
1491 		break;
1492 	case CQ_RX_ERROP_L4_CHK:
1493 		this_cpu_inc(nic->drv_stats->rx_l4_csum_errs);
1494 		break;
1495 	case CQ_RX_ERROP_UDP_LEN:
1496 		this_cpu_inc(nic->drv_stats->rx_udp_len_errs);
1497 		break;
1498 	case CQ_RX_ERROP_L4_PORT:
1499 		this_cpu_inc(nic->drv_stats->rx_l4_port_errs);
1500 		break;
1501 	case CQ_RX_ERROP_TCP_FLAG:
1502 		this_cpu_inc(nic->drv_stats->rx_tcp_flag_errs);
1503 		break;
1504 	case CQ_RX_ERROP_TCP_OFFSET:
1505 		this_cpu_inc(nic->drv_stats->rx_tcp_offset_errs);
1506 		break;
1507 	case CQ_RX_ERROP_L4_PCLP:
1508 		this_cpu_inc(nic->drv_stats->rx_l4_pclp);
1509 		break;
1510 	case CQ_RX_ERROP_RBDR_TRUNC:
1511 		this_cpu_inc(nic->drv_stats->rx_truncated_pkts);
1512 		break;
1513 	}
1514 
1515 	return 1;
1516 }
1517 
1518 /* Check for errors in the send cmp.queue entry */
1519 int nicvf_check_cqe_tx_errs(struct nicvf *nic, struct cqe_send_t *cqe_tx)
1520 {
1521 	switch (cqe_tx->send_status) {
1522 	case CQ_TX_ERROP_GOOD:
1523 		return 0;
1524 	case CQ_TX_ERROP_DESC_FAULT:
1525 		this_cpu_inc(nic->drv_stats->tx_desc_fault);
1526 		break;
1527 	case CQ_TX_ERROP_HDR_CONS_ERR:
1528 		this_cpu_inc(nic->drv_stats->tx_hdr_cons_err);
1529 		break;
1530 	case CQ_TX_ERROP_SUBDC_ERR:
1531 		this_cpu_inc(nic->drv_stats->tx_subdesc_err);
1532 		break;
1533 	case CQ_TX_ERROP_MAX_SIZE_VIOL:
1534 		this_cpu_inc(nic->drv_stats->tx_max_size_exceeded);
1535 		break;
1536 	case CQ_TX_ERROP_IMM_SIZE_OFLOW:
1537 		this_cpu_inc(nic->drv_stats->tx_imm_size_oflow);
1538 		break;
1539 	case CQ_TX_ERROP_DATA_SEQUENCE_ERR:
1540 		this_cpu_inc(nic->drv_stats->tx_data_seq_err);
1541 		break;
1542 	case CQ_TX_ERROP_MEM_SEQUENCE_ERR:
1543 		this_cpu_inc(nic->drv_stats->tx_mem_seq_err);
1544 		break;
1545 	case CQ_TX_ERROP_LOCK_VIOL:
1546 		this_cpu_inc(nic->drv_stats->tx_lock_viol);
1547 		break;
1548 	case CQ_TX_ERROP_DATA_FAULT:
1549 		this_cpu_inc(nic->drv_stats->tx_data_fault);
1550 		break;
1551 	case CQ_TX_ERROP_TSTMP_CONFLICT:
1552 		this_cpu_inc(nic->drv_stats->tx_tstmp_conflict);
1553 		break;
1554 	case CQ_TX_ERROP_TSTMP_TIMEOUT:
1555 		this_cpu_inc(nic->drv_stats->tx_tstmp_timeout);
1556 		break;
1557 	case CQ_TX_ERROP_MEM_FAULT:
1558 		this_cpu_inc(nic->drv_stats->tx_mem_fault);
1559 		break;
1560 	case CQ_TX_ERROP_CK_OVERLAP:
1561 		this_cpu_inc(nic->drv_stats->tx_csum_overlap);
1562 		break;
1563 	case CQ_TX_ERROP_CK_OFLOW:
1564 		this_cpu_inc(nic->drv_stats->tx_csum_overflow);
1565 		break;
1566 	}
1567 
1568 	return 1;
1569 }
1570