1 /**********************************************************************
2 * Author: Cavium, Inc.
3 *
4 * Contact: support@cavium.com
5 *          Please include "LiquidIO" in the subject.
6 *
7 * Copyright (c) 2003-2015 Cavium, Inc.
8 *
9 * This file is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License, Version 2, as
11 * published by the Free Software Foundation.
12 *
13 * This file is distributed in the hope that it will be useful, but
14 * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty
15 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or
16 * NONINFRINGEMENT.  See the GNU General Public License for more
17 * details.
18 *
19 * This file may also be available under a different license from Cavium.
20 * Contact Cavium, Inc. for more information
21 **********************************************************************/
22 #include <linux/pci.h>
23 #include <linux/netdevice.h>
24 #include <linux/vmalloc.h>
25 #include "liquidio_common.h"
26 #include "octeon_droq.h"
27 #include "octeon_iq.h"
28 #include "response_manager.h"
29 #include "octeon_device.h"
30 #include "octeon_main.h"
31 #include "octeon_network.h"
32 #include "cn66xx_regs.h"
33 #include "cn66xx_device.h"
34 #include "cn23xx_pf_device.h"
35 
36 #define     CVM_MIN(d1, d2)           (((d1) < (d2)) ? (d1) : (d2))
37 #define     CVM_MAX(d1, d2)           (((d1) > (d2)) ? (d1) : (d2))
38 
39 struct niclist {
40 	struct list_head list;
41 	void *ptr;
42 };
43 
44 struct __dispatch {
45 	struct list_head list;
46 	struct octeon_recv_info *rinfo;
47 	octeon_dispatch_fn_t disp_fn;
48 };
49 
50 /** Get the argument that the user set when registering dispatch
51  *  function for a given opcode/subcode.
52  *  @param  octeon_dev - the octeon device pointer.
53  *  @param  opcode     - the opcode for which the dispatch argument
54  *                       is to be checked.
55  *  @param  subcode    - the subcode for which the dispatch argument
56  *                       is to be checked.
57  *  @return  Success: void * (argument to the dispatch function)
58  *  @return  Failure: NULL
59  *
60  */
61 static inline void *octeon_get_dispatch_arg(struct octeon_device *octeon_dev,
62 					    u16 opcode, u16 subcode)
63 {
64 	int idx;
65 	struct list_head *dispatch;
66 	void *fn_arg = NULL;
67 	u16 combined_opcode = OPCODE_SUBCODE(opcode, subcode);
68 
69 	idx = combined_opcode & OCTEON_OPCODE_MASK;
70 
71 	spin_lock_bh(&octeon_dev->dispatch.lock);
72 
73 	if (octeon_dev->dispatch.count == 0) {
74 		spin_unlock_bh(&octeon_dev->dispatch.lock);
75 		return NULL;
76 	}
77 
78 	if (octeon_dev->dispatch.dlist[idx].opcode == combined_opcode) {
79 		fn_arg = octeon_dev->dispatch.dlist[idx].arg;
80 	} else {
81 		list_for_each(dispatch,
82 			      &octeon_dev->dispatch.dlist[idx].list) {
83 			if (((struct octeon_dispatch *)dispatch)->opcode ==
84 			    combined_opcode) {
85 				fn_arg = ((struct octeon_dispatch *)
86 					  dispatch)->arg;
87 				break;
88 			}
89 		}
90 	}
91 
92 	spin_unlock_bh(&octeon_dev->dispatch.lock);
93 	return fn_arg;
94 }
95 
96 /** Check for packets on Droq. This function should be called with lock held.
97  *  @param  droq - Droq on which count is checked.
98  *  @return Returns packet count.
99  */
100 u32 octeon_droq_check_hw_for_pkts(struct octeon_droq *droq)
101 {
102 	u32 pkt_count = 0;
103 	u32 last_count;
104 
105 	pkt_count = readl(droq->pkts_sent_reg);
106 
107 	last_count = pkt_count - droq->pkt_count;
108 	droq->pkt_count = pkt_count;
109 
110 	/* we shall write to cnts  at napi irq enable or end of droq tasklet */
111 	if (last_count)
112 		atomic_add(last_count, &droq->pkts_pending);
113 
114 	return last_count;
115 }
116 
117 static void octeon_droq_compute_max_packet_bufs(struct octeon_droq *droq)
118 {
119 	u32 count = 0;
120 
121 	/* max_empty_descs is the max. no. of descs that can have no buffers.
122 	 * If the empty desc count goes beyond this value, we cannot safely
123 	 * read in a 64K packet sent by Octeon
124 	 * (64K is max pkt size from Octeon)
125 	 */
126 	droq->max_empty_descs = 0;
127 
128 	do {
129 		droq->max_empty_descs++;
130 		count += droq->buffer_size;
131 	} while (count < (64 * 1024));
132 
133 	droq->max_empty_descs = droq->max_count - droq->max_empty_descs;
134 }
135 
136 static void octeon_droq_reset_indices(struct octeon_droq *droq)
137 {
138 	droq->read_idx = 0;
139 	droq->write_idx = 0;
140 	droq->refill_idx = 0;
141 	droq->refill_count = 0;
142 	atomic_set(&droq->pkts_pending, 0);
143 }
144 
145 static void
146 octeon_droq_destroy_ring_buffers(struct octeon_device *oct,
147 				 struct octeon_droq *droq)
148 {
149 	u32 i;
150 	struct octeon_skb_page_info *pg_info;
151 
152 	for (i = 0; i < droq->max_count; i++) {
153 		pg_info = &droq->recv_buf_list[i].pg_info;
154 
155 		if (pg_info->dma)
156 			lio_unmap_ring(oct->pci_dev,
157 				       (u64)pg_info->dma);
158 		pg_info->dma = 0;
159 
160 		if (pg_info->page)
161 			recv_buffer_destroy(droq->recv_buf_list[i].buffer,
162 					    pg_info);
163 
164 		if (droq->desc_ring && droq->desc_ring[i].info_ptr)
165 			lio_unmap_ring_info(oct->pci_dev,
166 					    (u64)droq->
167 					    desc_ring[i].info_ptr,
168 					    OCT_DROQ_INFO_SIZE);
169 		droq->recv_buf_list[i].buffer = NULL;
170 	}
171 
172 	octeon_droq_reset_indices(droq);
173 }
174 
175 static int
176 octeon_droq_setup_ring_buffers(struct octeon_device *oct,
177 			       struct octeon_droq *droq)
178 {
179 	u32 i;
180 	void *buf;
181 	struct octeon_droq_desc *desc_ring = droq->desc_ring;
182 
183 	for (i = 0; i < droq->max_count; i++) {
184 		buf = recv_buffer_alloc(oct, &droq->recv_buf_list[i].pg_info);
185 
186 		if (!buf) {
187 			dev_err(&oct->pci_dev->dev, "%s buffer alloc failed\n",
188 				__func__);
189 			droq->stats.rx_alloc_failure++;
190 			return -ENOMEM;
191 		}
192 
193 		droq->recv_buf_list[i].buffer = buf;
194 		droq->recv_buf_list[i].data = get_rbd(buf);
195 		droq->info_list[i].length = 0;
196 
197 		/* map ring buffers into memory */
198 		desc_ring[i].info_ptr = lio_map_ring_info(droq, i);
199 		desc_ring[i].buffer_ptr =
200 			lio_map_ring(droq->recv_buf_list[i].buffer);
201 	}
202 
203 	octeon_droq_reset_indices(droq);
204 
205 	octeon_droq_compute_max_packet_bufs(droq);
206 
207 	return 0;
208 }
209 
210 int octeon_delete_droq(struct octeon_device *oct, u32 q_no)
211 {
212 	struct octeon_droq *droq = oct->droq[q_no];
213 
214 	dev_dbg(&oct->pci_dev->dev, "%s[%d]\n", __func__, q_no);
215 
216 	octeon_droq_destroy_ring_buffers(oct, droq);
217 	vfree(droq->recv_buf_list);
218 
219 	if (droq->info_base_addr)
220 		cnnic_free_aligned_dma(oct->pci_dev, droq->info_list,
221 				       droq->info_alloc_size,
222 				       droq->info_base_addr,
223 				       droq->info_list_dma);
224 
225 	if (droq->desc_ring)
226 		lio_dma_free(oct, (droq->max_count * OCT_DROQ_DESC_SIZE),
227 			     droq->desc_ring, droq->desc_ring_dma);
228 
229 	memset(droq, 0, OCT_DROQ_SIZE);
230 
231 	return 0;
232 }
233 
234 int octeon_init_droq(struct octeon_device *oct,
235 		     u32 q_no,
236 		     u32 num_descs,
237 		     u32 desc_size,
238 		     void *app_ctx)
239 {
240 	struct octeon_droq *droq;
241 	u32 desc_ring_size = 0, c_num_descs = 0, c_buf_size = 0;
242 	u32 c_pkts_per_intr = 0, c_refill_threshold = 0;
243 	int orig_node = dev_to_node(&oct->pci_dev->dev);
244 	int numa_node = cpu_to_node(q_no % num_online_cpus());
245 
246 	dev_dbg(&oct->pci_dev->dev, "%s[%d]\n", __func__, q_no);
247 
248 	droq = oct->droq[q_no];
249 	memset(droq, 0, OCT_DROQ_SIZE);
250 
251 	droq->oct_dev = oct;
252 	droq->q_no = q_no;
253 	if (app_ctx)
254 		droq->app_ctx = app_ctx;
255 	else
256 		droq->app_ctx = (void *)(size_t)q_no;
257 
258 	c_num_descs = num_descs;
259 	c_buf_size = desc_size;
260 	if (OCTEON_CN6XXX(oct)) {
261 		struct octeon_config *conf6x = CHIP_FIELD(oct, cn6xxx, conf);
262 
263 		c_pkts_per_intr = (u32)CFG_GET_OQ_PKTS_PER_INTR(conf6x);
264 		c_refill_threshold =
265 			(u32)CFG_GET_OQ_REFILL_THRESHOLD(conf6x);
266 	} else if (OCTEON_CN23XX_PF(oct)) {
267 		struct octeon_config *conf23 = CHIP_FIELD(oct, cn23xx_pf, conf);
268 
269 		c_pkts_per_intr = (u32)CFG_GET_OQ_PKTS_PER_INTR(conf23);
270 		c_refill_threshold = (u32)CFG_GET_OQ_REFILL_THRESHOLD(conf23);
271 	} else {
272 		return 1;
273 	}
274 
275 	droq->max_count = c_num_descs;
276 	droq->buffer_size = c_buf_size;
277 
278 	desc_ring_size = droq->max_count * OCT_DROQ_DESC_SIZE;
279 	set_dev_node(&oct->pci_dev->dev, numa_node);
280 	droq->desc_ring = lio_dma_alloc(oct, desc_ring_size,
281 					(dma_addr_t *)&droq->desc_ring_dma);
282 	set_dev_node(&oct->pci_dev->dev, orig_node);
283 	if (!droq->desc_ring)
284 		droq->desc_ring = lio_dma_alloc(oct, desc_ring_size,
285 					(dma_addr_t *)&droq->desc_ring_dma);
286 
287 	if (!droq->desc_ring) {
288 		dev_err(&oct->pci_dev->dev,
289 			"Output queue %d ring alloc failed\n", q_no);
290 		return 1;
291 	}
292 
293 	dev_dbg(&oct->pci_dev->dev, "droq[%d]: desc_ring: virt: 0x%p, dma: %lx\n",
294 		q_no, droq->desc_ring, droq->desc_ring_dma);
295 	dev_dbg(&oct->pci_dev->dev, "droq[%d]: num_desc: %d\n", q_no,
296 		droq->max_count);
297 
298 	droq->info_list =
299 		cnnic_numa_alloc_aligned_dma((droq->max_count *
300 					      OCT_DROQ_INFO_SIZE),
301 					     &droq->info_alloc_size,
302 					     &droq->info_base_addr,
303 					     numa_node);
304 	if (!droq->info_list) {
305 		dev_err(&oct->pci_dev->dev, "Cannot allocate memory for info list.\n");
306 		lio_dma_free(oct, (droq->max_count * OCT_DROQ_DESC_SIZE),
307 			     droq->desc_ring, droq->desc_ring_dma);
308 		return 1;
309 	}
310 
311 	droq->recv_buf_list = (struct octeon_recv_buffer *)
312 			      vmalloc_node(droq->max_count *
313 						OCT_DROQ_RECVBUF_SIZE,
314 						numa_node);
315 	if (!droq->recv_buf_list)
316 		droq->recv_buf_list = (struct octeon_recv_buffer *)
317 				      vmalloc(droq->max_count *
318 						OCT_DROQ_RECVBUF_SIZE);
319 	if (!droq->recv_buf_list) {
320 		dev_err(&oct->pci_dev->dev, "Output queue recv buf list alloc failed\n");
321 		goto init_droq_fail;
322 	}
323 
324 	if (octeon_droq_setup_ring_buffers(oct, droq))
325 		goto init_droq_fail;
326 
327 	droq->pkts_per_intr = c_pkts_per_intr;
328 	droq->refill_threshold = c_refill_threshold;
329 
330 	dev_dbg(&oct->pci_dev->dev, "DROQ INIT: max_empty_descs: %d\n",
331 		droq->max_empty_descs);
332 
333 	spin_lock_init(&droq->lock);
334 
335 	INIT_LIST_HEAD(&droq->dispatch_list);
336 
337 	/* For 56xx Pass1, this function won't be called, so no checks. */
338 	oct->fn_list.setup_oq_regs(oct, q_no);
339 
340 	oct->io_qmask.oq |= (1ULL << q_no);
341 
342 	return 0;
343 
344 init_droq_fail:
345 	octeon_delete_droq(oct, q_no);
346 	return 1;
347 }
348 
349 /* octeon_create_recv_info
350  * Parameters:
351  *  octeon_dev - pointer to the octeon device structure
352  *  droq       - droq in which the packet arrived.
353  *  buf_cnt    - no. of buffers used by the packet.
354  *  idx        - index in the descriptor for the first buffer in the packet.
355  * Description:
356  *  Allocates a recv_info_t and copies the buffer addresses for packet data
357  *  into the recv_pkt space which starts at an 8B offset from recv_info_t.
358  *  Flags the descriptors for refill later. If available descriptors go
359  *  below the threshold to receive a 64K pkt, new buffers are first allocated
360  *  before the recv_pkt_t is created.
361  *  This routine will be called in interrupt context.
362  * Returns:
363  *  Success: Pointer to recv_info_t
364  *  Failure: NULL.
365  * Locks:
366  *  The droq->lock is held when this routine is called.
367  */
368 static inline struct octeon_recv_info *octeon_create_recv_info(
369 		struct octeon_device *octeon_dev,
370 		struct octeon_droq *droq,
371 		u32 buf_cnt,
372 		u32 idx)
373 {
374 	struct octeon_droq_info *info;
375 	struct octeon_recv_pkt *recv_pkt;
376 	struct octeon_recv_info *recv_info;
377 	u32 i, bytes_left;
378 	struct octeon_skb_page_info *pg_info;
379 
380 	info = &droq->info_list[idx];
381 
382 	recv_info = octeon_alloc_recv_info(sizeof(struct __dispatch));
383 	if (!recv_info)
384 		return NULL;
385 
386 	recv_pkt = recv_info->recv_pkt;
387 	recv_pkt->rh = info->rh;
388 	recv_pkt->length = (u32)info->length;
389 	recv_pkt->buffer_count = (u16)buf_cnt;
390 	recv_pkt->octeon_id = (u16)octeon_dev->octeon_id;
391 
392 	i = 0;
393 	bytes_left = (u32)info->length;
394 
395 	while (buf_cnt) {
396 		{
397 			pg_info = &droq->recv_buf_list[idx].pg_info;
398 
399 			lio_unmap_ring(octeon_dev->pci_dev,
400 				       (u64)pg_info->dma);
401 			pg_info->page = NULL;
402 			pg_info->dma = 0;
403 		}
404 
405 		recv_pkt->buffer_size[i] =
406 			(bytes_left >=
407 			 droq->buffer_size) ? droq->buffer_size : bytes_left;
408 
409 		recv_pkt->buffer_ptr[i] = droq->recv_buf_list[idx].buffer;
410 		droq->recv_buf_list[idx].buffer = NULL;
411 
412 		INCR_INDEX_BY1(idx, droq->max_count);
413 		bytes_left -= droq->buffer_size;
414 		i++;
415 		buf_cnt--;
416 	}
417 
418 	return recv_info;
419 }
420 
421 /* If we were not able to refill all buffers, try to move around
422  * the buffers that were not dispatched.
423  */
424 static inline u32
425 octeon_droq_refill_pullup_descs(struct octeon_droq *droq,
426 				struct octeon_droq_desc *desc_ring)
427 {
428 	u32 desc_refilled = 0;
429 
430 	u32 refill_index = droq->refill_idx;
431 
432 	while (refill_index != droq->read_idx) {
433 		if (droq->recv_buf_list[refill_index].buffer) {
434 			droq->recv_buf_list[droq->refill_idx].buffer =
435 				droq->recv_buf_list[refill_index].buffer;
436 			droq->recv_buf_list[droq->refill_idx].data =
437 				droq->recv_buf_list[refill_index].data;
438 			desc_ring[droq->refill_idx].buffer_ptr =
439 				desc_ring[refill_index].buffer_ptr;
440 			droq->recv_buf_list[refill_index].buffer = NULL;
441 			desc_ring[refill_index].buffer_ptr = 0;
442 			do {
443 				INCR_INDEX_BY1(droq->refill_idx,
444 					       droq->max_count);
445 				desc_refilled++;
446 				droq->refill_count--;
447 			} while (droq->recv_buf_list[droq->refill_idx].
448 				 buffer);
449 		}
450 		INCR_INDEX_BY1(refill_index, droq->max_count);
451 	}                       /* while */
452 	return desc_refilled;
453 }
454 
455 /* octeon_droq_refill
456  * Parameters:
457  *  droq       - droq in which descriptors require new buffers.
458  * Description:
459  *  Called during normal DROQ processing in interrupt mode or by the poll
460  *  thread to refill the descriptors from which buffers were dispatched
461  *  to upper layers. Attempts to allocate new buffers. If that fails, moves
462  *  up buffers (that were not dispatched) to form a contiguous ring.
463  * Returns:
464  *  No of descriptors refilled.
465  * Locks:
466  *  This routine is called with droq->lock held.
467  */
468 static u32
469 octeon_droq_refill(struct octeon_device *octeon_dev, struct octeon_droq *droq)
470 {
471 	struct octeon_droq_desc *desc_ring;
472 	void *buf = NULL;
473 	u8 *data;
474 	u32 desc_refilled = 0;
475 	struct octeon_skb_page_info *pg_info;
476 
477 	desc_ring = droq->desc_ring;
478 
479 	while (droq->refill_count && (desc_refilled < droq->max_count)) {
480 		/* If a valid buffer exists (happens if there is no dispatch),
481 		 * reuse
482 		 * the buffer, else allocate.
483 		 */
484 		if (!droq->recv_buf_list[droq->refill_idx].buffer) {
485 			pg_info =
486 				&droq->recv_buf_list[droq->refill_idx].pg_info;
487 			/* Either recycle the existing pages or go for
488 			 * new page alloc
489 			 */
490 			if (pg_info->page)
491 				buf = recv_buffer_reuse(octeon_dev, pg_info);
492 			else
493 				buf = recv_buffer_alloc(octeon_dev, pg_info);
494 			/* If a buffer could not be allocated, no point in
495 			 * continuing
496 			 */
497 			if (!buf) {
498 				droq->stats.rx_alloc_failure++;
499 				break;
500 			}
501 			droq->recv_buf_list[droq->refill_idx].buffer =
502 				buf;
503 			data = get_rbd(buf);
504 		} else {
505 			data = get_rbd(droq->recv_buf_list
506 				       [droq->refill_idx].buffer);
507 		}
508 
509 		droq->recv_buf_list[droq->refill_idx].data = data;
510 
511 		desc_ring[droq->refill_idx].buffer_ptr =
512 			lio_map_ring(droq->recv_buf_list[droq->
513 				     refill_idx].buffer);
514 		/* Reset any previous values in the length field. */
515 		droq->info_list[droq->refill_idx].length = 0;
516 
517 		INCR_INDEX_BY1(droq->refill_idx, droq->max_count);
518 		desc_refilled++;
519 		droq->refill_count--;
520 	}
521 
522 	if (droq->refill_count)
523 		desc_refilled +=
524 			octeon_droq_refill_pullup_descs(droq, desc_ring);
525 
526 	/* if droq->refill_count
527 	 * The refill count would not change in pass two. We only moved buffers
528 	 * to close the gap in the ring, but we would still have the same no. of
529 	 * buffers to refill.
530 	 */
531 	return desc_refilled;
532 }
533 
534 static inline u32
535 octeon_droq_get_bufcount(u32 buf_size, u32 total_len)
536 {
537 	u32 buf_cnt = 0;
538 
539 	while (total_len > (buf_size * buf_cnt))
540 		buf_cnt++;
541 	return buf_cnt;
542 }
543 
544 static int
545 octeon_droq_dispatch_pkt(struct octeon_device *oct,
546 			 struct octeon_droq *droq,
547 			 union octeon_rh *rh,
548 			 struct octeon_droq_info *info)
549 {
550 	u32 cnt;
551 	octeon_dispatch_fn_t disp_fn;
552 	struct octeon_recv_info *rinfo;
553 
554 	cnt = octeon_droq_get_bufcount(droq->buffer_size, (u32)info->length);
555 
556 	disp_fn = octeon_get_dispatch(oct, (u16)rh->r.opcode,
557 				      (u16)rh->r.subcode);
558 	if (disp_fn) {
559 		rinfo = octeon_create_recv_info(oct, droq, cnt, droq->read_idx);
560 		if (rinfo) {
561 			struct __dispatch *rdisp = rinfo->rsvd;
562 
563 			rdisp->rinfo = rinfo;
564 			rdisp->disp_fn = disp_fn;
565 			rinfo->recv_pkt->rh = *rh;
566 			list_add_tail(&rdisp->list,
567 				      &droq->dispatch_list);
568 		} else {
569 			droq->stats.dropped_nomem++;
570 		}
571 	} else {
572 		dev_err(&oct->pci_dev->dev, "DROQ: No dispatch function (opcode %u/%u)\n",
573 			(unsigned int)rh->r.opcode,
574 			(unsigned int)rh->r.subcode);
575 		droq->stats.dropped_nodispatch++;
576 	}
577 
578 	return cnt;
579 }
580 
581 static inline void octeon_droq_drop_packets(struct octeon_device *oct,
582 					    struct octeon_droq *droq,
583 					    u32 cnt)
584 {
585 	u32 i = 0, buf_cnt;
586 	struct octeon_droq_info *info;
587 
588 	for (i = 0; i < cnt; i++) {
589 		info = &droq->info_list[droq->read_idx];
590 		octeon_swap_8B_data((u64 *)info, 2);
591 
592 		if (info->length) {
593 			info->length -= OCT_RH_SIZE;
594 			droq->stats.bytes_received += info->length;
595 			buf_cnt = octeon_droq_get_bufcount(droq->buffer_size,
596 							   (u32)info->length);
597 		} else {
598 			dev_err(&oct->pci_dev->dev, "DROQ: In drop: pkt with len 0\n");
599 			buf_cnt = 1;
600 		}
601 
602 		INCR_INDEX(droq->read_idx, buf_cnt, droq->max_count);
603 		droq->refill_count += buf_cnt;
604 	}
605 }
606 
607 static u32
608 octeon_droq_fast_process_packets(struct octeon_device *oct,
609 				 struct octeon_droq *droq,
610 				 u32 pkts_to_process)
611 {
612 	struct octeon_droq_info *info;
613 	union octeon_rh *rh;
614 	u32 pkt, total_len = 0, pkt_count;
615 
616 	pkt_count = pkts_to_process;
617 
618 	for (pkt = 0; pkt < pkt_count; pkt++) {
619 		u32 pkt_len = 0;
620 		struct sk_buff *nicbuf = NULL;
621 		struct octeon_skb_page_info *pg_info;
622 		void *buf;
623 
624 		info = &droq->info_list[droq->read_idx];
625 		octeon_swap_8B_data((u64 *)info, 2);
626 
627 		if (!info->length) {
628 			dev_err(&oct->pci_dev->dev,
629 				"DROQ[%d] idx: %d len:0, pkt_cnt: %d\n",
630 				droq->q_no, droq->read_idx, pkt_count);
631 			print_hex_dump_bytes("", DUMP_PREFIX_ADDRESS,
632 					     (u8 *)info,
633 					     OCT_DROQ_INFO_SIZE);
634 			break;
635 		}
636 
637 		/* Len of resp hdr in included in the received data len. */
638 		info->length -= OCT_RH_SIZE;
639 		rh = &info->rh;
640 
641 		total_len += (u32)info->length;
642 		if (OPCODE_SLOW_PATH(rh)) {
643 			u32 buf_cnt;
644 
645 			buf_cnt = octeon_droq_dispatch_pkt(oct, droq, rh, info);
646 			INCR_INDEX(droq->read_idx, buf_cnt, droq->max_count);
647 			droq->refill_count += buf_cnt;
648 		} else {
649 			if (info->length <= droq->buffer_size) {
650 				pkt_len = (u32)info->length;
651 				nicbuf = droq->recv_buf_list[
652 					droq->read_idx].buffer;
653 				pg_info = &droq->recv_buf_list[
654 					droq->read_idx].pg_info;
655 				if (recv_buffer_recycle(oct, pg_info))
656 					pg_info->page = NULL;
657 				droq->recv_buf_list[droq->read_idx].buffer =
658 					NULL;
659 
660 				INCR_INDEX_BY1(droq->read_idx, droq->max_count);
661 				droq->refill_count++;
662 			} else {
663 				nicbuf = octeon_fast_packet_alloc((u32)
664 								  info->length);
665 				pkt_len = 0;
666 				/* nicbuf allocation can fail. We'll handle it
667 				 * inside the loop.
668 				 */
669 				while (pkt_len < info->length) {
670 					int cpy_len, idx = droq->read_idx;
671 
672 					cpy_len = ((pkt_len + droq->buffer_size)
673 						   > info->length) ?
674 						((u32)info->length - pkt_len) :
675 						droq->buffer_size;
676 
677 					if (nicbuf) {
678 						octeon_fast_packet_next(droq,
679 									nicbuf,
680 									cpy_len,
681 									idx);
682 						buf = droq->recv_buf_list[idx].
683 							buffer;
684 						recv_buffer_fast_free(buf);
685 						droq->recv_buf_list[idx].buffer
686 							= NULL;
687 					} else {
688 						droq->stats.rx_alloc_failure++;
689 					}
690 
691 					pkt_len += cpy_len;
692 					INCR_INDEX_BY1(droq->read_idx,
693 						       droq->max_count);
694 					droq->refill_count++;
695 				}
696 			}
697 
698 			if (nicbuf) {
699 				if (droq->ops.fptr) {
700 					droq->ops.fptr(oct->octeon_id,
701 						       nicbuf, pkt_len,
702 						       rh, &droq->napi,
703 						       droq->ops.farg);
704 				} else {
705 					recv_buffer_free(nicbuf);
706 				}
707 			}
708 		}
709 
710 		if (droq->refill_count >= droq->refill_threshold) {
711 			int desc_refilled = octeon_droq_refill(oct, droq);
712 
713 			/* Flush the droq descriptor data to memory to be sure
714 			 * that when we update the credits the data in memory
715 			 * is accurate.
716 			 */
717 			wmb();
718 			writel((desc_refilled), droq->pkts_credit_reg);
719 			/* make sure mmio write completes */
720 			mmiowb();
721 		}
722 
723 	}                       /* for (each packet)... */
724 
725 	/* Increment refill_count by the number of buffers processed. */
726 	droq->stats.pkts_received += pkt;
727 	droq->stats.bytes_received += total_len;
728 
729 	if ((droq->ops.drop_on_max) && (pkts_to_process - pkt)) {
730 		octeon_droq_drop_packets(oct, droq, (pkts_to_process - pkt));
731 
732 		droq->stats.dropped_toomany += (pkts_to_process - pkt);
733 		return pkts_to_process;
734 	}
735 
736 	return pkt;
737 }
738 
739 int
740 octeon_droq_process_packets(struct octeon_device *oct,
741 			    struct octeon_droq *droq,
742 			    u32 budget)
743 {
744 	u32 pkt_count = 0, pkts_processed = 0;
745 	struct list_head *tmp, *tmp2;
746 
747 	/* Grab the droq lock */
748 	spin_lock(&droq->lock);
749 
750 	octeon_droq_check_hw_for_pkts(droq);
751 	pkt_count = atomic_read(&droq->pkts_pending);
752 
753 	if (!pkt_count) {
754 		spin_unlock(&droq->lock);
755 		return 0;
756 	}
757 
758 	if (pkt_count > budget)
759 		pkt_count = budget;
760 
761 	pkts_processed = octeon_droq_fast_process_packets(oct, droq, pkt_count);
762 
763 	atomic_sub(pkts_processed, &droq->pkts_pending);
764 
765 	/* Release the spin lock */
766 	spin_unlock(&droq->lock);
767 
768 	list_for_each_safe(tmp, tmp2, &droq->dispatch_list) {
769 		struct __dispatch *rdisp = (struct __dispatch *)tmp;
770 
771 		list_del(tmp);
772 		rdisp->disp_fn(rdisp->rinfo,
773 			       octeon_get_dispatch_arg
774 			       (oct,
775 				(u16)rdisp->rinfo->recv_pkt->rh.r.opcode,
776 				(u16)rdisp->rinfo->recv_pkt->rh.r.subcode));
777 	}
778 
779 	/* If there are packets pending. schedule tasklet again */
780 	if (atomic_read(&droq->pkts_pending))
781 		return 1;
782 
783 	return 0;
784 }
785 
786 /**
787  * Utility function to poll for packets. check_hw_for_packets must be
788  * called before calling this routine.
789  */
790 
791 static int
792 octeon_droq_process_poll_pkts(struct octeon_device *oct,
793 			      struct octeon_droq *droq, u32 budget)
794 {
795 	struct list_head *tmp, *tmp2;
796 	u32 pkts_available = 0, pkts_processed = 0;
797 	u32 total_pkts_processed = 0;
798 
799 	if (budget > droq->max_count)
800 		budget = droq->max_count;
801 
802 	spin_lock(&droq->lock);
803 
804 	while (total_pkts_processed < budget) {
805 		octeon_droq_check_hw_for_pkts(droq);
806 
807 		pkts_available =
808 			CVM_MIN((budget - total_pkts_processed),
809 				(u32)(atomic_read(&droq->pkts_pending)));
810 
811 		if (pkts_available == 0)
812 			break;
813 
814 		pkts_processed =
815 			octeon_droq_fast_process_packets(oct, droq,
816 							 pkts_available);
817 
818 		atomic_sub(pkts_processed, &droq->pkts_pending);
819 
820 		total_pkts_processed += pkts_processed;
821 	}
822 
823 	spin_unlock(&droq->lock);
824 
825 	list_for_each_safe(tmp, tmp2, &droq->dispatch_list) {
826 		struct __dispatch *rdisp = (struct __dispatch *)tmp;
827 
828 		list_del(tmp);
829 		rdisp->disp_fn(rdisp->rinfo,
830 			       octeon_get_dispatch_arg
831 			       (oct,
832 				(u16)rdisp->rinfo->recv_pkt->rh.r.opcode,
833 				(u16)rdisp->rinfo->recv_pkt->rh.r.subcode));
834 	}
835 
836 	return total_pkts_processed;
837 }
838 
839 int
840 octeon_process_droq_poll_cmd(struct octeon_device *oct, u32 q_no, int cmd,
841 			     u32 arg)
842 {
843 	struct octeon_droq *droq;
844 
845 	droq = oct->droq[q_no];
846 
847 	if (cmd == POLL_EVENT_PROCESS_PKTS)
848 		return octeon_droq_process_poll_pkts(oct, droq, arg);
849 
850 	if (cmd == POLL_EVENT_PENDING_PKTS) {
851 		u32 pkt_cnt = atomic_read(&droq->pkts_pending);
852 
853 		return  octeon_droq_process_packets(oct, droq, pkt_cnt);
854 	}
855 
856 	if (cmd == POLL_EVENT_ENABLE_INTR) {
857 		u32 value;
858 		unsigned long flags;
859 
860 		/* Enable Pkt Interrupt */
861 		switch (oct->chip_id) {
862 		case OCTEON_CN66XX:
863 		case OCTEON_CN68XX: {
864 			struct octeon_cn6xxx *cn6xxx =
865 				(struct octeon_cn6xxx *)oct->chip;
866 			spin_lock_irqsave
867 				(&cn6xxx->lock_for_droq_int_enb_reg, flags);
868 			value =
869 				octeon_read_csr(oct,
870 						CN6XXX_SLI_PKT_TIME_INT_ENB);
871 			value |= (1 << q_no);
872 			octeon_write_csr(oct,
873 					 CN6XXX_SLI_PKT_TIME_INT_ENB,
874 					 value);
875 			value =
876 				octeon_read_csr(oct,
877 						CN6XXX_SLI_PKT_CNT_INT_ENB);
878 			value |= (1 << q_no);
879 			octeon_write_csr(oct,
880 					 CN6XXX_SLI_PKT_CNT_INT_ENB,
881 					 value);
882 
883 			/* don't bother flushing the enables */
884 
885 			spin_unlock_irqrestore
886 				(&cn6xxx->lock_for_droq_int_enb_reg, flags);
887 			return 0;
888 		}
889 		break;
890 		case OCTEON_CN23XX_PF_VID: {
891 			lio_enable_irq(oct->droq[q_no], oct->instr_queue[q_no]);
892 		}
893 		break;
894 		}
895 		return 0;
896 	}
897 
898 	dev_err(&oct->pci_dev->dev, "%s Unknown command: %d\n", __func__, cmd);
899 	return -EINVAL;
900 }
901 
902 int octeon_register_droq_ops(struct octeon_device *oct, u32 q_no,
903 			     struct octeon_droq_ops *ops)
904 {
905 	struct octeon_droq *droq;
906 	unsigned long flags;
907 	struct octeon_config *oct_cfg = NULL;
908 
909 	oct_cfg = octeon_get_conf(oct);
910 
911 	if (!oct_cfg)
912 		return -EINVAL;
913 
914 	if (!(ops)) {
915 		dev_err(&oct->pci_dev->dev, "%s: droq_ops pointer is NULL\n",
916 			__func__);
917 		return -EINVAL;
918 	}
919 
920 	if (q_no >= CFG_GET_OQ_MAX_Q(oct_cfg)) {
921 		dev_err(&oct->pci_dev->dev, "%s: droq id (%d) exceeds MAX (%d)\n",
922 			__func__, q_no, (oct->num_oqs - 1));
923 		return -EINVAL;
924 	}
925 
926 	droq = oct->droq[q_no];
927 
928 	spin_lock_irqsave(&droq->lock, flags);
929 
930 	memcpy(&droq->ops, ops, sizeof(struct octeon_droq_ops));
931 
932 	spin_unlock_irqrestore(&droq->lock, flags);
933 
934 	return 0;
935 }
936 
937 int octeon_unregister_droq_ops(struct octeon_device *oct, u32 q_no)
938 {
939 	unsigned long flags;
940 	struct octeon_droq *droq;
941 	struct octeon_config *oct_cfg = NULL;
942 
943 	oct_cfg = octeon_get_conf(oct);
944 
945 	if (!oct_cfg)
946 		return -EINVAL;
947 
948 	if (q_no >= CFG_GET_OQ_MAX_Q(oct_cfg)) {
949 		dev_err(&oct->pci_dev->dev, "%s: droq id (%d) exceeds MAX (%d)\n",
950 			__func__, q_no, oct->num_oqs - 1);
951 		return -EINVAL;
952 	}
953 
954 	droq = oct->droq[q_no];
955 
956 	if (!droq) {
957 		dev_info(&oct->pci_dev->dev,
958 			 "Droq id (%d) not available.\n", q_no);
959 		return 0;
960 	}
961 
962 	spin_lock_irqsave(&droq->lock, flags);
963 
964 	droq->ops.fptr = NULL;
965 	droq->ops.farg = NULL;
966 	droq->ops.drop_on_max = 0;
967 
968 	spin_unlock_irqrestore(&droq->lock, flags);
969 
970 	return 0;
971 }
972 
973 int octeon_create_droq(struct octeon_device *oct,
974 		       u32 q_no, u32 num_descs,
975 		       u32 desc_size, void *app_ctx)
976 {
977 	struct octeon_droq *droq;
978 	int numa_node = cpu_to_node(q_no % num_online_cpus());
979 
980 	if (oct->droq[q_no]) {
981 		dev_dbg(&oct->pci_dev->dev, "Droq already in use. Cannot create droq %d again\n",
982 			q_no);
983 		return 1;
984 	}
985 
986 	/* Allocate the DS for the new droq. */
987 	droq = vmalloc_node(sizeof(*droq), numa_node);
988 	if (!droq)
989 		droq = vmalloc(sizeof(*droq));
990 	if (!droq)
991 		return -1;
992 
993 	memset(droq, 0, sizeof(struct octeon_droq));
994 
995 	/*Disable the pkt o/p for this Q  */
996 	octeon_set_droq_pkt_op(oct, q_no, 0);
997 	oct->droq[q_no] = droq;
998 
999 	/* Initialize the Droq */
1000 	if (octeon_init_droq(oct, q_no, num_descs, desc_size, app_ctx)) {
1001 		vfree(oct->droq[q_no]);
1002 		oct->droq[q_no] = NULL;
1003 		return -1;
1004 	}
1005 
1006 	oct->num_oqs++;
1007 
1008 	dev_dbg(&oct->pci_dev->dev, "%s: Total number of OQ: %d\n", __func__,
1009 		oct->num_oqs);
1010 
1011 	/* Global Droq register settings */
1012 
1013 	/* As of now not required, as setting are done for all 32 Droqs at
1014 	 * the same time.
1015 	 */
1016 	return 0;
1017 }
1018