xref: /openbmc/linux/drivers/dma/mv_xor.c (revision ca79522c)
1 /*
2  * offload engine driver for the Marvell XOR engine
3  * Copyright (C) 2007, 2008, Marvell International Ltd.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  *
14  * You should have received a copy of the GNU General Public License along with
15  * this program; if not, write to the Free Software Foundation, Inc.,
16  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
17  */
18 
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/slab.h>
22 #include <linux/delay.h>
23 #include <linux/dma-mapping.h>
24 #include <linux/spinlock.h>
25 #include <linux/interrupt.h>
26 #include <linux/platform_device.h>
27 #include <linux/memory.h>
28 #include <linux/clk.h>
29 #include <linux/of.h>
30 #include <linux/of_irq.h>
31 #include <linux/irqdomain.h>
32 #include <linux/platform_data/dma-mv_xor.h>
33 
34 #include "dmaengine.h"
35 #include "mv_xor.h"
36 
37 static void mv_xor_issue_pending(struct dma_chan *chan);
38 
39 #define to_mv_xor_chan(chan)		\
40 	container_of(chan, struct mv_xor_chan, dmachan)
41 
42 #define to_mv_xor_slot(tx)		\
43 	container_of(tx, struct mv_xor_desc_slot, async_tx)
44 
45 #define mv_chan_to_devp(chan)           \
46 	((chan)->dmadev.dev)
47 
48 static void mv_desc_init(struct mv_xor_desc_slot *desc, unsigned long flags)
49 {
50 	struct mv_xor_desc *hw_desc = desc->hw_desc;
51 
52 	hw_desc->status = (1 << 31);
53 	hw_desc->phy_next_desc = 0;
54 	hw_desc->desc_command = (1 << 31);
55 }
56 
57 static u32 mv_desc_get_dest_addr(struct mv_xor_desc_slot *desc)
58 {
59 	struct mv_xor_desc *hw_desc = desc->hw_desc;
60 	return hw_desc->phy_dest_addr;
61 }
62 
63 static u32 mv_desc_get_src_addr(struct mv_xor_desc_slot *desc,
64 				int src_idx)
65 {
66 	struct mv_xor_desc *hw_desc = desc->hw_desc;
67 	return hw_desc->phy_src_addr[src_idx];
68 }
69 
70 
71 static void mv_desc_set_byte_count(struct mv_xor_desc_slot *desc,
72 				   u32 byte_count)
73 {
74 	struct mv_xor_desc *hw_desc = desc->hw_desc;
75 	hw_desc->byte_count = byte_count;
76 }
77 
78 static void mv_desc_set_next_desc(struct mv_xor_desc_slot *desc,
79 				  u32 next_desc_addr)
80 {
81 	struct mv_xor_desc *hw_desc = desc->hw_desc;
82 	BUG_ON(hw_desc->phy_next_desc);
83 	hw_desc->phy_next_desc = next_desc_addr;
84 }
85 
86 static void mv_desc_clear_next_desc(struct mv_xor_desc_slot *desc)
87 {
88 	struct mv_xor_desc *hw_desc = desc->hw_desc;
89 	hw_desc->phy_next_desc = 0;
90 }
91 
92 static void mv_desc_set_block_fill_val(struct mv_xor_desc_slot *desc, u32 val)
93 {
94 	desc->value = val;
95 }
96 
97 static void mv_desc_set_dest_addr(struct mv_xor_desc_slot *desc,
98 				  dma_addr_t addr)
99 {
100 	struct mv_xor_desc *hw_desc = desc->hw_desc;
101 	hw_desc->phy_dest_addr = addr;
102 }
103 
104 static int mv_chan_memset_slot_count(size_t len)
105 {
106 	return 1;
107 }
108 
109 #define mv_chan_memcpy_slot_count(c) mv_chan_memset_slot_count(c)
110 
111 static void mv_desc_set_src_addr(struct mv_xor_desc_slot *desc,
112 				 int index, dma_addr_t addr)
113 {
114 	struct mv_xor_desc *hw_desc = desc->hw_desc;
115 	hw_desc->phy_src_addr[index] = addr;
116 	if (desc->type == DMA_XOR)
117 		hw_desc->desc_command |= (1 << index);
118 }
119 
120 static u32 mv_chan_get_current_desc(struct mv_xor_chan *chan)
121 {
122 	return __raw_readl(XOR_CURR_DESC(chan));
123 }
124 
125 static void mv_chan_set_next_descriptor(struct mv_xor_chan *chan,
126 					u32 next_desc_addr)
127 {
128 	__raw_writel(next_desc_addr, XOR_NEXT_DESC(chan));
129 }
130 
131 static void mv_chan_set_dest_pointer(struct mv_xor_chan *chan, u32 desc_addr)
132 {
133 	__raw_writel(desc_addr, XOR_DEST_POINTER(chan));
134 }
135 
136 static void mv_chan_set_block_size(struct mv_xor_chan *chan, u32 block_size)
137 {
138 	__raw_writel(block_size, XOR_BLOCK_SIZE(chan));
139 }
140 
141 static void mv_chan_set_value(struct mv_xor_chan *chan, u32 value)
142 {
143 	__raw_writel(value, XOR_INIT_VALUE_LOW(chan));
144 	__raw_writel(value, XOR_INIT_VALUE_HIGH(chan));
145 }
146 
147 static void mv_chan_unmask_interrupts(struct mv_xor_chan *chan)
148 {
149 	u32 val = __raw_readl(XOR_INTR_MASK(chan));
150 	val |= XOR_INTR_MASK_VALUE << (chan->idx * 16);
151 	__raw_writel(val, XOR_INTR_MASK(chan));
152 }
153 
154 static u32 mv_chan_get_intr_cause(struct mv_xor_chan *chan)
155 {
156 	u32 intr_cause = __raw_readl(XOR_INTR_CAUSE(chan));
157 	intr_cause = (intr_cause >> (chan->idx * 16)) & 0xFFFF;
158 	return intr_cause;
159 }
160 
161 static int mv_is_err_intr(u32 intr_cause)
162 {
163 	if (intr_cause & ((1<<4)|(1<<5)|(1<<6)|(1<<7)|(1<<8)|(1<<9)))
164 		return 1;
165 
166 	return 0;
167 }
168 
169 static void mv_xor_device_clear_eoc_cause(struct mv_xor_chan *chan)
170 {
171 	u32 val = ~(1 << (chan->idx * 16));
172 	dev_dbg(mv_chan_to_devp(chan), "%s, val 0x%08x\n", __func__, val);
173 	__raw_writel(val, XOR_INTR_CAUSE(chan));
174 }
175 
176 static void mv_xor_device_clear_err_status(struct mv_xor_chan *chan)
177 {
178 	u32 val = 0xFFFF0000 >> (chan->idx * 16);
179 	__raw_writel(val, XOR_INTR_CAUSE(chan));
180 }
181 
182 static int mv_can_chain(struct mv_xor_desc_slot *desc)
183 {
184 	struct mv_xor_desc_slot *chain_old_tail = list_entry(
185 		desc->chain_node.prev, struct mv_xor_desc_slot, chain_node);
186 
187 	if (chain_old_tail->type != desc->type)
188 		return 0;
189 	if (desc->type == DMA_MEMSET)
190 		return 0;
191 
192 	return 1;
193 }
194 
195 static void mv_set_mode(struct mv_xor_chan *chan,
196 			       enum dma_transaction_type type)
197 {
198 	u32 op_mode;
199 	u32 config = __raw_readl(XOR_CONFIG(chan));
200 
201 	switch (type) {
202 	case DMA_XOR:
203 		op_mode = XOR_OPERATION_MODE_XOR;
204 		break;
205 	case DMA_MEMCPY:
206 		op_mode = XOR_OPERATION_MODE_MEMCPY;
207 		break;
208 	case DMA_MEMSET:
209 		op_mode = XOR_OPERATION_MODE_MEMSET;
210 		break;
211 	default:
212 		dev_err(mv_chan_to_devp(chan),
213 			"error: unsupported operation %d\n",
214 			type);
215 		BUG();
216 		return;
217 	}
218 
219 	config &= ~0x7;
220 	config |= op_mode;
221 	__raw_writel(config, XOR_CONFIG(chan));
222 	chan->current_type = type;
223 }
224 
225 static void mv_chan_activate(struct mv_xor_chan *chan)
226 {
227 	u32 activation;
228 
229 	dev_dbg(mv_chan_to_devp(chan), " activate chan.\n");
230 	activation = __raw_readl(XOR_ACTIVATION(chan));
231 	activation |= 0x1;
232 	__raw_writel(activation, XOR_ACTIVATION(chan));
233 }
234 
235 static char mv_chan_is_busy(struct mv_xor_chan *chan)
236 {
237 	u32 state = __raw_readl(XOR_ACTIVATION(chan));
238 
239 	state = (state >> 4) & 0x3;
240 
241 	return (state == 1) ? 1 : 0;
242 }
243 
244 static int mv_chan_xor_slot_count(size_t len, int src_cnt)
245 {
246 	return 1;
247 }
248 
249 /**
250  * mv_xor_free_slots - flags descriptor slots for reuse
251  * @slot: Slot to free
252  * Caller must hold &mv_chan->lock while calling this function
253  */
254 static void mv_xor_free_slots(struct mv_xor_chan *mv_chan,
255 			      struct mv_xor_desc_slot *slot)
256 {
257 	dev_dbg(mv_chan_to_devp(mv_chan), "%s %d slot %p\n",
258 		__func__, __LINE__, slot);
259 
260 	slot->slots_per_op = 0;
261 
262 }
263 
264 /*
265  * mv_xor_start_new_chain - program the engine to operate on new chain headed by
266  * sw_desc
267  * Caller must hold &mv_chan->lock while calling this function
268  */
269 static void mv_xor_start_new_chain(struct mv_xor_chan *mv_chan,
270 				   struct mv_xor_desc_slot *sw_desc)
271 {
272 	dev_dbg(mv_chan_to_devp(mv_chan), "%s %d: sw_desc %p\n",
273 		__func__, __LINE__, sw_desc);
274 	if (sw_desc->type != mv_chan->current_type)
275 		mv_set_mode(mv_chan, sw_desc->type);
276 
277 	if (sw_desc->type == DMA_MEMSET) {
278 		/* for memset requests we need to program the engine, no
279 		 * descriptors used.
280 		 */
281 		struct mv_xor_desc *hw_desc = sw_desc->hw_desc;
282 		mv_chan_set_dest_pointer(mv_chan, hw_desc->phy_dest_addr);
283 		mv_chan_set_block_size(mv_chan, sw_desc->unmap_len);
284 		mv_chan_set_value(mv_chan, sw_desc->value);
285 	} else {
286 		/* set the hardware chain */
287 		mv_chan_set_next_descriptor(mv_chan, sw_desc->async_tx.phys);
288 	}
289 	mv_chan->pending += sw_desc->slot_cnt;
290 	mv_xor_issue_pending(&mv_chan->dmachan);
291 }
292 
293 static dma_cookie_t
294 mv_xor_run_tx_complete_actions(struct mv_xor_desc_slot *desc,
295 	struct mv_xor_chan *mv_chan, dma_cookie_t cookie)
296 {
297 	BUG_ON(desc->async_tx.cookie < 0);
298 
299 	if (desc->async_tx.cookie > 0) {
300 		cookie = desc->async_tx.cookie;
301 
302 		/* call the callback (must not sleep or submit new
303 		 * operations to this channel)
304 		 */
305 		if (desc->async_tx.callback)
306 			desc->async_tx.callback(
307 				desc->async_tx.callback_param);
308 
309 		/* unmap dma addresses
310 		 * (unmap_single vs unmap_page?)
311 		 */
312 		if (desc->group_head && desc->unmap_len) {
313 			struct mv_xor_desc_slot *unmap = desc->group_head;
314 			struct device *dev = mv_chan_to_devp(mv_chan);
315 			u32 len = unmap->unmap_len;
316 			enum dma_ctrl_flags flags = desc->async_tx.flags;
317 			u32 src_cnt;
318 			dma_addr_t addr;
319 			dma_addr_t dest;
320 
321 			src_cnt = unmap->unmap_src_cnt;
322 			dest = mv_desc_get_dest_addr(unmap);
323 			if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
324 				enum dma_data_direction dir;
325 
326 				if (src_cnt > 1) /* is xor ? */
327 					dir = DMA_BIDIRECTIONAL;
328 				else
329 					dir = DMA_FROM_DEVICE;
330 				dma_unmap_page(dev, dest, len, dir);
331 			}
332 
333 			if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
334 				while (src_cnt--) {
335 					addr = mv_desc_get_src_addr(unmap,
336 								    src_cnt);
337 					if (addr == dest)
338 						continue;
339 					dma_unmap_page(dev, addr, len,
340 						       DMA_TO_DEVICE);
341 				}
342 			}
343 			desc->group_head = NULL;
344 		}
345 	}
346 
347 	/* run dependent operations */
348 	dma_run_dependencies(&desc->async_tx);
349 
350 	return cookie;
351 }
352 
353 static int
354 mv_xor_clean_completed_slots(struct mv_xor_chan *mv_chan)
355 {
356 	struct mv_xor_desc_slot *iter, *_iter;
357 
358 	dev_dbg(mv_chan_to_devp(mv_chan), "%s %d\n", __func__, __LINE__);
359 	list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
360 				 completed_node) {
361 
362 		if (async_tx_test_ack(&iter->async_tx)) {
363 			list_del(&iter->completed_node);
364 			mv_xor_free_slots(mv_chan, iter);
365 		}
366 	}
367 	return 0;
368 }
369 
370 static int
371 mv_xor_clean_slot(struct mv_xor_desc_slot *desc,
372 	struct mv_xor_chan *mv_chan)
373 {
374 	dev_dbg(mv_chan_to_devp(mv_chan), "%s %d: desc %p flags %d\n",
375 		__func__, __LINE__, desc, desc->async_tx.flags);
376 	list_del(&desc->chain_node);
377 	/* the client is allowed to attach dependent operations
378 	 * until 'ack' is set
379 	 */
380 	if (!async_tx_test_ack(&desc->async_tx)) {
381 		/* move this slot to the completed_slots */
382 		list_add_tail(&desc->completed_node, &mv_chan->completed_slots);
383 		return 0;
384 	}
385 
386 	mv_xor_free_slots(mv_chan, desc);
387 	return 0;
388 }
389 
390 static void __mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
391 {
392 	struct mv_xor_desc_slot *iter, *_iter;
393 	dma_cookie_t cookie = 0;
394 	int busy = mv_chan_is_busy(mv_chan);
395 	u32 current_desc = mv_chan_get_current_desc(mv_chan);
396 	int seen_current = 0;
397 
398 	dev_dbg(mv_chan_to_devp(mv_chan), "%s %d\n", __func__, __LINE__);
399 	dev_dbg(mv_chan_to_devp(mv_chan), "current_desc %x\n", current_desc);
400 	mv_xor_clean_completed_slots(mv_chan);
401 
402 	/* free completed slots from the chain starting with
403 	 * the oldest descriptor
404 	 */
405 
406 	list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
407 					chain_node) {
408 		prefetch(_iter);
409 		prefetch(&_iter->async_tx);
410 
411 		/* do not advance past the current descriptor loaded into the
412 		 * hardware channel, subsequent descriptors are either in
413 		 * process or have not been submitted
414 		 */
415 		if (seen_current)
416 			break;
417 
418 		/* stop the search if we reach the current descriptor and the
419 		 * channel is busy
420 		 */
421 		if (iter->async_tx.phys == current_desc) {
422 			seen_current = 1;
423 			if (busy)
424 				break;
425 		}
426 
427 		cookie = mv_xor_run_tx_complete_actions(iter, mv_chan, cookie);
428 
429 		if (mv_xor_clean_slot(iter, mv_chan))
430 			break;
431 	}
432 
433 	if ((busy == 0) && !list_empty(&mv_chan->chain)) {
434 		struct mv_xor_desc_slot *chain_head;
435 		chain_head = list_entry(mv_chan->chain.next,
436 					struct mv_xor_desc_slot,
437 					chain_node);
438 
439 		mv_xor_start_new_chain(mv_chan, chain_head);
440 	}
441 
442 	if (cookie > 0)
443 		mv_chan->dmachan.completed_cookie = cookie;
444 }
445 
446 static void
447 mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
448 {
449 	spin_lock_bh(&mv_chan->lock);
450 	__mv_xor_slot_cleanup(mv_chan);
451 	spin_unlock_bh(&mv_chan->lock);
452 }
453 
454 static void mv_xor_tasklet(unsigned long data)
455 {
456 	struct mv_xor_chan *chan = (struct mv_xor_chan *) data;
457 	mv_xor_slot_cleanup(chan);
458 }
459 
460 static struct mv_xor_desc_slot *
461 mv_xor_alloc_slots(struct mv_xor_chan *mv_chan, int num_slots,
462 		    int slots_per_op)
463 {
464 	struct mv_xor_desc_slot *iter, *_iter, *alloc_start = NULL;
465 	LIST_HEAD(chain);
466 	int slots_found, retry = 0;
467 
468 	/* start search from the last allocated descrtiptor
469 	 * if a contiguous allocation can not be found start searching
470 	 * from the beginning of the list
471 	 */
472 retry:
473 	slots_found = 0;
474 	if (retry == 0)
475 		iter = mv_chan->last_used;
476 	else
477 		iter = list_entry(&mv_chan->all_slots,
478 			struct mv_xor_desc_slot,
479 			slot_node);
480 
481 	list_for_each_entry_safe_continue(
482 		iter, _iter, &mv_chan->all_slots, slot_node) {
483 		prefetch(_iter);
484 		prefetch(&_iter->async_tx);
485 		if (iter->slots_per_op) {
486 			/* give up after finding the first busy slot
487 			 * on the second pass through the list
488 			 */
489 			if (retry)
490 				break;
491 
492 			slots_found = 0;
493 			continue;
494 		}
495 
496 		/* start the allocation if the slot is correctly aligned */
497 		if (!slots_found++)
498 			alloc_start = iter;
499 
500 		if (slots_found == num_slots) {
501 			struct mv_xor_desc_slot *alloc_tail = NULL;
502 			struct mv_xor_desc_slot *last_used = NULL;
503 			iter = alloc_start;
504 			while (num_slots) {
505 				int i;
506 
507 				/* pre-ack all but the last descriptor */
508 				async_tx_ack(&iter->async_tx);
509 
510 				list_add_tail(&iter->chain_node, &chain);
511 				alloc_tail = iter;
512 				iter->async_tx.cookie = 0;
513 				iter->slot_cnt = num_slots;
514 				iter->xor_check_result = NULL;
515 				for (i = 0; i < slots_per_op; i++) {
516 					iter->slots_per_op = slots_per_op - i;
517 					last_used = iter;
518 					iter = list_entry(iter->slot_node.next,
519 						struct mv_xor_desc_slot,
520 						slot_node);
521 				}
522 				num_slots -= slots_per_op;
523 			}
524 			alloc_tail->group_head = alloc_start;
525 			alloc_tail->async_tx.cookie = -EBUSY;
526 			list_splice(&chain, &alloc_tail->tx_list);
527 			mv_chan->last_used = last_used;
528 			mv_desc_clear_next_desc(alloc_start);
529 			mv_desc_clear_next_desc(alloc_tail);
530 			return alloc_tail;
531 		}
532 	}
533 	if (!retry++)
534 		goto retry;
535 
536 	/* try to free some slots if the allocation fails */
537 	tasklet_schedule(&mv_chan->irq_tasklet);
538 
539 	return NULL;
540 }
541 
542 /************************ DMA engine API functions ****************************/
543 static dma_cookie_t
544 mv_xor_tx_submit(struct dma_async_tx_descriptor *tx)
545 {
546 	struct mv_xor_desc_slot *sw_desc = to_mv_xor_slot(tx);
547 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(tx->chan);
548 	struct mv_xor_desc_slot *grp_start, *old_chain_tail;
549 	dma_cookie_t cookie;
550 	int new_hw_chain = 1;
551 
552 	dev_dbg(mv_chan_to_devp(mv_chan),
553 		"%s sw_desc %p: async_tx %p\n",
554 		__func__, sw_desc, &sw_desc->async_tx);
555 
556 	grp_start = sw_desc->group_head;
557 
558 	spin_lock_bh(&mv_chan->lock);
559 	cookie = dma_cookie_assign(tx);
560 
561 	if (list_empty(&mv_chan->chain))
562 		list_splice_init(&sw_desc->tx_list, &mv_chan->chain);
563 	else {
564 		new_hw_chain = 0;
565 
566 		old_chain_tail = list_entry(mv_chan->chain.prev,
567 					    struct mv_xor_desc_slot,
568 					    chain_node);
569 		list_splice_init(&grp_start->tx_list,
570 				 &old_chain_tail->chain_node);
571 
572 		if (!mv_can_chain(grp_start))
573 			goto submit_done;
574 
575 		dev_dbg(mv_chan_to_devp(mv_chan), "Append to last desc %x\n",
576 			old_chain_tail->async_tx.phys);
577 
578 		/* fix up the hardware chain */
579 		mv_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
580 
581 		/* if the channel is not busy */
582 		if (!mv_chan_is_busy(mv_chan)) {
583 			u32 current_desc = mv_chan_get_current_desc(mv_chan);
584 			/*
585 			 * and the curren desc is the end of the chain before
586 			 * the append, then we need to start the channel
587 			 */
588 			if (current_desc == old_chain_tail->async_tx.phys)
589 				new_hw_chain = 1;
590 		}
591 	}
592 
593 	if (new_hw_chain)
594 		mv_xor_start_new_chain(mv_chan, grp_start);
595 
596 submit_done:
597 	spin_unlock_bh(&mv_chan->lock);
598 
599 	return cookie;
600 }
601 
602 /* returns the number of allocated descriptors */
603 static int mv_xor_alloc_chan_resources(struct dma_chan *chan)
604 {
605 	char *hw_desc;
606 	int idx;
607 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
608 	struct mv_xor_desc_slot *slot = NULL;
609 	int num_descs_in_pool = MV_XOR_POOL_SIZE/MV_XOR_SLOT_SIZE;
610 
611 	/* Allocate descriptor slots */
612 	idx = mv_chan->slots_allocated;
613 	while (idx < num_descs_in_pool) {
614 		slot = kzalloc(sizeof(*slot), GFP_KERNEL);
615 		if (!slot) {
616 			printk(KERN_INFO "MV XOR Channel only initialized"
617 				" %d descriptor slots", idx);
618 			break;
619 		}
620 		hw_desc = (char *) mv_chan->dma_desc_pool_virt;
621 		slot->hw_desc = (void *) &hw_desc[idx * MV_XOR_SLOT_SIZE];
622 
623 		dma_async_tx_descriptor_init(&slot->async_tx, chan);
624 		slot->async_tx.tx_submit = mv_xor_tx_submit;
625 		INIT_LIST_HEAD(&slot->chain_node);
626 		INIT_LIST_HEAD(&slot->slot_node);
627 		INIT_LIST_HEAD(&slot->tx_list);
628 		hw_desc = (char *) mv_chan->dma_desc_pool;
629 		slot->async_tx.phys =
630 			(dma_addr_t) &hw_desc[idx * MV_XOR_SLOT_SIZE];
631 		slot->idx = idx++;
632 
633 		spin_lock_bh(&mv_chan->lock);
634 		mv_chan->slots_allocated = idx;
635 		list_add_tail(&slot->slot_node, &mv_chan->all_slots);
636 		spin_unlock_bh(&mv_chan->lock);
637 	}
638 
639 	if (mv_chan->slots_allocated && !mv_chan->last_used)
640 		mv_chan->last_used = list_entry(mv_chan->all_slots.next,
641 					struct mv_xor_desc_slot,
642 					slot_node);
643 
644 	dev_dbg(mv_chan_to_devp(mv_chan),
645 		"allocated %d descriptor slots last_used: %p\n",
646 		mv_chan->slots_allocated, mv_chan->last_used);
647 
648 	return mv_chan->slots_allocated ? : -ENOMEM;
649 }
650 
651 static struct dma_async_tx_descriptor *
652 mv_xor_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
653 		size_t len, unsigned long flags)
654 {
655 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
656 	struct mv_xor_desc_slot *sw_desc, *grp_start;
657 	int slot_cnt;
658 
659 	dev_dbg(mv_chan_to_devp(mv_chan),
660 		"%s dest: %x src %x len: %u flags: %ld\n",
661 		__func__, dest, src, len, flags);
662 	if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
663 		return NULL;
664 
665 	BUG_ON(len > MV_XOR_MAX_BYTE_COUNT);
666 
667 	spin_lock_bh(&mv_chan->lock);
668 	slot_cnt = mv_chan_memcpy_slot_count(len);
669 	sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
670 	if (sw_desc) {
671 		sw_desc->type = DMA_MEMCPY;
672 		sw_desc->async_tx.flags = flags;
673 		grp_start = sw_desc->group_head;
674 		mv_desc_init(grp_start, flags);
675 		mv_desc_set_byte_count(grp_start, len);
676 		mv_desc_set_dest_addr(sw_desc->group_head, dest);
677 		mv_desc_set_src_addr(grp_start, 0, src);
678 		sw_desc->unmap_src_cnt = 1;
679 		sw_desc->unmap_len = len;
680 	}
681 	spin_unlock_bh(&mv_chan->lock);
682 
683 	dev_dbg(mv_chan_to_devp(mv_chan),
684 		"%s sw_desc %p async_tx %p\n",
685 		__func__, sw_desc, sw_desc ? &sw_desc->async_tx : 0);
686 
687 	return sw_desc ? &sw_desc->async_tx : NULL;
688 }
689 
690 static struct dma_async_tx_descriptor *
691 mv_xor_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,
692 		       size_t len, unsigned long flags)
693 {
694 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
695 	struct mv_xor_desc_slot *sw_desc, *grp_start;
696 	int slot_cnt;
697 
698 	dev_dbg(mv_chan_to_devp(mv_chan),
699 		"%s dest: %x len: %u flags: %ld\n",
700 		__func__, dest, len, flags);
701 	if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
702 		return NULL;
703 
704 	BUG_ON(len > MV_XOR_MAX_BYTE_COUNT);
705 
706 	spin_lock_bh(&mv_chan->lock);
707 	slot_cnt = mv_chan_memset_slot_count(len);
708 	sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
709 	if (sw_desc) {
710 		sw_desc->type = DMA_MEMSET;
711 		sw_desc->async_tx.flags = flags;
712 		grp_start = sw_desc->group_head;
713 		mv_desc_init(grp_start, flags);
714 		mv_desc_set_byte_count(grp_start, len);
715 		mv_desc_set_dest_addr(sw_desc->group_head, dest);
716 		mv_desc_set_block_fill_val(grp_start, value);
717 		sw_desc->unmap_src_cnt = 1;
718 		sw_desc->unmap_len = len;
719 	}
720 	spin_unlock_bh(&mv_chan->lock);
721 	dev_dbg(mv_chan_to_devp(mv_chan),
722 		"%s sw_desc %p async_tx %p \n",
723 		__func__, sw_desc, &sw_desc->async_tx);
724 	return sw_desc ? &sw_desc->async_tx : NULL;
725 }
726 
727 static struct dma_async_tx_descriptor *
728 mv_xor_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
729 		    unsigned int src_cnt, size_t len, unsigned long flags)
730 {
731 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
732 	struct mv_xor_desc_slot *sw_desc, *grp_start;
733 	int slot_cnt;
734 
735 	if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
736 		return NULL;
737 
738 	BUG_ON(len > MV_XOR_MAX_BYTE_COUNT);
739 
740 	dev_dbg(mv_chan_to_devp(mv_chan),
741 		"%s src_cnt: %d len: dest %x %u flags: %ld\n",
742 		__func__, src_cnt, len, dest, flags);
743 
744 	spin_lock_bh(&mv_chan->lock);
745 	slot_cnt = mv_chan_xor_slot_count(len, src_cnt);
746 	sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
747 	if (sw_desc) {
748 		sw_desc->type = DMA_XOR;
749 		sw_desc->async_tx.flags = flags;
750 		grp_start = sw_desc->group_head;
751 		mv_desc_init(grp_start, flags);
752 		/* the byte count field is the same as in memcpy desc*/
753 		mv_desc_set_byte_count(grp_start, len);
754 		mv_desc_set_dest_addr(sw_desc->group_head, dest);
755 		sw_desc->unmap_src_cnt = src_cnt;
756 		sw_desc->unmap_len = len;
757 		while (src_cnt--)
758 			mv_desc_set_src_addr(grp_start, src_cnt, src[src_cnt]);
759 	}
760 	spin_unlock_bh(&mv_chan->lock);
761 	dev_dbg(mv_chan_to_devp(mv_chan),
762 		"%s sw_desc %p async_tx %p \n",
763 		__func__, sw_desc, &sw_desc->async_tx);
764 	return sw_desc ? &sw_desc->async_tx : NULL;
765 }
766 
767 static void mv_xor_free_chan_resources(struct dma_chan *chan)
768 {
769 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
770 	struct mv_xor_desc_slot *iter, *_iter;
771 	int in_use_descs = 0;
772 
773 	mv_xor_slot_cleanup(mv_chan);
774 
775 	spin_lock_bh(&mv_chan->lock);
776 	list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
777 					chain_node) {
778 		in_use_descs++;
779 		list_del(&iter->chain_node);
780 	}
781 	list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
782 				 completed_node) {
783 		in_use_descs++;
784 		list_del(&iter->completed_node);
785 	}
786 	list_for_each_entry_safe_reverse(
787 		iter, _iter, &mv_chan->all_slots, slot_node) {
788 		list_del(&iter->slot_node);
789 		kfree(iter);
790 		mv_chan->slots_allocated--;
791 	}
792 	mv_chan->last_used = NULL;
793 
794 	dev_dbg(mv_chan_to_devp(mv_chan), "%s slots_allocated %d\n",
795 		__func__, mv_chan->slots_allocated);
796 	spin_unlock_bh(&mv_chan->lock);
797 
798 	if (in_use_descs)
799 		dev_err(mv_chan_to_devp(mv_chan),
800 			"freeing %d in use descriptors!\n", in_use_descs);
801 }
802 
803 /**
804  * mv_xor_status - poll the status of an XOR transaction
805  * @chan: XOR channel handle
806  * @cookie: XOR transaction identifier
807  * @txstate: XOR transactions state holder (or NULL)
808  */
809 static enum dma_status mv_xor_status(struct dma_chan *chan,
810 					  dma_cookie_t cookie,
811 					  struct dma_tx_state *txstate)
812 {
813 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
814 	enum dma_status ret;
815 
816 	ret = dma_cookie_status(chan, cookie, txstate);
817 	if (ret == DMA_SUCCESS) {
818 		mv_xor_clean_completed_slots(mv_chan);
819 		return ret;
820 	}
821 	mv_xor_slot_cleanup(mv_chan);
822 
823 	return dma_cookie_status(chan, cookie, txstate);
824 }
825 
826 static void mv_dump_xor_regs(struct mv_xor_chan *chan)
827 {
828 	u32 val;
829 
830 	val = __raw_readl(XOR_CONFIG(chan));
831 	dev_err(mv_chan_to_devp(chan), "config       0x%08x\n", val);
832 
833 	val = __raw_readl(XOR_ACTIVATION(chan));
834 	dev_err(mv_chan_to_devp(chan), "activation   0x%08x\n", val);
835 
836 	val = __raw_readl(XOR_INTR_CAUSE(chan));
837 	dev_err(mv_chan_to_devp(chan), "intr cause   0x%08x\n", val);
838 
839 	val = __raw_readl(XOR_INTR_MASK(chan));
840 	dev_err(mv_chan_to_devp(chan), "intr mask    0x%08x\n", val);
841 
842 	val = __raw_readl(XOR_ERROR_CAUSE(chan));
843 	dev_err(mv_chan_to_devp(chan), "error cause  0x%08x\n", val);
844 
845 	val = __raw_readl(XOR_ERROR_ADDR(chan));
846 	dev_err(mv_chan_to_devp(chan), "error addr   0x%08x\n", val);
847 }
848 
849 static void mv_xor_err_interrupt_handler(struct mv_xor_chan *chan,
850 					 u32 intr_cause)
851 {
852 	if (intr_cause & (1 << 4)) {
853 	     dev_dbg(mv_chan_to_devp(chan),
854 		     "ignore this error\n");
855 	     return;
856 	}
857 
858 	dev_err(mv_chan_to_devp(chan),
859 		"error on chan %d. intr cause 0x%08x\n",
860 		chan->idx, intr_cause);
861 
862 	mv_dump_xor_regs(chan);
863 	BUG();
864 }
865 
866 static irqreturn_t mv_xor_interrupt_handler(int irq, void *data)
867 {
868 	struct mv_xor_chan *chan = data;
869 	u32 intr_cause = mv_chan_get_intr_cause(chan);
870 
871 	dev_dbg(mv_chan_to_devp(chan), "intr cause %x\n", intr_cause);
872 
873 	if (mv_is_err_intr(intr_cause))
874 		mv_xor_err_interrupt_handler(chan, intr_cause);
875 
876 	tasklet_schedule(&chan->irq_tasklet);
877 
878 	mv_xor_device_clear_eoc_cause(chan);
879 
880 	return IRQ_HANDLED;
881 }
882 
883 static void mv_xor_issue_pending(struct dma_chan *chan)
884 {
885 	struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
886 
887 	if (mv_chan->pending >= MV_XOR_THRESHOLD) {
888 		mv_chan->pending = 0;
889 		mv_chan_activate(mv_chan);
890 	}
891 }
892 
893 /*
894  * Perform a transaction to verify the HW works.
895  */
896 #define MV_XOR_TEST_SIZE 2000
897 
898 static int mv_xor_memcpy_self_test(struct mv_xor_chan *mv_chan)
899 {
900 	int i;
901 	void *src, *dest;
902 	dma_addr_t src_dma, dest_dma;
903 	struct dma_chan *dma_chan;
904 	dma_cookie_t cookie;
905 	struct dma_async_tx_descriptor *tx;
906 	int err = 0;
907 
908 	src = kmalloc(sizeof(u8) * MV_XOR_TEST_SIZE, GFP_KERNEL);
909 	if (!src)
910 		return -ENOMEM;
911 
912 	dest = kzalloc(sizeof(u8) * MV_XOR_TEST_SIZE, GFP_KERNEL);
913 	if (!dest) {
914 		kfree(src);
915 		return -ENOMEM;
916 	}
917 
918 	/* Fill in src buffer */
919 	for (i = 0; i < MV_XOR_TEST_SIZE; i++)
920 		((u8 *) src)[i] = (u8)i;
921 
922 	dma_chan = &mv_chan->dmachan;
923 	if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
924 		err = -ENODEV;
925 		goto out;
926 	}
927 
928 	dest_dma = dma_map_single(dma_chan->device->dev, dest,
929 				  MV_XOR_TEST_SIZE, DMA_FROM_DEVICE);
930 
931 	src_dma = dma_map_single(dma_chan->device->dev, src,
932 				 MV_XOR_TEST_SIZE, DMA_TO_DEVICE);
933 
934 	tx = mv_xor_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
935 				    MV_XOR_TEST_SIZE, 0);
936 	cookie = mv_xor_tx_submit(tx);
937 	mv_xor_issue_pending(dma_chan);
938 	async_tx_ack(tx);
939 	msleep(1);
940 
941 	if (mv_xor_status(dma_chan, cookie, NULL) !=
942 	    DMA_SUCCESS) {
943 		dev_err(dma_chan->device->dev,
944 			"Self-test copy timed out, disabling\n");
945 		err = -ENODEV;
946 		goto free_resources;
947 	}
948 
949 	dma_sync_single_for_cpu(dma_chan->device->dev, dest_dma,
950 				MV_XOR_TEST_SIZE, DMA_FROM_DEVICE);
951 	if (memcmp(src, dest, MV_XOR_TEST_SIZE)) {
952 		dev_err(dma_chan->device->dev,
953 			"Self-test copy failed compare, disabling\n");
954 		err = -ENODEV;
955 		goto free_resources;
956 	}
957 
958 free_resources:
959 	mv_xor_free_chan_resources(dma_chan);
960 out:
961 	kfree(src);
962 	kfree(dest);
963 	return err;
964 }
965 
966 #define MV_XOR_NUM_SRC_TEST 4 /* must be <= 15 */
967 static int
968 mv_xor_xor_self_test(struct mv_xor_chan *mv_chan)
969 {
970 	int i, src_idx;
971 	struct page *dest;
972 	struct page *xor_srcs[MV_XOR_NUM_SRC_TEST];
973 	dma_addr_t dma_srcs[MV_XOR_NUM_SRC_TEST];
974 	dma_addr_t dest_dma;
975 	struct dma_async_tx_descriptor *tx;
976 	struct dma_chan *dma_chan;
977 	dma_cookie_t cookie;
978 	u8 cmp_byte = 0;
979 	u32 cmp_word;
980 	int err = 0;
981 
982 	for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++) {
983 		xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
984 		if (!xor_srcs[src_idx]) {
985 			while (src_idx--)
986 				__free_page(xor_srcs[src_idx]);
987 			return -ENOMEM;
988 		}
989 	}
990 
991 	dest = alloc_page(GFP_KERNEL);
992 	if (!dest) {
993 		while (src_idx--)
994 			__free_page(xor_srcs[src_idx]);
995 		return -ENOMEM;
996 	}
997 
998 	/* Fill in src buffers */
999 	for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++) {
1000 		u8 *ptr = page_address(xor_srcs[src_idx]);
1001 		for (i = 0; i < PAGE_SIZE; i++)
1002 			ptr[i] = (1 << src_idx);
1003 	}
1004 
1005 	for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++)
1006 		cmp_byte ^= (u8) (1 << src_idx);
1007 
1008 	cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
1009 		(cmp_byte << 8) | cmp_byte;
1010 
1011 	memset(page_address(dest), 0, PAGE_SIZE);
1012 
1013 	dma_chan = &mv_chan->dmachan;
1014 	if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
1015 		err = -ENODEV;
1016 		goto out;
1017 	}
1018 
1019 	/* test xor */
1020 	dest_dma = dma_map_page(dma_chan->device->dev, dest, 0, PAGE_SIZE,
1021 				DMA_FROM_DEVICE);
1022 
1023 	for (i = 0; i < MV_XOR_NUM_SRC_TEST; i++)
1024 		dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
1025 					   0, PAGE_SIZE, DMA_TO_DEVICE);
1026 
1027 	tx = mv_xor_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
1028 				 MV_XOR_NUM_SRC_TEST, PAGE_SIZE, 0);
1029 
1030 	cookie = mv_xor_tx_submit(tx);
1031 	mv_xor_issue_pending(dma_chan);
1032 	async_tx_ack(tx);
1033 	msleep(8);
1034 
1035 	if (mv_xor_status(dma_chan, cookie, NULL) !=
1036 	    DMA_SUCCESS) {
1037 		dev_err(dma_chan->device->dev,
1038 			"Self-test xor timed out, disabling\n");
1039 		err = -ENODEV;
1040 		goto free_resources;
1041 	}
1042 
1043 	dma_sync_single_for_cpu(dma_chan->device->dev, dest_dma,
1044 				PAGE_SIZE, DMA_FROM_DEVICE);
1045 	for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1046 		u32 *ptr = page_address(dest);
1047 		if (ptr[i] != cmp_word) {
1048 			dev_err(dma_chan->device->dev,
1049 				"Self-test xor failed compare, disabling. index %d, data %x, expected %x\n",
1050 				i, ptr[i], cmp_word);
1051 			err = -ENODEV;
1052 			goto free_resources;
1053 		}
1054 	}
1055 
1056 free_resources:
1057 	mv_xor_free_chan_resources(dma_chan);
1058 out:
1059 	src_idx = MV_XOR_NUM_SRC_TEST;
1060 	while (src_idx--)
1061 		__free_page(xor_srcs[src_idx]);
1062 	__free_page(dest);
1063 	return err;
1064 }
1065 
1066 /* This driver does not implement any of the optional DMA operations. */
1067 static int
1068 mv_xor_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
1069 	       unsigned long arg)
1070 {
1071 	return -ENOSYS;
1072 }
1073 
1074 static int mv_xor_channel_remove(struct mv_xor_chan *mv_chan)
1075 {
1076 	struct dma_chan *chan, *_chan;
1077 	struct device *dev = mv_chan->dmadev.dev;
1078 
1079 	dma_async_device_unregister(&mv_chan->dmadev);
1080 
1081 	dma_free_coherent(dev, MV_XOR_POOL_SIZE,
1082 			  mv_chan->dma_desc_pool_virt, mv_chan->dma_desc_pool);
1083 
1084 	list_for_each_entry_safe(chan, _chan, &mv_chan->dmadev.channels,
1085 				 device_node) {
1086 		list_del(&chan->device_node);
1087 	}
1088 
1089 	free_irq(mv_chan->irq, mv_chan);
1090 
1091 	return 0;
1092 }
1093 
1094 static struct mv_xor_chan *
1095 mv_xor_channel_add(struct mv_xor_device *xordev,
1096 		   struct platform_device *pdev,
1097 		   int idx, dma_cap_mask_t cap_mask, int irq)
1098 {
1099 	int ret = 0;
1100 	struct mv_xor_chan *mv_chan;
1101 	struct dma_device *dma_dev;
1102 
1103 	mv_chan = devm_kzalloc(&pdev->dev, sizeof(*mv_chan), GFP_KERNEL);
1104 	if (!mv_chan) {
1105 		ret = -ENOMEM;
1106 		goto err_free_dma;
1107 	}
1108 
1109 	mv_chan->idx = idx;
1110 	mv_chan->irq = irq;
1111 
1112 	dma_dev = &mv_chan->dmadev;
1113 
1114 	/* allocate coherent memory for hardware descriptors
1115 	 * note: writecombine gives slightly better performance, but
1116 	 * requires that we explicitly flush the writes
1117 	 */
1118 	mv_chan->dma_desc_pool_virt =
1119 	  dma_alloc_writecombine(&pdev->dev, MV_XOR_POOL_SIZE,
1120 				 &mv_chan->dma_desc_pool, GFP_KERNEL);
1121 	if (!mv_chan->dma_desc_pool_virt)
1122 		return ERR_PTR(-ENOMEM);
1123 
1124 	/* discover transaction capabilites from the platform data */
1125 	dma_dev->cap_mask = cap_mask;
1126 
1127 	INIT_LIST_HEAD(&dma_dev->channels);
1128 
1129 	/* set base routines */
1130 	dma_dev->device_alloc_chan_resources = mv_xor_alloc_chan_resources;
1131 	dma_dev->device_free_chan_resources = mv_xor_free_chan_resources;
1132 	dma_dev->device_tx_status = mv_xor_status;
1133 	dma_dev->device_issue_pending = mv_xor_issue_pending;
1134 	dma_dev->device_control = mv_xor_control;
1135 	dma_dev->dev = &pdev->dev;
1136 
1137 	/* set prep routines based on capability */
1138 	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1139 		dma_dev->device_prep_dma_memcpy = mv_xor_prep_dma_memcpy;
1140 	if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1141 		dma_dev->device_prep_dma_memset = mv_xor_prep_dma_memset;
1142 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1143 		dma_dev->max_xor = 8;
1144 		dma_dev->device_prep_dma_xor = mv_xor_prep_dma_xor;
1145 	}
1146 
1147 	mv_chan->mmr_base = xordev->xor_base;
1148 	if (!mv_chan->mmr_base) {
1149 		ret = -ENOMEM;
1150 		goto err_free_dma;
1151 	}
1152 	tasklet_init(&mv_chan->irq_tasklet, mv_xor_tasklet, (unsigned long)
1153 		     mv_chan);
1154 
1155 	/* clear errors before enabling interrupts */
1156 	mv_xor_device_clear_err_status(mv_chan);
1157 
1158 	ret = request_irq(mv_chan->irq, mv_xor_interrupt_handler,
1159 			  0, dev_name(&pdev->dev), mv_chan);
1160 	if (ret)
1161 		goto err_free_dma;
1162 
1163 	mv_chan_unmask_interrupts(mv_chan);
1164 
1165 	mv_set_mode(mv_chan, DMA_MEMCPY);
1166 
1167 	spin_lock_init(&mv_chan->lock);
1168 	INIT_LIST_HEAD(&mv_chan->chain);
1169 	INIT_LIST_HEAD(&mv_chan->completed_slots);
1170 	INIT_LIST_HEAD(&mv_chan->all_slots);
1171 	mv_chan->dmachan.device = dma_dev;
1172 	dma_cookie_init(&mv_chan->dmachan);
1173 
1174 	list_add_tail(&mv_chan->dmachan.device_node, &dma_dev->channels);
1175 
1176 	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1177 		ret = mv_xor_memcpy_self_test(mv_chan);
1178 		dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1179 		if (ret)
1180 			goto err_free_irq;
1181 	}
1182 
1183 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1184 		ret = mv_xor_xor_self_test(mv_chan);
1185 		dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1186 		if (ret)
1187 			goto err_free_irq;
1188 	}
1189 
1190 	dev_info(&pdev->dev, "Marvell XOR: ( %s%s%s%s)\n",
1191 		 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1192 		 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)  ? "fill " : "",
1193 		 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1194 		 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1195 
1196 	dma_async_device_register(dma_dev);
1197 	return mv_chan;
1198 
1199 err_free_irq:
1200 	free_irq(mv_chan->irq, mv_chan);
1201  err_free_dma:
1202 	dma_free_coherent(&pdev->dev, MV_XOR_POOL_SIZE,
1203 			  mv_chan->dma_desc_pool_virt, mv_chan->dma_desc_pool);
1204 	return ERR_PTR(ret);
1205 }
1206 
1207 static void
1208 mv_xor_conf_mbus_windows(struct mv_xor_device *xordev,
1209 			 const struct mbus_dram_target_info *dram)
1210 {
1211 	void __iomem *base = xordev->xor_base;
1212 	u32 win_enable = 0;
1213 	int i;
1214 
1215 	for (i = 0; i < 8; i++) {
1216 		writel(0, base + WINDOW_BASE(i));
1217 		writel(0, base + WINDOW_SIZE(i));
1218 		if (i < 4)
1219 			writel(0, base + WINDOW_REMAP_HIGH(i));
1220 	}
1221 
1222 	for (i = 0; i < dram->num_cs; i++) {
1223 		const struct mbus_dram_window *cs = dram->cs + i;
1224 
1225 		writel((cs->base & 0xffff0000) |
1226 		       (cs->mbus_attr << 8) |
1227 		       dram->mbus_dram_target_id, base + WINDOW_BASE(i));
1228 		writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
1229 
1230 		win_enable |= (1 << i);
1231 		win_enable |= 3 << (16 + (2 * i));
1232 	}
1233 
1234 	writel(win_enable, base + WINDOW_BAR_ENABLE(0));
1235 	writel(win_enable, base + WINDOW_BAR_ENABLE(1));
1236 	writel(0, base + WINDOW_OVERRIDE_CTRL(0));
1237 	writel(0, base + WINDOW_OVERRIDE_CTRL(1));
1238 }
1239 
1240 static int mv_xor_probe(struct platform_device *pdev)
1241 {
1242 	const struct mbus_dram_target_info *dram;
1243 	struct mv_xor_device *xordev;
1244 	struct mv_xor_platform_data *pdata = pdev->dev.platform_data;
1245 	struct resource *res;
1246 	int i, ret;
1247 
1248 	dev_notice(&pdev->dev, "Marvell shared XOR driver\n");
1249 
1250 	xordev = devm_kzalloc(&pdev->dev, sizeof(*xordev), GFP_KERNEL);
1251 	if (!xordev)
1252 		return -ENOMEM;
1253 
1254 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1255 	if (!res)
1256 		return -ENODEV;
1257 
1258 	xordev->xor_base = devm_ioremap(&pdev->dev, res->start,
1259 					resource_size(res));
1260 	if (!xordev->xor_base)
1261 		return -EBUSY;
1262 
1263 	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1264 	if (!res)
1265 		return -ENODEV;
1266 
1267 	xordev->xor_high_base = devm_ioremap(&pdev->dev, res->start,
1268 					     resource_size(res));
1269 	if (!xordev->xor_high_base)
1270 		return -EBUSY;
1271 
1272 	platform_set_drvdata(pdev, xordev);
1273 
1274 	/*
1275 	 * (Re-)program MBUS remapping windows if we are asked to.
1276 	 */
1277 	dram = mv_mbus_dram_info();
1278 	if (dram)
1279 		mv_xor_conf_mbus_windows(xordev, dram);
1280 
1281 	/* Not all platforms can gate the clock, so it is not
1282 	 * an error if the clock does not exists.
1283 	 */
1284 	xordev->clk = clk_get(&pdev->dev, NULL);
1285 	if (!IS_ERR(xordev->clk))
1286 		clk_prepare_enable(xordev->clk);
1287 
1288 	if (pdev->dev.of_node) {
1289 		struct device_node *np;
1290 		int i = 0;
1291 
1292 		for_each_child_of_node(pdev->dev.of_node, np) {
1293 			dma_cap_mask_t cap_mask;
1294 			int irq;
1295 
1296 			dma_cap_zero(cap_mask);
1297 			if (of_property_read_bool(np, "dmacap,memcpy"))
1298 				dma_cap_set(DMA_MEMCPY, cap_mask);
1299 			if (of_property_read_bool(np, "dmacap,xor"))
1300 				dma_cap_set(DMA_XOR, cap_mask);
1301 			if (of_property_read_bool(np, "dmacap,memset"))
1302 				dma_cap_set(DMA_MEMSET, cap_mask);
1303 			if (of_property_read_bool(np, "dmacap,interrupt"))
1304 				dma_cap_set(DMA_INTERRUPT, cap_mask);
1305 
1306 			irq = irq_of_parse_and_map(np, 0);
1307 			if (!irq) {
1308 				ret = -ENODEV;
1309 				goto err_channel_add;
1310 			}
1311 
1312 			xordev->channels[i] =
1313 				mv_xor_channel_add(xordev, pdev, i,
1314 						   cap_mask, irq);
1315 			if (IS_ERR(xordev->channels[i])) {
1316 				ret = PTR_ERR(xordev->channels[i]);
1317 				xordev->channels[i] = NULL;
1318 				irq_dispose_mapping(irq);
1319 				goto err_channel_add;
1320 			}
1321 
1322 			i++;
1323 		}
1324 	} else if (pdata && pdata->channels) {
1325 		for (i = 0; i < MV_XOR_MAX_CHANNELS; i++) {
1326 			struct mv_xor_channel_data *cd;
1327 			int irq;
1328 
1329 			cd = &pdata->channels[i];
1330 			if (!cd) {
1331 				ret = -ENODEV;
1332 				goto err_channel_add;
1333 			}
1334 
1335 			irq = platform_get_irq(pdev, i);
1336 			if (irq < 0) {
1337 				ret = irq;
1338 				goto err_channel_add;
1339 			}
1340 
1341 			xordev->channels[i] =
1342 				mv_xor_channel_add(xordev, pdev, i,
1343 						   cd->cap_mask, irq);
1344 			if (IS_ERR(xordev->channels[i])) {
1345 				ret = PTR_ERR(xordev->channels[i]);
1346 				goto err_channel_add;
1347 			}
1348 		}
1349 	}
1350 
1351 	return 0;
1352 
1353 err_channel_add:
1354 	for (i = 0; i < MV_XOR_MAX_CHANNELS; i++)
1355 		if (xordev->channels[i]) {
1356 			mv_xor_channel_remove(xordev->channels[i]);
1357 			if (pdev->dev.of_node)
1358 				irq_dispose_mapping(xordev->channels[i]->irq);
1359 		}
1360 
1361 	if (!IS_ERR(xordev->clk)) {
1362 		clk_disable_unprepare(xordev->clk);
1363 		clk_put(xordev->clk);
1364 	}
1365 
1366 	return ret;
1367 }
1368 
1369 static int mv_xor_remove(struct platform_device *pdev)
1370 {
1371 	struct mv_xor_device *xordev = platform_get_drvdata(pdev);
1372 	int i;
1373 
1374 	for (i = 0; i < MV_XOR_MAX_CHANNELS; i++) {
1375 		if (xordev->channels[i])
1376 			mv_xor_channel_remove(xordev->channels[i]);
1377 	}
1378 
1379 	if (!IS_ERR(xordev->clk)) {
1380 		clk_disable_unprepare(xordev->clk);
1381 		clk_put(xordev->clk);
1382 	}
1383 
1384 	return 0;
1385 }
1386 
1387 #ifdef CONFIG_OF
1388 static struct of_device_id mv_xor_dt_ids[] = {
1389        { .compatible = "marvell,orion-xor", },
1390        {},
1391 };
1392 MODULE_DEVICE_TABLE(of, mv_xor_dt_ids);
1393 #endif
1394 
1395 static struct platform_driver mv_xor_driver = {
1396 	.probe		= mv_xor_probe,
1397 	.remove		= mv_xor_remove,
1398 	.driver		= {
1399 		.owner	        = THIS_MODULE,
1400 		.name	        = MV_XOR_NAME,
1401 		.of_match_table = of_match_ptr(mv_xor_dt_ids),
1402 	},
1403 };
1404 
1405 
1406 static int __init mv_xor_init(void)
1407 {
1408 	return platform_driver_register(&mv_xor_driver);
1409 }
1410 module_init(mv_xor_init);
1411 
1412 /* it's currently unsafe to unload this module */
1413 #if 0
1414 static void __exit mv_xor_exit(void)
1415 {
1416 	platform_driver_unregister(&mv_xor_driver);
1417 	return;
1418 }
1419 
1420 module_exit(mv_xor_exit);
1421 #endif
1422 
1423 MODULE_AUTHOR("Saeed Bishara <saeed@marvell.com>");
1424 MODULE_DESCRIPTION("DMA engine driver for Marvell's XOR engine");
1425 MODULE_LICENSE("GPL");
1426