• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * offload engine driver for the Intel Xscale series of i/o processors
4  * Copyright © 2006, Intel Corporation.
5  */
6 
7 /*
8  * This driver supports the asynchrounous DMA copy and RAID engines available
9  * on the Intel Xscale(R) family of I/O Processors (IOP 32x, 33x, 134x)
10  */
11 
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/delay.h>
15 #include <linux/dma-mapping.h>
16 #include <linux/spinlock.h>
17 #include <linux/interrupt.h>
18 #include <linux/platform_device.h>
19 #include <linux/prefetch.h>
20 #include <linux/memory.h>
21 #include <linux/ioport.h>
22 #include <linux/raid/pq.h>
23 #include <linux/slab.h>
24 
25 #include "iop-adma.h"
26 #include "dmaengine.h"
27 
28 #define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
29 #define to_iop_adma_device(dev) \
30 	container_of(dev, struct iop_adma_device, common)
31 #define tx_to_iop_adma_slot(tx) \
32 	container_of(tx, struct iop_adma_desc_slot, async_tx)
33 
34 /**
35  * iop_adma_free_slots - flags descriptor slots for reuse
36  * @slot: Slot to free
37  * Caller must hold &iop_chan->lock while calling this function
38  */
iop_adma_free_slots(struct iop_adma_desc_slot * slot)39 static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
40 {
41 	int stride = slot->slots_per_op;
42 
43 	while (stride--) {
44 		slot->slots_per_op = 0;
45 		slot = list_entry(slot->slot_node.next,
46 				struct iop_adma_desc_slot,
47 				slot_node);
48 	}
49 }
50 
51 static dma_cookie_t
iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot * desc,struct iop_adma_chan * iop_chan,dma_cookie_t cookie)52 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
53 	struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
54 {
55 	struct dma_async_tx_descriptor *tx = &desc->async_tx;
56 
57 	BUG_ON(tx->cookie < 0);
58 	if (tx->cookie > 0) {
59 		cookie = tx->cookie;
60 		tx->cookie = 0;
61 
62 		/* call the callback (must not sleep or submit new
63 		 * operations to this channel)
64 		 */
65 		dmaengine_desc_get_callback_invoke(tx, NULL);
66 
67 		dma_descriptor_unmap(tx);
68 		if (desc->group_head)
69 			desc->group_head = NULL;
70 	}
71 
72 	/* run dependent operations */
73 	dma_run_dependencies(tx);
74 
75 	return cookie;
76 }
77 
78 static int
iop_adma_clean_slot(struct iop_adma_desc_slot * desc,struct iop_adma_chan * iop_chan)79 iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
80 	struct iop_adma_chan *iop_chan)
81 {
82 	/* the client is allowed to attach dependent operations
83 	 * until 'ack' is set
84 	 */
85 	if (!async_tx_test_ack(&desc->async_tx))
86 		return 0;
87 
88 	/* leave the last descriptor in the chain
89 	 * so we can append to it
90 	 */
91 	if (desc->chain_node.next == &iop_chan->chain)
92 		return 1;
93 
94 	dev_dbg(iop_chan->device->common.dev,
95 		"\tfree slot: %d slots_per_op: %d\n",
96 		desc->idx, desc->slots_per_op);
97 
98 	list_del(&desc->chain_node);
99 	iop_adma_free_slots(desc);
100 
101 	return 0;
102 }
103 
__iop_adma_slot_cleanup(struct iop_adma_chan * iop_chan)104 static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
105 {
106 	struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
107 	dma_cookie_t cookie = 0;
108 	u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
109 	int busy = iop_chan_is_busy(iop_chan);
110 	int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
111 
112 	dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
113 	/* free completed slots from the chain starting with
114 	 * the oldest descriptor
115 	 */
116 	list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
117 					chain_node) {
118 		pr_debug("\tcookie: %d slot: %d busy: %d "
119 			"this_desc: %pad next_desc: %#llx ack: %d\n",
120 			iter->async_tx.cookie, iter->idx, busy,
121 			&iter->async_tx.phys, (u64)iop_desc_get_next_desc(iter),
122 			async_tx_test_ack(&iter->async_tx));
123 		prefetch(_iter);
124 		prefetch(&_iter->async_tx);
125 
126 		/* do not advance past the current descriptor loaded into the
127 		 * hardware channel, subsequent descriptors are either in
128 		 * process or have not been submitted
129 		 */
130 		if (seen_current)
131 			break;
132 
133 		/* stop the search if we reach the current descriptor and the
134 		 * channel is busy, or if it appears that the current descriptor
135 		 * needs to be re-read (i.e. has been appended to)
136 		 */
137 		if (iter->async_tx.phys == current_desc) {
138 			BUG_ON(seen_current++);
139 			if (busy || iop_desc_get_next_desc(iter))
140 				break;
141 		}
142 
143 		/* detect the start of a group transaction */
144 		if (!slot_cnt && !slots_per_op) {
145 			slot_cnt = iter->slot_cnt;
146 			slots_per_op = iter->slots_per_op;
147 			if (slot_cnt <= slots_per_op) {
148 				slot_cnt = 0;
149 				slots_per_op = 0;
150 			}
151 		}
152 
153 		if (slot_cnt) {
154 			pr_debug("\tgroup++\n");
155 			if (!grp_start)
156 				grp_start = iter;
157 			slot_cnt -= slots_per_op;
158 		}
159 
160 		/* all the members of a group are complete */
161 		if (slots_per_op != 0 && slot_cnt == 0) {
162 			struct iop_adma_desc_slot *grp_iter, *_grp_iter;
163 			int end_of_chain = 0;
164 			pr_debug("\tgroup end\n");
165 
166 			/* collect the total results */
167 			if (grp_start->xor_check_result) {
168 				u32 zero_sum_result = 0;
169 				slot_cnt = grp_start->slot_cnt;
170 				grp_iter = grp_start;
171 
172 				list_for_each_entry_from(grp_iter,
173 					&iop_chan->chain, chain_node) {
174 					zero_sum_result |=
175 					    iop_desc_get_zero_result(grp_iter);
176 					    pr_debug("\titer%d result: %d\n",
177 					    grp_iter->idx, zero_sum_result);
178 					slot_cnt -= slots_per_op;
179 					if (slot_cnt == 0)
180 						break;
181 				}
182 				pr_debug("\tgrp_start->xor_check_result: %p\n",
183 					grp_start->xor_check_result);
184 				*grp_start->xor_check_result = zero_sum_result;
185 			}
186 
187 			/* clean up the group */
188 			slot_cnt = grp_start->slot_cnt;
189 			grp_iter = grp_start;
190 			list_for_each_entry_safe_from(grp_iter, _grp_iter,
191 				&iop_chan->chain, chain_node) {
192 				cookie = iop_adma_run_tx_complete_actions(
193 					grp_iter, iop_chan, cookie);
194 
195 				slot_cnt -= slots_per_op;
196 				end_of_chain = iop_adma_clean_slot(grp_iter,
197 					iop_chan);
198 
199 				if (slot_cnt == 0 || end_of_chain)
200 					break;
201 			}
202 
203 			/* the group should be complete at this point */
204 			BUG_ON(slot_cnt);
205 
206 			slots_per_op = 0;
207 			grp_start = NULL;
208 			if (end_of_chain)
209 				break;
210 			else
211 				continue;
212 		} else if (slots_per_op) /* wait for group completion */
213 			continue;
214 
215 		/* write back zero sum results (single descriptor case) */
216 		if (iter->xor_check_result && iter->async_tx.cookie)
217 			*iter->xor_check_result =
218 				iop_desc_get_zero_result(iter);
219 
220 		cookie = iop_adma_run_tx_complete_actions(
221 					iter, iop_chan, cookie);
222 
223 		if (iop_adma_clean_slot(iter, iop_chan))
224 			break;
225 	}
226 
227 	if (cookie > 0) {
228 		iop_chan->common.completed_cookie = cookie;
229 		pr_debug("\tcompleted cookie %d\n", cookie);
230 	}
231 }
232 
233 static void
iop_adma_slot_cleanup(struct iop_adma_chan * iop_chan)234 iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
235 {
236 	spin_lock_bh(&iop_chan->lock);
237 	__iop_adma_slot_cleanup(iop_chan);
238 	spin_unlock_bh(&iop_chan->lock);
239 }
240 
iop_adma_tasklet(unsigned long data)241 static void iop_adma_tasklet(unsigned long data)
242 {
243 	struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data;
244 
245 	/* lockdep will flag depedency submissions as potentially
246 	 * recursive locking, this is not the case as a dependency
247 	 * submission will never recurse a channels submit routine.
248 	 * There are checks in async_tx.c to prevent this.
249 	 */
250 	spin_lock_nested(&iop_chan->lock, SINGLE_DEPTH_NESTING);
251 	__iop_adma_slot_cleanup(iop_chan);
252 	spin_unlock(&iop_chan->lock);
253 }
254 
255 static struct iop_adma_desc_slot *
iop_adma_alloc_slots(struct iop_adma_chan * iop_chan,int num_slots,int slots_per_op)256 iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
257 			int slots_per_op)
258 {
259 	struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
260 	LIST_HEAD(chain);
261 	int slots_found, retry = 0;
262 
263 	/* start search from the last allocated descrtiptor
264 	 * if a contiguous allocation can not be found start searching
265 	 * from the beginning of the list
266 	 */
267 retry:
268 	slots_found = 0;
269 	if (retry == 0)
270 		iter = iop_chan->last_used;
271 	else
272 		iter = list_entry(&iop_chan->all_slots,
273 			struct iop_adma_desc_slot,
274 			slot_node);
275 
276 	list_for_each_entry_safe_continue(
277 		iter, _iter, &iop_chan->all_slots, slot_node) {
278 		prefetch(_iter);
279 		prefetch(&_iter->async_tx);
280 		if (iter->slots_per_op) {
281 			/* give up after finding the first busy slot
282 			 * on the second pass through the list
283 			 */
284 			if (retry)
285 				break;
286 
287 			slots_found = 0;
288 			continue;
289 		}
290 
291 		/* start the allocation if the slot is correctly aligned */
292 		if (!slots_found++) {
293 			if (iop_desc_is_aligned(iter, slots_per_op))
294 				alloc_start = iter;
295 			else {
296 				slots_found = 0;
297 				continue;
298 			}
299 		}
300 
301 		if (slots_found == num_slots) {
302 			struct iop_adma_desc_slot *alloc_tail = NULL;
303 			struct iop_adma_desc_slot *last_used = NULL;
304 			iter = alloc_start;
305 			while (num_slots) {
306 				int i;
307 				dev_dbg(iop_chan->device->common.dev,
308 					"allocated slot: %d "
309 					"(desc %p phys: %#llx) slots_per_op %d\n",
310 					iter->idx, iter->hw_desc,
311 					(u64)iter->async_tx.phys, slots_per_op);
312 
313 				/* pre-ack all but the last descriptor */
314 				if (num_slots != slots_per_op)
315 					async_tx_ack(&iter->async_tx);
316 
317 				list_add_tail(&iter->chain_node, &chain);
318 				alloc_tail = iter;
319 				iter->async_tx.cookie = 0;
320 				iter->slot_cnt = num_slots;
321 				iter->xor_check_result = NULL;
322 				for (i = 0; i < slots_per_op; i++) {
323 					iter->slots_per_op = slots_per_op - i;
324 					last_used = iter;
325 					iter = list_entry(iter->slot_node.next,
326 						struct iop_adma_desc_slot,
327 						slot_node);
328 				}
329 				num_slots -= slots_per_op;
330 			}
331 			alloc_tail->group_head = alloc_start;
332 			alloc_tail->async_tx.cookie = -EBUSY;
333 			list_splice(&chain, &alloc_tail->tx_list);
334 			iop_chan->last_used = last_used;
335 			iop_desc_clear_next_desc(alloc_start);
336 			iop_desc_clear_next_desc(alloc_tail);
337 			return alloc_tail;
338 		}
339 	}
340 	if (!retry++)
341 		goto retry;
342 
343 	/* perform direct reclaim if the allocation fails */
344 	__iop_adma_slot_cleanup(iop_chan);
345 
346 	return NULL;
347 }
348 
iop_adma_check_threshold(struct iop_adma_chan * iop_chan)349 static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
350 {
351 	dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
352 		iop_chan->pending);
353 
354 	if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
355 		iop_chan->pending = 0;
356 		iop_chan_append(iop_chan);
357 	}
358 }
359 
360 static dma_cookie_t
iop_adma_tx_submit(struct dma_async_tx_descriptor * tx)361 iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
362 {
363 	struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
364 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
365 	struct iop_adma_desc_slot *grp_start, *old_chain_tail;
366 	int slot_cnt;
367 	dma_cookie_t cookie;
368 	dma_addr_t next_dma;
369 
370 	grp_start = sw_desc->group_head;
371 	slot_cnt = grp_start->slot_cnt;
372 
373 	spin_lock_bh(&iop_chan->lock);
374 	cookie = dma_cookie_assign(tx);
375 
376 	old_chain_tail = list_entry(iop_chan->chain.prev,
377 		struct iop_adma_desc_slot, chain_node);
378 	list_splice_init(&sw_desc->tx_list,
379 			 &old_chain_tail->chain_node);
380 
381 	/* fix up the hardware chain */
382 	next_dma = grp_start->async_tx.phys;
383 	iop_desc_set_next_desc(old_chain_tail, next_dma);
384 	BUG_ON(iop_desc_get_next_desc(old_chain_tail) != next_dma); /* flush */
385 
386 	/* check for pre-chained descriptors */
387 	iop_paranoia(iop_desc_get_next_desc(sw_desc));
388 
389 	/* increment the pending count by the number of slots
390 	 * memcpy operations have a 1:1 (slot:operation) relation
391 	 * other operations are heavier and will pop the threshold
392 	 * more often.
393 	 */
394 	iop_chan->pending += slot_cnt;
395 	iop_adma_check_threshold(iop_chan);
396 	spin_unlock_bh(&iop_chan->lock);
397 
398 	dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
399 		__func__, sw_desc->async_tx.cookie, sw_desc->idx);
400 
401 	return cookie;
402 }
403 
404 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
405 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
406 
407 /**
408  * iop_adma_alloc_chan_resources -  returns the number of allocated descriptors
409  * @chan - allocate descriptor resources for this channel
410  * @client - current client requesting the channel be ready for requests
411  *
412  * Note: We keep the slots for 1 operation on iop_chan->chain at all times.  To
413  * avoid deadlock, via async_xor, num_descs_in_pool must at a minimum be
414  * greater than 2x the number slots needed to satisfy a device->max_xor
415  * request.
416  * */
iop_adma_alloc_chan_resources(struct dma_chan * chan)417 static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
418 {
419 	char *hw_desc;
420 	int idx;
421 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
422 	struct iop_adma_desc_slot *slot = NULL;
423 	int init = iop_chan->slots_allocated ? 0 : 1;
424 	struct iop_adma_platform_data *plat_data =
425 		dev_get_platdata(&iop_chan->device->pdev->dev);
426 	int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
427 
428 	/* Allocate descriptor slots */
429 	do {
430 		idx = iop_chan->slots_allocated;
431 		if (idx == num_descs_in_pool)
432 			break;
433 
434 		slot = kzalloc(sizeof(*slot), GFP_KERNEL);
435 		if (!slot) {
436 			printk(KERN_INFO "IOP ADMA Channel only initialized"
437 				" %d descriptor slots", idx);
438 			break;
439 		}
440 		hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
441 		slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
442 
443 		dma_async_tx_descriptor_init(&slot->async_tx, chan);
444 		slot->async_tx.tx_submit = iop_adma_tx_submit;
445 		INIT_LIST_HEAD(&slot->tx_list);
446 		INIT_LIST_HEAD(&slot->chain_node);
447 		INIT_LIST_HEAD(&slot->slot_node);
448 		hw_desc = (char *) iop_chan->device->dma_desc_pool;
449 		slot->async_tx.phys =
450 			(dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
451 		slot->idx = idx;
452 
453 		spin_lock_bh(&iop_chan->lock);
454 		iop_chan->slots_allocated++;
455 		list_add_tail(&slot->slot_node, &iop_chan->all_slots);
456 		spin_unlock_bh(&iop_chan->lock);
457 	} while (iop_chan->slots_allocated < num_descs_in_pool);
458 
459 	if (idx && !iop_chan->last_used)
460 		iop_chan->last_used = list_entry(iop_chan->all_slots.next,
461 					struct iop_adma_desc_slot,
462 					slot_node);
463 
464 	dev_dbg(iop_chan->device->common.dev,
465 		"allocated %d descriptor slots last_used: %p\n",
466 		iop_chan->slots_allocated, iop_chan->last_used);
467 
468 	/* initialize the channel and the chain with a null operation */
469 	if (init) {
470 		if (dma_has_cap(DMA_MEMCPY,
471 			iop_chan->device->common.cap_mask))
472 			iop_chan_start_null_memcpy(iop_chan);
473 		else if (dma_has_cap(DMA_XOR,
474 			iop_chan->device->common.cap_mask))
475 			iop_chan_start_null_xor(iop_chan);
476 		else
477 			BUG();
478 	}
479 
480 	return (idx > 0) ? idx : -ENOMEM;
481 }
482 
483 static struct dma_async_tx_descriptor *
iop_adma_prep_dma_interrupt(struct dma_chan * chan,unsigned long flags)484 iop_adma_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags)
485 {
486 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
487 	struct iop_adma_desc_slot *sw_desc, *grp_start;
488 	int slot_cnt, slots_per_op;
489 
490 	dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
491 
492 	spin_lock_bh(&iop_chan->lock);
493 	slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
494 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
495 	if (sw_desc) {
496 		grp_start = sw_desc->group_head;
497 		iop_desc_init_interrupt(grp_start, iop_chan);
498 		sw_desc->async_tx.flags = flags;
499 	}
500 	spin_unlock_bh(&iop_chan->lock);
501 
502 	return sw_desc ? &sw_desc->async_tx : NULL;
503 }
504 
505 static struct dma_async_tx_descriptor *
iop_adma_prep_dma_memcpy(struct dma_chan * chan,dma_addr_t dma_dest,dma_addr_t dma_src,size_t len,unsigned long flags)506 iop_adma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
507 			 dma_addr_t dma_src, size_t len, unsigned long flags)
508 {
509 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
510 	struct iop_adma_desc_slot *sw_desc, *grp_start;
511 	int slot_cnt, slots_per_op;
512 
513 	if (unlikely(!len))
514 		return NULL;
515 	BUG_ON(len > IOP_ADMA_MAX_BYTE_COUNT);
516 
517 	dev_dbg(iop_chan->device->common.dev, "%s len: %zu\n",
518 		__func__, len);
519 
520 	spin_lock_bh(&iop_chan->lock);
521 	slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
522 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
523 	if (sw_desc) {
524 		grp_start = sw_desc->group_head;
525 		iop_desc_init_memcpy(grp_start, flags);
526 		iop_desc_set_byte_count(grp_start, iop_chan, len);
527 		iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
528 		iop_desc_set_memcpy_src_addr(grp_start, dma_src);
529 		sw_desc->async_tx.flags = flags;
530 	}
531 	spin_unlock_bh(&iop_chan->lock);
532 
533 	return sw_desc ? &sw_desc->async_tx : NULL;
534 }
535 
536 static struct dma_async_tx_descriptor *
iop_adma_prep_dma_xor(struct dma_chan * chan,dma_addr_t dma_dest,dma_addr_t * dma_src,unsigned int src_cnt,size_t len,unsigned long flags)537 iop_adma_prep_dma_xor(struct dma_chan *chan, dma_addr_t dma_dest,
538 		      dma_addr_t *dma_src, unsigned int src_cnt, size_t len,
539 		      unsigned long flags)
540 {
541 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
542 	struct iop_adma_desc_slot *sw_desc, *grp_start;
543 	int slot_cnt, slots_per_op;
544 
545 	if (unlikely(!len))
546 		return NULL;
547 	BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
548 
549 	dev_dbg(iop_chan->device->common.dev,
550 		"%s src_cnt: %d len: %zu flags: %lx\n",
551 		__func__, src_cnt, len, flags);
552 
553 	spin_lock_bh(&iop_chan->lock);
554 	slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
555 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
556 	if (sw_desc) {
557 		grp_start = sw_desc->group_head;
558 		iop_desc_init_xor(grp_start, src_cnt, flags);
559 		iop_desc_set_byte_count(grp_start, iop_chan, len);
560 		iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
561 		sw_desc->async_tx.flags = flags;
562 		while (src_cnt--)
563 			iop_desc_set_xor_src_addr(grp_start, src_cnt,
564 						  dma_src[src_cnt]);
565 	}
566 	spin_unlock_bh(&iop_chan->lock);
567 
568 	return sw_desc ? &sw_desc->async_tx : NULL;
569 }
570 
571 static struct dma_async_tx_descriptor *
iop_adma_prep_dma_xor_val(struct dma_chan * chan,dma_addr_t * dma_src,unsigned int src_cnt,size_t len,u32 * result,unsigned long flags)572 iop_adma_prep_dma_xor_val(struct dma_chan *chan, dma_addr_t *dma_src,
573 			  unsigned int src_cnt, size_t len, u32 *result,
574 			  unsigned long flags)
575 {
576 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
577 	struct iop_adma_desc_slot *sw_desc, *grp_start;
578 	int slot_cnt, slots_per_op;
579 
580 	if (unlikely(!len))
581 		return NULL;
582 
583 	dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %zu\n",
584 		__func__, src_cnt, len);
585 
586 	spin_lock_bh(&iop_chan->lock);
587 	slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
588 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
589 	if (sw_desc) {
590 		grp_start = sw_desc->group_head;
591 		iop_desc_init_zero_sum(grp_start, src_cnt, flags);
592 		iop_desc_set_zero_sum_byte_count(grp_start, len);
593 		grp_start->xor_check_result = result;
594 		pr_debug("\t%s: grp_start->xor_check_result: %p\n",
595 			__func__, grp_start->xor_check_result);
596 		sw_desc->async_tx.flags = flags;
597 		while (src_cnt--)
598 			iop_desc_set_zero_sum_src_addr(grp_start, src_cnt,
599 						       dma_src[src_cnt]);
600 	}
601 	spin_unlock_bh(&iop_chan->lock);
602 
603 	return sw_desc ? &sw_desc->async_tx : NULL;
604 }
605 
606 static struct dma_async_tx_descriptor *
iop_adma_prep_dma_pq(struct dma_chan * chan,dma_addr_t * dst,dma_addr_t * src,unsigned int src_cnt,const unsigned char * scf,size_t len,unsigned long flags)607 iop_adma_prep_dma_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
608 		     unsigned int src_cnt, const unsigned char *scf, size_t len,
609 		     unsigned long flags)
610 {
611 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
612 	struct iop_adma_desc_slot *sw_desc, *g;
613 	int slot_cnt, slots_per_op;
614 	int continue_srcs;
615 
616 	if (unlikely(!len))
617 		return NULL;
618 	BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
619 
620 	dev_dbg(iop_chan->device->common.dev,
621 		"%s src_cnt: %d len: %zu flags: %lx\n",
622 		__func__, src_cnt, len, flags);
623 
624 	if (dmaf_p_disabled_continue(flags))
625 		continue_srcs = 1+src_cnt;
626 	else if (dmaf_continue(flags))
627 		continue_srcs = 3+src_cnt;
628 	else
629 		continue_srcs = 0+src_cnt;
630 
631 	spin_lock_bh(&iop_chan->lock);
632 	slot_cnt = iop_chan_pq_slot_count(len, continue_srcs, &slots_per_op);
633 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
634 	if (sw_desc) {
635 		int i;
636 
637 		g = sw_desc->group_head;
638 		iop_desc_set_byte_count(g, iop_chan, len);
639 
640 		/* even if P is disabled its destination address (bits
641 		 * [3:0]) must match Q.  It is ok if P points to an
642 		 * invalid address, it won't be written.
643 		 */
644 		if (flags & DMA_PREP_PQ_DISABLE_P)
645 			dst[0] = dst[1] & 0x7;
646 
647 		iop_desc_set_pq_addr(g, dst);
648 		sw_desc->async_tx.flags = flags;
649 		for (i = 0; i < src_cnt; i++)
650 			iop_desc_set_pq_src_addr(g, i, src[i], scf[i]);
651 
652 		/* if we are continuing a previous operation factor in
653 		 * the old p and q values, see the comment for dma_maxpq
654 		 * in include/linux/dmaengine.h
655 		 */
656 		if (dmaf_p_disabled_continue(flags))
657 			iop_desc_set_pq_src_addr(g, i++, dst[1], 1);
658 		else if (dmaf_continue(flags)) {
659 			iop_desc_set_pq_src_addr(g, i++, dst[0], 0);
660 			iop_desc_set_pq_src_addr(g, i++, dst[1], 1);
661 			iop_desc_set_pq_src_addr(g, i++, dst[1], 0);
662 		}
663 		iop_desc_init_pq(g, i, flags);
664 	}
665 	spin_unlock_bh(&iop_chan->lock);
666 
667 	return sw_desc ? &sw_desc->async_tx : NULL;
668 }
669 
670 static struct dma_async_tx_descriptor *
iop_adma_prep_dma_pq_val(struct dma_chan * chan,dma_addr_t * pq,dma_addr_t * src,unsigned int src_cnt,const unsigned char * scf,size_t len,enum sum_check_flags * pqres,unsigned long flags)671 iop_adma_prep_dma_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
672 			 unsigned int src_cnt, const unsigned char *scf,
673 			 size_t len, enum sum_check_flags *pqres,
674 			 unsigned long flags)
675 {
676 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
677 	struct iop_adma_desc_slot *sw_desc, *g;
678 	int slot_cnt, slots_per_op;
679 
680 	if (unlikely(!len))
681 		return NULL;
682 	BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
683 
684 	dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %zu\n",
685 		__func__, src_cnt, len);
686 
687 	spin_lock_bh(&iop_chan->lock);
688 	slot_cnt = iop_chan_pq_zero_sum_slot_count(len, src_cnt + 2, &slots_per_op);
689 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
690 	if (sw_desc) {
691 		/* for validate operations p and q are tagged onto the
692 		 * end of the source list
693 		 */
694 		int pq_idx = src_cnt;
695 
696 		g = sw_desc->group_head;
697 		iop_desc_init_pq_zero_sum(g, src_cnt+2, flags);
698 		iop_desc_set_pq_zero_sum_byte_count(g, len);
699 		g->pq_check_result = pqres;
700 		pr_debug("\t%s: g->pq_check_result: %p\n",
701 			__func__, g->pq_check_result);
702 		sw_desc->async_tx.flags = flags;
703 		while (src_cnt--)
704 			iop_desc_set_pq_zero_sum_src_addr(g, src_cnt,
705 							  src[src_cnt],
706 							  scf[src_cnt]);
707 		iop_desc_set_pq_zero_sum_addr(g, pq_idx, src);
708 	}
709 	spin_unlock_bh(&iop_chan->lock);
710 
711 	return sw_desc ? &sw_desc->async_tx : NULL;
712 }
713 
iop_adma_free_chan_resources(struct dma_chan * chan)714 static void iop_adma_free_chan_resources(struct dma_chan *chan)
715 {
716 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
717 	struct iop_adma_desc_slot *iter, *_iter;
718 	int in_use_descs = 0;
719 
720 	iop_adma_slot_cleanup(iop_chan);
721 
722 	spin_lock_bh(&iop_chan->lock);
723 	list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
724 					chain_node) {
725 		in_use_descs++;
726 		list_del(&iter->chain_node);
727 	}
728 	list_for_each_entry_safe_reverse(
729 		iter, _iter, &iop_chan->all_slots, slot_node) {
730 		list_del(&iter->slot_node);
731 		kfree(iter);
732 		iop_chan->slots_allocated--;
733 	}
734 	iop_chan->last_used = NULL;
735 
736 	dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
737 		__func__, iop_chan->slots_allocated);
738 	spin_unlock_bh(&iop_chan->lock);
739 
740 	/* one is ok since we left it on there on purpose */
741 	if (in_use_descs > 1)
742 		printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
743 			in_use_descs - 1);
744 }
745 
746 /**
747  * iop_adma_status - poll the status of an ADMA transaction
748  * @chan: ADMA channel handle
749  * @cookie: ADMA transaction identifier
750  * @txstate: a holder for the current state of the channel or NULL
751  */
iop_adma_status(struct dma_chan * chan,dma_cookie_t cookie,struct dma_tx_state * txstate)752 static enum dma_status iop_adma_status(struct dma_chan *chan,
753 					dma_cookie_t cookie,
754 					struct dma_tx_state *txstate)
755 {
756 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
757 	int ret;
758 
759 	ret = dma_cookie_status(chan, cookie, txstate);
760 	if (ret == DMA_COMPLETE)
761 		return ret;
762 
763 	iop_adma_slot_cleanup(iop_chan);
764 
765 	return dma_cookie_status(chan, cookie, txstate);
766 }
767 
iop_adma_eot_handler(int irq,void * data)768 static irqreturn_t iop_adma_eot_handler(int irq, void *data)
769 {
770 	struct iop_adma_chan *chan = data;
771 
772 	dev_dbg(chan->device->common.dev, "%s\n", __func__);
773 
774 	tasklet_schedule(&chan->irq_tasklet);
775 
776 	iop_adma_device_clear_eot_status(chan);
777 
778 	return IRQ_HANDLED;
779 }
780 
iop_adma_eoc_handler(int irq,void * data)781 static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
782 {
783 	struct iop_adma_chan *chan = data;
784 
785 	dev_dbg(chan->device->common.dev, "%s\n", __func__);
786 
787 	tasklet_schedule(&chan->irq_tasklet);
788 
789 	iop_adma_device_clear_eoc_status(chan);
790 
791 	return IRQ_HANDLED;
792 }
793 
iop_adma_err_handler(int irq,void * data)794 static irqreturn_t iop_adma_err_handler(int irq, void *data)
795 {
796 	struct iop_adma_chan *chan = data;
797 	unsigned long status = iop_chan_get_status(chan);
798 
799 	dev_err(chan->device->common.dev,
800 		"error ( %s%s%s%s%s%s%s)\n",
801 		iop_is_err_int_parity(status, chan) ? "int_parity " : "",
802 		iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
803 		iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
804 		iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
805 		iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
806 		iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
807 		iop_is_err_split_tx(status, chan) ? "split_tx " : "");
808 
809 	iop_adma_device_clear_err_status(chan);
810 
811 	BUG();
812 
813 	return IRQ_HANDLED;
814 }
815 
iop_adma_issue_pending(struct dma_chan * chan)816 static void iop_adma_issue_pending(struct dma_chan *chan)
817 {
818 	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
819 
820 	if (iop_chan->pending) {
821 		iop_chan->pending = 0;
822 		iop_chan_append(iop_chan);
823 	}
824 }
825 
826 /*
827  * Perform a transaction to verify the HW works.
828  */
829 #define IOP_ADMA_TEST_SIZE 2000
830 
iop_adma_memcpy_self_test(struct iop_adma_device * device)831 static int iop_adma_memcpy_self_test(struct iop_adma_device *device)
832 {
833 	int i;
834 	void *src, *dest;
835 	dma_addr_t src_dma, dest_dma;
836 	struct dma_chan *dma_chan;
837 	dma_cookie_t cookie;
838 	struct dma_async_tx_descriptor *tx;
839 	int err = 0;
840 	struct iop_adma_chan *iop_chan;
841 
842 	dev_dbg(device->common.dev, "%s\n", __func__);
843 
844 	src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
845 	if (!src)
846 		return -ENOMEM;
847 	dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
848 	if (!dest) {
849 		kfree(src);
850 		return -ENOMEM;
851 	}
852 
853 	/* Fill in src buffer */
854 	for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
855 		((u8 *) src)[i] = (u8)i;
856 
857 	/* Start copy, using first DMA channel */
858 	dma_chan = container_of(device->common.channels.next,
859 				struct dma_chan,
860 				device_node);
861 	if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
862 		err = -ENODEV;
863 		goto out;
864 	}
865 
866 	dest_dma = dma_map_single(dma_chan->device->dev, dest,
867 				IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
868 	src_dma = dma_map_single(dma_chan->device->dev, src,
869 				IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
870 	tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
871 				      IOP_ADMA_TEST_SIZE,
872 				      DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
873 
874 	cookie = iop_adma_tx_submit(tx);
875 	iop_adma_issue_pending(dma_chan);
876 	msleep(1);
877 
878 	if (iop_adma_status(dma_chan, cookie, NULL) !=
879 			DMA_COMPLETE) {
880 		dev_err(dma_chan->device->dev,
881 			"Self-test copy timed out, disabling\n");
882 		err = -ENODEV;
883 		goto free_resources;
884 	}
885 
886 	iop_chan = to_iop_adma_chan(dma_chan);
887 	dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
888 		IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
889 	if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
890 		dev_err(dma_chan->device->dev,
891 			"Self-test copy failed compare, disabling\n");
892 		err = -ENODEV;
893 		goto free_resources;
894 	}
895 
896 free_resources:
897 	iop_adma_free_chan_resources(dma_chan);
898 out:
899 	kfree(src);
900 	kfree(dest);
901 	return err;
902 }
903 
904 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
905 static int
iop_adma_xor_val_self_test(struct iop_adma_device * device)906 iop_adma_xor_val_self_test(struct iop_adma_device *device)
907 {
908 	int i, src_idx;
909 	struct page *dest;
910 	struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
911 	struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
912 	dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
913 	dma_addr_t dest_dma;
914 	struct dma_async_tx_descriptor *tx;
915 	struct dma_chan *dma_chan;
916 	dma_cookie_t cookie;
917 	u8 cmp_byte = 0;
918 	u32 cmp_word;
919 	u32 zero_sum_result;
920 	int err = 0;
921 	struct iop_adma_chan *iop_chan;
922 
923 	dev_dbg(device->common.dev, "%s\n", __func__);
924 
925 	for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
926 		xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
927 		if (!xor_srcs[src_idx]) {
928 			while (src_idx--)
929 				__free_page(xor_srcs[src_idx]);
930 			return -ENOMEM;
931 		}
932 	}
933 
934 	dest = alloc_page(GFP_KERNEL);
935 	if (!dest) {
936 		while (src_idx--)
937 			__free_page(xor_srcs[src_idx]);
938 		return -ENOMEM;
939 	}
940 
941 	/* Fill in src buffers */
942 	for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
943 		u8 *ptr = page_address(xor_srcs[src_idx]);
944 		for (i = 0; i < PAGE_SIZE; i++)
945 			ptr[i] = (1 << src_idx);
946 	}
947 
948 	for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
949 		cmp_byte ^= (u8) (1 << src_idx);
950 
951 	cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
952 			(cmp_byte << 8) | cmp_byte;
953 
954 	memset(page_address(dest), 0, PAGE_SIZE);
955 
956 	dma_chan = container_of(device->common.channels.next,
957 				struct dma_chan,
958 				device_node);
959 	if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
960 		err = -ENODEV;
961 		goto out;
962 	}
963 
964 	/* test xor */
965 	dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
966 				PAGE_SIZE, DMA_FROM_DEVICE);
967 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
968 		dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
969 					   0, PAGE_SIZE, DMA_TO_DEVICE);
970 	tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
971 				   IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE,
972 				   DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
973 
974 	cookie = iop_adma_tx_submit(tx);
975 	iop_adma_issue_pending(dma_chan);
976 	msleep(8);
977 
978 	if (iop_adma_status(dma_chan, cookie, NULL) !=
979 		DMA_COMPLETE) {
980 		dev_err(dma_chan->device->dev,
981 			"Self-test xor timed out, disabling\n");
982 		err = -ENODEV;
983 		goto free_resources;
984 	}
985 
986 	iop_chan = to_iop_adma_chan(dma_chan);
987 	dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
988 		PAGE_SIZE, DMA_FROM_DEVICE);
989 	for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
990 		u32 *ptr = page_address(dest);
991 		if (ptr[i] != cmp_word) {
992 			dev_err(dma_chan->device->dev,
993 				"Self-test xor failed compare, disabling\n");
994 			err = -ENODEV;
995 			goto free_resources;
996 		}
997 	}
998 	dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
999 		PAGE_SIZE, DMA_TO_DEVICE);
1000 
1001 	/* skip zero sum if the capability is not present */
1002 	if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask))
1003 		goto free_resources;
1004 
1005 	/* zero sum the sources with the destintation page */
1006 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1007 		zero_sum_srcs[i] = xor_srcs[i];
1008 	zero_sum_srcs[i] = dest;
1009 
1010 	zero_sum_result = 1;
1011 
1012 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1013 		dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1014 					   zero_sum_srcs[i], 0, PAGE_SIZE,
1015 					   DMA_TO_DEVICE);
1016 	tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1017 				       IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1018 				       &zero_sum_result,
1019 				       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1020 
1021 	cookie = iop_adma_tx_submit(tx);
1022 	iop_adma_issue_pending(dma_chan);
1023 	msleep(8);
1024 
1025 	if (iop_adma_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
1026 		dev_err(dma_chan->device->dev,
1027 			"Self-test zero sum timed out, disabling\n");
1028 		err = -ENODEV;
1029 		goto free_resources;
1030 	}
1031 
1032 	if (zero_sum_result != 0) {
1033 		dev_err(dma_chan->device->dev,
1034 			"Self-test zero sum failed compare, disabling\n");
1035 		err = -ENODEV;
1036 		goto free_resources;
1037 	}
1038 
1039 	/* test for non-zero parity sum */
1040 	zero_sum_result = 0;
1041 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1042 		dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1043 					   zero_sum_srcs[i], 0, PAGE_SIZE,
1044 					   DMA_TO_DEVICE);
1045 	tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1046 				       IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1047 				       &zero_sum_result,
1048 				       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1049 
1050 	cookie = iop_adma_tx_submit(tx);
1051 	iop_adma_issue_pending(dma_chan);
1052 	msleep(8);
1053 
1054 	if (iop_adma_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
1055 		dev_err(dma_chan->device->dev,
1056 			"Self-test non-zero sum timed out, disabling\n");
1057 		err = -ENODEV;
1058 		goto free_resources;
1059 	}
1060 
1061 	if (zero_sum_result != 1) {
1062 		dev_err(dma_chan->device->dev,
1063 			"Self-test non-zero sum failed compare, disabling\n");
1064 		err = -ENODEV;
1065 		goto free_resources;
1066 	}
1067 
1068 free_resources:
1069 	iop_adma_free_chan_resources(dma_chan);
1070 out:
1071 	src_idx = IOP_ADMA_NUM_SRC_TEST;
1072 	while (src_idx--)
1073 		__free_page(xor_srcs[src_idx]);
1074 	__free_page(dest);
1075 	return err;
1076 }
1077 
1078 #ifdef CONFIG_RAID6_PQ
1079 static int
iop_adma_pq_zero_sum_self_test(struct iop_adma_device * device)1080 iop_adma_pq_zero_sum_self_test(struct iop_adma_device *device)
1081 {
1082 	/* combined sources, software pq results, and extra hw pq results */
1083 	struct page *pq[IOP_ADMA_NUM_SRC_TEST+2+2];
1084 	/* ptr to the extra hw pq buffers defined above */
1085 	struct page **pq_hw = &pq[IOP_ADMA_NUM_SRC_TEST+2];
1086 	/* address conversion buffers (dma_map / page_address) */
1087 	void *pq_sw[IOP_ADMA_NUM_SRC_TEST+2];
1088 	dma_addr_t pq_src[IOP_ADMA_NUM_SRC_TEST+2];
1089 	dma_addr_t *pq_dest = &pq_src[IOP_ADMA_NUM_SRC_TEST];
1090 
1091 	int i;
1092 	struct dma_async_tx_descriptor *tx;
1093 	struct dma_chan *dma_chan;
1094 	dma_cookie_t cookie;
1095 	u32 zero_sum_result;
1096 	int err = 0;
1097 	struct device *dev;
1098 
1099 	dev_dbg(device->common.dev, "%s\n", __func__);
1100 
1101 	for (i = 0; i < ARRAY_SIZE(pq); i++) {
1102 		pq[i] = alloc_page(GFP_KERNEL);
1103 		if (!pq[i]) {
1104 			while (i--)
1105 				__free_page(pq[i]);
1106 			return -ENOMEM;
1107 		}
1108 	}
1109 
1110 	/* Fill in src buffers */
1111 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) {
1112 		pq_sw[i] = page_address(pq[i]);
1113 		memset(pq_sw[i], 0x11111111 * (1<<i), PAGE_SIZE);
1114 	}
1115 	pq_sw[i] = page_address(pq[i]);
1116 	pq_sw[i+1] = page_address(pq[i+1]);
1117 
1118 	dma_chan = container_of(device->common.channels.next,
1119 				struct dma_chan,
1120 				device_node);
1121 	if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1122 		err = -ENODEV;
1123 		goto out;
1124 	}
1125 
1126 	dev = dma_chan->device->dev;
1127 
1128 	/* initialize the dests */
1129 	memset(page_address(pq_hw[0]), 0 , PAGE_SIZE);
1130 	memset(page_address(pq_hw[1]), 0 , PAGE_SIZE);
1131 
1132 	/* test pq */
1133 	pq_dest[0] = dma_map_page(dev, pq_hw[0], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1134 	pq_dest[1] = dma_map_page(dev, pq_hw[1], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1135 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1136 		pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1137 					 DMA_TO_DEVICE);
1138 
1139 	tx = iop_adma_prep_dma_pq(dma_chan, pq_dest, pq_src,
1140 				  IOP_ADMA_NUM_SRC_TEST, (u8 *)raid6_gfexp,
1141 				  PAGE_SIZE,
1142 				  DMA_PREP_INTERRUPT |
1143 				  DMA_CTRL_ACK);
1144 
1145 	cookie = iop_adma_tx_submit(tx);
1146 	iop_adma_issue_pending(dma_chan);
1147 	msleep(8);
1148 
1149 	if (iop_adma_status(dma_chan, cookie, NULL) !=
1150 		DMA_COMPLETE) {
1151 		dev_err(dev, "Self-test pq timed out, disabling\n");
1152 		err = -ENODEV;
1153 		goto free_resources;
1154 	}
1155 
1156 	raid6_call.gen_syndrome(IOP_ADMA_NUM_SRC_TEST+2, PAGE_SIZE, pq_sw);
1157 
1158 	if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST],
1159 		   page_address(pq_hw[0]), PAGE_SIZE) != 0) {
1160 		dev_err(dev, "Self-test p failed compare, disabling\n");
1161 		err = -ENODEV;
1162 		goto free_resources;
1163 	}
1164 	if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST+1],
1165 		   page_address(pq_hw[1]), PAGE_SIZE) != 0) {
1166 		dev_err(dev, "Self-test q failed compare, disabling\n");
1167 		err = -ENODEV;
1168 		goto free_resources;
1169 	}
1170 
1171 	/* test correct zero sum using the software generated pq values */
1172 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1173 		pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1174 					 DMA_TO_DEVICE);
1175 
1176 	zero_sum_result = ~0;
1177 	tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1178 				      pq_src, IOP_ADMA_NUM_SRC_TEST,
1179 				      raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1180 				      DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1181 
1182 	cookie = iop_adma_tx_submit(tx);
1183 	iop_adma_issue_pending(dma_chan);
1184 	msleep(8);
1185 
1186 	if (iop_adma_status(dma_chan, cookie, NULL) !=
1187 		DMA_COMPLETE) {
1188 		dev_err(dev, "Self-test pq-zero-sum timed out, disabling\n");
1189 		err = -ENODEV;
1190 		goto free_resources;
1191 	}
1192 
1193 	if (zero_sum_result != 0) {
1194 		dev_err(dev, "Self-test pq-zero-sum failed to validate: %x\n",
1195 			zero_sum_result);
1196 		err = -ENODEV;
1197 		goto free_resources;
1198 	}
1199 
1200 	/* test incorrect zero sum */
1201 	i = IOP_ADMA_NUM_SRC_TEST;
1202 	memset(pq_sw[i] + 100, 0, 100);
1203 	memset(pq_sw[i+1] + 200, 0, 200);
1204 	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1205 		pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1206 					 DMA_TO_DEVICE);
1207 
1208 	zero_sum_result = 0;
1209 	tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1210 				      pq_src, IOP_ADMA_NUM_SRC_TEST,
1211 				      raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1212 				      DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1213 
1214 	cookie = iop_adma_tx_submit(tx);
1215 	iop_adma_issue_pending(dma_chan);
1216 	msleep(8);
1217 
1218 	if (iop_adma_status(dma_chan, cookie, NULL) !=
1219 		DMA_COMPLETE) {
1220 		dev_err(dev, "Self-test !pq-zero-sum timed out, disabling\n");
1221 		err = -ENODEV;
1222 		goto free_resources;
1223 	}
1224 
1225 	if (zero_sum_result != (SUM_CHECK_P_RESULT | SUM_CHECK_Q_RESULT)) {
1226 		dev_err(dev, "Self-test !pq-zero-sum failed to validate: %x\n",
1227 			zero_sum_result);
1228 		err = -ENODEV;
1229 		goto free_resources;
1230 	}
1231 
1232 free_resources:
1233 	iop_adma_free_chan_resources(dma_chan);
1234 out:
1235 	i = ARRAY_SIZE(pq);
1236 	while (i--)
1237 		__free_page(pq[i]);
1238 	return err;
1239 }
1240 #endif
1241 
iop_adma_remove(struct platform_device * dev)1242 static int iop_adma_remove(struct platform_device *dev)
1243 {
1244 	struct iop_adma_device *device = platform_get_drvdata(dev);
1245 	struct dma_chan *chan, *_chan;
1246 	struct iop_adma_chan *iop_chan;
1247 	struct iop_adma_platform_data *plat_data = dev_get_platdata(&dev->dev);
1248 
1249 	dma_async_device_unregister(&device->common);
1250 
1251 	dma_free_coherent(&dev->dev, plat_data->pool_size,
1252 			device->dma_desc_pool_virt, device->dma_desc_pool);
1253 
1254 	list_for_each_entry_safe(chan, _chan, &device->common.channels,
1255 				device_node) {
1256 		iop_chan = to_iop_adma_chan(chan);
1257 		list_del(&chan->device_node);
1258 		kfree(iop_chan);
1259 	}
1260 	kfree(device);
1261 
1262 	return 0;
1263 }
1264 
iop_adma_probe(struct platform_device * pdev)1265 static int iop_adma_probe(struct platform_device *pdev)
1266 {
1267 	struct resource *res;
1268 	int ret = 0, i;
1269 	struct iop_adma_device *adev;
1270 	struct iop_adma_chan *iop_chan;
1271 	struct dma_device *dma_dev;
1272 	struct iop_adma_platform_data *plat_data = dev_get_platdata(&pdev->dev);
1273 
1274 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1275 	if (!res)
1276 		return -ENODEV;
1277 
1278 	if (!devm_request_mem_region(&pdev->dev, res->start,
1279 				resource_size(res), pdev->name))
1280 		return -EBUSY;
1281 
1282 	adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1283 	if (!adev)
1284 		return -ENOMEM;
1285 	dma_dev = &adev->common;
1286 
1287 	/* allocate coherent memory for hardware descriptors
1288 	 * note: writecombine gives slightly better performance, but
1289 	 * requires that we explicitly flush the writes
1290 	 */
1291 	adev->dma_desc_pool_virt = dma_alloc_wc(&pdev->dev,
1292 						plat_data->pool_size,
1293 						&adev->dma_desc_pool,
1294 						GFP_KERNEL);
1295 	if (!adev->dma_desc_pool_virt) {
1296 		ret = -ENOMEM;
1297 		goto err_free_adev;
1298 	}
1299 
1300 	dev_dbg(&pdev->dev, "%s: allocated descriptor pool virt %p phys %p\n",
1301 		__func__, adev->dma_desc_pool_virt,
1302 		(void *) adev->dma_desc_pool);
1303 
1304 	adev->id = plat_data->hw_id;
1305 
1306 	/* discover transaction capabilites from the platform data */
1307 	dma_dev->cap_mask = plat_data->cap_mask;
1308 
1309 	adev->pdev = pdev;
1310 	platform_set_drvdata(pdev, adev);
1311 
1312 	INIT_LIST_HEAD(&dma_dev->channels);
1313 
1314 	/* set base routines */
1315 	dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1316 	dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1317 	dma_dev->device_tx_status = iop_adma_status;
1318 	dma_dev->device_issue_pending = iop_adma_issue_pending;
1319 	dma_dev->dev = &pdev->dev;
1320 
1321 	/* set prep routines based on capability */
1322 	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1323 		dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1324 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1325 		dma_dev->max_xor = iop_adma_get_max_xor();
1326 		dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1327 	}
1328 	if (dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask))
1329 		dma_dev->device_prep_dma_xor_val =
1330 			iop_adma_prep_dma_xor_val;
1331 	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1332 		dma_set_maxpq(dma_dev, iop_adma_get_max_pq(), 0);
1333 		dma_dev->device_prep_dma_pq = iop_adma_prep_dma_pq;
1334 	}
1335 	if (dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask))
1336 		dma_dev->device_prep_dma_pq_val =
1337 			iop_adma_prep_dma_pq_val;
1338 	if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1339 		dma_dev->device_prep_dma_interrupt =
1340 			iop_adma_prep_dma_interrupt;
1341 
1342 	iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1343 	if (!iop_chan) {
1344 		ret = -ENOMEM;
1345 		goto err_free_dma;
1346 	}
1347 	iop_chan->device = adev;
1348 
1349 	iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1350 					resource_size(res));
1351 	if (!iop_chan->mmr_base) {
1352 		ret = -ENOMEM;
1353 		goto err_free_iop_chan;
1354 	}
1355 	tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1356 		iop_chan);
1357 
1358 	/* clear errors before enabling interrupts */
1359 	iop_adma_device_clear_err_status(iop_chan);
1360 
1361 	for (i = 0; i < 3; i++) {
1362 		irq_handler_t handler[] = { iop_adma_eot_handler,
1363 					iop_adma_eoc_handler,
1364 					iop_adma_err_handler };
1365 		int irq = platform_get_irq(pdev, i);
1366 		if (irq < 0) {
1367 			ret = -ENXIO;
1368 			goto err_free_iop_chan;
1369 		} else {
1370 			ret = devm_request_irq(&pdev->dev, irq,
1371 					handler[i], 0, pdev->name, iop_chan);
1372 			if (ret)
1373 				goto err_free_iop_chan;
1374 		}
1375 	}
1376 
1377 	spin_lock_init(&iop_chan->lock);
1378 	INIT_LIST_HEAD(&iop_chan->chain);
1379 	INIT_LIST_HEAD(&iop_chan->all_slots);
1380 	iop_chan->common.device = dma_dev;
1381 	dma_cookie_init(&iop_chan->common);
1382 	list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1383 
1384 	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1385 		ret = iop_adma_memcpy_self_test(adev);
1386 		dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1387 		if (ret)
1388 			goto err_free_iop_chan;
1389 	}
1390 
1391 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1392 		ret = iop_adma_xor_val_self_test(adev);
1393 		dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1394 		if (ret)
1395 			goto err_free_iop_chan;
1396 	}
1397 
1398 	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask) &&
1399 	    dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask)) {
1400 		#ifdef CONFIG_RAID6_PQ
1401 		ret = iop_adma_pq_zero_sum_self_test(adev);
1402 		dev_dbg(&pdev->dev, "pq self test returned %d\n", ret);
1403 		#else
1404 		/* can not test raid6, so do not publish capability */
1405 		dma_cap_clear(DMA_PQ, dma_dev->cap_mask);
1406 		dma_cap_clear(DMA_PQ_VAL, dma_dev->cap_mask);
1407 		ret = 0;
1408 		#endif
1409 		if (ret)
1410 			goto err_free_iop_chan;
1411 	}
1412 
1413 	dev_info(&pdev->dev, "Intel(R) IOP: ( %s%s%s%s%s%s)\n",
1414 		 dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "",
1415 		 dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask) ? "pq_val " : "",
1416 		 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1417 		 dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask) ? "xor_val " : "",
1418 		 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1419 		 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1420 
1421 	dma_async_device_register(dma_dev);
1422 	goto out;
1423 
1424  err_free_iop_chan:
1425 	kfree(iop_chan);
1426  err_free_dma:
1427 	dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1428 			adev->dma_desc_pool_virt, adev->dma_desc_pool);
1429  err_free_adev:
1430 	kfree(adev);
1431  out:
1432 	return ret;
1433 }
1434 
iop_chan_start_null_memcpy(struct iop_adma_chan * iop_chan)1435 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1436 {
1437 	struct iop_adma_desc_slot *sw_desc, *grp_start;
1438 	dma_cookie_t cookie;
1439 	int slot_cnt, slots_per_op;
1440 
1441 	dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1442 
1443 	spin_lock_bh(&iop_chan->lock);
1444 	slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1445 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1446 	if (sw_desc) {
1447 		grp_start = sw_desc->group_head;
1448 
1449 		list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1450 		async_tx_ack(&sw_desc->async_tx);
1451 		iop_desc_init_memcpy(grp_start, 0);
1452 		iop_desc_set_byte_count(grp_start, iop_chan, 0);
1453 		iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1454 		iop_desc_set_memcpy_src_addr(grp_start, 0);
1455 
1456 		cookie = dma_cookie_assign(&sw_desc->async_tx);
1457 
1458 		/* initialize the completed cookie to be less than
1459 		 * the most recently used cookie
1460 		 */
1461 		iop_chan->common.completed_cookie = cookie - 1;
1462 
1463 		/* channel should not be busy */
1464 		BUG_ON(iop_chan_is_busy(iop_chan));
1465 
1466 		/* clear any prior error-status bits */
1467 		iop_adma_device_clear_err_status(iop_chan);
1468 
1469 		/* disable operation */
1470 		iop_chan_disable(iop_chan);
1471 
1472 		/* set the descriptor address */
1473 		iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1474 
1475 		/* 1/ don't add pre-chained descriptors
1476 		 * 2/ dummy read to flush next_desc write
1477 		 */
1478 		BUG_ON(iop_desc_get_next_desc(sw_desc));
1479 
1480 		/* run the descriptor */
1481 		iop_chan_enable(iop_chan);
1482 	} else
1483 		dev_err(iop_chan->device->common.dev,
1484 			"failed to allocate null descriptor\n");
1485 	spin_unlock_bh(&iop_chan->lock);
1486 }
1487 
iop_chan_start_null_xor(struct iop_adma_chan * iop_chan)1488 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1489 {
1490 	struct iop_adma_desc_slot *sw_desc, *grp_start;
1491 	dma_cookie_t cookie;
1492 	int slot_cnt, slots_per_op;
1493 
1494 	dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1495 
1496 	spin_lock_bh(&iop_chan->lock);
1497 	slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1498 	sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1499 	if (sw_desc) {
1500 		grp_start = sw_desc->group_head;
1501 		list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1502 		async_tx_ack(&sw_desc->async_tx);
1503 		iop_desc_init_null_xor(grp_start, 2, 0);
1504 		iop_desc_set_byte_count(grp_start, iop_chan, 0);
1505 		iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1506 		iop_desc_set_xor_src_addr(grp_start, 0, 0);
1507 		iop_desc_set_xor_src_addr(grp_start, 1, 0);
1508 
1509 		cookie = dma_cookie_assign(&sw_desc->async_tx);
1510 
1511 		/* initialize the completed cookie to be less than
1512 		 * the most recently used cookie
1513 		 */
1514 		iop_chan->common.completed_cookie = cookie - 1;
1515 
1516 		/* channel should not be busy */
1517 		BUG_ON(iop_chan_is_busy(iop_chan));
1518 
1519 		/* clear any prior error-status bits */
1520 		iop_adma_device_clear_err_status(iop_chan);
1521 
1522 		/* disable operation */
1523 		iop_chan_disable(iop_chan);
1524 
1525 		/* set the descriptor address */
1526 		iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1527 
1528 		/* 1/ don't add pre-chained descriptors
1529 		 * 2/ dummy read to flush next_desc write
1530 		 */
1531 		BUG_ON(iop_desc_get_next_desc(sw_desc));
1532 
1533 		/* run the descriptor */
1534 		iop_chan_enable(iop_chan);
1535 	} else
1536 		dev_err(iop_chan->device->common.dev,
1537 			"failed to allocate null descriptor\n");
1538 	spin_unlock_bh(&iop_chan->lock);
1539 }
1540 
1541 static struct platform_driver iop_adma_driver = {
1542 	.probe		= iop_adma_probe,
1543 	.remove		= iop_adma_remove,
1544 	.driver		= {
1545 		.name	= "iop-adma",
1546 	},
1547 };
1548 
1549 module_platform_driver(iop_adma_driver);
1550 
1551 MODULE_AUTHOR("Intel Corporation");
1552 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1553 MODULE_LICENSE("GPL");
1554 MODULE_ALIAS("platform:iop-adma");
1555