Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[linux-2.6] / drivers / dma / iop-adma.c
1 /*
2  * offload engine driver for the Intel Xscale series of i/o processors
3  * Copyright © 2006, Intel Corporation.
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
20 /*
21  * This driver supports the asynchrounous DMA copy and RAID engines available
22  * on the Intel Xscale(R) family of I/O Processors (IOP 32x, 33x, 134x)
23  */
24
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/async_tx.h>
28 #include <linux/delay.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/spinlock.h>
31 #include <linux/interrupt.h>
32 #include <linux/platform_device.h>
33 #include <linux/memory.h>
34 #include <linux/ioport.h>
35
36 #include <mach/adma.h>
37
38 #define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
39 #define to_iop_adma_device(dev) \
40         container_of(dev, struct iop_adma_device, common)
41 #define tx_to_iop_adma_slot(tx) \
42         container_of(tx, struct iop_adma_desc_slot, async_tx)
43
44 /**
45  * iop_adma_free_slots - flags descriptor slots for reuse
46  * @slot: Slot to free
47  * Caller must hold &iop_chan->lock while calling this function
48  */
49 static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
50 {
51         int stride = slot->slots_per_op;
52
53         while (stride--) {
54                 slot->slots_per_op = 0;
55                 slot = list_entry(slot->slot_node.next,
56                                 struct iop_adma_desc_slot,
57                                 slot_node);
58         }
59 }
60
61 static dma_cookie_t
62 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
63         struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
64 {
65         BUG_ON(desc->async_tx.cookie < 0);
66         if (desc->async_tx.cookie > 0) {
67                 cookie = desc->async_tx.cookie;
68                 desc->async_tx.cookie = 0;
69
70                 /* call the callback (must not sleep or submit new
71                  * operations to this channel)
72                  */
73                 if (desc->async_tx.callback)
74                         desc->async_tx.callback(
75                                 desc->async_tx.callback_param);
76
77                 /* unmap dma addresses
78                  * (unmap_single vs unmap_page?)
79                  */
80                 if (desc->group_head && desc->unmap_len) {
81                         struct iop_adma_desc_slot *unmap = desc->group_head;
82                         struct device *dev =
83                                 &iop_chan->device->pdev->dev;
84                         u32 len = unmap->unmap_len;
85                         enum dma_ctrl_flags flags = desc->async_tx.flags;
86                         u32 src_cnt;
87                         dma_addr_t addr;
88
89                         if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
90                                 addr = iop_desc_get_dest_addr(unmap, iop_chan);
91                                 dma_unmap_page(dev, addr, len, DMA_FROM_DEVICE);
92                         }
93
94                         if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
95                                 src_cnt = unmap->unmap_src_cnt;
96                                 while (src_cnt--) {
97                                         addr = iop_desc_get_src_addr(unmap,
98                                                                      iop_chan,
99                                                                      src_cnt);
100                                         dma_unmap_page(dev, addr, len,
101                                                        DMA_TO_DEVICE);
102                                 }
103                         }
104                         desc->group_head = NULL;
105                 }
106         }
107
108         /* run dependent operations */
109         async_tx_run_dependencies(&desc->async_tx);
110
111         return cookie;
112 }
113
114 static int
115 iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
116         struct iop_adma_chan *iop_chan)
117 {
118         /* the client is allowed to attach dependent operations
119          * until 'ack' is set
120          */
121         if (!async_tx_test_ack(&desc->async_tx))
122                 return 0;
123
124         /* leave the last descriptor in the chain
125          * so we can append to it
126          */
127         if (desc->chain_node.next == &iop_chan->chain)
128                 return 1;
129
130         dev_dbg(iop_chan->device->common.dev,
131                 "\tfree slot: %d slots_per_op: %d\n",
132                 desc->idx, desc->slots_per_op);
133
134         list_del(&desc->chain_node);
135         iop_adma_free_slots(desc);
136
137         return 0;
138 }
139
140 static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
141 {
142         struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
143         dma_cookie_t cookie = 0;
144         u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
145         int busy = iop_chan_is_busy(iop_chan);
146         int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
147
148         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
149         /* free completed slots from the chain starting with
150          * the oldest descriptor
151          */
152         list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
153                                         chain_node) {
154                 pr_debug("\tcookie: %d slot: %d busy: %d "
155                         "this_desc: %#x next_desc: %#x ack: %d\n",
156                         iter->async_tx.cookie, iter->idx, busy,
157                         iter->async_tx.phys, iop_desc_get_next_desc(iter),
158                         async_tx_test_ack(&iter->async_tx));
159                 prefetch(_iter);
160                 prefetch(&_iter->async_tx);
161
162                 /* do not advance past the current descriptor loaded into the
163                  * hardware channel, subsequent descriptors are either in
164                  * process or have not been submitted
165                  */
166                 if (seen_current)
167                         break;
168
169                 /* stop the search if we reach the current descriptor and the
170                  * channel is busy, or if it appears that the current descriptor
171                  * needs to be re-read (i.e. has been appended to)
172                  */
173                 if (iter->async_tx.phys == current_desc) {
174                         BUG_ON(seen_current++);
175                         if (busy || iop_desc_get_next_desc(iter))
176                                 break;
177                 }
178
179                 /* detect the start of a group transaction */
180                 if (!slot_cnt && !slots_per_op) {
181                         slot_cnt = iter->slot_cnt;
182                         slots_per_op = iter->slots_per_op;
183                         if (slot_cnt <= slots_per_op) {
184                                 slot_cnt = 0;
185                                 slots_per_op = 0;
186                         }
187                 }
188
189                 if (slot_cnt) {
190                         pr_debug("\tgroup++\n");
191                         if (!grp_start)
192                                 grp_start = iter;
193                         slot_cnt -= slots_per_op;
194                 }
195
196                 /* all the members of a group are complete */
197                 if (slots_per_op != 0 && slot_cnt == 0) {
198                         struct iop_adma_desc_slot *grp_iter, *_grp_iter;
199                         int end_of_chain = 0;
200                         pr_debug("\tgroup end\n");
201
202                         /* collect the total results */
203                         if (grp_start->xor_check_result) {
204                                 u32 zero_sum_result = 0;
205                                 slot_cnt = grp_start->slot_cnt;
206                                 grp_iter = grp_start;
207
208                                 list_for_each_entry_from(grp_iter,
209                                         &iop_chan->chain, chain_node) {
210                                         zero_sum_result |=
211                                             iop_desc_get_zero_result(grp_iter);
212                                             pr_debug("\titer%d result: %d\n",
213                                             grp_iter->idx, zero_sum_result);
214                                         slot_cnt -= slots_per_op;
215                                         if (slot_cnt == 0)
216                                                 break;
217                                 }
218                                 pr_debug("\tgrp_start->xor_check_result: %p\n",
219                                         grp_start->xor_check_result);
220                                 *grp_start->xor_check_result = zero_sum_result;
221                         }
222
223                         /* clean up the group */
224                         slot_cnt = grp_start->slot_cnt;
225                         grp_iter = grp_start;
226                         list_for_each_entry_safe_from(grp_iter, _grp_iter,
227                                 &iop_chan->chain, chain_node) {
228                                 cookie = iop_adma_run_tx_complete_actions(
229                                         grp_iter, iop_chan, cookie);
230
231                                 slot_cnt -= slots_per_op;
232                                 end_of_chain = iop_adma_clean_slot(grp_iter,
233                                         iop_chan);
234
235                                 if (slot_cnt == 0 || end_of_chain)
236                                         break;
237                         }
238
239                         /* the group should be complete at this point */
240                         BUG_ON(slot_cnt);
241
242                         slots_per_op = 0;
243                         grp_start = NULL;
244                         if (end_of_chain)
245                                 break;
246                         else
247                                 continue;
248                 } else if (slots_per_op) /* wait for group completion */
249                         continue;
250
251                 /* write back zero sum results (single descriptor case) */
252                 if (iter->xor_check_result && iter->async_tx.cookie)
253                         *iter->xor_check_result =
254                                 iop_desc_get_zero_result(iter);
255
256                 cookie = iop_adma_run_tx_complete_actions(
257                                         iter, iop_chan, cookie);
258
259                 if (iop_adma_clean_slot(iter, iop_chan))
260                         break;
261         }
262
263         BUG_ON(!seen_current);
264
265         if (cookie > 0) {
266                 iop_chan->completed_cookie = cookie;
267                 pr_debug("\tcompleted cookie %d\n", cookie);
268         }
269 }
270
271 static void
272 iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
273 {
274         spin_lock_bh(&iop_chan->lock);
275         __iop_adma_slot_cleanup(iop_chan);
276         spin_unlock_bh(&iop_chan->lock);
277 }
278
279 static void iop_adma_tasklet(unsigned long data)
280 {
281         struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data;
282
283         spin_lock(&iop_chan->lock);
284         __iop_adma_slot_cleanup(iop_chan);
285         spin_unlock(&iop_chan->lock);
286 }
287
288 static struct iop_adma_desc_slot *
289 iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
290                         int slots_per_op)
291 {
292         struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
293         LIST_HEAD(chain);
294         int slots_found, retry = 0;
295
296         /* start search from the last allocated descrtiptor
297          * if a contiguous allocation can not be found start searching
298          * from the beginning of the list
299          */
300 retry:
301         slots_found = 0;
302         if (retry == 0)
303                 iter = iop_chan->last_used;
304         else
305                 iter = list_entry(&iop_chan->all_slots,
306                         struct iop_adma_desc_slot,
307                         slot_node);
308
309         list_for_each_entry_safe_continue(
310                 iter, _iter, &iop_chan->all_slots, slot_node) {
311                 prefetch(_iter);
312                 prefetch(&_iter->async_tx);
313                 if (iter->slots_per_op) {
314                         /* give up after finding the first busy slot
315                          * on the second pass through the list
316                          */
317                         if (retry)
318                                 break;
319
320                         slots_found = 0;
321                         continue;
322                 }
323
324                 /* start the allocation if the slot is correctly aligned */
325                 if (!slots_found++) {
326                         if (iop_desc_is_aligned(iter, slots_per_op))
327                                 alloc_start = iter;
328                         else {
329                                 slots_found = 0;
330                                 continue;
331                         }
332                 }
333
334                 if (slots_found == num_slots) {
335                         struct iop_adma_desc_slot *alloc_tail = NULL;
336                         struct iop_adma_desc_slot *last_used = NULL;
337                         iter = alloc_start;
338                         while (num_slots) {
339                                 int i;
340                                 dev_dbg(iop_chan->device->common.dev,
341                                         "allocated slot: %d "
342                                         "(desc %p phys: %#x) slots_per_op %d\n",
343                                         iter->idx, iter->hw_desc,
344                                         iter->async_tx.phys, slots_per_op);
345
346                                 /* pre-ack all but the last descriptor */
347                                 if (num_slots != slots_per_op)
348                                         async_tx_ack(&iter->async_tx);
349
350                                 list_add_tail(&iter->chain_node, &chain);
351                                 alloc_tail = iter;
352                                 iter->async_tx.cookie = 0;
353                                 iter->slot_cnt = num_slots;
354                                 iter->xor_check_result = NULL;
355                                 for (i = 0; i < slots_per_op; i++) {
356                                         iter->slots_per_op = slots_per_op - i;
357                                         last_used = iter;
358                                         iter = list_entry(iter->slot_node.next,
359                                                 struct iop_adma_desc_slot,
360                                                 slot_node);
361                                 }
362                                 num_slots -= slots_per_op;
363                         }
364                         alloc_tail->group_head = alloc_start;
365                         alloc_tail->async_tx.cookie = -EBUSY;
366                         list_splice(&chain, &alloc_tail->async_tx.tx_list);
367                         iop_chan->last_used = last_used;
368                         iop_desc_clear_next_desc(alloc_start);
369                         iop_desc_clear_next_desc(alloc_tail);
370                         return alloc_tail;
371                 }
372         }
373         if (!retry++)
374                 goto retry;
375
376         /* perform direct reclaim if the allocation fails */
377         __iop_adma_slot_cleanup(iop_chan);
378
379         return NULL;
380 }
381
382 static dma_cookie_t
383 iop_desc_assign_cookie(struct iop_adma_chan *iop_chan,
384         struct iop_adma_desc_slot *desc)
385 {
386         dma_cookie_t cookie = iop_chan->common.cookie;
387         cookie++;
388         if (cookie < 0)
389                 cookie = 1;
390         iop_chan->common.cookie = desc->async_tx.cookie = cookie;
391         return cookie;
392 }
393
394 static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
395 {
396         dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
397                 iop_chan->pending);
398
399         if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
400                 iop_chan->pending = 0;
401                 iop_chan_append(iop_chan);
402         }
403 }
404
405 static dma_cookie_t
406 iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
407 {
408         struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
409         struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
410         struct iop_adma_desc_slot *grp_start, *old_chain_tail;
411         int slot_cnt;
412         int slots_per_op;
413         dma_cookie_t cookie;
414
415         grp_start = sw_desc->group_head;
416         slot_cnt = grp_start->slot_cnt;
417         slots_per_op = grp_start->slots_per_op;
418
419         spin_lock_bh(&iop_chan->lock);
420         cookie = iop_desc_assign_cookie(iop_chan, sw_desc);
421
422         old_chain_tail = list_entry(iop_chan->chain.prev,
423                 struct iop_adma_desc_slot, chain_node);
424         list_splice_init(&sw_desc->async_tx.tx_list,
425                          &old_chain_tail->chain_node);
426
427         /* fix up the hardware chain */
428         iop_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
429
430         /* 1/ don't add pre-chained descriptors
431          * 2/ dummy read to flush next_desc write
432          */
433         BUG_ON(iop_desc_get_next_desc(sw_desc));
434
435         /* increment the pending count by the number of slots
436          * memcpy operations have a 1:1 (slot:operation) relation
437          * other operations are heavier and will pop the threshold
438          * more often.
439          */
440         iop_chan->pending += slot_cnt;
441         iop_adma_check_threshold(iop_chan);
442         spin_unlock_bh(&iop_chan->lock);
443
444         dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
445                 __func__, sw_desc->async_tx.cookie, sw_desc->idx);
446
447         return cookie;
448 }
449
450 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
451 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
452
453 /**
454  * iop_adma_alloc_chan_resources -  returns the number of allocated descriptors
455  * @chan - allocate descriptor resources for this channel
456  * @client - current client requesting the channel be ready for requests
457  *
458  * Note: We keep the slots for 1 operation on iop_chan->chain at all times.  To
459  * avoid deadlock, via async_xor, num_descs_in_pool must at a minimum be
460  * greater than 2x the number slots needed to satisfy a device->max_xor
461  * request.
462  * */
463 static int iop_adma_alloc_chan_resources(struct dma_chan *chan,
464                                          struct dma_client *client)
465 {
466         char *hw_desc;
467         int idx;
468         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
469         struct iop_adma_desc_slot *slot = NULL;
470         int init = iop_chan->slots_allocated ? 0 : 1;
471         struct iop_adma_platform_data *plat_data =
472                 iop_chan->device->pdev->dev.platform_data;
473         int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
474
475         /* Allocate descriptor slots */
476         do {
477                 idx = iop_chan->slots_allocated;
478                 if (idx == num_descs_in_pool)
479                         break;
480
481                 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
482                 if (!slot) {
483                         printk(KERN_INFO "IOP ADMA Channel only initialized"
484                                 " %d descriptor slots", idx);
485                         break;
486                 }
487                 hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
488                 slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
489
490                 dma_async_tx_descriptor_init(&slot->async_tx, chan);
491                 slot->async_tx.tx_submit = iop_adma_tx_submit;
492                 INIT_LIST_HEAD(&slot->chain_node);
493                 INIT_LIST_HEAD(&slot->slot_node);
494                 INIT_LIST_HEAD(&slot->async_tx.tx_list);
495                 hw_desc = (char *) iop_chan->device->dma_desc_pool;
496                 slot->async_tx.phys =
497                         (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
498                 slot->idx = idx;
499
500                 spin_lock_bh(&iop_chan->lock);
501                 iop_chan->slots_allocated++;
502                 list_add_tail(&slot->slot_node, &iop_chan->all_slots);
503                 spin_unlock_bh(&iop_chan->lock);
504         } while (iop_chan->slots_allocated < num_descs_in_pool);
505
506         if (idx && !iop_chan->last_used)
507                 iop_chan->last_used = list_entry(iop_chan->all_slots.next,
508                                         struct iop_adma_desc_slot,
509                                         slot_node);
510
511         dev_dbg(iop_chan->device->common.dev,
512                 "allocated %d descriptor slots last_used: %p\n",
513                 iop_chan->slots_allocated, iop_chan->last_used);
514
515         /* initialize the channel and the chain with a null operation */
516         if (init) {
517                 if (dma_has_cap(DMA_MEMCPY,
518                         iop_chan->device->common.cap_mask))
519                         iop_chan_start_null_memcpy(iop_chan);
520                 else if (dma_has_cap(DMA_XOR,
521                         iop_chan->device->common.cap_mask))
522                         iop_chan_start_null_xor(iop_chan);
523                 else
524                         BUG();
525         }
526
527         return (idx > 0) ? idx : -ENOMEM;
528 }
529
530 static struct dma_async_tx_descriptor *
531 iop_adma_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags)
532 {
533         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
534         struct iop_adma_desc_slot *sw_desc, *grp_start;
535         int slot_cnt, slots_per_op;
536
537         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
538
539         spin_lock_bh(&iop_chan->lock);
540         slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
541         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
542         if (sw_desc) {
543                 grp_start = sw_desc->group_head;
544                 iop_desc_init_interrupt(grp_start, iop_chan);
545                 grp_start->unmap_len = 0;
546                 sw_desc->async_tx.flags = flags;
547         }
548         spin_unlock_bh(&iop_chan->lock);
549
550         return sw_desc ? &sw_desc->async_tx : NULL;
551 }
552
553 static struct dma_async_tx_descriptor *
554 iop_adma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
555                          dma_addr_t dma_src, size_t len, unsigned long flags)
556 {
557         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
558         struct iop_adma_desc_slot *sw_desc, *grp_start;
559         int slot_cnt, slots_per_op;
560
561         if (unlikely(!len))
562                 return NULL;
563         BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
564
565         dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
566                 __func__, len);
567
568         spin_lock_bh(&iop_chan->lock);
569         slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
570         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
571         if (sw_desc) {
572                 grp_start = sw_desc->group_head;
573                 iop_desc_init_memcpy(grp_start, flags);
574                 iop_desc_set_byte_count(grp_start, iop_chan, len);
575                 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
576                 iop_desc_set_memcpy_src_addr(grp_start, dma_src);
577                 sw_desc->unmap_src_cnt = 1;
578                 sw_desc->unmap_len = len;
579                 sw_desc->async_tx.flags = flags;
580         }
581         spin_unlock_bh(&iop_chan->lock);
582
583         return sw_desc ? &sw_desc->async_tx : NULL;
584 }
585
586 static struct dma_async_tx_descriptor *
587 iop_adma_prep_dma_memset(struct dma_chan *chan, dma_addr_t dma_dest,
588                          int value, size_t len, unsigned long flags)
589 {
590         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
591         struct iop_adma_desc_slot *sw_desc, *grp_start;
592         int slot_cnt, slots_per_op;
593
594         if (unlikely(!len))
595                 return NULL;
596         BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
597
598         dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
599                 __func__, len);
600
601         spin_lock_bh(&iop_chan->lock);
602         slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op);
603         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
604         if (sw_desc) {
605                 grp_start = sw_desc->group_head;
606                 iop_desc_init_memset(grp_start, flags);
607                 iop_desc_set_byte_count(grp_start, iop_chan, len);
608                 iop_desc_set_block_fill_val(grp_start, value);
609                 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
610                 sw_desc->unmap_src_cnt = 1;
611                 sw_desc->unmap_len = len;
612                 sw_desc->async_tx.flags = flags;
613         }
614         spin_unlock_bh(&iop_chan->lock);
615
616         return sw_desc ? &sw_desc->async_tx : NULL;
617 }
618
619 static struct dma_async_tx_descriptor *
620 iop_adma_prep_dma_xor(struct dma_chan *chan, dma_addr_t dma_dest,
621                       dma_addr_t *dma_src, unsigned int src_cnt, size_t len,
622                       unsigned long flags)
623 {
624         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
625         struct iop_adma_desc_slot *sw_desc, *grp_start;
626         int slot_cnt, slots_per_op;
627
628         if (unlikely(!len))
629                 return NULL;
630         BUG_ON(unlikely(len > IOP_ADMA_XOR_MAX_BYTE_COUNT));
631
632         dev_dbg(iop_chan->device->common.dev,
633                 "%s src_cnt: %d len: %u flags: %lx\n",
634                 __func__, src_cnt, len, flags);
635
636         spin_lock_bh(&iop_chan->lock);
637         slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
638         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
639         if (sw_desc) {
640                 grp_start = sw_desc->group_head;
641                 iop_desc_init_xor(grp_start, src_cnt, flags);
642                 iop_desc_set_byte_count(grp_start, iop_chan, len);
643                 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
644                 sw_desc->unmap_src_cnt = src_cnt;
645                 sw_desc->unmap_len = len;
646                 sw_desc->async_tx.flags = flags;
647                 while (src_cnt--)
648                         iop_desc_set_xor_src_addr(grp_start, src_cnt,
649                                                   dma_src[src_cnt]);
650         }
651         spin_unlock_bh(&iop_chan->lock);
652
653         return sw_desc ? &sw_desc->async_tx : NULL;
654 }
655
656 static struct dma_async_tx_descriptor *
657 iop_adma_prep_dma_zero_sum(struct dma_chan *chan, dma_addr_t *dma_src,
658                            unsigned int src_cnt, size_t len, u32 *result,
659                            unsigned long flags)
660 {
661         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
662         struct iop_adma_desc_slot *sw_desc, *grp_start;
663         int slot_cnt, slots_per_op;
664
665         if (unlikely(!len))
666                 return NULL;
667
668         dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
669                 __func__, src_cnt, len);
670
671         spin_lock_bh(&iop_chan->lock);
672         slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
673         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
674         if (sw_desc) {
675                 grp_start = sw_desc->group_head;
676                 iop_desc_init_zero_sum(grp_start, src_cnt, flags);
677                 iop_desc_set_zero_sum_byte_count(grp_start, len);
678                 grp_start->xor_check_result = result;
679                 pr_debug("\t%s: grp_start->xor_check_result: %p\n",
680                         __func__, grp_start->xor_check_result);
681                 sw_desc->unmap_src_cnt = src_cnt;
682                 sw_desc->unmap_len = len;
683                 sw_desc->async_tx.flags = flags;
684                 while (src_cnt--)
685                         iop_desc_set_zero_sum_src_addr(grp_start, src_cnt,
686                                                        dma_src[src_cnt]);
687         }
688         spin_unlock_bh(&iop_chan->lock);
689
690         return sw_desc ? &sw_desc->async_tx : NULL;
691 }
692
693 static void iop_adma_free_chan_resources(struct dma_chan *chan)
694 {
695         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
696         struct iop_adma_desc_slot *iter, *_iter;
697         int in_use_descs = 0;
698
699         iop_adma_slot_cleanup(iop_chan);
700
701         spin_lock_bh(&iop_chan->lock);
702         list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
703                                         chain_node) {
704                 in_use_descs++;
705                 list_del(&iter->chain_node);
706         }
707         list_for_each_entry_safe_reverse(
708                 iter, _iter, &iop_chan->all_slots, slot_node) {
709                 list_del(&iter->slot_node);
710                 kfree(iter);
711                 iop_chan->slots_allocated--;
712         }
713         iop_chan->last_used = NULL;
714
715         dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
716                 __func__, iop_chan->slots_allocated);
717         spin_unlock_bh(&iop_chan->lock);
718
719         /* one is ok since we left it on there on purpose */
720         if (in_use_descs > 1)
721                 printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
722                         in_use_descs - 1);
723 }
724
725 /**
726  * iop_adma_is_complete - poll the status of an ADMA transaction
727  * @chan: ADMA channel handle
728  * @cookie: ADMA transaction identifier
729  */
730 static enum dma_status iop_adma_is_complete(struct dma_chan *chan,
731                                         dma_cookie_t cookie,
732                                         dma_cookie_t *done,
733                                         dma_cookie_t *used)
734 {
735         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
736         dma_cookie_t last_used;
737         dma_cookie_t last_complete;
738         enum dma_status ret;
739
740         last_used = chan->cookie;
741         last_complete = iop_chan->completed_cookie;
742
743         if (done)
744                 *done = last_complete;
745         if (used)
746                 *used = last_used;
747
748         ret = dma_async_is_complete(cookie, last_complete, last_used);
749         if (ret == DMA_SUCCESS)
750                 return ret;
751
752         iop_adma_slot_cleanup(iop_chan);
753
754         last_used = chan->cookie;
755         last_complete = iop_chan->completed_cookie;
756
757         if (done)
758                 *done = last_complete;
759         if (used)
760                 *used = last_used;
761
762         return dma_async_is_complete(cookie, last_complete, last_used);
763 }
764
765 static irqreturn_t iop_adma_eot_handler(int irq, void *data)
766 {
767         struct iop_adma_chan *chan = data;
768
769         dev_dbg(chan->device->common.dev, "%s\n", __func__);
770
771         tasklet_schedule(&chan->irq_tasklet);
772
773         iop_adma_device_clear_eot_status(chan);
774
775         return IRQ_HANDLED;
776 }
777
778 static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
779 {
780         struct iop_adma_chan *chan = data;
781
782         dev_dbg(chan->device->common.dev, "%s\n", __func__);
783
784         tasklet_schedule(&chan->irq_tasklet);
785
786         iop_adma_device_clear_eoc_status(chan);
787
788         return IRQ_HANDLED;
789 }
790
791 static irqreturn_t iop_adma_err_handler(int irq, void *data)
792 {
793         struct iop_adma_chan *chan = data;
794         unsigned long status = iop_chan_get_status(chan);
795
796         dev_printk(KERN_ERR, chan->device->common.dev,
797                 "error ( %s%s%s%s%s%s%s)\n",
798                 iop_is_err_int_parity(status, chan) ? "int_parity " : "",
799                 iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
800                 iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
801                 iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
802                 iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
803                 iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
804                 iop_is_err_split_tx(status, chan) ? "split_tx " : "");
805
806         iop_adma_device_clear_err_status(chan);
807
808         BUG();
809
810         return IRQ_HANDLED;
811 }
812
813 static void iop_adma_issue_pending(struct dma_chan *chan)
814 {
815         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
816
817         if (iop_chan->pending) {
818                 iop_chan->pending = 0;
819                 iop_chan_append(iop_chan);
820         }
821 }
822
823 /*
824  * Perform a transaction to verify the HW works.
825  */
826 #define IOP_ADMA_TEST_SIZE 2000
827
828 static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device)
829 {
830         int i;
831         void *src, *dest;
832         dma_addr_t src_dma, dest_dma;
833         struct dma_chan *dma_chan;
834         dma_cookie_t cookie;
835         struct dma_async_tx_descriptor *tx;
836         int err = 0;
837         struct iop_adma_chan *iop_chan;
838
839         dev_dbg(device->common.dev, "%s\n", __func__);
840
841         src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
842         if (!src)
843                 return -ENOMEM;
844         dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
845         if (!dest) {
846                 kfree(src);
847                 return -ENOMEM;
848         }
849
850         /* Fill in src buffer */
851         for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
852                 ((u8 *) src)[i] = (u8)i;
853
854         /* Start copy, using first DMA channel */
855         dma_chan = container_of(device->common.channels.next,
856                                 struct dma_chan,
857                                 device_node);
858         if (iop_adma_alloc_chan_resources(dma_chan, NULL) < 1) {
859                 err = -ENODEV;
860                 goto out;
861         }
862
863         dest_dma = dma_map_single(dma_chan->device->dev, dest,
864                                 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
865         src_dma = dma_map_single(dma_chan->device->dev, src,
866                                 IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
867         tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
868                                       IOP_ADMA_TEST_SIZE,
869                                       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
870
871         cookie = iop_adma_tx_submit(tx);
872         iop_adma_issue_pending(dma_chan);
873         msleep(1);
874
875         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
876                         DMA_SUCCESS) {
877                 dev_printk(KERN_ERR, dma_chan->device->dev,
878                         "Self-test copy timed out, disabling\n");
879                 err = -ENODEV;
880                 goto free_resources;
881         }
882
883         iop_chan = to_iop_adma_chan(dma_chan);
884         dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
885                 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
886         if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
887                 dev_printk(KERN_ERR, dma_chan->device->dev,
888                         "Self-test copy failed compare, disabling\n");
889                 err = -ENODEV;
890                 goto free_resources;
891         }
892
893 free_resources:
894         iop_adma_free_chan_resources(dma_chan);
895 out:
896         kfree(src);
897         kfree(dest);
898         return err;
899 }
900
901 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
902 static int __devinit
903 iop_adma_xor_zero_sum_self_test(struct iop_adma_device *device)
904 {
905         int i, src_idx;
906         struct page *dest;
907         struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
908         struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
909         dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
910         dma_addr_t dma_addr, dest_dma;
911         struct dma_async_tx_descriptor *tx;
912         struct dma_chan *dma_chan;
913         dma_cookie_t cookie;
914         u8 cmp_byte = 0;
915         u32 cmp_word;
916         u32 zero_sum_result;
917         int err = 0;
918         struct iop_adma_chan *iop_chan;
919
920         dev_dbg(device->common.dev, "%s\n", __func__);
921
922         for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
923                 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
924                 if (!xor_srcs[src_idx])
925                         while (src_idx--) {
926                                 __free_page(xor_srcs[src_idx]);
927                                 return -ENOMEM;
928                         }
929         }
930
931         dest = alloc_page(GFP_KERNEL);
932         if (!dest)
933                 while (src_idx--) {
934                         __free_page(xor_srcs[src_idx]);
935                         return -ENOMEM;
936                 }
937
938         /* Fill in src buffers */
939         for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
940                 u8 *ptr = page_address(xor_srcs[src_idx]);
941                 for (i = 0; i < PAGE_SIZE; i++)
942                         ptr[i] = (1 << src_idx);
943         }
944
945         for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
946                 cmp_byte ^= (u8) (1 << src_idx);
947
948         cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
949                         (cmp_byte << 8) | cmp_byte;
950
951         memset(page_address(dest), 0, PAGE_SIZE);
952
953         dma_chan = container_of(device->common.channels.next,
954                                 struct dma_chan,
955                                 device_node);
956         if (iop_adma_alloc_chan_resources(dma_chan, NULL) < 1) {
957                 err = -ENODEV;
958                 goto out;
959         }
960
961         /* test xor */
962         dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
963                                 PAGE_SIZE, DMA_FROM_DEVICE);
964         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
965                 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
966                                            0, PAGE_SIZE, DMA_TO_DEVICE);
967         tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
968                                    IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE,
969                                    DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
970
971         cookie = iop_adma_tx_submit(tx);
972         iop_adma_issue_pending(dma_chan);
973         msleep(8);
974
975         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
976                 DMA_SUCCESS) {
977                 dev_printk(KERN_ERR, dma_chan->device->dev,
978                         "Self-test xor timed out, disabling\n");
979                 err = -ENODEV;
980                 goto free_resources;
981         }
982
983         iop_chan = to_iop_adma_chan(dma_chan);
984         dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
985                 PAGE_SIZE, DMA_FROM_DEVICE);
986         for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
987                 u32 *ptr = page_address(dest);
988                 if (ptr[i] != cmp_word) {
989                         dev_printk(KERN_ERR, dma_chan->device->dev,
990                                 "Self-test xor failed compare, disabling\n");
991                         err = -ENODEV;
992                         goto free_resources;
993                 }
994         }
995         dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
996                 PAGE_SIZE, DMA_TO_DEVICE);
997
998         /* skip zero sum if the capability is not present */
999         if (!dma_has_cap(DMA_ZERO_SUM, dma_chan->device->cap_mask))
1000                 goto free_resources;
1001
1002         /* zero sum the sources with the destintation page */
1003         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1004                 zero_sum_srcs[i] = xor_srcs[i];
1005         zero_sum_srcs[i] = dest;
1006
1007         zero_sum_result = 1;
1008
1009         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1010                 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1011                                            zero_sum_srcs[i], 0, PAGE_SIZE,
1012                                            DMA_TO_DEVICE);
1013         tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs,
1014                                         IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1015                                         &zero_sum_result,
1016                                         DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1017
1018         cookie = iop_adma_tx_submit(tx);
1019         iop_adma_issue_pending(dma_chan);
1020         msleep(8);
1021
1022         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1023                 dev_printk(KERN_ERR, dma_chan->device->dev,
1024                         "Self-test zero sum timed out, disabling\n");
1025                 err = -ENODEV;
1026                 goto free_resources;
1027         }
1028
1029         if (zero_sum_result != 0) {
1030                 dev_printk(KERN_ERR, dma_chan->device->dev,
1031                         "Self-test zero sum failed compare, disabling\n");
1032                 err = -ENODEV;
1033                 goto free_resources;
1034         }
1035
1036         /* test memset */
1037         dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
1038                         PAGE_SIZE, DMA_FROM_DEVICE);
1039         tx = iop_adma_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE,
1040                                       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1041
1042         cookie = iop_adma_tx_submit(tx);
1043         iop_adma_issue_pending(dma_chan);
1044         msleep(8);
1045
1046         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1047                 dev_printk(KERN_ERR, dma_chan->device->dev,
1048                         "Self-test memset timed out, disabling\n");
1049                 err = -ENODEV;
1050                 goto free_resources;
1051         }
1052
1053         for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1054                 u32 *ptr = page_address(dest);
1055                 if (ptr[i]) {
1056                         dev_printk(KERN_ERR, dma_chan->device->dev,
1057                                 "Self-test memset failed compare, disabling\n");
1058                         err = -ENODEV;
1059                         goto free_resources;
1060                 }
1061         }
1062
1063         /* test for non-zero parity sum */
1064         zero_sum_result = 0;
1065         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1066                 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1067                                            zero_sum_srcs[i], 0, PAGE_SIZE,
1068                                            DMA_TO_DEVICE);
1069         tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs,
1070                                         IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1071                                         &zero_sum_result,
1072                                         DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1073
1074         cookie = iop_adma_tx_submit(tx);
1075         iop_adma_issue_pending(dma_chan);
1076         msleep(8);
1077
1078         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1079                 dev_printk(KERN_ERR, dma_chan->device->dev,
1080                         "Self-test non-zero sum timed out, disabling\n");
1081                 err = -ENODEV;
1082                 goto free_resources;
1083         }
1084
1085         if (zero_sum_result != 1) {
1086                 dev_printk(KERN_ERR, dma_chan->device->dev,
1087                         "Self-test non-zero sum failed compare, disabling\n");
1088                 err = -ENODEV;
1089                 goto free_resources;
1090         }
1091
1092 free_resources:
1093         iop_adma_free_chan_resources(dma_chan);
1094 out:
1095         src_idx = IOP_ADMA_NUM_SRC_TEST;
1096         while (src_idx--)
1097                 __free_page(xor_srcs[src_idx]);
1098         __free_page(dest);
1099         return err;
1100 }
1101
1102 static int __devexit iop_adma_remove(struct platform_device *dev)
1103 {
1104         struct iop_adma_device *device = platform_get_drvdata(dev);
1105         struct dma_chan *chan, *_chan;
1106         struct iop_adma_chan *iop_chan;
1107         int i;
1108         struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
1109
1110         dma_async_device_unregister(&device->common);
1111
1112         for (i = 0; i < 3; i++) {
1113                 unsigned int irq;
1114                 irq = platform_get_irq(dev, i);
1115                 free_irq(irq, device);
1116         }
1117
1118         dma_free_coherent(&dev->dev, plat_data->pool_size,
1119                         device->dma_desc_pool_virt, device->dma_desc_pool);
1120
1121         do {
1122                 struct resource *res;
1123                 res = platform_get_resource(dev, IORESOURCE_MEM, 0);
1124                 release_mem_region(res->start, res->end - res->start);
1125         } while (0);
1126
1127         list_for_each_entry_safe(chan, _chan, &device->common.channels,
1128                                 device_node) {
1129                 iop_chan = to_iop_adma_chan(chan);
1130                 list_del(&chan->device_node);
1131                 kfree(iop_chan);
1132         }
1133         kfree(device);
1134
1135         return 0;
1136 }
1137
1138 static int __devinit iop_adma_probe(struct platform_device *pdev)
1139 {
1140         struct resource *res;
1141         int ret = 0, i;
1142         struct iop_adma_device *adev;
1143         struct iop_adma_chan *iop_chan;
1144         struct dma_device *dma_dev;
1145         struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
1146
1147         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1148         if (!res)
1149                 return -ENODEV;
1150
1151         if (!devm_request_mem_region(&pdev->dev, res->start,
1152                                 res->end - res->start, pdev->name))
1153                 return -EBUSY;
1154
1155         adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1156         if (!adev)
1157                 return -ENOMEM;
1158         dma_dev = &adev->common;
1159
1160         /* allocate coherent memory for hardware descriptors
1161          * note: writecombine gives slightly better performance, but
1162          * requires that we explicitly flush the writes
1163          */
1164         if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1165                                         plat_data->pool_size,
1166                                         &adev->dma_desc_pool,
1167                                         GFP_KERNEL)) == NULL) {
1168                 ret = -ENOMEM;
1169                 goto err_free_adev;
1170         }
1171
1172         dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n",
1173                 __func__, adev->dma_desc_pool_virt,
1174                 (void *) adev->dma_desc_pool);
1175
1176         adev->id = plat_data->hw_id;
1177
1178         /* discover transaction capabilites from the platform data */
1179         dma_dev->cap_mask = plat_data->cap_mask;
1180
1181         adev->pdev = pdev;
1182         platform_set_drvdata(pdev, adev);
1183
1184         INIT_LIST_HEAD(&dma_dev->channels);
1185
1186         /* set base routines */
1187         dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1188         dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1189         dma_dev->device_is_tx_complete = iop_adma_is_complete;
1190         dma_dev->device_issue_pending = iop_adma_issue_pending;
1191         dma_dev->dev = &pdev->dev;
1192
1193         /* set prep routines based on capability */
1194         if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1195                 dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1196         if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1197                 dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
1198         if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1199                 dma_dev->max_xor = iop_adma_get_max_xor();
1200                 dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1201         }
1202         if (dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask))
1203                 dma_dev->device_prep_dma_zero_sum =
1204                         iop_adma_prep_dma_zero_sum;
1205         if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1206                 dma_dev->device_prep_dma_interrupt =
1207                         iop_adma_prep_dma_interrupt;
1208
1209         iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1210         if (!iop_chan) {
1211                 ret = -ENOMEM;
1212                 goto err_free_dma;
1213         }
1214         iop_chan->device = adev;
1215
1216         iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1217                                         res->end - res->start);
1218         if (!iop_chan->mmr_base) {
1219                 ret = -ENOMEM;
1220                 goto err_free_iop_chan;
1221         }
1222         tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1223                 iop_chan);
1224
1225         /* clear errors before enabling interrupts */
1226         iop_adma_device_clear_err_status(iop_chan);
1227
1228         for (i = 0; i < 3; i++) {
1229                 irq_handler_t handler[] = { iop_adma_eot_handler,
1230                                         iop_adma_eoc_handler,
1231                                         iop_adma_err_handler };
1232                 int irq = platform_get_irq(pdev, i);
1233                 if (irq < 0) {
1234                         ret = -ENXIO;
1235                         goto err_free_iop_chan;
1236                 } else {
1237                         ret = devm_request_irq(&pdev->dev, irq,
1238                                         handler[i], 0, pdev->name, iop_chan);
1239                         if (ret)
1240                                 goto err_free_iop_chan;
1241                 }
1242         }
1243
1244         spin_lock_init(&iop_chan->lock);
1245         INIT_LIST_HEAD(&iop_chan->chain);
1246         INIT_LIST_HEAD(&iop_chan->all_slots);
1247         INIT_RCU_HEAD(&iop_chan->common.rcu);
1248         iop_chan->common.device = dma_dev;
1249         list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1250
1251         if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1252                 ret = iop_adma_memcpy_self_test(adev);
1253                 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1254                 if (ret)
1255                         goto err_free_iop_chan;
1256         }
1257
1258         if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
1259                 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1260                 ret = iop_adma_xor_zero_sum_self_test(adev);
1261                 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1262                 if (ret)
1263                         goto err_free_iop_chan;
1264         }
1265
1266         dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
1267           "( %s%s%s%s%s%s%s%s%s%s)\n",
1268           dma_has_cap(DMA_PQ_XOR, dma_dev->cap_mask) ? "pq_xor " : "",
1269           dma_has_cap(DMA_PQ_UPDATE, dma_dev->cap_mask) ? "pq_update " : "",
1270           dma_has_cap(DMA_PQ_ZERO_SUM, dma_dev->cap_mask) ? "pq_zero_sum " : "",
1271           dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1272           dma_has_cap(DMA_DUAL_XOR, dma_dev->cap_mask) ? "dual_xor " : "",
1273           dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask) ? "xor_zero_sum " : "",
1274           dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)  ? "fill " : "",
1275           dma_has_cap(DMA_MEMCPY_CRC32C, dma_dev->cap_mask) ? "cpy+crc " : "",
1276           dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1277           dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1278
1279         dma_async_device_register(dma_dev);
1280         goto out;
1281
1282  err_free_iop_chan:
1283         kfree(iop_chan);
1284  err_free_dma:
1285         dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1286                         adev->dma_desc_pool_virt, adev->dma_desc_pool);
1287  err_free_adev:
1288         kfree(adev);
1289  out:
1290         return ret;
1291 }
1292
1293 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1294 {
1295         struct iop_adma_desc_slot *sw_desc, *grp_start;
1296         dma_cookie_t cookie;
1297         int slot_cnt, slots_per_op;
1298
1299         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1300
1301         spin_lock_bh(&iop_chan->lock);
1302         slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1303         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1304         if (sw_desc) {
1305                 grp_start = sw_desc->group_head;
1306
1307                 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1308                 async_tx_ack(&sw_desc->async_tx);
1309                 iop_desc_init_memcpy(grp_start, 0);
1310                 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1311                 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1312                 iop_desc_set_memcpy_src_addr(grp_start, 0);
1313
1314                 cookie = iop_chan->common.cookie;
1315                 cookie++;
1316                 if (cookie <= 1)
1317                         cookie = 2;
1318
1319                 /* initialize the completed cookie to be less than
1320                  * the most recently used cookie
1321                  */
1322                 iop_chan->completed_cookie = cookie - 1;
1323                 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1324
1325                 /* channel should not be busy */
1326                 BUG_ON(iop_chan_is_busy(iop_chan));
1327
1328                 /* clear any prior error-status bits */
1329                 iop_adma_device_clear_err_status(iop_chan);
1330
1331                 /* disable operation */
1332                 iop_chan_disable(iop_chan);
1333
1334                 /* set the descriptor address */
1335                 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1336
1337                 /* 1/ don't add pre-chained descriptors
1338                  * 2/ dummy read to flush next_desc write
1339                  */
1340                 BUG_ON(iop_desc_get_next_desc(sw_desc));
1341
1342                 /* run the descriptor */
1343                 iop_chan_enable(iop_chan);
1344         } else
1345                 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1346                          "failed to allocate null descriptor\n");
1347         spin_unlock_bh(&iop_chan->lock);
1348 }
1349
1350 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1351 {
1352         struct iop_adma_desc_slot *sw_desc, *grp_start;
1353         dma_cookie_t cookie;
1354         int slot_cnt, slots_per_op;
1355
1356         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1357
1358         spin_lock_bh(&iop_chan->lock);
1359         slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1360         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1361         if (sw_desc) {
1362                 grp_start = sw_desc->group_head;
1363                 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1364                 async_tx_ack(&sw_desc->async_tx);
1365                 iop_desc_init_null_xor(grp_start, 2, 0);
1366                 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1367                 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1368                 iop_desc_set_xor_src_addr(grp_start, 0, 0);
1369                 iop_desc_set_xor_src_addr(grp_start, 1, 0);
1370
1371                 cookie = iop_chan->common.cookie;
1372                 cookie++;
1373                 if (cookie <= 1)
1374                         cookie = 2;
1375
1376                 /* initialize the completed cookie to be less than
1377                  * the most recently used cookie
1378                  */
1379                 iop_chan->completed_cookie = cookie - 1;
1380                 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1381
1382                 /* channel should not be busy */
1383                 BUG_ON(iop_chan_is_busy(iop_chan));
1384
1385                 /* clear any prior error-status bits */
1386                 iop_adma_device_clear_err_status(iop_chan);
1387
1388                 /* disable operation */
1389                 iop_chan_disable(iop_chan);
1390
1391                 /* set the descriptor address */
1392                 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1393
1394                 /* 1/ don't add pre-chained descriptors
1395                  * 2/ dummy read to flush next_desc write
1396                  */
1397                 BUG_ON(iop_desc_get_next_desc(sw_desc));
1398
1399                 /* run the descriptor */
1400                 iop_chan_enable(iop_chan);
1401         } else
1402                 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1403                         "failed to allocate null descriptor\n");
1404         spin_unlock_bh(&iop_chan->lock);
1405 }
1406
1407 MODULE_ALIAS("platform:iop-adma");
1408
1409 static struct platform_driver iop_adma_driver = {
1410         .probe          = iop_adma_probe,
1411         .remove         = iop_adma_remove,
1412         .driver         = {
1413                 .owner  = THIS_MODULE,
1414                 .name   = "iop-adma",
1415         },
1416 };
1417
1418 static int __init iop_adma_init (void)
1419 {
1420         return platform_driver_register(&iop_adma_driver);
1421 }
1422
1423 /* it's currently unsafe to unload this module */
1424 #if 0
1425 static void __exit iop_adma_exit (void)
1426 {
1427         platform_driver_unregister(&iop_adma_driver);
1428         return;
1429 }
1430 module_exit(iop_adma_exit);
1431 #endif
1432
1433 module_init(iop_adma_init);
1434
1435 MODULE_AUTHOR("Intel Corporation");
1436 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1437 MODULE_LICENSE("GPL");