mm-only debug patch...
[mmotm.git] / drivers / dma / iop-adma.c
blob645ca8d54ec43350059bd8d7c8802f017b4a4abd
1 /*
2 * offload engine driver for the Intel Xscale series of i/o processors
3 * Copyright © 2006, Intel Corporation.
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.
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.
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.
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)
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/delay.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/spinlock.h>
30 #include <linux/interrupt.h>
31 #include <linux/platform_device.h>
32 #include <linux/memory.h>
33 #include <linux/ioport.h>
34 #include <linux/raid/pq.h>
36 #include <mach/adma.h>
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)
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
49 static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
51 int stride = slot->slots_per_op;
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);
61 static void
62 iop_desc_unmap(struct iop_adma_chan *iop_chan, struct iop_adma_desc_slot *desc)
64 struct dma_async_tx_descriptor *tx = &desc->async_tx;
65 struct iop_adma_desc_slot *unmap = desc->group_head;
66 struct device *dev = &iop_chan->device->pdev->dev;
67 u32 len = unmap->unmap_len;
68 enum dma_ctrl_flags flags = tx->flags;
69 u32 src_cnt;
70 dma_addr_t addr;
71 dma_addr_t dest;
73 src_cnt = unmap->unmap_src_cnt;
74 dest = iop_desc_get_dest_addr(unmap, iop_chan);
75 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
76 enum dma_data_direction dir;
78 if (src_cnt > 1) /* is xor? */
79 dir = DMA_BIDIRECTIONAL;
80 else
81 dir = DMA_FROM_DEVICE;
83 dma_unmap_page(dev, dest, len, dir);
86 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
87 while (src_cnt--) {
88 addr = iop_desc_get_src_addr(unmap, iop_chan, src_cnt);
89 if (addr == dest)
90 continue;
91 dma_unmap_page(dev, addr, len, DMA_TO_DEVICE);
94 desc->group_head = NULL;
97 static void
98 iop_desc_unmap_pq(struct iop_adma_chan *iop_chan, struct iop_adma_desc_slot *desc)
100 struct dma_async_tx_descriptor *tx = &desc->async_tx;
101 struct iop_adma_desc_slot *unmap = desc->group_head;
102 struct device *dev = &iop_chan->device->pdev->dev;
103 u32 len = unmap->unmap_len;
104 enum dma_ctrl_flags flags = tx->flags;
105 u32 src_cnt = unmap->unmap_src_cnt;
106 dma_addr_t pdest = iop_desc_get_dest_addr(unmap, iop_chan);
107 dma_addr_t qdest = iop_desc_get_qdest_addr(unmap, iop_chan);
108 int i;
110 if (tx->flags & DMA_PREP_CONTINUE)
111 src_cnt -= 3;
113 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP) && !desc->pq_check_result) {
114 dma_unmap_page(dev, pdest, len, DMA_BIDIRECTIONAL);
115 dma_unmap_page(dev, qdest, len, DMA_BIDIRECTIONAL);
118 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
119 dma_addr_t addr;
121 for (i = 0; i < src_cnt; i++) {
122 addr = iop_desc_get_src_addr(unmap, iop_chan, i);
123 dma_unmap_page(dev, addr, len, DMA_TO_DEVICE);
125 if (desc->pq_check_result) {
126 dma_unmap_page(dev, pdest, len, DMA_TO_DEVICE);
127 dma_unmap_page(dev, qdest, len, DMA_TO_DEVICE);
131 desc->group_head = NULL;
135 static dma_cookie_t
136 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
137 struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
139 struct dma_async_tx_descriptor *tx = &desc->async_tx;
141 BUG_ON(tx->cookie < 0);
142 if (tx->cookie > 0) {
143 cookie = tx->cookie;
144 tx->cookie = 0;
146 /* call the callback (must not sleep or submit new
147 * operations to this channel)
149 if (tx->callback)
150 tx->callback(tx->callback_param);
152 /* unmap dma addresses
153 * (unmap_single vs unmap_page?)
155 if (desc->group_head && desc->unmap_len) {
156 if (iop_desc_is_pq(desc))
157 iop_desc_unmap_pq(iop_chan, desc);
158 else
159 iop_desc_unmap(iop_chan, desc);
163 /* run dependent operations */
164 dma_run_dependencies(tx);
166 return cookie;
169 static int
170 iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
171 struct iop_adma_chan *iop_chan)
173 /* the client is allowed to attach dependent operations
174 * until 'ack' is set
176 if (!async_tx_test_ack(&desc->async_tx))
177 return 0;
179 /* leave the last descriptor in the chain
180 * so we can append to it
182 if (desc->chain_node.next == &iop_chan->chain)
183 return 1;
185 dev_dbg(iop_chan->device->common.dev,
186 "\tfree slot: %d slots_per_op: %d\n",
187 desc->idx, desc->slots_per_op);
189 list_del(&desc->chain_node);
190 iop_adma_free_slots(desc);
192 return 0;
195 static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
197 struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
198 dma_cookie_t cookie = 0;
199 u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
200 int busy = iop_chan_is_busy(iop_chan);
201 int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
203 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
204 /* free completed slots from the chain starting with
205 * the oldest descriptor
207 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
208 chain_node) {
209 pr_debug("\tcookie: %d slot: %d busy: %d "
210 "this_desc: %#x next_desc: %#x ack: %d\n",
211 iter->async_tx.cookie, iter->idx, busy,
212 iter->async_tx.phys, iop_desc_get_next_desc(iter),
213 async_tx_test_ack(&iter->async_tx));
214 prefetch(_iter);
215 prefetch(&_iter->async_tx);
217 /* do not advance past the current descriptor loaded into the
218 * hardware channel, subsequent descriptors are either in
219 * process or have not been submitted
221 if (seen_current)
222 break;
224 /* stop the search if we reach the current descriptor and the
225 * channel is busy, or if it appears that the current descriptor
226 * needs to be re-read (i.e. has been appended to)
228 if (iter->async_tx.phys == current_desc) {
229 BUG_ON(seen_current++);
230 if (busy || iop_desc_get_next_desc(iter))
231 break;
234 /* detect the start of a group transaction */
235 if (!slot_cnt && !slots_per_op) {
236 slot_cnt = iter->slot_cnt;
237 slots_per_op = iter->slots_per_op;
238 if (slot_cnt <= slots_per_op) {
239 slot_cnt = 0;
240 slots_per_op = 0;
244 if (slot_cnt) {
245 pr_debug("\tgroup++\n");
246 if (!grp_start)
247 grp_start = iter;
248 slot_cnt -= slots_per_op;
251 /* all the members of a group are complete */
252 if (slots_per_op != 0 && slot_cnt == 0) {
253 struct iop_adma_desc_slot *grp_iter, *_grp_iter;
254 int end_of_chain = 0;
255 pr_debug("\tgroup end\n");
257 /* collect the total results */
258 if (grp_start->xor_check_result) {
259 u32 zero_sum_result = 0;
260 slot_cnt = grp_start->slot_cnt;
261 grp_iter = grp_start;
263 list_for_each_entry_from(grp_iter,
264 &iop_chan->chain, chain_node) {
265 zero_sum_result |=
266 iop_desc_get_zero_result(grp_iter);
267 pr_debug("\titer%d result: %d\n",
268 grp_iter->idx, zero_sum_result);
269 slot_cnt -= slots_per_op;
270 if (slot_cnt == 0)
271 break;
273 pr_debug("\tgrp_start->xor_check_result: %p\n",
274 grp_start->xor_check_result);
275 *grp_start->xor_check_result = zero_sum_result;
278 /* clean up the group */
279 slot_cnt = grp_start->slot_cnt;
280 grp_iter = grp_start;
281 list_for_each_entry_safe_from(grp_iter, _grp_iter,
282 &iop_chan->chain, chain_node) {
283 cookie = iop_adma_run_tx_complete_actions(
284 grp_iter, iop_chan, cookie);
286 slot_cnt -= slots_per_op;
287 end_of_chain = iop_adma_clean_slot(grp_iter,
288 iop_chan);
290 if (slot_cnt == 0 || end_of_chain)
291 break;
294 /* the group should be complete at this point */
295 BUG_ON(slot_cnt);
297 slots_per_op = 0;
298 grp_start = NULL;
299 if (end_of_chain)
300 break;
301 else
302 continue;
303 } else if (slots_per_op) /* wait for group completion */
304 continue;
306 /* write back zero sum results (single descriptor case) */
307 if (iter->xor_check_result && iter->async_tx.cookie)
308 *iter->xor_check_result =
309 iop_desc_get_zero_result(iter);
311 cookie = iop_adma_run_tx_complete_actions(
312 iter, iop_chan, cookie);
314 if (iop_adma_clean_slot(iter, iop_chan))
315 break;
318 if (cookie > 0) {
319 iop_chan->completed_cookie = cookie;
320 pr_debug("\tcompleted cookie %d\n", cookie);
324 static void
325 iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
327 spin_lock_bh(&iop_chan->lock);
328 __iop_adma_slot_cleanup(iop_chan);
329 spin_unlock_bh(&iop_chan->lock);
332 static void iop_adma_tasklet(unsigned long data)
334 struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data;
336 /* lockdep will flag depedency submissions as potentially
337 * recursive locking, this is not the case as a dependency
338 * submission will never recurse a channels submit routine.
339 * There are checks in async_tx.c to prevent this.
341 spin_lock_nested(&iop_chan->lock, SINGLE_DEPTH_NESTING);
342 __iop_adma_slot_cleanup(iop_chan);
343 spin_unlock(&iop_chan->lock);
346 static struct iop_adma_desc_slot *
347 iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
348 int slots_per_op)
350 struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
351 LIST_HEAD(chain);
352 int slots_found, retry = 0;
354 /* start search from the last allocated descrtiptor
355 * if a contiguous allocation can not be found start searching
356 * from the beginning of the list
358 retry:
359 slots_found = 0;
360 if (retry == 0)
361 iter = iop_chan->last_used;
362 else
363 iter = list_entry(&iop_chan->all_slots,
364 struct iop_adma_desc_slot,
365 slot_node);
367 list_for_each_entry_safe_continue(
368 iter, _iter, &iop_chan->all_slots, slot_node) {
369 prefetch(_iter);
370 prefetch(&_iter->async_tx);
371 if (iter->slots_per_op) {
372 /* give up after finding the first busy slot
373 * on the second pass through the list
375 if (retry)
376 break;
378 slots_found = 0;
379 continue;
382 /* start the allocation if the slot is correctly aligned */
383 if (!slots_found++) {
384 if (iop_desc_is_aligned(iter, slots_per_op))
385 alloc_start = iter;
386 else {
387 slots_found = 0;
388 continue;
392 if (slots_found == num_slots) {
393 struct iop_adma_desc_slot *alloc_tail = NULL;
394 struct iop_adma_desc_slot *last_used = NULL;
395 iter = alloc_start;
396 while (num_slots) {
397 int i;
398 dev_dbg(iop_chan->device->common.dev,
399 "allocated slot: %d "
400 "(desc %p phys: %#x) slots_per_op %d\n",
401 iter->idx, iter->hw_desc,
402 iter->async_tx.phys, slots_per_op);
404 /* pre-ack all but the last descriptor */
405 if (num_slots != slots_per_op)
406 async_tx_ack(&iter->async_tx);
408 list_add_tail(&iter->chain_node, &chain);
409 alloc_tail = iter;
410 iter->async_tx.cookie = 0;
411 iter->slot_cnt = num_slots;
412 iter->xor_check_result = NULL;
413 for (i = 0; i < slots_per_op; i++) {
414 iter->slots_per_op = slots_per_op - i;
415 last_used = iter;
416 iter = list_entry(iter->slot_node.next,
417 struct iop_adma_desc_slot,
418 slot_node);
420 num_slots -= slots_per_op;
422 alloc_tail->group_head = alloc_start;
423 alloc_tail->async_tx.cookie = -EBUSY;
424 list_splice(&chain, &alloc_tail->tx_list);
425 iop_chan->last_used = last_used;
426 iop_desc_clear_next_desc(alloc_start);
427 iop_desc_clear_next_desc(alloc_tail);
428 return alloc_tail;
431 if (!retry++)
432 goto retry;
434 /* perform direct reclaim if the allocation fails */
435 __iop_adma_slot_cleanup(iop_chan);
437 return NULL;
440 static dma_cookie_t
441 iop_desc_assign_cookie(struct iop_adma_chan *iop_chan,
442 struct iop_adma_desc_slot *desc)
444 dma_cookie_t cookie = iop_chan->common.cookie;
445 cookie++;
446 if (cookie < 0)
447 cookie = 1;
448 iop_chan->common.cookie = desc->async_tx.cookie = cookie;
449 return cookie;
452 static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
454 dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
455 iop_chan->pending);
457 if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
458 iop_chan->pending = 0;
459 iop_chan_append(iop_chan);
463 static dma_cookie_t
464 iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
466 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
467 struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
468 struct iop_adma_desc_slot *grp_start, *old_chain_tail;
469 int slot_cnt;
470 int slots_per_op;
471 dma_cookie_t cookie;
472 dma_addr_t next_dma;
474 grp_start = sw_desc->group_head;
475 slot_cnt = grp_start->slot_cnt;
476 slots_per_op = grp_start->slots_per_op;
478 spin_lock_bh(&iop_chan->lock);
479 cookie = iop_desc_assign_cookie(iop_chan, sw_desc);
481 old_chain_tail = list_entry(iop_chan->chain.prev,
482 struct iop_adma_desc_slot, chain_node);
483 list_splice_init(&sw_desc->tx_list,
484 &old_chain_tail->chain_node);
486 /* fix up the hardware chain */
487 next_dma = grp_start->async_tx.phys;
488 iop_desc_set_next_desc(old_chain_tail, next_dma);
489 BUG_ON(iop_desc_get_next_desc(old_chain_tail) != next_dma); /* flush */
491 /* check for pre-chained descriptors */
492 iop_paranoia(iop_desc_get_next_desc(sw_desc));
494 /* increment the pending count by the number of slots
495 * memcpy operations have a 1:1 (slot:operation) relation
496 * other operations are heavier and will pop the threshold
497 * more often.
499 iop_chan->pending += slot_cnt;
500 iop_adma_check_threshold(iop_chan);
501 spin_unlock_bh(&iop_chan->lock);
503 dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
504 __func__, sw_desc->async_tx.cookie, sw_desc->idx);
506 return cookie;
509 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
510 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
513 * iop_adma_alloc_chan_resources - returns the number of allocated descriptors
514 * @chan - allocate descriptor resources for this channel
515 * @client - current client requesting the channel be ready for requests
517 * Note: We keep the slots for 1 operation on iop_chan->chain at all times. To
518 * avoid deadlock, via async_xor, num_descs_in_pool must at a minimum be
519 * greater than 2x the number slots needed to satisfy a device->max_xor
520 * request.
521 * */
522 static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
524 char *hw_desc;
525 int idx;
526 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
527 struct iop_adma_desc_slot *slot = NULL;
528 int init = iop_chan->slots_allocated ? 0 : 1;
529 struct iop_adma_platform_data *plat_data =
530 iop_chan->device->pdev->dev.platform_data;
531 int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
533 /* Allocate descriptor slots */
534 do {
535 idx = iop_chan->slots_allocated;
536 if (idx == num_descs_in_pool)
537 break;
539 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
540 if (!slot) {
541 printk(KERN_INFO "IOP ADMA Channel only initialized"
542 " %d descriptor slots", idx);
543 break;
545 hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
546 slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
548 dma_async_tx_descriptor_init(&slot->async_tx, chan);
549 slot->async_tx.tx_submit = iop_adma_tx_submit;
550 INIT_LIST_HEAD(&slot->tx_list);
551 INIT_LIST_HEAD(&slot->chain_node);
552 INIT_LIST_HEAD(&slot->slot_node);
553 hw_desc = (char *) iop_chan->device->dma_desc_pool;
554 slot->async_tx.phys =
555 (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
556 slot->idx = idx;
558 spin_lock_bh(&iop_chan->lock);
559 iop_chan->slots_allocated++;
560 list_add_tail(&slot->slot_node, &iop_chan->all_slots);
561 spin_unlock_bh(&iop_chan->lock);
562 } while (iop_chan->slots_allocated < num_descs_in_pool);
564 if (idx && !iop_chan->last_used)
565 iop_chan->last_used = list_entry(iop_chan->all_slots.next,
566 struct iop_adma_desc_slot,
567 slot_node);
569 dev_dbg(iop_chan->device->common.dev,
570 "allocated %d descriptor slots last_used: %p\n",
571 iop_chan->slots_allocated, iop_chan->last_used);
573 /* initialize the channel and the chain with a null operation */
574 if (init) {
575 if (dma_has_cap(DMA_MEMCPY,
576 iop_chan->device->common.cap_mask))
577 iop_chan_start_null_memcpy(iop_chan);
578 else if (dma_has_cap(DMA_XOR,
579 iop_chan->device->common.cap_mask))
580 iop_chan_start_null_xor(iop_chan);
581 else
582 BUG();
585 return (idx > 0) ? idx : -ENOMEM;
588 static struct dma_async_tx_descriptor *
589 iop_adma_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags)
591 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
592 struct iop_adma_desc_slot *sw_desc, *grp_start;
593 int slot_cnt, slots_per_op;
595 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
597 spin_lock_bh(&iop_chan->lock);
598 slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
599 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
600 if (sw_desc) {
601 grp_start = sw_desc->group_head;
602 iop_desc_init_interrupt(grp_start, iop_chan);
603 grp_start->unmap_len = 0;
604 sw_desc->async_tx.flags = flags;
606 spin_unlock_bh(&iop_chan->lock);
608 return sw_desc ? &sw_desc->async_tx : NULL;
611 static struct dma_async_tx_descriptor *
612 iop_adma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
613 dma_addr_t dma_src, size_t len, unsigned long flags)
615 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
616 struct iop_adma_desc_slot *sw_desc, *grp_start;
617 int slot_cnt, slots_per_op;
619 if (unlikely(!len))
620 return NULL;
621 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
623 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
624 __func__, len);
626 spin_lock_bh(&iop_chan->lock);
627 slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
628 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
629 if (sw_desc) {
630 grp_start = sw_desc->group_head;
631 iop_desc_init_memcpy(grp_start, flags);
632 iop_desc_set_byte_count(grp_start, iop_chan, len);
633 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
634 iop_desc_set_memcpy_src_addr(grp_start, dma_src);
635 sw_desc->unmap_src_cnt = 1;
636 sw_desc->unmap_len = len;
637 sw_desc->async_tx.flags = flags;
639 spin_unlock_bh(&iop_chan->lock);
641 return sw_desc ? &sw_desc->async_tx : NULL;
644 static struct dma_async_tx_descriptor *
645 iop_adma_prep_dma_memset(struct dma_chan *chan, dma_addr_t dma_dest,
646 int value, size_t len, unsigned long flags)
648 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
649 struct iop_adma_desc_slot *sw_desc, *grp_start;
650 int slot_cnt, slots_per_op;
652 if (unlikely(!len))
653 return NULL;
654 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
656 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
657 __func__, len);
659 spin_lock_bh(&iop_chan->lock);
660 slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op);
661 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
662 if (sw_desc) {
663 grp_start = sw_desc->group_head;
664 iop_desc_init_memset(grp_start, flags);
665 iop_desc_set_byte_count(grp_start, iop_chan, len);
666 iop_desc_set_block_fill_val(grp_start, value);
667 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
668 sw_desc->unmap_src_cnt = 1;
669 sw_desc->unmap_len = len;
670 sw_desc->async_tx.flags = flags;
672 spin_unlock_bh(&iop_chan->lock);
674 return sw_desc ? &sw_desc->async_tx : NULL;
677 static struct dma_async_tx_descriptor *
678 iop_adma_prep_dma_xor(struct dma_chan *chan, dma_addr_t dma_dest,
679 dma_addr_t *dma_src, unsigned int src_cnt, size_t len,
680 unsigned long flags)
682 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
683 struct iop_adma_desc_slot *sw_desc, *grp_start;
684 int slot_cnt, slots_per_op;
686 if (unlikely(!len))
687 return NULL;
688 BUG_ON(unlikely(len > IOP_ADMA_XOR_MAX_BYTE_COUNT));
690 dev_dbg(iop_chan->device->common.dev,
691 "%s src_cnt: %d len: %u flags: %lx\n",
692 __func__, src_cnt, len, flags);
694 spin_lock_bh(&iop_chan->lock);
695 slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
696 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
697 if (sw_desc) {
698 grp_start = sw_desc->group_head;
699 iop_desc_init_xor(grp_start, src_cnt, flags);
700 iop_desc_set_byte_count(grp_start, iop_chan, len);
701 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
702 sw_desc->unmap_src_cnt = src_cnt;
703 sw_desc->unmap_len = len;
704 sw_desc->async_tx.flags = flags;
705 while (src_cnt--)
706 iop_desc_set_xor_src_addr(grp_start, src_cnt,
707 dma_src[src_cnt]);
709 spin_unlock_bh(&iop_chan->lock);
711 return sw_desc ? &sw_desc->async_tx : NULL;
714 static struct dma_async_tx_descriptor *
715 iop_adma_prep_dma_xor_val(struct dma_chan *chan, dma_addr_t *dma_src,
716 unsigned int src_cnt, size_t len, u32 *result,
717 unsigned long flags)
719 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
720 struct iop_adma_desc_slot *sw_desc, *grp_start;
721 int slot_cnt, slots_per_op;
723 if (unlikely(!len))
724 return NULL;
726 dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
727 __func__, src_cnt, len);
729 spin_lock_bh(&iop_chan->lock);
730 slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
731 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
732 if (sw_desc) {
733 grp_start = sw_desc->group_head;
734 iop_desc_init_zero_sum(grp_start, src_cnt, flags);
735 iop_desc_set_zero_sum_byte_count(grp_start, len);
736 grp_start->xor_check_result = result;
737 pr_debug("\t%s: grp_start->xor_check_result: %p\n",
738 __func__, grp_start->xor_check_result);
739 sw_desc->unmap_src_cnt = src_cnt;
740 sw_desc->unmap_len = len;
741 sw_desc->async_tx.flags = flags;
742 while (src_cnt--)
743 iop_desc_set_zero_sum_src_addr(grp_start, src_cnt,
744 dma_src[src_cnt]);
746 spin_unlock_bh(&iop_chan->lock);
748 return sw_desc ? &sw_desc->async_tx : NULL;
751 static struct dma_async_tx_descriptor *
752 iop_adma_prep_dma_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
753 unsigned int src_cnt, const unsigned char *scf, size_t len,
754 unsigned long flags)
756 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
757 struct iop_adma_desc_slot *sw_desc, *g;
758 int slot_cnt, slots_per_op;
759 int continue_srcs;
761 if (unlikely(!len))
762 return NULL;
763 BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
765 dev_dbg(iop_chan->device->common.dev,
766 "%s src_cnt: %d len: %u flags: %lx\n",
767 __func__, src_cnt, len, flags);
769 if (dmaf_p_disabled_continue(flags))
770 continue_srcs = 1+src_cnt;
771 else if (dmaf_continue(flags))
772 continue_srcs = 3+src_cnt;
773 else
774 continue_srcs = 0+src_cnt;
776 spin_lock_bh(&iop_chan->lock);
777 slot_cnt = iop_chan_pq_slot_count(len, continue_srcs, &slots_per_op);
778 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
779 if (sw_desc) {
780 int i;
782 g = sw_desc->group_head;
783 iop_desc_set_byte_count(g, iop_chan, len);
785 /* even if P is disabled its destination address (bits
786 * [3:0]) must match Q. It is ok if P points to an
787 * invalid address, it won't be written.
789 if (flags & DMA_PREP_PQ_DISABLE_P)
790 dst[0] = dst[1] & 0x7;
792 iop_desc_set_pq_addr(g, dst);
793 sw_desc->unmap_src_cnt = src_cnt;
794 sw_desc->unmap_len = len;
795 sw_desc->async_tx.flags = flags;
796 for (i = 0; i < src_cnt; i++)
797 iop_desc_set_pq_src_addr(g, i, src[i], scf[i]);
799 /* if we are continuing a previous operation factor in
800 * the old p and q values, see the comment for dma_maxpq
801 * in include/linux/dmaengine.h
803 if (dmaf_p_disabled_continue(flags))
804 iop_desc_set_pq_src_addr(g, i++, dst[1], 1);
805 else if (dmaf_continue(flags)) {
806 iop_desc_set_pq_src_addr(g, i++, dst[0], 0);
807 iop_desc_set_pq_src_addr(g, i++, dst[1], 1);
808 iop_desc_set_pq_src_addr(g, i++, dst[1], 0);
810 iop_desc_init_pq(g, i, flags);
812 spin_unlock_bh(&iop_chan->lock);
814 return sw_desc ? &sw_desc->async_tx : NULL;
817 static struct dma_async_tx_descriptor *
818 iop_adma_prep_dma_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
819 unsigned int src_cnt, const unsigned char *scf,
820 size_t len, enum sum_check_flags *pqres,
821 unsigned long flags)
823 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
824 struct iop_adma_desc_slot *sw_desc, *g;
825 int slot_cnt, slots_per_op;
827 if (unlikely(!len))
828 return NULL;
829 BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
831 dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
832 __func__, src_cnt, len);
834 spin_lock_bh(&iop_chan->lock);
835 slot_cnt = iop_chan_pq_zero_sum_slot_count(len, src_cnt + 2, &slots_per_op);
836 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
837 if (sw_desc) {
838 /* for validate operations p and q are tagged onto the
839 * end of the source list
841 int pq_idx = src_cnt;
843 g = sw_desc->group_head;
844 iop_desc_init_pq_zero_sum(g, src_cnt+2, flags);
845 iop_desc_set_pq_zero_sum_byte_count(g, len);
846 g->pq_check_result = pqres;
847 pr_debug("\t%s: g->pq_check_result: %p\n",
848 __func__, g->pq_check_result);
849 sw_desc->unmap_src_cnt = src_cnt+2;
850 sw_desc->unmap_len = len;
851 sw_desc->async_tx.flags = flags;
852 while (src_cnt--)
853 iop_desc_set_pq_zero_sum_src_addr(g, src_cnt,
854 src[src_cnt],
855 scf[src_cnt]);
856 iop_desc_set_pq_zero_sum_addr(g, pq_idx, src);
858 spin_unlock_bh(&iop_chan->lock);
860 return sw_desc ? &sw_desc->async_tx : NULL;
863 static void iop_adma_free_chan_resources(struct dma_chan *chan)
865 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
866 struct iop_adma_desc_slot *iter, *_iter;
867 int in_use_descs = 0;
869 iop_adma_slot_cleanup(iop_chan);
871 spin_lock_bh(&iop_chan->lock);
872 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
873 chain_node) {
874 in_use_descs++;
875 list_del(&iter->chain_node);
877 list_for_each_entry_safe_reverse(
878 iter, _iter, &iop_chan->all_slots, slot_node) {
879 list_del(&iter->slot_node);
880 kfree(iter);
881 iop_chan->slots_allocated--;
883 iop_chan->last_used = NULL;
885 dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
886 __func__, iop_chan->slots_allocated);
887 spin_unlock_bh(&iop_chan->lock);
889 /* one is ok since we left it on there on purpose */
890 if (in_use_descs > 1)
891 printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
892 in_use_descs - 1);
896 * iop_adma_is_complete - poll the status of an ADMA transaction
897 * @chan: ADMA channel handle
898 * @cookie: ADMA transaction identifier
900 static enum dma_status iop_adma_is_complete(struct dma_chan *chan,
901 dma_cookie_t cookie,
902 dma_cookie_t *done,
903 dma_cookie_t *used)
905 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
906 dma_cookie_t last_used;
907 dma_cookie_t last_complete;
908 enum dma_status ret;
910 last_used = chan->cookie;
911 last_complete = iop_chan->completed_cookie;
913 if (done)
914 *done = last_complete;
915 if (used)
916 *used = last_used;
918 ret = dma_async_is_complete(cookie, last_complete, last_used);
919 if (ret == DMA_SUCCESS)
920 return ret;
922 iop_adma_slot_cleanup(iop_chan);
924 last_used = chan->cookie;
925 last_complete = iop_chan->completed_cookie;
927 if (done)
928 *done = last_complete;
929 if (used)
930 *used = last_used;
932 return dma_async_is_complete(cookie, last_complete, last_used);
935 static irqreturn_t iop_adma_eot_handler(int irq, void *data)
937 struct iop_adma_chan *chan = data;
939 dev_dbg(chan->device->common.dev, "%s\n", __func__);
941 tasklet_schedule(&chan->irq_tasklet);
943 iop_adma_device_clear_eot_status(chan);
945 return IRQ_HANDLED;
948 static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
950 struct iop_adma_chan *chan = data;
952 dev_dbg(chan->device->common.dev, "%s\n", __func__);
954 tasklet_schedule(&chan->irq_tasklet);
956 iop_adma_device_clear_eoc_status(chan);
958 return IRQ_HANDLED;
961 static irqreturn_t iop_adma_err_handler(int irq, void *data)
963 struct iop_adma_chan *chan = data;
964 unsigned long status = iop_chan_get_status(chan);
966 dev_printk(KERN_ERR, chan->device->common.dev,
967 "error ( %s%s%s%s%s%s%s)\n",
968 iop_is_err_int_parity(status, chan) ? "int_parity " : "",
969 iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
970 iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
971 iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
972 iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
973 iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
974 iop_is_err_split_tx(status, chan) ? "split_tx " : "");
976 iop_adma_device_clear_err_status(chan);
978 BUG();
980 return IRQ_HANDLED;
983 static void iop_adma_issue_pending(struct dma_chan *chan)
985 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
987 if (iop_chan->pending) {
988 iop_chan->pending = 0;
989 iop_chan_append(iop_chan);
994 * Perform a transaction to verify the HW works.
996 #define IOP_ADMA_TEST_SIZE 2000
998 static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device)
1000 int i;
1001 void *src, *dest;
1002 dma_addr_t src_dma, dest_dma;
1003 struct dma_chan *dma_chan;
1004 dma_cookie_t cookie;
1005 struct dma_async_tx_descriptor *tx;
1006 int err = 0;
1007 struct iop_adma_chan *iop_chan;
1009 dev_dbg(device->common.dev, "%s\n", __func__);
1011 src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
1012 if (!src)
1013 return -ENOMEM;
1014 dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
1015 if (!dest) {
1016 kfree(src);
1017 return -ENOMEM;
1020 /* Fill in src buffer */
1021 for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
1022 ((u8 *) src)[i] = (u8)i;
1024 /* Start copy, using first DMA channel */
1025 dma_chan = container_of(device->common.channels.next,
1026 struct dma_chan,
1027 device_node);
1028 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1029 err = -ENODEV;
1030 goto out;
1033 dest_dma = dma_map_single(dma_chan->device->dev, dest,
1034 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
1035 src_dma = dma_map_single(dma_chan->device->dev, src,
1036 IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
1037 tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
1038 IOP_ADMA_TEST_SIZE,
1039 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1041 cookie = iop_adma_tx_submit(tx);
1042 iop_adma_issue_pending(dma_chan);
1043 msleep(1);
1045 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
1046 DMA_SUCCESS) {
1047 dev_printk(KERN_ERR, dma_chan->device->dev,
1048 "Self-test copy timed out, disabling\n");
1049 err = -ENODEV;
1050 goto free_resources;
1053 iop_chan = to_iop_adma_chan(dma_chan);
1054 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
1055 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
1056 if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
1057 dev_printk(KERN_ERR, dma_chan->device->dev,
1058 "Self-test copy failed compare, disabling\n");
1059 err = -ENODEV;
1060 goto free_resources;
1063 free_resources:
1064 iop_adma_free_chan_resources(dma_chan);
1065 out:
1066 kfree(src);
1067 kfree(dest);
1068 return err;
1071 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
1072 static int __devinit
1073 iop_adma_xor_val_self_test(struct iop_adma_device *device)
1075 int i, src_idx;
1076 struct page *dest;
1077 struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
1078 struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
1079 dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
1080 dma_addr_t dma_addr, dest_dma;
1081 struct dma_async_tx_descriptor *tx;
1082 struct dma_chan *dma_chan;
1083 dma_cookie_t cookie;
1084 u8 cmp_byte = 0;
1085 u32 cmp_word;
1086 u32 zero_sum_result;
1087 int err = 0;
1088 struct iop_adma_chan *iop_chan;
1090 dev_dbg(device->common.dev, "%s\n", __func__);
1092 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
1093 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
1094 if (!xor_srcs[src_idx]) {
1095 while (src_idx--)
1096 __free_page(xor_srcs[src_idx]);
1097 return -ENOMEM;
1101 dest = alloc_page(GFP_KERNEL);
1102 if (!dest) {
1103 while (src_idx--)
1104 __free_page(xor_srcs[src_idx]);
1105 return -ENOMEM;
1108 /* Fill in src buffers */
1109 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
1110 u8 *ptr = page_address(xor_srcs[src_idx]);
1111 for (i = 0; i < PAGE_SIZE; i++)
1112 ptr[i] = (1 << src_idx);
1115 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
1116 cmp_byte ^= (u8) (1 << src_idx);
1118 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
1119 (cmp_byte << 8) | cmp_byte;
1121 memset(page_address(dest), 0, PAGE_SIZE);
1123 dma_chan = container_of(device->common.channels.next,
1124 struct dma_chan,
1125 device_node);
1126 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1127 err = -ENODEV;
1128 goto out;
1131 /* test xor */
1132 dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
1133 PAGE_SIZE, DMA_FROM_DEVICE);
1134 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1135 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
1136 0, PAGE_SIZE, DMA_TO_DEVICE);
1137 tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
1138 IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE,
1139 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1141 cookie = iop_adma_tx_submit(tx);
1142 iop_adma_issue_pending(dma_chan);
1143 msleep(8);
1145 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
1146 DMA_SUCCESS) {
1147 dev_printk(KERN_ERR, dma_chan->device->dev,
1148 "Self-test xor timed out, disabling\n");
1149 err = -ENODEV;
1150 goto free_resources;
1153 iop_chan = to_iop_adma_chan(dma_chan);
1154 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
1155 PAGE_SIZE, DMA_FROM_DEVICE);
1156 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1157 u32 *ptr = page_address(dest);
1158 if (ptr[i] != cmp_word) {
1159 dev_printk(KERN_ERR, dma_chan->device->dev,
1160 "Self-test xor failed compare, disabling\n");
1161 err = -ENODEV;
1162 goto free_resources;
1165 dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
1166 PAGE_SIZE, DMA_TO_DEVICE);
1168 /* skip zero sum if the capability is not present */
1169 if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask))
1170 goto free_resources;
1172 /* zero sum the sources with the destintation page */
1173 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1174 zero_sum_srcs[i] = xor_srcs[i];
1175 zero_sum_srcs[i] = dest;
1177 zero_sum_result = 1;
1179 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1180 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1181 zero_sum_srcs[i], 0, PAGE_SIZE,
1182 DMA_TO_DEVICE);
1183 tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1184 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1185 &zero_sum_result,
1186 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1188 cookie = iop_adma_tx_submit(tx);
1189 iop_adma_issue_pending(dma_chan);
1190 msleep(8);
1192 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1193 dev_printk(KERN_ERR, dma_chan->device->dev,
1194 "Self-test zero sum timed out, disabling\n");
1195 err = -ENODEV;
1196 goto free_resources;
1199 if (zero_sum_result != 0) {
1200 dev_printk(KERN_ERR, dma_chan->device->dev,
1201 "Self-test zero sum failed compare, disabling\n");
1202 err = -ENODEV;
1203 goto free_resources;
1206 /* test memset */
1207 dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
1208 PAGE_SIZE, DMA_FROM_DEVICE);
1209 tx = iop_adma_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE,
1210 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1212 cookie = iop_adma_tx_submit(tx);
1213 iop_adma_issue_pending(dma_chan);
1214 msleep(8);
1216 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1217 dev_printk(KERN_ERR, dma_chan->device->dev,
1218 "Self-test memset timed out, disabling\n");
1219 err = -ENODEV;
1220 goto free_resources;
1223 for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1224 u32 *ptr = page_address(dest);
1225 if (ptr[i]) {
1226 dev_printk(KERN_ERR, dma_chan->device->dev,
1227 "Self-test memset failed compare, disabling\n");
1228 err = -ENODEV;
1229 goto free_resources;
1233 /* test for non-zero parity sum */
1234 zero_sum_result = 0;
1235 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1236 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1237 zero_sum_srcs[i], 0, PAGE_SIZE,
1238 DMA_TO_DEVICE);
1239 tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1240 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1241 &zero_sum_result,
1242 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1244 cookie = iop_adma_tx_submit(tx);
1245 iop_adma_issue_pending(dma_chan);
1246 msleep(8);
1248 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1249 dev_printk(KERN_ERR, dma_chan->device->dev,
1250 "Self-test non-zero sum timed out, disabling\n");
1251 err = -ENODEV;
1252 goto free_resources;
1255 if (zero_sum_result != 1) {
1256 dev_printk(KERN_ERR, dma_chan->device->dev,
1257 "Self-test non-zero sum failed compare, disabling\n");
1258 err = -ENODEV;
1259 goto free_resources;
1262 free_resources:
1263 iop_adma_free_chan_resources(dma_chan);
1264 out:
1265 src_idx = IOP_ADMA_NUM_SRC_TEST;
1266 while (src_idx--)
1267 __free_page(xor_srcs[src_idx]);
1268 __free_page(dest);
1269 return err;
1272 #ifdef CONFIG_MD_RAID6_PQ
1273 static int __devinit
1274 iop_adma_pq_zero_sum_self_test(struct iop_adma_device *device)
1276 /* combined sources, software pq results, and extra hw pq results */
1277 struct page *pq[IOP_ADMA_NUM_SRC_TEST+2+2];
1278 /* ptr to the extra hw pq buffers defined above */
1279 struct page **pq_hw = &pq[IOP_ADMA_NUM_SRC_TEST+2];
1280 /* address conversion buffers (dma_map / page_address) */
1281 void *pq_sw[IOP_ADMA_NUM_SRC_TEST+2];
1282 dma_addr_t pq_src[IOP_ADMA_NUM_SRC_TEST];
1283 dma_addr_t pq_dest[2];
1285 int i;
1286 struct dma_async_tx_descriptor *tx;
1287 struct dma_chan *dma_chan;
1288 dma_cookie_t cookie;
1289 u32 zero_sum_result;
1290 int err = 0;
1291 struct device *dev;
1293 dev_dbg(device->common.dev, "%s\n", __func__);
1295 for (i = 0; i < ARRAY_SIZE(pq); i++) {
1296 pq[i] = alloc_page(GFP_KERNEL);
1297 if (!pq[i]) {
1298 while (i--)
1299 __free_page(pq[i]);
1300 return -ENOMEM;
1304 /* Fill in src buffers */
1305 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) {
1306 pq_sw[i] = page_address(pq[i]);
1307 memset(pq_sw[i], 0x11111111 * (1<<i), PAGE_SIZE);
1309 pq_sw[i] = page_address(pq[i]);
1310 pq_sw[i+1] = page_address(pq[i+1]);
1312 dma_chan = container_of(device->common.channels.next,
1313 struct dma_chan,
1314 device_node);
1315 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1316 err = -ENODEV;
1317 goto out;
1320 dev = dma_chan->device->dev;
1322 /* initialize the dests */
1323 memset(page_address(pq_hw[0]), 0 , PAGE_SIZE);
1324 memset(page_address(pq_hw[1]), 0 , PAGE_SIZE);
1326 /* test pq */
1327 pq_dest[0] = dma_map_page(dev, pq_hw[0], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1328 pq_dest[1] = dma_map_page(dev, pq_hw[1], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1329 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1330 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1331 DMA_TO_DEVICE);
1333 tx = iop_adma_prep_dma_pq(dma_chan, pq_dest, pq_src,
1334 IOP_ADMA_NUM_SRC_TEST, (u8 *)raid6_gfexp,
1335 PAGE_SIZE,
1336 DMA_PREP_INTERRUPT |
1337 DMA_CTRL_ACK);
1339 cookie = iop_adma_tx_submit(tx);
1340 iop_adma_issue_pending(dma_chan);
1341 msleep(8);
1343 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
1344 DMA_SUCCESS) {
1345 dev_err(dev, "Self-test pq timed out, disabling\n");
1346 err = -ENODEV;
1347 goto free_resources;
1350 raid6_call.gen_syndrome(IOP_ADMA_NUM_SRC_TEST+2, PAGE_SIZE, pq_sw);
1352 if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST],
1353 page_address(pq_hw[0]), PAGE_SIZE) != 0) {
1354 dev_err(dev, "Self-test p failed compare, disabling\n");
1355 err = -ENODEV;
1356 goto free_resources;
1358 if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST+1],
1359 page_address(pq_hw[1]), PAGE_SIZE) != 0) {
1360 dev_err(dev, "Self-test q failed compare, disabling\n");
1361 err = -ENODEV;
1362 goto free_resources;
1365 /* test correct zero sum using the software generated pq values */
1366 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1367 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1368 DMA_TO_DEVICE);
1370 zero_sum_result = ~0;
1371 tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1372 pq_src, IOP_ADMA_NUM_SRC_TEST,
1373 raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1374 DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1376 cookie = iop_adma_tx_submit(tx);
1377 iop_adma_issue_pending(dma_chan);
1378 msleep(8);
1380 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
1381 DMA_SUCCESS) {
1382 dev_err(dev, "Self-test pq-zero-sum timed out, disabling\n");
1383 err = -ENODEV;
1384 goto free_resources;
1387 if (zero_sum_result != 0) {
1388 dev_err(dev, "Self-test pq-zero-sum failed to validate: %x\n",
1389 zero_sum_result);
1390 err = -ENODEV;
1391 goto free_resources;
1394 /* test incorrect zero sum */
1395 i = IOP_ADMA_NUM_SRC_TEST;
1396 memset(pq_sw[i] + 100, 0, 100);
1397 memset(pq_sw[i+1] + 200, 0, 200);
1398 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1399 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1400 DMA_TO_DEVICE);
1402 zero_sum_result = 0;
1403 tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1404 pq_src, IOP_ADMA_NUM_SRC_TEST,
1405 raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1406 DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1408 cookie = iop_adma_tx_submit(tx);
1409 iop_adma_issue_pending(dma_chan);
1410 msleep(8);
1412 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
1413 DMA_SUCCESS) {
1414 dev_err(dev, "Self-test !pq-zero-sum timed out, disabling\n");
1415 err = -ENODEV;
1416 goto free_resources;
1419 if (zero_sum_result != (SUM_CHECK_P_RESULT | SUM_CHECK_Q_RESULT)) {
1420 dev_err(dev, "Self-test !pq-zero-sum failed to validate: %x\n",
1421 zero_sum_result);
1422 err = -ENODEV;
1423 goto free_resources;
1426 free_resources:
1427 iop_adma_free_chan_resources(dma_chan);
1428 out:
1429 i = ARRAY_SIZE(pq);
1430 while (i--)
1431 __free_page(pq[i]);
1432 return err;
1434 #endif
1436 static int __devexit iop_adma_remove(struct platform_device *dev)
1438 struct iop_adma_device *device = platform_get_drvdata(dev);
1439 struct dma_chan *chan, *_chan;
1440 struct iop_adma_chan *iop_chan;
1441 struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
1443 dma_async_device_unregister(&device->common);
1445 dma_free_coherent(&dev->dev, plat_data->pool_size,
1446 device->dma_desc_pool_virt, device->dma_desc_pool);
1448 list_for_each_entry_safe(chan, _chan, &device->common.channels,
1449 device_node) {
1450 iop_chan = to_iop_adma_chan(chan);
1451 list_del(&chan->device_node);
1452 kfree(iop_chan);
1454 kfree(device);
1456 return 0;
1459 static int __devinit iop_adma_probe(struct platform_device *pdev)
1461 struct resource *res;
1462 int ret = 0, i;
1463 struct iop_adma_device *adev;
1464 struct iop_adma_chan *iop_chan;
1465 struct dma_device *dma_dev;
1466 struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
1468 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1469 if (!res)
1470 return -ENODEV;
1472 if (!devm_request_mem_region(&pdev->dev, res->start,
1473 res->end - res->start, pdev->name))
1474 return -EBUSY;
1476 adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1477 if (!adev)
1478 return -ENOMEM;
1479 dma_dev = &adev->common;
1481 /* allocate coherent memory for hardware descriptors
1482 * note: writecombine gives slightly better performance, but
1483 * requires that we explicitly flush the writes
1485 if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1486 plat_data->pool_size,
1487 &adev->dma_desc_pool,
1488 GFP_KERNEL)) == NULL) {
1489 ret = -ENOMEM;
1490 goto err_free_adev;
1493 dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n",
1494 __func__, adev->dma_desc_pool_virt,
1495 (void *) adev->dma_desc_pool);
1497 adev->id = plat_data->hw_id;
1499 /* discover transaction capabilites from the platform data */
1500 dma_dev->cap_mask = plat_data->cap_mask;
1502 adev->pdev = pdev;
1503 platform_set_drvdata(pdev, adev);
1505 INIT_LIST_HEAD(&dma_dev->channels);
1507 /* set base routines */
1508 dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1509 dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1510 dma_dev->device_is_tx_complete = iop_adma_is_complete;
1511 dma_dev->device_issue_pending = iop_adma_issue_pending;
1512 dma_dev->dev = &pdev->dev;
1514 /* set prep routines based on capability */
1515 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1516 dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1517 if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1518 dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
1519 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1520 dma_dev->max_xor = iop_adma_get_max_xor();
1521 dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1523 if (dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask))
1524 dma_dev->device_prep_dma_xor_val =
1525 iop_adma_prep_dma_xor_val;
1526 if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1527 dma_set_maxpq(dma_dev, iop_adma_get_max_pq(), 0);
1528 dma_dev->device_prep_dma_pq = iop_adma_prep_dma_pq;
1530 if (dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask))
1531 dma_dev->device_prep_dma_pq_val =
1532 iop_adma_prep_dma_pq_val;
1533 if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1534 dma_dev->device_prep_dma_interrupt =
1535 iop_adma_prep_dma_interrupt;
1537 iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1538 if (!iop_chan) {
1539 ret = -ENOMEM;
1540 goto err_free_dma;
1542 iop_chan->device = adev;
1544 iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1545 res->end - res->start);
1546 if (!iop_chan->mmr_base) {
1547 ret = -ENOMEM;
1548 goto err_free_iop_chan;
1550 tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1551 iop_chan);
1553 /* clear errors before enabling interrupts */
1554 iop_adma_device_clear_err_status(iop_chan);
1556 for (i = 0; i < 3; i++) {
1557 irq_handler_t handler[] = { iop_adma_eot_handler,
1558 iop_adma_eoc_handler,
1559 iop_adma_err_handler };
1560 int irq = platform_get_irq(pdev, i);
1561 if (irq < 0) {
1562 ret = -ENXIO;
1563 goto err_free_iop_chan;
1564 } else {
1565 ret = devm_request_irq(&pdev->dev, irq,
1566 handler[i], 0, pdev->name, iop_chan);
1567 if (ret)
1568 goto err_free_iop_chan;
1572 spin_lock_init(&iop_chan->lock);
1573 INIT_LIST_HEAD(&iop_chan->chain);
1574 INIT_LIST_HEAD(&iop_chan->all_slots);
1575 iop_chan->common.device = dma_dev;
1576 list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1578 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1579 ret = iop_adma_memcpy_self_test(adev);
1580 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1581 if (ret)
1582 goto err_free_iop_chan;
1585 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
1586 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1587 ret = iop_adma_xor_val_self_test(adev);
1588 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1589 if (ret)
1590 goto err_free_iop_chan;
1593 if (dma_has_cap(DMA_PQ, dma_dev->cap_mask) &&
1594 dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask)) {
1595 #ifdef CONFIG_MD_RAID6_PQ
1596 ret = iop_adma_pq_zero_sum_self_test(adev);
1597 dev_dbg(&pdev->dev, "pq self test returned %d\n", ret);
1598 #else
1599 /* can not test raid6, so do not publish capability */
1600 dma_cap_clear(DMA_PQ, dma_dev->cap_mask);
1601 dma_cap_clear(DMA_PQ_VAL, dma_dev->cap_mask);
1602 ret = 0;
1603 #endif
1604 if (ret)
1605 goto err_free_iop_chan;
1608 dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
1609 "( %s%s%s%s%s%s%s)\n",
1610 dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "",
1611 dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask) ? "pq_val " : "",
1612 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1613 dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask) ? "xor_val " : "",
1614 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "",
1615 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1616 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1618 dma_async_device_register(dma_dev);
1619 goto out;
1621 err_free_iop_chan:
1622 kfree(iop_chan);
1623 err_free_dma:
1624 dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1625 adev->dma_desc_pool_virt, adev->dma_desc_pool);
1626 err_free_adev:
1627 kfree(adev);
1628 out:
1629 return ret;
1632 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1634 struct iop_adma_desc_slot *sw_desc, *grp_start;
1635 dma_cookie_t cookie;
1636 int slot_cnt, slots_per_op;
1638 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1640 spin_lock_bh(&iop_chan->lock);
1641 slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1642 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1643 if (sw_desc) {
1644 grp_start = sw_desc->group_head;
1646 list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1647 async_tx_ack(&sw_desc->async_tx);
1648 iop_desc_init_memcpy(grp_start, 0);
1649 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1650 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1651 iop_desc_set_memcpy_src_addr(grp_start, 0);
1653 cookie = iop_chan->common.cookie;
1654 cookie++;
1655 if (cookie <= 1)
1656 cookie = 2;
1658 /* initialize the completed cookie to be less than
1659 * the most recently used cookie
1661 iop_chan->completed_cookie = cookie - 1;
1662 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1664 /* channel should not be busy */
1665 BUG_ON(iop_chan_is_busy(iop_chan));
1667 /* clear any prior error-status bits */
1668 iop_adma_device_clear_err_status(iop_chan);
1670 /* disable operation */
1671 iop_chan_disable(iop_chan);
1673 /* set the descriptor address */
1674 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1676 /* 1/ don't add pre-chained descriptors
1677 * 2/ dummy read to flush next_desc write
1679 BUG_ON(iop_desc_get_next_desc(sw_desc));
1681 /* run the descriptor */
1682 iop_chan_enable(iop_chan);
1683 } else
1684 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1685 "failed to allocate null descriptor\n");
1686 spin_unlock_bh(&iop_chan->lock);
1689 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1691 struct iop_adma_desc_slot *sw_desc, *grp_start;
1692 dma_cookie_t cookie;
1693 int slot_cnt, slots_per_op;
1695 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1697 spin_lock_bh(&iop_chan->lock);
1698 slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1699 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1700 if (sw_desc) {
1701 grp_start = sw_desc->group_head;
1702 list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1703 async_tx_ack(&sw_desc->async_tx);
1704 iop_desc_init_null_xor(grp_start, 2, 0);
1705 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1706 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1707 iop_desc_set_xor_src_addr(grp_start, 0, 0);
1708 iop_desc_set_xor_src_addr(grp_start, 1, 0);
1710 cookie = iop_chan->common.cookie;
1711 cookie++;
1712 if (cookie <= 1)
1713 cookie = 2;
1715 /* initialize the completed cookie to be less than
1716 * the most recently used cookie
1718 iop_chan->completed_cookie = cookie - 1;
1719 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1721 /* channel should not be busy */
1722 BUG_ON(iop_chan_is_busy(iop_chan));
1724 /* clear any prior error-status bits */
1725 iop_adma_device_clear_err_status(iop_chan);
1727 /* disable operation */
1728 iop_chan_disable(iop_chan);
1730 /* set the descriptor address */
1731 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1733 /* 1/ don't add pre-chained descriptors
1734 * 2/ dummy read to flush next_desc write
1736 BUG_ON(iop_desc_get_next_desc(sw_desc));
1738 /* run the descriptor */
1739 iop_chan_enable(iop_chan);
1740 } else
1741 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1742 "failed to allocate null descriptor\n");
1743 spin_unlock_bh(&iop_chan->lock);
1746 MODULE_ALIAS("platform:iop-adma");
1748 static struct platform_driver iop_adma_driver = {
1749 .probe = iop_adma_probe,
1750 .remove = __devexit_p(iop_adma_remove),
1751 .driver = {
1752 .owner = THIS_MODULE,
1753 .name = "iop-adma",
1757 static int __init iop_adma_init (void)
1759 return platform_driver_register(&iop_adma_driver);
1762 static void __exit iop_adma_exit (void)
1764 platform_driver_unregister(&iop_adma_driver);
1765 return;
1767 module_exit(iop_adma_exit);
1768 module_init(iop_adma_init);
1770 MODULE_AUTHOR("Intel Corporation");
1771 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1772 MODULE_LICENSE("GPL");