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
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)
45 * iop_adma_free_slots - flags descriptor slots for reuse
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;
54 slot->slots_per_op = 0;
55 slot = list_entry(slot->slot_node.next,
56 struct iop_adma_desc_slot,
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;
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;
81 dir = DMA_FROM_DEVICE;
83 dma_unmap_page(dev, dest, len, dir);
86 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
88 addr = iop_desc_get_src_addr(unmap, iop_chan, src_cnt);
91 dma_unmap_page(dev, addr, len, DMA_TO_DEVICE);
94 desc->group_head = NULL;
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);
110 if (tx->flags & DMA_PREP_CONTINUE)
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)) {
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;
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) {
146 /* call the callback (must not sleep or submit new
147 * operations to this channel)
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);
159 iop_desc_unmap(iop_chan, desc);
163 /* run dependent operations */
164 dma_run_dependencies(tx);
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
176 if (!async_tx_test_ack(&desc->async_tx))
179 /* leave the last descriptor in the chain
180 * so we can append to it
182 if (desc->chain_node.next == &iop_chan->chain)
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);
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,
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));
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
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))
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) {
245 pr_debug("\tgroup++\n");
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) {
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;
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,
290 if (slot_cnt == 0 || end_of_chain)
294 /* the group should be complete at this point */
303 } else if (slots_per_op) /* wait for group completion */
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))
319 iop_chan->completed_cookie = cookie;
320 pr_debug("\tcompleted cookie %d\n", cookie);
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,
350 struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
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
361 iter = iop_chan->last_used;
363 iter = list_entry(&iop_chan->all_slots,
364 struct iop_adma_desc_slot,
367 list_for_each_entry_safe_continue(
368 iter, _iter, &iop_chan->all_slots, slot_node) {
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
382 /* start the allocation if the slot is correctly aligned */
383 if (!slots_found++) {
384 if (iop_desc_is_aligned(iter, slots_per_op))
392 if (slots_found == num_slots) {
393 struct iop_adma_desc_slot *alloc_tail = NULL;
394 struct iop_adma_desc_slot *last_used = NULL;
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);
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;
416 iter = list_entry(iter->slot_node.next,
417 struct iop_adma_desc_slot,
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);
434 /* perform direct reclaim if the allocation fails */
435 __iop_adma_slot_cleanup(iop_chan);
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;
448 iop_chan->common.cookie = desc->async_tx.cookie = 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",
457 if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
458 iop_chan->pending = 0;
459 iop_chan_append(iop_chan);
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;
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
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);
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
522 static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
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 */
535 idx = iop_chan->slots_allocated;
536 if (idx == num_descs_in_pool)
539 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
541 printk(KERN_INFO "IOP ADMA Channel only initialized"
542 " %d descriptor slots", idx);
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];
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,
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 */
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);
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);
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;
621 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
623 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
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);
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;
654 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
656 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
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);
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,
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;
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);
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;
706 iop_desc_set_xor_src_addr(grp_start, 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,
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;
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);
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;
743 iop_desc_set_zero_sum_src_addr(grp_start, 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,
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;
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;
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);
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,
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;
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);
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;
853 iop_desc_set_pq_zero_sum_src_addr(g, 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,
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);
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",
896 * iop_adma_status - poll the status of an ADMA transaction
897 * @chan: ADMA channel handle
898 * @cookie: ADMA transaction identifier
899 * @txstate: a holder for the current state of the channel or NULL
901 static enum dma_status iop_adma_status(struct dma_chan *chan,
903 struct dma_tx_state *txstate)
905 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
906 dma_cookie_t last_used;
907 dma_cookie_t last_complete;
910 last_used = chan->cookie;
911 last_complete = iop_chan->completed_cookie;
912 dma_set_tx_state(txstate, last_complete, last_used, 0);
913 ret = dma_async_is_complete(cookie, last_complete, last_used);
914 if (ret == DMA_SUCCESS)
917 iop_adma_slot_cleanup(iop_chan);
919 last_used = chan->cookie;
920 last_complete = iop_chan->completed_cookie;
921 dma_set_tx_state(txstate, last_complete, last_used, 0);
923 return dma_async_is_complete(cookie, last_complete, last_used);
926 static irqreturn_t iop_adma_eot_handler(int irq, void *data)
928 struct iop_adma_chan *chan = data;
930 dev_dbg(chan->device->common.dev, "%s\n", __func__);
932 tasklet_schedule(&chan->irq_tasklet);
934 iop_adma_device_clear_eot_status(chan);
939 static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
941 struct iop_adma_chan *chan = data;
943 dev_dbg(chan->device->common.dev, "%s\n", __func__);
945 tasklet_schedule(&chan->irq_tasklet);
947 iop_adma_device_clear_eoc_status(chan);
952 static irqreturn_t iop_adma_err_handler(int irq, void *data)
954 struct iop_adma_chan *chan = data;
955 unsigned long status = iop_chan_get_status(chan);
957 dev_printk(KERN_ERR, chan->device->common.dev,
958 "error ( %s%s%s%s%s%s%s)\n",
959 iop_is_err_int_parity(status, chan) ? "int_parity " : "",
960 iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
961 iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
962 iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
963 iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
964 iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
965 iop_is_err_split_tx(status, chan) ? "split_tx " : "");
967 iop_adma_device_clear_err_status(chan);
974 static void iop_adma_issue_pending(struct dma_chan *chan)
976 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
978 if (iop_chan->pending) {
979 iop_chan->pending = 0;
980 iop_chan_append(iop_chan);
985 * Perform a transaction to verify the HW works.
987 #define IOP_ADMA_TEST_SIZE 2000
989 static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device)
993 dma_addr_t src_dma, dest_dma;
994 struct dma_chan *dma_chan;
996 struct dma_async_tx_descriptor *tx;
998 struct iop_adma_chan *iop_chan;
1000 dev_dbg(device->common.dev, "%s\n", __func__);
1002 src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
1005 dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
1011 /* Fill in src buffer */
1012 for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
1013 ((u8 *) src)[i] = (u8)i;
1015 /* Start copy, using first DMA channel */
1016 dma_chan = container_of(device->common.channels.next,
1019 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1024 dest_dma = dma_map_single(dma_chan->device->dev, dest,
1025 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
1026 src_dma = dma_map_single(dma_chan->device->dev, src,
1027 IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
1028 tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
1030 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1032 cookie = iop_adma_tx_submit(tx);
1033 iop_adma_issue_pending(dma_chan);
1036 if (iop_adma_status(dma_chan, cookie, NULL) !=
1038 dev_printk(KERN_ERR, dma_chan->device->dev,
1039 "Self-test copy timed out, disabling\n");
1041 goto free_resources;
1044 iop_chan = to_iop_adma_chan(dma_chan);
1045 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
1046 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
1047 if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
1048 dev_printk(KERN_ERR, dma_chan->device->dev,
1049 "Self-test copy failed compare, disabling\n");
1051 goto free_resources;
1055 iop_adma_free_chan_resources(dma_chan);
1062 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
1063 static int __devinit
1064 iop_adma_xor_val_self_test(struct iop_adma_device *device)
1068 struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
1069 struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
1070 dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
1071 dma_addr_t dma_addr, dest_dma;
1072 struct dma_async_tx_descriptor *tx;
1073 struct dma_chan *dma_chan;
1074 dma_cookie_t cookie;
1077 u32 zero_sum_result;
1079 struct iop_adma_chan *iop_chan;
1081 dev_dbg(device->common.dev, "%s\n", __func__);
1083 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
1084 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
1085 if (!xor_srcs[src_idx]) {
1087 __free_page(xor_srcs[src_idx]);
1092 dest = alloc_page(GFP_KERNEL);
1095 __free_page(xor_srcs[src_idx]);
1099 /* Fill in src buffers */
1100 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
1101 u8 *ptr = page_address(xor_srcs[src_idx]);
1102 for (i = 0; i < PAGE_SIZE; i++)
1103 ptr[i] = (1 << src_idx);
1106 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
1107 cmp_byte ^= (u8) (1 << src_idx);
1109 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
1110 (cmp_byte << 8) | cmp_byte;
1112 memset(page_address(dest), 0, PAGE_SIZE);
1114 dma_chan = container_of(device->common.channels.next,
1117 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1123 dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
1124 PAGE_SIZE, DMA_FROM_DEVICE);
1125 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1126 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
1127 0, PAGE_SIZE, DMA_TO_DEVICE);
1128 tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
1129 IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE,
1130 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1132 cookie = iop_adma_tx_submit(tx);
1133 iop_adma_issue_pending(dma_chan);
1136 if (iop_adma_status(dma_chan, cookie, NULL) !=
1138 dev_printk(KERN_ERR, dma_chan->device->dev,
1139 "Self-test xor timed out, disabling\n");
1141 goto free_resources;
1144 iop_chan = to_iop_adma_chan(dma_chan);
1145 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
1146 PAGE_SIZE, DMA_FROM_DEVICE);
1147 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1148 u32 *ptr = page_address(dest);
1149 if (ptr[i] != cmp_word) {
1150 dev_printk(KERN_ERR, dma_chan->device->dev,
1151 "Self-test xor failed compare, disabling\n");
1153 goto free_resources;
1156 dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
1157 PAGE_SIZE, DMA_TO_DEVICE);
1159 /* skip zero sum if the capability is not present */
1160 if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask))
1161 goto free_resources;
1163 /* zero sum the sources with the destintation page */
1164 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1165 zero_sum_srcs[i] = xor_srcs[i];
1166 zero_sum_srcs[i] = dest;
1168 zero_sum_result = 1;
1170 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1171 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1172 zero_sum_srcs[i], 0, PAGE_SIZE,
1174 tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1175 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1177 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1179 cookie = iop_adma_tx_submit(tx);
1180 iop_adma_issue_pending(dma_chan);
1183 if (iop_adma_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1184 dev_printk(KERN_ERR, dma_chan->device->dev,
1185 "Self-test zero sum timed out, disabling\n");
1187 goto free_resources;
1190 if (zero_sum_result != 0) {
1191 dev_printk(KERN_ERR, dma_chan->device->dev,
1192 "Self-test zero sum failed compare, disabling\n");
1194 goto free_resources;
1198 dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
1199 PAGE_SIZE, DMA_FROM_DEVICE);
1200 tx = iop_adma_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE,
1201 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1203 cookie = iop_adma_tx_submit(tx);
1204 iop_adma_issue_pending(dma_chan);
1207 if (iop_adma_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1208 dev_printk(KERN_ERR, dma_chan->device->dev,
1209 "Self-test memset timed out, disabling\n");
1211 goto free_resources;
1214 for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1215 u32 *ptr = page_address(dest);
1217 dev_printk(KERN_ERR, dma_chan->device->dev,
1218 "Self-test memset failed compare, disabling\n");
1220 goto free_resources;
1224 /* test for non-zero parity sum */
1225 zero_sum_result = 0;
1226 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1227 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1228 zero_sum_srcs[i], 0, PAGE_SIZE,
1230 tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1231 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1233 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1235 cookie = iop_adma_tx_submit(tx);
1236 iop_adma_issue_pending(dma_chan);
1239 if (iop_adma_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1240 dev_printk(KERN_ERR, dma_chan->device->dev,
1241 "Self-test non-zero sum timed out, disabling\n");
1243 goto free_resources;
1246 if (zero_sum_result != 1) {
1247 dev_printk(KERN_ERR, dma_chan->device->dev,
1248 "Self-test non-zero sum failed compare, disabling\n");
1250 goto free_resources;
1254 iop_adma_free_chan_resources(dma_chan);
1256 src_idx = IOP_ADMA_NUM_SRC_TEST;
1258 __free_page(xor_srcs[src_idx]);
1263 #ifdef CONFIG_MD_RAID6_PQ
1264 static int __devinit
1265 iop_adma_pq_zero_sum_self_test(struct iop_adma_device *device)
1267 /* combined sources, software pq results, and extra hw pq results */
1268 struct page *pq[IOP_ADMA_NUM_SRC_TEST+2+2];
1269 /* ptr to the extra hw pq buffers defined above */
1270 struct page **pq_hw = &pq[IOP_ADMA_NUM_SRC_TEST+2];
1271 /* address conversion buffers (dma_map / page_address) */
1272 void *pq_sw[IOP_ADMA_NUM_SRC_TEST+2];
1273 dma_addr_t pq_src[IOP_ADMA_NUM_SRC_TEST];
1274 dma_addr_t pq_dest[2];
1277 struct dma_async_tx_descriptor *tx;
1278 struct dma_chan *dma_chan;
1279 dma_cookie_t cookie;
1280 u32 zero_sum_result;
1284 dev_dbg(device->common.dev, "%s\n", __func__);
1286 for (i = 0; i < ARRAY_SIZE(pq); i++) {
1287 pq[i] = alloc_page(GFP_KERNEL);
1295 /* Fill in src buffers */
1296 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) {
1297 pq_sw[i] = page_address(pq[i]);
1298 memset(pq_sw[i], 0x11111111 * (1<<i), PAGE_SIZE);
1300 pq_sw[i] = page_address(pq[i]);
1301 pq_sw[i+1] = page_address(pq[i+1]);
1303 dma_chan = container_of(device->common.channels.next,
1306 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1311 dev = dma_chan->device->dev;
1313 /* initialize the dests */
1314 memset(page_address(pq_hw[0]), 0 , PAGE_SIZE);
1315 memset(page_address(pq_hw[1]), 0 , PAGE_SIZE);
1318 pq_dest[0] = dma_map_page(dev, pq_hw[0], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1319 pq_dest[1] = dma_map_page(dev, pq_hw[1], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1320 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1321 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1324 tx = iop_adma_prep_dma_pq(dma_chan, pq_dest, pq_src,
1325 IOP_ADMA_NUM_SRC_TEST, (u8 *)raid6_gfexp,
1327 DMA_PREP_INTERRUPT |
1330 cookie = iop_adma_tx_submit(tx);
1331 iop_adma_issue_pending(dma_chan);
1334 if (iop_adma_status(dma_chan, cookie, NULL) !=
1336 dev_err(dev, "Self-test pq timed out, disabling\n");
1338 goto free_resources;
1341 raid6_call.gen_syndrome(IOP_ADMA_NUM_SRC_TEST+2, PAGE_SIZE, pq_sw);
1343 if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST],
1344 page_address(pq_hw[0]), PAGE_SIZE) != 0) {
1345 dev_err(dev, "Self-test p failed compare, disabling\n");
1347 goto free_resources;
1349 if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST+1],
1350 page_address(pq_hw[1]), PAGE_SIZE) != 0) {
1351 dev_err(dev, "Self-test q failed compare, disabling\n");
1353 goto free_resources;
1356 /* test correct zero sum using the software generated pq values */
1357 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1358 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1361 zero_sum_result = ~0;
1362 tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1363 pq_src, IOP_ADMA_NUM_SRC_TEST,
1364 raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1365 DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1367 cookie = iop_adma_tx_submit(tx);
1368 iop_adma_issue_pending(dma_chan);
1371 if (iop_adma_status(dma_chan, cookie, NULL) !=
1373 dev_err(dev, "Self-test pq-zero-sum timed out, disabling\n");
1375 goto free_resources;
1378 if (zero_sum_result != 0) {
1379 dev_err(dev, "Self-test pq-zero-sum failed to validate: %x\n",
1382 goto free_resources;
1385 /* test incorrect zero sum */
1386 i = IOP_ADMA_NUM_SRC_TEST;
1387 memset(pq_sw[i] + 100, 0, 100);
1388 memset(pq_sw[i+1] + 200, 0, 200);
1389 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1390 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1393 zero_sum_result = 0;
1394 tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1395 pq_src, IOP_ADMA_NUM_SRC_TEST,
1396 raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1397 DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1399 cookie = iop_adma_tx_submit(tx);
1400 iop_adma_issue_pending(dma_chan);
1403 if (iop_adma_status(dma_chan, cookie, NULL) !=
1405 dev_err(dev, "Self-test !pq-zero-sum timed out, disabling\n");
1407 goto free_resources;
1410 if (zero_sum_result != (SUM_CHECK_P_RESULT | SUM_CHECK_Q_RESULT)) {
1411 dev_err(dev, "Self-test !pq-zero-sum failed to validate: %x\n",
1414 goto free_resources;
1418 iop_adma_free_chan_resources(dma_chan);
1427 static int __devexit iop_adma_remove(struct platform_device *dev)
1429 struct iop_adma_device *device = platform_get_drvdata(dev);
1430 struct dma_chan *chan, *_chan;
1431 struct iop_adma_chan *iop_chan;
1432 struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
1434 dma_async_device_unregister(&device->common);
1436 dma_free_coherent(&dev->dev, plat_data->pool_size,
1437 device->dma_desc_pool_virt, device->dma_desc_pool);
1439 list_for_each_entry_safe(chan, _chan, &device->common.channels,
1441 iop_chan = to_iop_adma_chan(chan);
1442 list_del(&chan->device_node);
1450 static int __devinit iop_adma_probe(struct platform_device *pdev)
1452 struct resource *res;
1454 struct iop_adma_device *adev;
1455 struct iop_adma_chan *iop_chan;
1456 struct dma_device *dma_dev;
1457 struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
1459 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1463 if (!devm_request_mem_region(&pdev->dev, res->start,
1464 resource_size(res), pdev->name))
1467 adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1470 dma_dev = &adev->common;
1472 /* allocate coherent memory for hardware descriptors
1473 * note: writecombine gives slightly better performance, but
1474 * requires that we explicitly flush the writes
1476 if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1477 plat_data->pool_size,
1478 &adev->dma_desc_pool,
1479 GFP_KERNEL)) == NULL) {
1484 dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n",
1485 __func__, adev->dma_desc_pool_virt,
1486 (void *) adev->dma_desc_pool);
1488 adev->id = plat_data->hw_id;
1490 /* discover transaction capabilites from the platform data */
1491 dma_dev->cap_mask = plat_data->cap_mask;
1494 platform_set_drvdata(pdev, adev);
1496 INIT_LIST_HEAD(&dma_dev->channels);
1498 /* set base routines */
1499 dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1500 dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1501 dma_dev->device_tx_status = iop_adma_status;
1502 dma_dev->device_issue_pending = iop_adma_issue_pending;
1503 dma_dev->dev = &pdev->dev;
1505 /* set prep routines based on capability */
1506 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1507 dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1508 if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1509 dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
1510 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1511 dma_dev->max_xor = iop_adma_get_max_xor();
1512 dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1514 if (dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask))
1515 dma_dev->device_prep_dma_xor_val =
1516 iop_adma_prep_dma_xor_val;
1517 if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1518 dma_set_maxpq(dma_dev, iop_adma_get_max_pq(), 0);
1519 dma_dev->device_prep_dma_pq = iop_adma_prep_dma_pq;
1521 if (dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask))
1522 dma_dev->device_prep_dma_pq_val =
1523 iop_adma_prep_dma_pq_val;
1524 if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1525 dma_dev->device_prep_dma_interrupt =
1526 iop_adma_prep_dma_interrupt;
1528 iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1533 iop_chan->device = adev;
1535 iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1536 resource_size(res));
1537 if (!iop_chan->mmr_base) {
1539 goto err_free_iop_chan;
1541 tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1544 /* clear errors before enabling interrupts */
1545 iop_adma_device_clear_err_status(iop_chan);
1547 for (i = 0; i < 3; i++) {
1548 irq_handler_t handler[] = { iop_adma_eot_handler,
1549 iop_adma_eoc_handler,
1550 iop_adma_err_handler };
1551 int irq = platform_get_irq(pdev, i);
1554 goto err_free_iop_chan;
1556 ret = devm_request_irq(&pdev->dev, irq,
1557 handler[i], 0, pdev->name, iop_chan);
1559 goto err_free_iop_chan;
1563 spin_lock_init(&iop_chan->lock);
1564 INIT_LIST_HEAD(&iop_chan->chain);
1565 INIT_LIST_HEAD(&iop_chan->all_slots);
1566 iop_chan->common.device = dma_dev;
1567 list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1569 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1570 ret = iop_adma_memcpy_self_test(adev);
1571 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1573 goto err_free_iop_chan;
1576 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
1577 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1578 ret = iop_adma_xor_val_self_test(adev);
1579 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1581 goto err_free_iop_chan;
1584 if (dma_has_cap(DMA_PQ, dma_dev->cap_mask) &&
1585 dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask)) {
1586 #ifdef CONFIG_MD_RAID6_PQ
1587 ret = iop_adma_pq_zero_sum_self_test(adev);
1588 dev_dbg(&pdev->dev, "pq self test returned %d\n", ret);
1590 /* can not test raid6, so do not publish capability */
1591 dma_cap_clear(DMA_PQ, dma_dev->cap_mask);
1592 dma_cap_clear(DMA_PQ_VAL, dma_dev->cap_mask);
1596 goto err_free_iop_chan;
1599 dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
1600 "( %s%s%s%s%s%s%s)\n",
1601 dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "",
1602 dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask) ? "pq_val " : "",
1603 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1604 dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask) ? "xor_val " : "",
1605 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "",
1606 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1607 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1609 dma_async_device_register(dma_dev);
1615 dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1616 adev->dma_desc_pool_virt, adev->dma_desc_pool);
1623 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1625 struct iop_adma_desc_slot *sw_desc, *grp_start;
1626 dma_cookie_t cookie;
1627 int slot_cnt, slots_per_op;
1629 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1631 spin_lock_bh(&iop_chan->lock);
1632 slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1633 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1635 grp_start = sw_desc->group_head;
1637 list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1638 async_tx_ack(&sw_desc->async_tx);
1639 iop_desc_init_memcpy(grp_start, 0);
1640 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1641 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1642 iop_desc_set_memcpy_src_addr(grp_start, 0);
1644 cookie = iop_chan->common.cookie;
1649 /* initialize the completed cookie to be less than
1650 * the most recently used cookie
1652 iop_chan->completed_cookie = cookie - 1;
1653 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1655 /* channel should not be busy */
1656 BUG_ON(iop_chan_is_busy(iop_chan));
1658 /* clear any prior error-status bits */
1659 iop_adma_device_clear_err_status(iop_chan);
1661 /* disable operation */
1662 iop_chan_disable(iop_chan);
1664 /* set the descriptor address */
1665 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1667 /* 1/ don't add pre-chained descriptors
1668 * 2/ dummy read to flush next_desc write
1670 BUG_ON(iop_desc_get_next_desc(sw_desc));
1672 /* run the descriptor */
1673 iop_chan_enable(iop_chan);
1675 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1676 "failed to allocate null descriptor\n");
1677 spin_unlock_bh(&iop_chan->lock);
1680 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1682 struct iop_adma_desc_slot *sw_desc, *grp_start;
1683 dma_cookie_t cookie;
1684 int slot_cnt, slots_per_op;
1686 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1688 spin_lock_bh(&iop_chan->lock);
1689 slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1690 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1692 grp_start = sw_desc->group_head;
1693 list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1694 async_tx_ack(&sw_desc->async_tx);
1695 iop_desc_init_null_xor(grp_start, 2, 0);
1696 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1697 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1698 iop_desc_set_xor_src_addr(grp_start, 0, 0);
1699 iop_desc_set_xor_src_addr(grp_start, 1, 0);
1701 cookie = iop_chan->common.cookie;
1706 /* initialize the completed cookie to be less than
1707 * the most recently used cookie
1709 iop_chan->completed_cookie = cookie - 1;
1710 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1712 /* channel should not be busy */
1713 BUG_ON(iop_chan_is_busy(iop_chan));
1715 /* clear any prior error-status bits */
1716 iop_adma_device_clear_err_status(iop_chan);
1718 /* disable operation */
1719 iop_chan_disable(iop_chan);
1721 /* set the descriptor address */
1722 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1724 /* 1/ don't add pre-chained descriptors
1725 * 2/ dummy read to flush next_desc write
1727 BUG_ON(iop_desc_get_next_desc(sw_desc));
1729 /* run the descriptor */
1730 iop_chan_enable(iop_chan);
1732 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1733 "failed to allocate null descriptor\n");
1734 spin_unlock_bh(&iop_chan->lock);
1737 MODULE_ALIAS("platform:iop-adma");
1739 static struct platform_driver iop_adma_driver = {
1740 .probe = iop_adma_probe,
1741 .remove = __devexit_p(iop_adma_remove),
1743 .owner = THIS_MODULE,
1748 static int __init iop_adma_init (void)
1750 return platform_driver_register(&iop_adma_driver);
1753 static void __exit iop_adma_exit (void)
1755 platform_driver_unregister(&iop_adma_driver);
1758 module_exit(iop_adma_exit);
1759 module_init(iop_adma_init);
1761 MODULE_AUTHOR("Intel Corporation");
1762 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1763 MODULE_LICENSE("GPL");