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>
35 #include <linux/slab.h>
37 #include <mach/adma.h>
39 #include "dmaengine.h"
41 #define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
42 #define to_iop_adma_device(dev) \
43 container_of(dev, struct iop_adma_device, common)
44 #define tx_to_iop_adma_slot(tx) \
45 container_of(tx, struct iop_adma_desc_slot, async_tx)
48 * iop_adma_free_slots - flags descriptor slots for reuse
50 * Caller must hold &iop_chan->lock while calling this function
52 static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
54 int stride = slot->slots_per_op;
57 slot->slots_per_op = 0;
58 slot = list_entry(slot->slot_node.next,
59 struct iop_adma_desc_slot,
65 iop_desc_unmap(struct iop_adma_chan *iop_chan, struct iop_adma_desc_slot *desc)
67 struct dma_async_tx_descriptor *tx = &desc->async_tx;
68 struct iop_adma_desc_slot *unmap = desc->group_head;
69 struct device *dev = &iop_chan->device->pdev->dev;
70 u32 len = unmap->unmap_len;
71 enum dma_ctrl_flags flags = tx->flags;
76 src_cnt = unmap->unmap_src_cnt;
77 dest = iop_desc_get_dest_addr(unmap, iop_chan);
78 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
79 enum dma_data_direction dir;
81 if (src_cnt > 1) /* is xor? */
82 dir = DMA_BIDIRECTIONAL;
84 dir = DMA_FROM_DEVICE;
86 dma_unmap_page(dev, dest, len, dir);
89 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
91 addr = iop_desc_get_src_addr(unmap, iop_chan, src_cnt);
94 dma_unmap_page(dev, addr, len, DMA_TO_DEVICE);
97 desc->group_head = NULL;
101 iop_desc_unmap_pq(struct iop_adma_chan *iop_chan, struct iop_adma_desc_slot *desc)
103 struct dma_async_tx_descriptor *tx = &desc->async_tx;
104 struct iop_adma_desc_slot *unmap = desc->group_head;
105 struct device *dev = &iop_chan->device->pdev->dev;
106 u32 len = unmap->unmap_len;
107 enum dma_ctrl_flags flags = tx->flags;
108 u32 src_cnt = unmap->unmap_src_cnt;
109 dma_addr_t pdest = iop_desc_get_dest_addr(unmap, iop_chan);
110 dma_addr_t qdest = iop_desc_get_qdest_addr(unmap, iop_chan);
113 if (tx->flags & DMA_PREP_CONTINUE)
116 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP) && !desc->pq_check_result) {
117 dma_unmap_page(dev, pdest, len, DMA_BIDIRECTIONAL);
118 dma_unmap_page(dev, qdest, len, DMA_BIDIRECTIONAL);
121 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
124 for (i = 0; i < src_cnt; i++) {
125 addr = iop_desc_get_src_addr(unmap, iop_chan, i);
126 dma_unmap_page(dev, addr, len, DMA_TO_DEVICE);
128 if (desc->pq_check_result) {
129 dma_unmap_page(dev, pdest, len, DMA_TO_DEVICE);
130 dma_unmap_page(dev, qdest, len, DMA_TO_DEVICE);
134 desc->group_head = NULL;
139 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
140 struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
142 struct dma_async_tx_descriptor *tx = &desc->async_tx;
144 BUG_ON(tx->cookie < 0);
145 if (tx->cookie > 0) {
149 /* call the callback (must not sleep or submit new
150 * operations to this channel)
153 tx->callback(tx->callback_param);
155 /* unmap dma addresses
156 * (unmap_single vs unmap_page?)
158 if (desc->group_head && desc->unmap_len) {
159 if (iop_desc_is_pq(desc))
160 iop_desc_unmap_pq(iop_chan, desc);
162 iop_desc_unmap(iop_chan, desc);
166 /* run dependent operations */
167 dma_run_dependencies(tx);
173 iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
174 struct iop_adma_chan *iop_chan)
176 /* the client is allowed to attach dependent operations
179 if (!async_tx_test_ack(&desc->async_tx))
182 /* leave the last descriptor in the chain
183 * so we can append to it
185 if (desc->chain_node.next == &iop_chan->chain)
188 dev_dbg(iop_chan->device->common.dev,
189 "\tfree slot: %d slots_per_op: %d\n",
190 desc->idx, desc->slots_per_op);
192 list_del(&desc->chain_node);
193 iop_adma_free_slots(desc);
198 static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
200 struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
201 dma_cookie_t cookie = 0;
202 u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
203 int busy = iop_chan_is_busy(iop_chan);
204 int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
206 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
207 /* free completed slots from the chain starting with
208 * the oldest descriptor
210 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
212 pr_debug("\tcookie: %d slot: %d busy: %d "
213 "this_desc: %#x next_desc: %#x ack: %d\n",
214 iter->async_tx.cookie, iter->idx, busy,
215 iter->async_tx.phys, iop_desc_get_next_desc(iter),
216 async_tx_test_ack(&iter->async_tx));
218 prefetch(&_iter->async_tx);
220 /* do not advance past the current descriptor loaded into the
221 * hardware channel, subsequent descriptors are either in
222 * process or have not been submitted
227 /* stop the search if we reach the current descriptor and the
228 * channel is busy, or if it appears that the current descriptor
229 * needs to be re-read (i.e. has been appended to)
231 if (iter->async_tx.phys == current_desc) {
232 BUG_ON(seen_current++);
233 if (busy || iop_desc_get_next_desc(iter))
237 /* detect the start of a group transaction */
238 if (!slot_cnt && !slots_per_op) {
239 slot_cnt = iter->slot_cnt;
240 slots_per_op = iter->slots_per_op;
241 if (slot_cnt <= slots_per_op) {
248 pr_debug("\tgroup++\n");
251 slot_cnt -= slots_per_op;
254 /* all the members of a group are complete */
255 if (slots_per_op != 0 && slot_cnt == 0) {
256 struct iop_adma_desc_slot *grp_iter, *_grp_iter;
257 int end_of_chain = 0;
258 pr_debug("\tgroup end\n");
260 /* collect the total results */
261 if (grp_start->xor_check_result) {
262 u32 zero_sum_result = 0;
263 slot_cnt = grp_start->slot_cnt;
264 grp_iter = grp_start;
266 list_for_each_entry_from(grp_iter,
267 &iop_chan->chain, chain_node) {
269 iop_desc_get_zero_result(grp_iter);
270 pr_debug("\titer%d result: %d\n",
271 grp_iter->idx, zero_sum_result);
272 slot_cnt -= slots_per_op;
276 pr_debug("\tgrp_start->xor_check_result: %p\n",
277 grp_start->xor_check_result);
278 *grp_start->xor_check_result = zero_sum_result;
281 /* clean up the group */
282 slot_cnt = grp_start->slot_cnt;
283 grp_iter = grp_start;
284 list_for_each_entry_safe_from(grp_iter, _grp_iter,
285 &iop_chan->chain, chain_node) {
286 cookie = iop_adma_run_tx_complete_actions(
287 grp_iter, iop_chan, cookie);
289 slot_cnt -= slots_per_op;
290 end_of_chain = iop_adma_clean_slot(grp_iter,
293 if (slot_cnt == 0 || end_of_chain)
297 /* the group should be complete at this point */
306 } else if (slots_per_op) /* wait for group completion */
309 /* write back zero sum results (single descriptor case) */
310 if (iter->xor_check_result && iter->async_tx.cookie)
311 *iter->xor_check_result =
312 iop_desc_get_zero_result(iter);
314 cookie = iop_adma_run_tx_complete_actions(
315 iter, iop_chan, cookie);
317 if (iop_adma_clean_slot(iter, iop_chan))
322 iop_chan->common.completed_cookie = cookie;
323 pr_debug("\tcompleted cookie %d\n", cookie);
328 iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
330 spin_lock_bh(&iop_chan->lock);
331 __iop_adma_slot_cleanup(iop_chan);
332 spin_unlock_bh(&iop_chan->lock);
335 static void iop_adma_tasklet(unsigned long data)
337 struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data;
339 /* lockdep will flag depedency submissions as potentially
340 * recursive locking, this is not the case as a dependency
341 * submission will never recurse a channels submit routine.
342 * There are checks in async_tx.c to prevent this.
344 spin_lock_nested(&iop_chan->lock, SINGLE_DEPTH_NESTING);
345 __iop_adma_slot_cleanup(iop_chan);
346 spin_unlock(&iop_chan->lock);
349 static struct iop_adma_desc_slot *
350 iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
353 struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
355 int slots_found, retry = 0;
357 /* start search from the last allocated descrtiptor
358 * if a contiguous allocation can not be found start searching
359 * from the beginning of the list
364 iter = iop_chan->last_used;
366 iter = list_entry(&iop_chan->all_slots,
367 struct iop_adma_desc_slot,
370 list_for_each_entry_safe_continue(
371 iter, _iter, &iop_chan->all_slots, slot_node) {
373 prefetch(&_iter->async_tx);
374 if (iter->slots_per_op) {
375 /* give up after finding the first busy slot
376 * on the second pass through the list
385 /* start the allocation if the slot is correctly aligned */
386 if (!slots_found++) {
387 if (iop_desc_is_aligned(iter, slots_per_op))
395 if (slots_found == num_slots) {
396 struct iop_adma_desc_slot *alloc_tail = NULL;
397 struct iop_adma_desc_slot *last_used = NULL;
401 dev_dbg(iop_chan->device->common.dev,
402 "allocated slot: %d "
403 "(desc %p phys: %#x) slots_per_op %d\n",
404 iter->idx, iter->hw_desc,
405 iter->async_tx.phys, slots_per_op);
407 /* pre-ack all but the last descriptor */
408 if (num_slots != slots_per_op)
409 async_tx_ack(&iter->async_tx);
411 list_add_tail(&iter->chain_node, &chain);
413 iter->async_tx.cookie = 0;
414 iter->slot_cnt = num_slots;
415 iter->xor_check_result = NULL;
416 for (i = 0; i < slots_per_op; i++) {
417 iter->slots_per_op = slots_per_op - i;
419 iter = list_entry(iter->slot_node.next,
420 struct iop_adma_desc_slot,
423 num_slots -= slots_per_op;
425 alloc_tail->group_head = alloc_start;
426 alloc_tail->async_tx.cookie = -EBUSY;
427 list_splice(&chain, &alloc_tail->tx_list);
428 iop_chan->last_used = last_used;
429 iop_desc_clear_next_desc(alloc_start);
430 iop_desc_clear_next_desc(alloc_tail);
437 /* perform direct reclaim if the allocation fails */
438 __iop_adma_slot_cleanup(iop_chan);
443 static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
445 dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
448 if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
449 iop_chan->pending = 0;
450 iop_chan_append(iop_chan);
455 iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
457 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
458 struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
459 struct iop_adma_desc_slot *grp_start, *old_chain_tail;
465 grp_start = sw_desc->group_head;
466 slot_cnt = grp_start->slot_cnt;
467 slots_per_op = grp_start->slots_per_op;
469 spin_lock_bh(&iop_chan->lock);
470 cookie = dma_cookie_assign(tx);
472 old_chain_tail = list_entry(iop_chan->chain.prev,
473 struct iop_adma_desc_slot, chain_node);
474 list_splice_init(&sw_desc->tx_list,
475 &old_chain_tail->chain_node);
477 /* fix up the hardware chain */
478 next_dma = grp_start->async_tx.phys;
479 iop_desc_set_next_desc(old_chain_tail, next_dma);
480 BUG_ON(iop_desc_get_next_desc(old_chain_tail) != next_dma); /* flush */
482 /* check for pre-chained descriptors */
483 iop_paranoia(iop_desc_get_next_desc(sw_desc));
485 /* increment the pending count by the number of slots
486 * memcpy operations have a 1:1 (slot:operation) relation
487 * other operations are heavier and will pop the threshold
490 iop_chan->pending += slot_cnt;
491 iop_adma_check_threshold(iop_chan);
492 spin_unlock_bh(&iop_chan->lock);
494 dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
495 __func__, sw_desc->async_tx.cookie, sw_desc->idx);
500 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
501 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
504 * iop_adma_alloc_chan_resources - returns the number of allocated descriptors
505 * @chan - allocate descriptor resources for this channel
506 * @client - current client requesting the channel be ready for requests
508 * Note: We keep the slots for 1 operation on iop_chan->chain at all times. To
509 * avoid deadlock, via async_xor, num_descs_in_pool must at a minimum be
510 * greater than 2x the number slots needed to satisfy a device->max_xor
513 static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
517 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
518 struct iop_adma_desc_slot *slot = NULL;
519 int init = iop_chan->slots_allocated ? 0 : 1;
520 struct iop_adma_platform_data *plat_data =
521 iop_chan->device->pdev->dev.platform_data;
522 int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
524 /* Allocate descriptor slots */
526 idx = iop_chan->slots_allocated;
527 if (idx == num_descs_in_pool)
530 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
532 printk(KERN_INFO "IOP ADMA Channel only initialized"
533 " %d descriptor slots", idx);
536 hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
537 slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
539 dma_async_tx_descriptor_init(&slot->async_tx, chan);
540 slot->async_tx.tx_submit = iop_adma_tx_submit;
541 INIT_LIST_HEAD(&slot->tx_list);
542 INIT_LIST_HEAD(&slot->chain_node);
543 INIT_LIST_HEAD(&slot->slot_node);
544 hw_desc = (char *) iop_chan->device->dma_desc_pool;
545 slot->async_tx.phys =
546 (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
549 spin_lock_bh(&iop_chan->lock);
550 iop_chan->slots_allocated++;
551 list_add_tail(&slot->slot_node, &iop_chan->all_slots);
552 spin_unlock_bh(&iop_chan->lock);
553 } while (iop_chan->slots_allocated < num_descs_in_pool);
555 if (idx && !iop_chan->last_used)
556 iop_chan->last_used = list_entry(iop_chan->all_slots.next,
557 struct iop_adma_desc_slot,
560 dev_dbg(iop_chan->device->common.dev,
561 "allocated %d descriptor slots last_used: %p\n",
562 iop_chan->slots_allocated, iop_chan->last_used);
564 /* initialize the channel and the chain with a null operation */
566 if (dma_has_cap(DMA_MEMCPY,
567 iop_chan->device->common.cap_mask))
568 iop_chan_start_null_memcpy(iop_chan);
569 else if (dma_has_cap(DMA_XOR,
570 iop_chan->device->common.cap_mask))
571 iop_chan_start_null_xor(iop_chan);
576 return (idx > 0) ? idx : -ENOMEM;
579 static struct dma_async_tx_descriptor *
580 iop_adma_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags)
582 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
583 struct iop_adma_desc_slot *sw_desc, *grp_start;
584 int slot_cnt, slots_per_op;
586 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
588 spin_lock_bh(&iop_chan->lock);
589 slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
590 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
592 grp_start = sw_desc->group_head;
593 iop_desc_init_interrupt(grp_start, iop_chan);
594 grp_start->unmap_len = 0;
595 sw_desc->async_tx.flags = flags;
597 spin_unlock_bh(&iop_chan->lock);
599 return sw_desc ? &sw_desc->async_tx : NULL;
602 static struct dma_async_tx_descriptor *
603 iop_adma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
604 dma_addr_t dma_src, size_t len, unsigned long flags)
606 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
607 struct iop_adma_desc_slot *sw_desc, *grp_start;
608 int slot_cnt, slots_per_op;
612 BUG_ON(len > IOP_ADMA_MAX_BYTE_COUNT);
614 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
617 spin_lock_bh(&iop_chan->lock);
618 slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
619 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
621 grp_start = sw_desc->group_head;
622 iop_desc_init_memcpy(grp_start, flags);
623 iop_desc_set_byte_count(grp_start, iop_chan, len);
624 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
625 iop_desc_set_memcpy_src_addr(grp_start, dma_src);
626 sw_desc->unmap_src_cnt = 1;
627 sw_desc->unmap_len = len;
628 sw_desc->async_tx.flags = flags;
630 spin_unlock_bh(&iop_chan->lock);
632 return sw_desc ? &sw_desc->async_tx : NULL;
635 static struct dma_async_tx_descriptor *
636 iop_adma_prep_dma_memset(struct dma_chan *chan, dma_addr_t dma_dest,
637 int value, size_t len, unsigned long flags)
639 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
640 struct iop_adma_desc_slot *sw_desc, *grp_start;
641 int slot_cnt, slots_per_op;
645 BUG_ON(len > IOP_ADMA_MAX_BYTE_COUNT);
647 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
650 spin_lock_bh(&iop_chan->lock);
651 slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op);
652 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
654 grp_start = sw_desc->group_head;
655 iop_desc_init_memset(grp_start, flags);
656 iop_desc_set_byte_count(grp_start, iop_chan, len);
657 iop_desc_set_block_fill_val(grp_start, value);
658 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
659 sw_desc->unmap_src_cnt = 1;
660 sw_desc->unmap_len = len;
661 sw_desc->async_tx.flags = flags;
663 spin_unlock_bh(&iop_chan->lock);
665 return sw_desc ? &sw_desc->async_tx : NULL;
668 static struct dma_async_tx_descriptor *
669 iop_adma_prep_dma_xor(struct dma_chan *chan, dma_addr_t dma_dest,
670 dma_addr_t *dma_src, unsigned int src_cnt, size_t len,
673 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
674 struct iop_adma_desc_slot *sw_desc, *grp_start;
675 int slot_cnt, slots_per_op;
679 BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
681 dev_dbg(iop_chan->device->common.dev,
682 "%s src_cnt: %d len: %u flags: %lx\n",
683 __func__, src_cnt, len, flags);
685 spin_lock_bh(&iop_chan->lock);
686 slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
687 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
689 grp_start = sw_desc->group_head;
690 iop_desc_init_xor(grp_start, src_cnt, flags);
691 iop_desc_set_byte_count(grp_start, iop_chan, len);
692 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
693 sw_desc->unmap_src_cnt = src_cnt;
694 sw_desc->unmap_len = len;
695 sw_desc->async_tx.flags = flags;
697 iop_desc_set_xor_src_addr(grp_start, src_cnt,
700 spin_unlock_bh(&iop_chan->lock);
702 return sw_desc ? &sw_desc->async_tx : NULL;
705 static struct dma_async_tx_descriptor *
706 iop_adma_prep_dma_xor_val(struct dma_chan *chan, dma_addr_t *dma_src,
707 unsigned int src_cnt, size_t len, u32 *result,
710 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
711 struct iop_adma_desc_slot *sw_desc, *grp_start;
712 int slot_cnt, slots_per_op;
717 dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
718 __func__, src_cnt, len);
720 spin_lock_bh(&iop_chan->lock);
721 slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
722 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
724 grp_start = sw_desc->group_head;
725 iop_desc_init_zero_sum(grp_start, src_cnt, flags);
726 iop_desc_set_zero_sum_byte_count(grp_start, len);
727 grp_start->xor_check_result = result;
728 pr_debug("\t%s: grp_start->xor_check_result: %p\n",
729 __func__, grp_start->xor_check_result);
730 sw_desc->unmap_src_cnt = src_cnt;
731 sw_desc->unmap_len = len;
732 sw_desc->async_tx.flags = flags;
734 iop_desc_set_zero_sum_src_addr(grp_start, src_cnt,
737 spin_unlock_bh(&iop_chan->lock);
739 return sw_desc ? &sw_desc->async_tx : NULL;
742 static struct dma_async_tx_descriptor *
743 iop_adma_prep_dma_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
744 unsigned int src_cnt, const unsigned char *scf, size_t len,
747 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
748 struct iop_adma_desc_slot *sw_desc, *g;
749 int slot_cnt, slots_per_op;
754 BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
756 dev_dbg(iop_chan->device->common.dev,
757 "%s src_cnt: %d len: %u flags: %lx\n",
758 __func__, src_cnt, len, flags);
760 if (dmaf_p_disabled_continue(flags))
761 continue_srcs = 1+src_cnt;
762 else if (dmaf_continue(flags))
763 continue_srcs = 3+src_cnt;
765 continue_srcs = 0+src_cnt;
767 spin_lock_bh(&iop_chan->lock);
768 slot_cnt = iop_chan_pq_slot_count(len, continue_srcs, &slots_per_op);
769 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
773 g = sw_desc->group_head;
774 iop_desc_set_byte_count(g, iop_chan, len);
776 /* even if P is disabled its destination address (bits
777 * [3:0]) must match Q. It is ok if P points to an
778 * invalid address, it won't be written.
780 if (flags & DMA_PREP_PQ_DISABLE_P)
781 dst[0] = dst[1] & 0x7;
783 iop_desc_set_pq_addr(g, dst);
784 sw_desc->unmap_src_cnt = src_cnt;
785 sw_desc->unmap_len = len;
786 sw_desc->async_tx.flags = flags;
787 for (i = 0; i < src_cnt; i++)
788 iop_desc_set_pq_src_addr(g, i, src[i], scf[i]);
790 /* if we are continuing a previous operation factor in
791 * the old p and q values, see the comment for dma_maxpq
792 * in include/linux/dmaengine.h
794 if (dmaf_p_disabled_continue(flags))
795 iop_desc_set_pq_src_addr(g, i++, dst[1], 1);
796 else if (dmaf_continue(flags)) {
797 iop_desc_set_pq_src_addr(g, i++, dst[0], 0);
798 iop_desc_set_pq_src_addr(g, i++, dst[1], 1);
799 iop_desc_set_pq_src_addr(g, i++, dst[1], 0);
801 iop_desc_init_pq(g, i, flags);
803 spin_unlock_bh(&iop_chan->lock);
805 return sw_desc ? &sw_desc->async_tx : NULL;
808 static struct dma_async_tx_descriptor *
809 iop_adma_prep_dma_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
810 unsigned int src_cnt, const unsigned char *scf,
811 size_t len, enum sum_check_flags *pqres,
814 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
815 struct iop_adma_desc_slot *sw_desc, *g;
816 int slot_cnt, slots_per_op;
820 BUG_ON(len > IOP_ADMA_XOR_MAX_BYTE_COUNT);
822 dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
823 __func__, src_cnt, len);
825 spin_lock_bh(&iop_chan->lock);
826 slot_cnt = iop_chan_pq_zero_sum_slot_count(len, src_cnt + 2, &slots_per_op);
827 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
829 /* for validate operations p and q are tagged onto the
830 * end of the source list
832 int pq_idx = src_cnt;
834 g = sw_desc->group_head;
835 iop_desc_init_pq_zero_sum(g, src_cnt+2, flags);
836 iop_desc_set_pq_zero_sum_byte_count(g, len);
837 g->pq_check_result = pqres;
838 pr_debug("\t%s: g->pq_check_result: %p\n",
839 __func__, g->pq_check_result);
840 sw_desc->unmap_src_cnt = src_cnt+2;
841 sw_desc->unmap_len = len;
842 sw_desc->async_tx.flags = flags;
844 iop_desc_set_pq_zero_sum_src_addr(g, src_cnt,
847 iop_desc_set_pq_zero_sum_addr(g, pq_idx, src);
849 spin_unlock_bh(&iop_chan->lock);
851 return sw_desc ? &sw_desc->async_tx : NULL;
854 static void iop_adma_free_chan_resources(struct dma_chan *chan)
856 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
857 struct iop_adma_desc_slot *iter, *_iter;
858 int in_use_descs = 0;
860 iop_adma_slot_cleanup(iop_chan);
862 spin_lock_bh(&iop_chan->lock);
863 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
866 list_del(&iter->chain_node);
868 list_for_each_entry_safe_reverse(
869 iter, _iter, &iop_chan->all_slots, slot_node) {
870 list_del(&iter->slot_node);
872 iop_chan->slots_allocated--;
874 iop_chan->last_used = NULL;
876 dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
877 __func__, iop_chan->slots_allocated);
878 spin_unlock_bh(&iop_chan->lock);
880 /* one is ok since we left it on there on purpose */
881 if (in_use_descs > 1)
882 printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
887 * iop_adma_status - poll the status of an ADMA transaction
888 * @chan: ADMA channel handle
889 * @cookie: ADMA transaction identifier
890 * @txstate: a holder for the current state of the channel or NULL
892 static enum dma_status iop_adma_status(struct dma_chan *chan,
894 struct dma_tx_state *txstate)
896 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
899 ret = dma_cookie_status(chan, cookie, txstate);
900 if (ret == DMA_SUCCESS)
903 iop_adma_slot_cleanup(iop_chan);
905 return dma_cookie_status(chan, cookie, txstate);
908 static irqreturn_t iop_adma_eot_handler(int irq, void *data)
910 struct iop_adma_chan *chan = data;
912 dev_dbg(chan->device->common.dev, "%s\n", __func__);
914 tasklet_schedule(&chan->irq_tasklet);
916 iop_adma_device_clear_eot_status(chan);
921 static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
923 struct iop_adma_chan *chan = data;
925 dev_dbg(chan->device->common.dev, "%s\n", __func__);
927 tasklet_schedule(&chan->irq_tasklet);
929 iop_adma_device_clear_eoc_status(chan);
934 static irqreturn_t iop_adma_err_handler(int irq, void *data)
936 struct iop_adma_chan *chan = data;
937 unsigned long status = iop_chan_get_status(chan);
939 dev_err(chan->device->common.dev,
940 "error ( %s%s%s%s%s%s%s)\n",
941 iop_is_err_int_parity(status, chan) ? "int_parity " : "",
942 iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
943 iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
944 iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
945 iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
946 iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
947 iop_is_err_split_tx(status, chan) ? "split_tx " : "");
949 iop_adma_device_clear_err_status(chan);
956 static void iop_adma_issue_pending(struct dma_chan *chan)
958 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
960 if (iop_chan->pending) {
961 iop_chan->pending = 0;
962 iop_chan_append(iop_chan);
967 * Perform a transaction to verify the HW works.
969 #define IOP_ADMA_TEST_SIZE 2000
971 static int iop_adma_memcpy_self_test(struct iop_adma_device *device)
975 dma_addr_t src_dma, dest_dma;
976 struct dma_chan *dma_chan;
978 struct dma_async_tx_descriptor *tx;
980 struct iop_adma_chan *iop_chan;
982 dev_dbg(device->common.dev, "%s\n", __func__);
984 src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
987 dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
993 /* Fill in src buffer */
994 for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
995 ((u8 *) src)[i] = (u8)i;
997 /* Start copy, using first DMA channel */
998 dma_chan = container_of(device->common.channels.next,
1001 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1006 dest_dma = dma_map_single(dma_chan->device->dev, dest,
1007 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
1008 src_dma = dma_map_single(dma_chan->device->dev, src,
1009 IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
1010 tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
1012 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1014 cookie = iop_adma_tx_submit(tx);
1015 iop_adma_issue_pending(dma_chan);
1018 if (iop_adma_status(dma_chan, cookie, NULL) !=
1020 dev_err(dma_chan->device->dev,
1021 "Self-test copy timed out, disabling\n");
1023 goto free_resources;
1026 iop_chan = to_iop_adma_chan(dma_chan);
1027 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
1028 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
1029 if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
1030 dev_err(dma_chan->device->dev,
1031 "Self-test copy failed compare, disabling\n");
1033 goto free_resources;
1037 iop_adma_free_chan_resources(dma_chan);
1044 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
1046 iop_adma_xor_val_self_test(struct iop_adma_device *device)
1050 struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
1051 struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
1052 dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
1053 dma_addr_t dma_addr, dest_dma;
1054 struct dma_async_tx_descriptor *tx;
1055 struct dma_chan *dma_chan;
1056 dma_cookie_t cookie;
1059 u32 zero_sum_result;
1061 struct iop_adma_chan *iop_chan;
1063 dev_dbg(device->common.dev, "%s\n", __func__);
1065 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
1066 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
1067 if (!xor_srcs[src_idx]) {
1069 __free_page(xor_srcs[src_idx]);
1074 dest = alloc_page(GFP_KERNEL);
1077 __free_page(xor_srcs[src_idx]);
1081 /* Fill in src buffers */
1082 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
1083 u8 *ptr = page_address(xor_srcs[src_idx]);
1084 for (i = 0; i < PAGE_SIZE; i++)
1085 ptr[i] = (1 << src_idx);
1088 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
1089 cmp_byte ^= (u8) (1 << src_idx);
1091 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
1092 (cmp_byte << 8) | cmp_byte;
1094 memset(page_address(dest), 0, PAGE_SIZE);
1096 dma_chan = container_of(device->common.channels.next,
1099 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1105 dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
1106 PAGE_SIZE, DMA_FROM_DEVICE);
1107 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1108 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
1109 0, PAGE_SIZE, DMA_TO_DEVICE);
1110 tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
1111 IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE,
1112 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1114 cookie = iop_adma_tx_submit(tx);
1115 iop_adma_issue_pending(dma_chan);
1118 if (iop_adma_status(dma_chan, cookie, NULL) !=
1120 dev_err(dma_chan->device->dev,
1121 "Self-test xor timed out, disabling\n");
1123 goto free_resources;
1126 iop_chan = to_iop_adma_chan(dma_chan);
1127 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
1128 PAGE_SIZE, DMA_FROM_DEVICE);
1129 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1130 u32 *ptr = page_address(dest);
1131 if (ptr[i] != cmp_word) {
1132 dev_err(dma_chan->device->dev,
1133 "Self-test xor failed compare, disabling\n");
1135 goto free_resources;
1138 dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
1139 PAGE_SIZE, DMA_TO_DEVICE);
1141 /* skip zero sum if the capability is not present */
1142 if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask))
1143 goto free_resources;
1145 /* zero sum the sources with the destintation page */
1146 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1147 zero_sum_srcs[i] = xor_srcs[i];
1148 zero_sum_srcs[i] = dest;
1150 zero_sum_result = 1;
1152 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1153 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1154 zero_sum_srcs[i], 0, PAGE_SIZE,
1156 tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1157 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1159 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1161 cookie = iop_adma_tx_submit(tx);
1162 iop_adma_issue_pending(dma_chan);
1165 if (iop_adma_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1166 dev_err(dma_chan->device->dev,
1167 "Self-test zero sum timed out, disabling\n");
1169 goto free_resources;
1172 if (zero_sum_result != 0) {
1173 dev_err(dma_chan->device->dev,
1174 "Self-test zero sum failed compare, disabling\n");
1176 goto free_resources;
1180 dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
1181 PAGE_SIZE, DMA_FROM_DEVICE);
1182 tx = iop_adma_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE,
1183 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1185 cookie = iop_adma_tx_submit(tx);
1186 iop_adma_issue_pending(dma_chan);
1189 if (iop_adma_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1190 dev_err(dma_chan->device->dev,
1191 "Self-test memset timed out, disabling\n");
1193 goto free_resources;
1196 for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1197 u32 *ptr = page_address(dest);
1199 dev_err(dma_chan->device->dev,
1200 "Self-test memset failed compare, disabling\n");
1202 goto free_resources;
1206 /* test for non-zero parity sum */
1207 zero_sum_result = 0;
1208 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1209 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1210 zero_sum_srcs[i], 0, PAGE_SIZE,
1212 tx = iop_adma_prep_dma_xor_val(dma_chan, dma_srcs,
1213 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1215 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1217 cookie = iop_adma_tx_submit(tx);
1218 iop_adma_issue_pending(dma_chan);
1221 if (iop_adma_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1222 dev_err(dma_chan->device->dev,
1223 "Self-test non-zero sum timed out, disabling\n");
1225 goto free_resources;
1228 if (zero_sum_result != 1) {
1229 dev_err(dma_chan->device->dev,
1230 "Self-test non-zero sum failed compare, disabling\n");
1232 goto free_resources;
1236 iop_adma_free_chan_resources(dma_chan);
1238 src_idx = IOP_ADMA_NUM_SRC_TEST;
1240 __free_page(xor_srcs[src_idx]);
1245 #ifdef CONFIG_RAID6_PQ
1247 iop_adma_pq_zero_sum_self_test(struct iop_adma_device *device)
1249 /* combined sources, software pq results, and extra hw pq results */
1250 struct page *pq[IOP_ADMA_NUM_SRC_TEST+2+2];
1251 /* ptr to the extra hw pq buffers defined above */
1252 struct page **pq_hw = &pq[IOP_ADMA_NUM_SRC_TEST+2];
1253 /* address conversion buffers (dma_map / page_address) */
1254 void *pq_sw[IOP_ADMA_NUM_SRC_TEST+2];
1255 dma_addr_t pq_src[IOP_ADMA_NUM_SRC_TEST+2];
1256 dma_addr_t *pq_dest = &pq_src[IOP_ADMA_NUM_SRC_TEST];
1259 struct dma_async_tx_descriptor *tx;
1260 struct dma_chan *dma_chan;
1261 dma_cookie_t cookie;
1262 u32 zero_sum_result;
1266 dev_dbg(device->common.dev, "%s\n", __func__);
1268 for (i = 0; i < ARRAY_SIZE(pq); i++) {
1269 pq[i] = alloc_page(GFP_KERNEL);
1277 /* Fill in src buffers */
1278 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) {
1279 pq_sw[i] = page_address(pq[i]);
1280 memset(pq_sw[i], 0x11111111 * (1<<i), PAGE_SIZE);
1282 pq_sw[i] = page_address(pq[i]);
1283 pq_sw[i+1] = page_address(pq[i+1]);
1285 dma_chan = container_of(device->common.channels.next,
1288 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
1293 dev = dma_chan->device->dev;
1295 /* initialize the dests */
1296 memset(page_address(pq_hw[0]), 0 , PAGE_SIZE);
1297 memset(page_address(pq_hw[1]), 0 , PAGE_SIZE);
1300 pq_dest[0] = dma_map_page(dev, pq_hw[0], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1301 pq_dest[1] = dma_map_page(dev, pq_hw[1], 0, PAGE_SIZE, DMA_FROM_DEVICE);
1302 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1303 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1306 tx = iop_adma_prep_dma_pq(dma_chan, pq_dest, pq_src,
1307 IOP_ADMA_NUM_SRC_TEST, (u8 *)raid6_gfexp,
1309 DMA_PREP_INTERRUPT |
1312 cookie = iop_adma_tx_submit(tx);
1313 iop_adma_issue_pending(dma_chan);
1316 if (iop_adma_status(dma_chan, cookie, NULL) !=
1318 dev_err(dev, "Self-test pq timed out, disabling\n");
1320 goto free_resources;
1323 raid6_call.gen_syndrome(IOP_ADMA_NUM_SRC_TEST+2, PAGE_SIZE, pq_sw);
1325 if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST],
1326 page_address(pq_hw[0]), PAGE_SIZE) != 0) {
1327 dev_err(dev, "Self-test p failed compare, disabling\n");
1329 goto free_resources;
1331 if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST+1],
1332 page_address(pq_hw[1]), PAGE_SIZE) != 0) {
1333 dev_err(dev, "Self-test q failed compare, disabling\n");
1335 goto free_resources;
1338 /* test correct zero sum using the software generated pq values */
1339 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1340 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1343 zero_sum_result = ~0;
1344 tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1345 pq_src, IOP_ADMA_NUM_SRC_TEST,
1346 raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1347 DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1349 cookie = iop_adma_tx_submit(tx);
1350 iop_adma_issue_pending(dma_chan);
1353 if (iop_adma_status(dma_chan, cookie, NULL) !=
1355 dev_err(dev, "Self-test pq-zero-sum timed out, disabling\n");
1357 goto free_resources;
1360 if (zero_sum_result != 0) {
1361 dev_err(dev, "Self-test pq-zero-sum failed to validate: %x\n",
1364 goto free_resources;
1367 /* test incorrect zero sum */
1368 i = IOP_ADMA_NUM_SRC_TEST;
1369 memset(pq_sw[i] + 100, 0, 100);
1370 memset(pq_sw[i+1] + 200, 0, 200);
1371 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
1372 pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
1375 zero_sum_result = 0;
1376 tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
1377 pq_src, IOP_ADMA_NUM_SRC_TEST,
1378 raid6_gfexp, PAGE_SIZE, &zero_sum_result,
1379 DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
1381 cookie = iop_adma_tx_submit(tx);
1382 iop_adma_issue_pending(dma_chan);
1385 if (iop_adma_status(dma_chan, cookie, NULL) !=
1387 dev_err(dev, "Self-test !pq-zero-sum timed out, disabling\n");
1389 goto free_resources;
1392 if (zero_sum_result != (SUM_CHECK_P_RESULT | SUM_CHECK_Q_RESULT)) {
1393 dev_err(dev, "Self-test !pq-zero-sum failed to validate: %x\n",
1396 goto free_resources;
1400 iop_adma_free_chan_resources(dma_chan);
1409 static int iop_adma_remove(struct platform_device *dev)
1411 struct iop_adma_device *device = platform_get_drvdata(dev);
1412 struct dma_chan *chan, *_chan;
1413 struct iop_adma_chan *iop_chan;
1414 struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
1416 dma_async_device_unregister(&device->common);
1418 dma_free_coherent(&dev->dev, plat_data->pool_size,
1419 device->dma_desc_pool_virt, device->dma_desc_pool);
1421 list_for_each_entry_safe(chan, _chan, &device->common.channels,
1423 iop_chan = to_iop_adma_chan(chan);
1424 list_del(&chan->device_node);
1432 static int iop_adma_probe(struct platform_device *pdev)
1434 struct resource *res;
1436 struct iop_adma_device *adev;
1437 struct iop_adma_chan *iop_chan;
1438 struct dma_device *dma_dev;
1439 struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
1441 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1445 if (!devm_request_mem_region(&pdev->dev, res->start,
1446 resource_size(res), pdev->name))
1449 adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1452 dma_dev = &adev->common;
1454 /* allocate coherent memory for hardware descriptors
1455 * note: writecombine gives slightly better performance, but
1456 * requires that we explicitly flush the writes
1458 if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1459 plat_data->pool_size,
1460 &adev->dma_desc_pool,
1461 GFP_KERNEL)) == NULL) {
1466 dev_dbg(&pdev->dev, "%s: allocated descriptor pool virt %p phys %p\n",
1467 __func__, adev->dma_desc_pool_virt,
1468 (void *) adev->dma_desc_pool);
1470 adev->id = plat_data->hw_id;
1472 /* discover transaction capabilites from the platform data */
1473 dma_dev->cap_mask = plat_data->cap_mask;
1476 platform_set_drvdata(pdev, adev);
1478 INIT_LIST_HEAD(&dma_dev->channels);
1480 /* set base routines */
1481 dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1482 dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1483 dma_dev->device_tx_status = iop_adma_status;
1484 dma_dev->device_issue_pending = iop_adma_issue_pending;
1485 dma_dev->dev = &pdev->dev;
1487 /* set prep routines based on capability */
1488 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1489 dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1490 if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1491 dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
1492 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1493 dma_dev->max_xor = iop_adma_get_max_xor();
1494 dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1496 if (dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask))
1497 dma_dev->device_prep_dma_xor_val =
1498 iop_adma_prep_dma_xor_val;
1499 if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1500 dma_set_maxpq(dma_dev, iop_adma_get_max_pq(), 0);
1501 dma_dev->device_prep_dma_pq = iop_adma_prep_dma_pq;
1503 if (dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask))
1504 dma_dev->device_prep_dma_pq_val =
1505 iop_adma_prep_dma_pq_val;
1506 if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1507 dma_dev->device_prep_dma_interrupt =
1508 iop_adma_prep_dma_interrupt;
1510 iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1515 iop_chan->device = adev;
1517 iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1518 resource_size(res));
1519 if (!iop_chan->mmr_base) {
1521 goto err_free_iop_chan;
1523 tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1526 /* clear errors before enabling interrupts */
1527 iop_adma_device_clear_err_status(iop_chan);
1529 for (i = 0; i < 3; i++) {
1530 irq_handler_t handler[] = { iop_adma_eot_handler,
1531 iop_adma_eoc_handler,
1532 iop_adma_err_handler };
1533 int irq = platform_get_irq(pdev, i);
1536 goto err_free_iop_chan;
1538 ret = devm_request_irq(&pdev->dev, irq,
1539 handler[i], 0, pdev->name, iop_chan);
1541 goto err_free_iop_chan;
1545 spin_lock_init(&iop_chan->lock);
1546 INIT_LIST_HEAD(&iop_chan->chain);
1547 INIT_LIST_HEAD(&iop_chan->all_slots);
1548 iop_chan->common.device = dma_dev;
1549 dma_cookie_init(&iop_chan->common);
1550 list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1552 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1553 ret = iop_adma_memcpy_self_test(adev);
1554 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1556 goto err_free_iop_chan;
1559 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
1560 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1561 ret = iop_adma_xor_val_self_test(adev);
1562 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1564 goto err_free_iop_chan;
1567 if (dma_has_cap(DMA_PQ, dma_dev->cap_mask) &&
1568 dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask)) {
1569 #ifdef CONFIG_RAID6_PQ
1570 ret = iop_adma_pq_zero_sum_self_test(adev);
1571 dev_dbg(&pdev->dev, "pq self test returned %d\n", ret);
1573 /* can not test raid6, so do not publish capability */
1574 dma_cap_clear(DMA_PQ, dma_dev->cap_mask);
1575 dma_cap_clear(DMA_PQ_VAL, dma_dev->cap_mask);
1579 goto err_free_iop_chan;
1582 dev_info(&pdev->dev, "Intel(R) IOP: ( %s%s%s%s%s%s%s)\n",
1583 dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "",
1584 dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask) ? "pq_val " : "",
1585 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1586 dma_has_cap(DMA_XOR_VAL, dma_dev->cap_mask) ? "xor_val " : "",
1587 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "",
1588 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1589 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1591 dma_async_device_register(dma_dev);
1597 dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1598 adev->dma_desc_pool_virt, adev->dma_desc_pool);
1605 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1607 struct iop_adma_desc_slot *sw_desc, *grp_start;
1608 dma_cookie_t cookie;
1609 int slot_cnt, slots_per_op;
1611 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1613 spin_lock_bh(&iop_chan->lock);
1614 slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1615 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1617 grp_start = sw_desc->group_head;
1619 list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1620 async_tx_ack(&sw_desc->async_tx);
1621 iop_desc_init_memcpy(grp_start, 0);
1622 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1623 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1624 iop_desc_set_memcpy_src_addr(grp_start, 0);
1626 cookie = dma_cookie_assign(&sw_desc->async_tx);
1628 /* initialize the completed cookie to be less than
1629 * the most recently used cookie
1631 iop_chan->common.completed_cookie = cookie - 1;
1633 /* channel should not be busy */
1634 BUG_ON(iop_chan_is_busy(iop_chan));
1636 /* clear any prior error-status bits */
1637 iop_adma_device_clear_err_status(iop_chan);
1639 /* disable operation */
1640 iop_chan_disable(iop_chan);
1642 /* set the descriptor address */
1643 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1645 /* 1/ don't add pre-chained descriptors
1646 * 2/ dummy read to flush next_desc write
1648 BUG_ON(iop_desc_get_next_desc(sw_desc));
1650 /* run the descriptor */
1651 iop_chan_enable(iop_chan);
1653 dev_err(iop_chan->device->common.dev,
1654 "failed to allocate null descriptor\n");
1655 spin_unlock_bh(&iop_chan->lock);
1658 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1660 struct iop_adma_desc_slot *sw_desc, *grp_start;
1661 dma_cookie_t cookie;
1662 int slot_cnt, slots_per_op;
1664 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1666 spin_lock_bh(&iop_chan->lock);
1667 slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1668 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1670 grp_start = sw_desc->group_head;
1671 list_splice_init(&sw_desc->tx_list, &iop_chan->chain);
1672 async_tx_ack(&sw_desc->async_tx);
1673 iop_desc_init_null_xor(grp_start, 2, 0);
1674 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1675 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1676 iop_desc_set_xor_src_addr(grp_start, 0, 0);
1677 iop_desc_set_xor_src_addr(grp_start, 1, 0);
1679 cookie = dma_cookie_assign(&sw_desc->async_tx);
1681 /* initialize the completed cookie to be less than
1682 * the most recently used cookie
1684 iop_chan->common.completed_cookie = cookie - 1;
1686 /* channel should not be busy */
1687 BUG_ON(iop_chan_is_busy(iop_chan));
1689 /* clear any prior error-status bits */
1690 iop_adma_device_clear_err_status(iop_chan);
1692 /* disable operation */
1693 iop_chan_disable(iop_chan);
1695 /* set the descriptor address */
1696 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1698 /* 1/ don't add pre-chained descriptors
1699 * 2/ dummy read to flush next_desc write
1701 BUG_ON(iop_desc_get_next_desc(sw_desc));
1703 /* run the descriptor */
1704 iop_chan_enable(iop_chan);
1706 dev_err(iop_chan->device->common.dev,
1707 "failed to allocate null descriptor\n");
1708 spin_unlock_bh(&iop_chan->lock);
1711 static struct platform_driver iop_adma_driver = {
1712 .probe = iop_adma_probe,
1713 .remove = iop_adma_remove,
1715 .owner = THIS_MODULE,
1720 module_platform_driver(iop_adma_driver);
1722 MODULE_AUTHOR("Intel Corporation");
1723 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1724 MODULE_LICENSE("GPL");
1725 MODULE_ALIAS("platform:iop-adma");