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>
35 #include <mach/adma.h>
37 #define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
38 #define to_iop_adma_device(dev) \
39 container_of(dev, struct iop_adma_device, common)
40 #define tx_to_iop_adma_slot(tx) \
41 container_of(tx, struct iop_adma_desc_slot, async_tx)
44 * iop_adma_free_slots - flags descriptor slots for reuse
46 * Caller must hold &iop_chan->lock while calling this function
48 static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
50 int stride = slot->slots_per_op;
53 slot->slots_per_op = 0;
54 slot = list_entry(slot->slot_node.next,
55 struct iop_adma_desc_slot,
61 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
62 struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
64 BUG_ON(desc->async_tx.cookie < 0);
65 if (desc->async_tx.cookie > 0) {
66 cookie = desc->async_tx.cookie;
67 desc->async_tx.cookie = 0;
69 /* call the callback (must not sleep or submit new
70 * operations to this channel)
72 if (desc->async_tx.callback)
73 desc->async_tx.callback(
74 desc->async_tx.callback_param);
76 /* unmap dma addresses
77 * (unmap_single vs unmap_page?)
79 if (desc->group_head && desc->unmap_len) {
80 struct iop_adma_desc_slot *unmap = desc->group_head;
82 &iop_chan->device->pdev->dev;
83 u32 len = unmap->unmap_len;
84 enum dma_ctrl_flags flags = desc->async_tx.flags;
89 src_cnt = unmap->unmap_src_cnt;
90 dest = iop_desc_get_dest_addr(unmap, iop_chan);
91 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
92 enum dma_data_direction dir;
94 if (src_cnt > 1) /* is xor? */
95 dir = DMA_BIDIRECTIONAL;
97 dir = DMA_FROM_DEVICE;
99 dma_unmap_page(dev, dest, len, dir);
102 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
104 addr = iop_desc_get_src_addr(unmap,
109 dma_unmap_page(dev, addr, len,
113 desc->group_head = NULL;
117 /* run dependent operations */
118 dma_run_dependencies(&desc->async_tx);
124 iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
125 struct iop_adma_chan *iop_chan)
127 /* the client is allowed to attach dependent operations
130 if (!async_tx_test_ack(&desc->async_tx))
133 /* leave the last descriptor in the chain
134 * so we can append to it
136 if (desc->chain_node.next == &iop_chan->chain)
139 dev_dbg(iop_chan->device->common.dev,
140 "\tfree slot: %d slots_per_op: %d\n",
141 desc->idx, desc->slots_per_op);
143 list_del(&desc->chain_node);
144 iop_adma_free_slots(desc);
149 static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
151 struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
152 dma_cookie_t cookie = 0;
153 u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
154 int busy = iop_chan_is_busy(iop_chan);
155 int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
157 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
158 /* free completed slots from the chain starting with
159 * the oldest descriptor
161 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
163 pr_debug("\tcookie: %d slot: %d busy: %d "
164 "this_desc: %#x next_desc: %#x ack: %d\n",
165 iter->async_tx.cookie, iter->idx, busy,
166 iter->async_tx.phys, iop_desc_get_next_desc(iter),
167 async_tx_test_ack(&iter->async_tx));
169 prefetch(&_iter->async_tx);
171 /* do not advance past the current descriptor loaded into the
172 * hardware channel, subsequent descriptors are either in
173 * process or have not been submitted
178 /* stop the search if we reach the current descriptor and the
179 * channel is busy, or if it appears that the current descriptor
180 * needs to be re-read (i.e. has been appended to)
182 if (iter->async_tx.phys == current_desc) {
183 BUG_ON(seen_current++);
184 if (busy || iop_desc_get_next_desc(iter))
188 /* detect the start of a group transaction */
189 if (!slot_cnt && !slots_per_op) {
190 slot_cnt = iter->slot_cnt;
191 slots_per_op = iter->slots_per_op;
192 if (slot_cnt <= slots_per_op) {
199 pr_debug("\tgroup++\n");
202 slot_cnt -= slots_per_op;
205 /* all the members of a group are complete */
206 if (slots_per_op != 0 && slot_cnt == 0) {
207 struct iop_adma_desc_slot *grp_iter, *_grp_iter;
208 int end_of_chain = 0;
209 pr_debug("\tgroup end\n");
211 /* collect the total results */
212 if (grp_start->xor_check_result) {
213 u32 zero_sum_result = 0;
214 slot_cnt = grp_start->slot_cnt;
215 grp_iter = grp_start;
217 list_for_each_entry_from(grp_iter,
218 &iop_chan->chain, chain_node) {
220 iop_desc_get_zero_result(grp_iter);
221 pr_debug("\titer%d result: %d\n",
222 grp_iter->idx, zero_sum_result);
223 slot_cnt -= slots_per_op;
227 pr_debug("\tgrp_start->xor_check_result: %p\n",
228 grp_start->xor_check_result);
229 *grp_start->xor_check_result = zero_sum_result;
232 /* clean up the group */
233 slot_cnt = grp_start->slot_cnt;
234 grp_iter = grp_start;
235 list_for_each_entry_safe_from(grp_iter, _grp_iter,
236 &iop_chan->chain, chain_node) {
237 cookie = iop_adma_run_tx_complete_actions(
238 grp_iter, iop_chan, cookie);
240 slot_cnt -= slots_per_op;
241 end_of_chain = iop_adma_clean_slot(grp_iter,
244 if (slot_cnt == 0 || end_of_chain)
248 /* the group should be complete at this point */
257 } else if (slots_per_op) /* wait for group completion */
260 /* write back zero sum results (single descriptor case) */
261 if (iter->xor_check_result && iter->async_tx.cookie)
262 *iter->xor_check_result =
263 iop_desc_get_zero_result(iter);
265 cookie = iop_adma_run_tx_complete_actions(
266 iter, iop_chan, cookie);
268 if (iop_adma_clean_slot(iter, iop_chan))
273 iop_chan->completed_cookie = cookie;
274 pr_debug("\tcompleted cookie %d\n", cookie);
279 iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
281 spin_lock_bh(&iop_chan->lock);
282 __iop_adma_slot_cleanup(iop_chan);
283 spin_unlock_bh(&iop_chan->lock);
286 static void iop_adma_tasklet(unsigned long data)
288 struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data;
290 spin_lock(&iop_chan->lock);
291 __iop_adma_slot_cleanup(iop_chan);
292 spin_unlock(&iop_chan->lock);
295 static struct iop_adma_desc_slot *
296 iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
299 struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
301 int slots_found, retry = 0;
303 /* start search from the last allocated descrtiptor
304 * if a contiguous allocation can not be found start searching
305 * from the beginning of the list
310 iter = iop_chan->last_used;
312 iter = list_entry(&iop_chan->all_slots,
313 struct iop_adma_desc_slot,
316 list_for_each_entry_safe_continue(
317 iter, _iter, &iop_chan->all_slots, slot_node) {
319 prefetch(&_iter->async_tx);
320 if (iter->slots_per_op) {
321 /* give up after finding the first busy slot
322 * on the second pass through the list
331 /* start the allocation if the slot is correctly aligned */
332 if (!slots_found++) {
333 if (iop_desc_is_aligned(iter, slots_per_op))
341 if (slots_found == num_slots) {
342 struct iop_adma_desc_slot *alloc_tail = NULL;
343 struct iop_adma_desc_slot *last_used = NULL;
347 dev_dbg(iop_chan->device->common.dev,
348 "allocated slot: %d "
349 "(desc %p phys: %#x) slots_per_op %d\n",
350 iter->idx, iter->hw_desc,
351 iter->async_tx.phys, slots_per_op);
353 /* pre-ack all but the last descriptor */
354 if (num_slots != slots_per_op)
355 async_tx_ack(&iter->async_tx);
357 list_add_tail(&iter->chain_node, &chain);
359 iter->async_tx.cookie = 0;
360 iter->slot_cnt = num_slots;
361 iter->xor_check_result = NULL;
362 for (i = 0; i < slots_per_op; i++) {
363 iter->slots_per_op = slots_per_op - i;
365 iter = list_entry(iter->slot_node.next,
366 struct iop_adma_desc_slot,
369 num_slots -= slots_per_op;
371 alloc_tail->group_head = alloc_start;
372 alloc_tail->async_tx.cookie = -EBUSY;
373 list_splice(&chain, &alloc_tail->async_tx.tx_list);
374 iop_chan->last_used = last_used;
375 iop_desc_clear_next_desc(alloc_start);
376 iop_desc_clear_next_desc(alloc_tail);
383 /* perform direct reclaim if the allocation fails */
384 __iop_adma_slot_cleanup(iop_chan);
390 iop_desc_assign_cookie(struct iop_adma_chan *iop_chan,
391 struct iop_adma_desc_slot *desc)
393 dma_cookie_t cookie = iop_chan->common.cookie;
397 iop_chan->common.cookie = desc->async_tx.cookie = cookie;
401 static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
403 dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
406 if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
407 iop_chan->pending = 0;
408 iop_chan_append(iop_chan);
413 iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
415 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
416 struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
417 struct iop_adma_desc_slot *grp_start, *old_chain_tail;
423 grp_start = sw_desc->group_head;
424 slot_cnt = grp_start->slot_cnt;
425 slots_per_op = grp_start->slots_per_op;
427 spin_lock_bh(&iop_chan->lock);
428 cookie = iop_desc_assign_cookie(iop_chan, sw_desc);
430 old_chain_tail = list_entry(iop_chan->chain.prev,
431 struct iop_adma_desc_slot, chain_node);
432 list_splice_init(&sw_desc->async_tx.tx_list,
433 &old_chain_tail->chain_node);
435 /* fix up the hardware chain */
436 next_dma = grp_start->async_tx.phys;
437 iop_desc_set_next_desc(old_chain_tail, next_dma);
438 BUG_ON(iop_desc_get_next_desc(old_chain_tail) != next_dma); /* flush */
440 /* check for pre-chained descriptors */
441 iop_paranoia(iop_desc_get_next_desc(sw_desc));
443 /* increment the pending count by the number of slots
444 * memcpy operations have a 1:1 (slot:operation) relation
445 * other operations are heavier and will pop the threshold
448 iop_chan->pending += slot_cnt;
449 iop_adma_check_threshold(iop_chan);
450 spin_unlock_bh(&iop_chan->lock);
452 dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
453 __func__, sw_desc->async_tx.cookie, sw_desc->idx);
458 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
459 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
462 * iop_adma_alloc_chan_resources - returns the number of allocated descriptors
463 * @chan - allocate descriptor resources for this channel
464 * @client - current client requesting the channel be ready for requests
466 * Note: We keep the slots for 1 operation on iop_chan->chain at all times. To
467 * avoid deadlock, via async_xor, num_descs_in_pool must at a minimum be
468 * greater than 2x the number slots needed to satisfy a device->max_xor
471 static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
475 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
476 struct iop_adma_desc_slot *slot = NULL;
477 int init = iop_chan->slots_allocated ? 0 : 1;
478 struct iop_adma_platform_data *plat_data =
479 iop_chan->device->pdev->dev.platform_data;
480 int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
482 /* Allocate descriptor slots */
484 idx = iop_chan->slots_allocated;
485 if (idx == num_descs_in_pool)
488 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
490 printk(KERN_INFO "IOP ADMA Channel only initialized"
491 " %d descriptor slots", idx);
494 hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
495 slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
497 dma_async_tx_descriptor_init(&slot->async_tx, chan);
498 slot->async_tx.tx_submit = iop_adma_tx_submit;
499 INIT_LIST_HEAD(&slot->chain_node);
500 INIT_LIST_HEAD(&slot->slot_node);
501 INIT_LIST_HEAD(&slot->async_tx.tx_list);
502 hw_desc = (char *) iop_chan->device->dma_desc_pool;
503 slot->async_tx.phys =
504 (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
507 spin_lock_bh(&iop_chan->lock);
508 iop_chan->slots_allocated++;
509 list_add_tail(&slot->slot_node, &iop_chan->all_slots);
510 spin_unlock_bh(&iop_chan->lock);
511 } while (iop_chan->slots_allocated < num_descs_in_pool);
513 if (idx && !iop_chan->last_used)
514 iop_chan->last_used = list_entry(iop_chan->all_slots.next,
515 struct iop_adma_desc_slot,
518 dev_dbg(iop_chan->device->common.dev,
519 "allocated %d descriptor slots last_used: %p\n",
520 iop_chan->slots_allocated, iop_chan->last_used);
522 /* initialize the channel and the chain with a null operation */
524 if (dma_has_cap(DMA_MEMCPY,
525 iop_chan->device->common.cap_mask))
526 iop_chan_start_null_memcpy(iop_chan);
527 else if (dma_has_cap(DMA_XOR,
528 iop_chan->device->common.cap_mask))
529 iop_chan_start_null_xor(iop_chan);
534 return (idx > 0) ? idx : -ENOMEM;
537 static struct dma_async_tx_descriptor *
538 iop_adma_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags)
540 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
541 struct iop_adma_desc_slot *sw_desc, *grp_start;
542 int slot_cnt, slots_per_op;
544 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
546 spin_lock_bh(&iop_chan->lock);
547 slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
548 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
550 grp_start = sw_desc->group_head;
551 iop_desc_init_interrupt(grp_start, iop_chan);
552 grp_start->unmap_len = 0;
553 sw_desc->async_tx.flags = flags;
555 spin_unlock_bh(&iop_chan->lock);
557 return sw_desc ? &sw_desc->async_tx : NULL;
560 static struct dma_async_tx_descriptor *
561 iop_adma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
562 dma_addr_t dma_src, size_t len, unsigned long flags)
564 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
565 struct iop_adma_desc_slot *sw_desc, *grp_start;
566 int slot_cnt, slots_per_op;
570 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
572 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
575 spin_lock_bh(&iop_chan->lock);
576 slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
577 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
579 grp_start = sw_desc->group_head;
580 iop_desc_init_memcpy(grp_start, flags);
581 iop_desc_set_byte_count(grp_start, iop_chan, len);
582 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
583 iop_desc_set_memcpy_src_addr(grp_start, dma_src);
584 sw_desc->unmap_src_cnt = 1;
585 sw_desc->unmap_len = len;
586 sw_desc->async_tx.flags = flags;
588 spin_unlock_bh(&iop_chan->lock);
590 return sw_desc ? &sw_desc->async_tx : NULL;
593 static struct dma_async_tx_descriptor *
594 iop_adma_prep_dma_memset(struct dma_chan *chan, dma_addr_t dma_dest,
595 int value, size_t len, unsigned long flags)
597 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
598 struct iop_adma_desc_slot *sw_desc, *grp_start;
599 int slot_cnt, slots_per_op;
603 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
605 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
608 spin_lock_bh(&iop_chan->lock);
609 slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op);
610 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
612 grp_start = sw_desc->group_head;
613 iop_desc_init_memset(grp_start, flags);
614 iop_desc_set_byte_count(grp_start, iop_chan, len);
615 iop_desc_set_block_fill_val(grp_start, value);
616 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
617 sw_desc->unmap_src_cnt = 1;
618 sw_desc->unmap_len = len;
619 sw_desc->async_tx.flags = flags;
621 spin_unlock_bh(&iop_chan->lock);
623 return sw_desc ? &sw_desc->async_tx : NULL;
626 static struct dma_async_tx_descriptor *
627 iop_adma_prep_dma_xor(struct dma_chan *chan, dma_addr_t dma_dest,
628 dma_addr_t *dma_src, unsigned int src_cnt, size_t len,
631 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
632 struct iop_adma_desc_slot *sw_desc, *grp_start;
633 int slot_cnt, slots_per_op;
637 BUG_ON(unlikely(len > IOP_ADMA_XOR_MAX_BYTE_COUNT));
639 dev_dbg(iop_chan->device->common.dev,
640 "%s src_cnt: %d len: %u flags: %lx\n",
641 __func__, src_cnt, len, flags);
643 spin_lock_bh(&iop_chan->lock);
644 slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
645 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
647 grp_start = sw_desc->group_head;
648 iop_desc_init_xor(grp_start, src_cnt, flags);
649 iop_desc_set_byte_count(grp_start, iop_chan, len);
650 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
651 sw_desc->unmap_src_cnt = src_cnt;
652 sw_desc->unmap_len = len;
653 sw_desc->async_tx.flags = flags;
655 iop_desc_set_xor_src_addr(grp_start, src_cnt,
658 spin_unlock_bh(&iop_chan->lock);
660 return sw_desc ? &sw_desc->async_tx : NULL;
663 static struct dma_async_tx_descriptor *
664 iop_adma_prep_dma_zero_sum(struct dma_chan *chan, dma_addr_t *dma_src,
665 unsigned int src_cnt, size_t len, u32 *result,
668 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
669 struct iop_adma_desc_slot *sw_desc, *grp_start;
670 int slot_cnt, slots_per_op;
675 dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
676 __func__, src_cnt, len);
678 spin_lock_bh(&iop_chan->lock);
679 slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
680 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
682 grp_start = sw_desc->group_head;
683 iop_desc_init_zero_sum(grp_start, src_cnt, flags);
684 iop_desc_set_zero_sum_byte_count(grp_start, len);
685 grp_start->xor_check_result = result;
686 pr_debug("\t%s: grp_start->xor_check_result: %p\n",
687 __func__, grp_start->xor_check_result);
688 sw_desc->unmap_src_cnt = src_cnt;
689 sw_desc->unmap_len = len;
690 sw_desc->async_tx.flags = flags;
692 iop_desc_set_zero_sum_src_addr(grp_start, src_cnt,
695 spin_unlock_bh(&iop_chan->lock);
697 return sw_desc ? &sw_desc->async_tx : NULL;
700 static void iop_adma_free_chan_resources(struct dma_chan *chan)
702 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
703 struct iop_adma_desc_slot *iter, *_iter;
704 int in_use_descs = 0;
706 iop_adma_slot_cleanup(iop_chan);
708 spin_lock_bh(&iop_chan->lock);
709 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
712 list_del(&iter->chain_node);
714 list_for_each_entry_safe_reverse(
715 iter, _iter, &iop_chan->all_slots, slot_node) {
716 list_del(&iter->slot_node);
718 iop_chan->slots_allocated--;
720 iop_chan->last_used = NULL;
722 dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
723 __func__, iop_chan->slots_allocated);
724 spin_unlock_bh(&iop_chan->lock);
726 /* one is ok since we left it on there on purpose */
727 if (in_use_descs > 1)
728 printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
733 * iop_adma_is_complete - poll the status of an ADMA transaction
734 * @chan: ADMA channel handle
735 * @cookie: ADMA transaction identifier
737 static enum dma_status iop_adma_is_complete(struct dma_chan *chan,
742 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
743 dma_cookie_t last_used;
744 dma_cookie_t last_complete;
747 last_used = chan->cookie;
748 last_complete = iop_chan->completed_cookie;
751 *done = last_complete;
755 ret = dma_async_is_complete(cookie, last_complete, last_used);
756 if (ret == DMA_SUCCESS)
759 iop_adma_slot_cleanup(iop_chan);
761 last_used = chan->cookie;
762 last_complete = iop_chan->completed_cookie;
765 *done = last_complete;
769 return dma_async_is_complete(cookie, last_complete, last_used);
772 static irqreturn_t iop_adma_eot_handler(int irq, void *data)
774 struct iop_adma_chan *chan = data;
776 dev_dbg(chan->device->common.dev, "%s\n", __func__);
778 tasklet_schedule(&chan->irq_tasklet);
780 iop_adma_device_clear_eot_status(chan);
785 static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
787 struct iop_adma_chan *chan = data;
789 dev_dbg(chan->device->common.dev, "%s\n", __func__);
791 tasklet_schedule(&chan->irq_tasklet);
793 iop_adma_device_clear_eoc_status(chan);
798 static irqreturn_t iop_adma_err_handler(int irq, void *data)
800 struct iop_adma_chan *chan = data;
801 unsigned long status = iop_chan_get_status(chan);
803 dev_printk(KERN_ERR, chan->device->common.dev,
804 "error ( %s%s%s%s%s%s%s)\n",
805 iop_is_err_int_parity(status, chan) ? "int_parity " : "",
806 iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
807 iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
808 iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
809 iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
810 iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
811 iop_is_err_split_tx(status, chan) ? "split_tx " : "");
813 iop_adma_device_clear_err_status(chan);
820 static void iop_adma_issue_pending(struct dma_chan *chan)
822 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
824 if (iop_chan->pending) {
825 iop_chan->pending = 0;
826 iop_chan_append(iop_chan);
831 * Perform a transaction to verify the HW works.
833 #define IOP_ADMA_TEST_SIZE 2000
835 static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device)
839 dma_addr_t src_dma, dest_dma;
840 struct dma_chan *dma_chan;
842 struct dma_async_tx_descriptor *tx;
844 struct iop_adma_chan *iop_chan;
846 dev_dbg(device->common.dev, "%s\n", __func__);
848 src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
851 dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
857 /* Fill in src buffer */
858 for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
859 ((u8 *) src)[i] = (u8)i;
861 /* Start copy, using first DMA channel */
862 dma_chan = container_of(device->common.channels.next,
865 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
870 dest_dma = dma_map_single(dma_chan->device->dev, dest,
871 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
872 src_dma = dma_map_single(dma_chan->device->dev, src,
873 IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
874 tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
876 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
878 cookie = iop_adma_tx_submit(tx);
879 iop_adma_issue_pending(dma_chan);
882 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
884 dev_printk(KERN_ERR, dma_chan->device->dev,
885 "Self-test copy timed out, disabling\n");
890 iop_chan = to_iop_adma_chan(dma_chan);
891 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
892 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
893 if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
894 dev_printk(KERN_ERR, dma_chan->device->dev,
895 "Self-test copy failed compare, disabling\n");
901 iop_adma_free_chan_resources(dma_chan);
908 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
910 iop_adma_xor_zero_sum_self_test(struct iop_adma_device *device)
914 struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
915 struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
916 dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
917 dma_addr_t dma_addr, dest_dma;
918 struct dma_async_tx_descriptor *tx;
919 struct dma_chan *dma_chan;
925 struct iop_adma_chan *iop_chan;
927 dev_dbg(device->common.dev, "%s\n", __func__);
929 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
930 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
931 if (!xor_srcs[src_idx])
933 __free_page(xor_srcs[src_idx]);
938 dest = alloc_page(GFP_KERNEL);
941 __free_page(xor_srcs[src_idx]);
945 /* Fill in src buffers */
946 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
947 u8 *ptr = page_address(xor_srcs[src_idx]);
948 for (i = 0; i < PAGE_SIZE; i++)
949 ptr[i] = (1 << src_idx);
952 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
953 cmp_byte ^= (u8) (1 << src_idx);
955 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
956 (cmp_byte << 8) | cmp_byte;
958 memset(page_address(dest), 0, PAGE_SIZE);
960 dma_chan = container_of(device->common.channels.next,
963 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
969 dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
970 PAGE_SIZE, DMA_FROM_DEVICE);
971 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
972 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
973 0, PAGE_SIZE, DMA_TO_DEVICE);
974 tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
975 IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE,
976 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
978 cookie = iop_adma_tx_submit(tx);
979 iop_adma_issue_pending(dma_chan);
982 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
984 dev_printk(KERN_ERR, dma_chan->device->dev,
985 "Self-test xor timed out, disabling\n");
990 iop_chan = to_iop_adma_chan(dma_chan);
991 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
992 PAGE_SIZE, DMA_FROM_DEVICE);
993 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
994 u32 *ptr = page_address(dest);
995 if (ptr[i] != cmp_word) {
996 dev_printk(KERN_ERR, dma_chan->device->dev,
997 "Self-test xor failed compare, disabling\n");
1002 dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
1003 PAGE_SIZE, DMA_TO_DEVICE);
1005 /* skip zero sum if the capability is not present */
1006 if (!dma_has_cap(DMA_ZERO_SUM, dma_chan->device->cap_mask))
1007 goto free_resources;
1009 /* zero sum the sources with the destintation page */
1010 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1011 zero_sum_srcs[i] = xor_srcs[i];
1012 zero_sum_srcs[i] = dest;
1014 zero_sum_result = 1;
1016 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1017 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1018 zero_sum_srcs[i], 0, PAGE_SIZE,
1020 tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs,
1021 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1023 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1025 cookie = iop_adma_tx_submit(tx);
1026 iop_adma_issue_pending(dma_chan);
1029 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1030 dev_printk(KERN_ERR, dma_chan->device->dev,
1031 "Self-test zero sum timed out, disabling\n");
1033 goto free_resources;
1036 if (zero_sum_result != 0) {
1037 dev_printk(KERN_ERR, dma_chan->device->dev,
1038 "Self-test zero sum failed compare, disabling\n");
1040 goto free_resources;
1044 dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
1045 PAGE_SIZE, DMA_FROM_DEVICE);
1046 tx = iop_adma_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE,
1047 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1049 cookie = iop_adma_tx_submit(tx);
1050 iop_adma_issue_pending(dma_chan);
1053 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1054 dev_printk(KERN_ERR, dma_chan->device->dev,
1055 "Self-test memset timed out, disabling\n");
1057 goto free_resources;
1060 for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1061 u32 *ptr = page_address(dest);
1063 dev_printk(KERN_ERR, dma_chan->device->dev,
1064 "Self-test memset failed compare, disabling\n");
1066 goto free_resources;
1070 /* test for non-zero parity sum */
1071 zero_sum_result = 0;
1072 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1073 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1074 zero_sum_srcs[i], 0, PAGE_SIZE,
1076 tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs,
1077 IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1079 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1081 cookie = iop_adma_tx_submit(tx);
1082 iop_adma_issue_pending(dma_chan);
1085 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1086 dev_printk(KERN_ERR, dma_chan->device->dev,
1087 "Self-test non-zero sum timed out, disabling\n");
1089 goto free_resources;
1092 if (zero_sum_result != 1) {
1093 dev_printk(KERN_ERR, dma_chan->device->dev,
1094 "Self-test non-zero sum failed compare, disabling\n");
1096 goto free_resources;
1100 iop_adma_free_chan_resources(dma_chan);
1102 src_idx = IOP_ADMA_NUM_SRC_TEST;
1104 __free_page(xor_srcs[src_idx]);
1109 static int __devexit iop_adma_remove(struct platform_device *dev)
1111 struct iop_adma_device *device = platform_get_drvdata(dev);
1112 struct dma_chan *chan, *_chan;
1113 struct iop_adma_chan *iop_chan;
1114 struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
1116 dma_async_device_unregister(&device->common);
1118 dma_free_coherent(&dev->dev, plat_data->pool_size,
1119 device->dma_desc_pool_virt, device->dma_desc_pool);
1121 list_for_each_entry_safe(chan, _chan, &device->common.channels,
1123 iop_chan = to_iop_adma_chan(chan);
1124 list_del(&chan->device_node);
1132 static int __devinit iop_adma_probe(struct platform_device *pdev)
1134 struct resource *res;
1136 struct iop_adma_device *adev;
1137 struct iop_adma_chan *iop_chan;
1138 struct dma_device *dma_dev;
1139 struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
1141 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1145 if (!devm_request_mem_region(&pdev->dev, res->start,
1146 res->end - res->start, pdev->name))
1149 adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1152 dma_dev = &adev->common;
1154 /* allocate coherent memory for hardware descriptors
1155 * note: writecombine gives slightly better performance, but
1156 * requires that we explicitly flush the writes
1158 if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1159 plat_data->pool_size,
1160 &adev->dma_desc_pool,
1161 GFP_KERNEL)) == NULL) {
1166 dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n",
1167 __func__, adev->dma_desc_pool_virt,
1168 (void *) adev->dma_desc_pool);
1170 adev->id = plat_data->hw_id;
1172 /* discover transaction capabilites from the platform data */
1173 dma_dev->cap_mask = plat_data->cap_mask;
1176 platform_set_drvdata(pdev, adev);
1178 INIT_LIST_HEAD(&dma_dev->channels);
1180 /* set base routines */
1181 dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1182 dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1183 dma_dev->device_is_tx_complete = iop_adma_is_complete;
1184 dma_dev->device_issue_pending = iop_adma_issue_pending;
1185 dma_dev->dev = &pdev->dev;
1187 /* set prep routines based on capability */
1188 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1189 dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1190 if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1191 dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
1192 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1193 dma_dev->max_xor = iop_adma_get_max_xor();
1194 dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1196 if (dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask))
1197 dma_dev->device_prep_dma_zero_sum =
1198 iop_adma_prep_dma_zero_sum;
1199 if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1200 dma_dev->device_prep_dma_interrupt =
1201 iop_adma_prep_dma_interrupt;
1203 iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1208 iop_chan->device = adev;
1210 iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1211 res->end - res->start);
1212 if (!iop_chan->mmr_base) {
1214 goto err_free_iop_chan;
1216 tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1219 /* clear errors before enabling interrupts */
1220 iop_adma_device_clear_err_status(iop_chan);
1222 for (i = 0; i < 3; i++) {
1223 irq_handler_t handler[] = { iop_adma_eot_handler,
1224 iop_adma_eoc_handler,
1225 iop_adma_err_handler };
1226 int irq = platform_get_irq(pdev, i);
1229 goto err_free_iop_chan;
1231 ret = devm_request_irq(&pdev->dev, irq,
1232 handler[i], 0, pdev->name, iop_chan);
1234 goto err_free_iop_chan;
1238 spin_lock_init(&iop_chan->lock);
1239 INIT_LIST_HEAD(&iop_chan->chain);
1240 INIT_LIST_HEAD(&iop_chan->all_slots);
1241 iop_chan->common.device = dma_dev;
1242 list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1244 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1245 ret = iop_adma_memcpy_self_test(adev);
1246 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1248 goto err_free_iop_chan;
1251 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
1252 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1253 ret = iop_adma_xor_zero_sum_self_test(adev);
1254 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1256 goto err_free_iop_chan;
1259 dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
1260 "( %s%s%s%s%s%s%s%s%s%s)\n",
1261 dma_has_cap(DMA_PQ_XOR, dma_dev->cap_mask) ? "pq_xor " : "",
1262 dma_has_cap(DMA_PQ_UPDATE, dma_dev->cap_mask) ? "pq_update " : "",
1263 dma_has_cap(DMA_PQ_ZERO_SUM, dma_dev->cap_mask) ? "pq_zero_sum " : "",
1264 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1265 dma_has_cap(DMA_DUAL_XOR, dma_dev->cap_mask) ? "dual_xor " : "",
1266 dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask) ? "xor_zero_sum " : "",
1267 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "",
1268 dma_has_cap(DMA_MEMCPY_CRC32C, dma_dev->cap_mask) ? "cpy+crc " : "",
1269 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1270 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1272 dma_async_device_register(dma_dev);
1278 dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1279 adev->dma_desc_pool_virt, adev->dma_desc_pool);
1286 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1288 struct iop_adma_desc_slot *sw_desc, *grp_start;
1289 dma_cookie_t cookie;
1290 int slot_cnt, slots_per_op;
1292 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1294 spin_lock_bh(&iop_chan->lock);
1295 slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1296 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1298 grp_start = sw_desc->group_head;
1300 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1301 async_tx_ack(&sw_desc->async_tx);
1302 iop_desc_init_memcpy(grp_start, 0);
1303 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1304 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1305 iop_desc_set_memcpy_src_addr(grp_start, 0);
1307 cookie = iop_chan->common.cookie;
1312 /* initialize the completed cookie to be less than
1313 * the most recently used cookie
1315 iop_chan->completed_cookie = cookie - 1;
1316 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1318 /* channel should not be busy */
1319 BUG_ON(iop_chan_is_busy(iop_chan));
1321 /* clear any prior error-status bits */
1322 iop_adma_device_clear_err_status(iop_chan);
1324 /* disable operation */
1325 iop_chan_disable(iop_chan);
1327 /* set the descriptor address */
1328 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1330 /* 1/ don't add pre-chained descriptors
1331 * 2/ dummy read to flush next_desc write
1333 BUG_ON(iop_desc_get_next_desc(sw_desc));
1335 /* run the descriptor */
1336 iop_chan_enable(iop_chan);
1338 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1339 "failed to allocate null descriptor\n");
1340 spin_unlock_bh(&iop_chan->lock);
1343 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1345 struct iop_adma_desc_slot *sw_desc, *grp_start;
1346 dma_cookie_t cookie;
1347 int slot_cnt, slots_per_op;
1349 dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1351 spin_lock_bh(&iop_chan->lock);
1352 slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1353 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1355 grp_start = sw_desc->group_head;
1356 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1357 async_tx_ack(&sw_desc->async_tx);
1358 iop_desc_init_null_xor(grp_start, 2, 0);
1359 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1360 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1361 iop_desc_set_xor_src_addr(grp_start, 0, 0);
1362 iop_desc_set_xor_src_addr(grp_start, 1, 0);
1364 cookie = iop_chan->common.cookie;
1369 /* initialize the completed cookie to be less than
1370 * the most recently used cookie
1372 iop_chan->completed_cookie = cookie - 1;
1373 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1375 /* channel should not be busy */
1376 BUG_ON(iop_chan_is_busy(iop_chan));
1378 /* clear any prior error-status bits */
1379 iop_adma_device_clear_err_status(iop_chan);
1381 /* disable operation */
1382 iop_chan_disable(iop_chan);
1384 /* set the descriptor address */
1385 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1387 /* 1/ don't add pre-chained descriptors
1388 * 2/ dummy read to flush next_desc write
1390 BUG_ON(iop_desc_get_next_desc(sw_desc));
1392 /* run the descriptor */
1393 iop_chan_enable(iop_chan);
1395 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1396 "failed to allocate null descriptor\n");
1397 spin_unlock_bh(&iop_chan->lock);
1400 MODULE_ALIAS("platform:iop-adma");
1402 static struct platform_driver iop_adma_driver = {
1403 .probe = iop_adma_probe,
1404 .remove = __devexit_p(iop_adma_remove),
1406 .owner = THIS_MODULE,
1411 static int __init iop_adma_init (void)
1413 return platform_driver_register(&iop_adma_driver);
1416 static void __exit iop_adma_exit (void)
1418 platform_driver_unregister(&iop_adma_driver);
1421 module_exit(iop_adma_exit);
1422 module_init(iop_adma_init);
1424 MODULE_AUTHOR("Intel Corporation");
1425 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1426 MODULE_LICENSE("GPL");