2 * offload engine driver for the Marvell XOR engine
3 * Copyright (C) 2007, 2008, Marvell International Ltd.
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.
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/delay.h>
22 #include <linux/dma-mapping.h>
23 #include <linux/spinlock.h>
24 #include <linux/interrupt.h>
25 #include <linux/platform_device.h>
26 #include <linux/memory.h>
27 #include <plat/mv_xor.h>
30 static void mv_xor_issue_pending(struct dma_chan *chan);
32 #define to_mv_xor_chan(chan) \
33 container_of(chan, struct mv_xor_chan, common)
35 #define to_mv_xor_device(dev) \
36 container_of(dev, struct mv_xor_device, common)
38 #define to_mv_xor_slot(tx) \
39 container_of(tx, struct mv_xor_desc_slot, async_tx)
41 static void mv_desc_init(struct mv_xor_desc_slot *desc, unsigned long flags)
43 struct mv_xor_desc *hw_desc = desc->hw_desc;
45 hw_desc->status = (1 << 31);
46 hw_desc->phy_next_desc = 0;
47 hw_desc->desc_command = (1 << 31);
50 static u32 mv_desc_get_dest_addr(struct mv_xor_desc_slot *desc)
52 struct mv_xor_desc *hw_desc = desc->hw_desc;
53 return hw_desc->phy_dest_addr;
56 static u32 mv_desc_get_src_addr(struct mv_xor_desc_slot *desc,
59 struct mv_xor_desc *hw_desc = desc->hw_desc;
60 return hw_desc->phy_src_addr[src_idx];
64 static void mv_desc_set_byte_count(struct mv_xor_desc_slot *desc,
67 struct mv_xor_desc *hw_desc = desc->hw_desc;
68 hw_desc->byte_count = byte_count;
71 static void mv_desc_set_next_desc(struct mv_xor_desc_slot *desc,
74 struct mv_xor_desc *hw_desc = desc->hw_desc;
75 BUG_ON(hw_desc->phy_next_desc);
76 hw_desc->phy_next_desc = next_desc_addr;
79 static void mv_desc_clear_next_desc(struct mv_xor_desc_slot *desc)
81 struct mv_xor_desc *hw_desc = desc->hw_desc;
82 hw_desc->phy_next_desc = 0;
85 static void mv_desc_set_block_fill_val(struct mv_xor_desc_slot *desc, u32 val)
90 static void mv_desc_set_dest_addr(struct mv_xor_desc_slot *desc,
93 struct mv_xor_desc *hw_desc = desc->hw_desc;
94 hw_desc->phy_dest_addr = addr;
97 static int mv_chan_memset_slot_count(size_t len)
102 #define mv_chan_memcpy_slot_count(c) mv_chan_memset_slot_count(c)
104 static void mv_desc_set_src_addr(struct mv_xor_desc_slot *desc,
105 int index, dma_addr_t addr)
107 struct mv_xor_desc *hw_desc = desc->hw_desc;
108 hw_desc->phy_src_addr[index] = addr;
109 if (desc->type == DMA_XOR)
110 hw_desc->desc_command |= (1 << index);
113 static u32 mv_chan_get_current_desc(struct mv_xor_chan *chan)
115 return __raw_readl(XOR_CURR_DESC(chan));
118 static void mv_chan_set_next_descriptor(struct mv_xor_chan *chan,
121 __raw_writel(next_desc_addr, XOR_NEXT_DESC(chan));
124 static void mv_chan_set_dest_pointer(struct mv_xor_chan *chan, u32 desc_addr)
126 __raw_writel(desc_addr, XOR_DEST_POINTER(chan));
129 static void mv_chan_set_block_size(struct mv_xor_chan *chan, u32 block_size)
131 __raw_writel(block_size, XOR_BLOCK_SIZE(chan));
134 static void mv_chan_set_value(struct mv_xor_chan *chan, u32 value)
136 __raw_writel(value, XOR_INIT_VALUE_LOW(chan));
137 __raw_writel(value, XOR_INIT_VALUE_HIGH(chan));
140 static void mv_chan_unmask_interrupts(struct mv_xor_chan *chan)
142 u32 val = __raw_readl(XOR_INTR_MASK(chan));
143 val |= XOR_INTR_MASK_VALUE << (chan->idx * 16);
144 __raw_writel(val, XOR_INTR_MASK(chan));
147 static u32 mv_chan_get_intr_cause(struct mv_xor_chan *chan)
149 u32 intr_cause = __raw_readl(XOR_INTR_CAUSE(chan));
150 intr_cause = (intr_cause >> (chan->idx * 16)) & 0xFFFF;
154 static int mv_is_err_intr(u32 intr_cause)
156 if (intr_cause & ((1<<4)|(1<<5)|(1<<6)|(1<<7)|(1<<8)|(1<<9)))
162 static void mv_xor_device_clear_eoc_cause(struct mv_xor_chan *chan)
164 u32 val = (1 << (1 + (chan->idx * 16)));
165 dev_dbg(chan->device->common.dev, "%s, val 0x%08x\n", __func__, val);
166 __raw_writel(val, XOR_INTR_CAUSE(chan));
169 static void mv_xor_device_clear_err_status(struct mv_xor_chan *chan)
171 u32 val = 0xFFFF0000 >> (chan->idx * 16);
172 __raw_writel(val, XOR_INTR_CAUSE(chan));
175 static int mv_can_chain(struct mv_xor_desc_slot *desc)
177 struct mv_xor_desc_slot *chain_old_tail = list_entry(
178 desc->chain_node.prev, struct mv_xor_desc_slot, chain_node);
180 if (chain_old_tail->type != desc->type)
182 if (desc->type == DMA_MEMSET)
188 static void mv_set_mode(struct mv_xor_chan *chan,
189 enum dma_transaction_type type)
192 u32 config = __raw_readl(XOR_CONFIG(chan));
196 op_mode = XOR_OPERATION_MODE_XOR;
199 op_mode = XOR_OPERATION_MODE_MEMCPY;
202 op_mode = XOR_OPERATION_MODE_MEMSET;
205 dev_printk(KERN_ERR, chan->device->common.dev,
206 "error: unsupported operation %d.\n",
214 __raw_writel(config, XOR_CONFIG(chan));
215 chan->current_type = type;
218 static void mv_chan_activate(struct mv_xor_chan *chan)
222 dev_dbg(chan->device->common.dev, " activate chan.\n");
223 activation = __raw_readl(XOR_ACTIVATION(chan));
225 __raw_writel(activation, XOR_ACTIVATION(chan));
228 static char mv_chan_is_busy(struct mv_xor_chan *chan)
230 u32 state = __raw_readl(XOR_ACTIVATION(chan));
232 state = (state >> 4) & 0x3;
234 return (state == 1) ? 1 : 0;
237 static int mv_chan_xor_slot_count(size_t len, int src_cnt)
243 * mv_xor_free_slots - flags descriptor slots for reuse
244 * @slot: Slot to free
245 * Caller must hold &mv_chan->lock while calling this function
247 static void mv_xor_free_slots(struct mv_xor_chan *mv_chan,
248 struct mv_xor_desc_slot *slot)
250 dev_dbg(mv_chan->device->common.dev, "%s %d slot %p\n",
251 __func__, __LINE__, slot);
253 slot->slots_per_op = 0;
258 * mv_xor_start_new_chain - program the engine to operate on new chain headed by
260 * Caller must hold &mv_chan->lock while calling this function
262 static void mv_xor_start_new_chain(struct mv_xor_chan *mv_chan,
263 struct mv_xor_desc_slot *sw_desc)
265 dev_dbg(mv_chan->device->common.dev, "%s %d: sw_desc %p\n",
266 __func__, __LINE__, sw_desc);
267 if (sw_desc->type != mv_chan->current_type)
268 mv_set_mode(mv_chan, sw_desc->type);
270 if (sw_desc->type == DMA_MEMSET) {
271 /* for memset requests we need to program the engine, no
274 struct mv_xor_desc *hw_desc = sw_desc->hw_desc;
275 mv_chan_set_dest_pointer(mv_chan, hw_desc->phy_dest_addr);
276 mv_chan_set_block_size(mv_chan, sw_desc->unmap_len);
277 mv_chan_set_value(mv_chan, sw_desc->value);
279 /* set the hardware chain */
280 mv_chan_set_next_descriptor(mv_chan, sw_desc->async_tx.phys);
282 mv_chan->pending += sw_desc->slot_cnt;
283 mv_xor_issue_pending(&mv_chan->common);
287 mv_xor_run_tx_complete_actions(struct mv_xor_desc_slot *desc,
288 struct mv_xor_chan *mv_chan, dma_cookie_t cookie)
290 BUG_ON(desc->async_tx.cookie < 0);
292 if (desc->async_tx.cookie > 0) {
293 cookie = desc->async_tx.cookie;
295 /* call the callback (must not sleep or submit new
296 * operations to this channel)
298 if (desc->async_tx.callback)
299 desc->async_tx.callback(
300 desc->async_tx.callback_param);
302 /* unmap dma addresses
303 * (unmap_single vs unmap_page?)
305 if (desc->group_head && desc->unmap_len) {
306 struct mv_xor_desc_slot *unmap = desc->group_head;
308 &mv_chan->device->pdev->dev;
309 u32 len = unmap->unmap_len;
310 enum dma_ctrl_flags flags = desc->async_tx.flags;
315 src_cnt = unmap->unmap_src_cnt;
316 dest = mv_desc_get_dest_addr(unmap);
317 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
318 enum dma_data_direction dir;
320 if (src_cnt > 1) /* is xor ? */
321 dir = DMA_BIDIRECTIONAL;
323 dir = DMA_FROM_DEVICE;
324 dma_unmap_page(dev, dest, len, dir);
327 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
329 addr = mv_desc_get_src_addr(unmap,
333 dma_unmap_page(dev, addr, len,
337 desc->group_head = NULL;
341 /* run dependent operations */
342 dma_run_dependencies(&desc->async_tx);
348 mv_xor_clean_completed_slots(struct mv_xor_chan *mv_chan)
350 struct mv_xor_desc_slot *iter, *_iter;
352 dev_dbg(mv_chan->device->common.dev, "%s %d\n", __func__, __LINE__);
353 list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
356 if (async_tx_test_ack(&iter->async_tx)) {
357 list_del(&iter->completed_node);
358 mv_xor_free_slots(mv_chan, iter);
365 mv_xor_clean_slot(struct mv_xor_desc_slot *desc,
366 struct mv_xor_chan *mv_chan)
368 dev_dbg(mv_chan->device->common.dev, "%s %d: desc %p flags %d\n",
369 __func__, __LINE__, desc, desc->async_tx.flags);
370 list_del(&desc->chain_node);
371 /* the client is allowed to attach dependent operations
374 if (!async_tx_test_ack(&desc->async_tx)) {
375 /* move this slot to the completed_slots */
376 list_add_tail(&desc->completed_node, &mv_chan->completed_slots);
380 mv_xor_free_slots(mv_chan, desc);
384 static void __mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
386 struct mv_xor_desc_slot *iter, *_iter;
387 dma_cookie_t cookie = 0;
388 int busy = mv_chan_is_busy(mv_chan);
389 u32 current_desc = mv_chan_get_current_desc(mv_chan);
390 int seen_current = 0;
392 dev_dbg(mv_chan->device->common.dev, "%s %d\n", __func__, __LINE__);
393 dev_dbg(mv_chan->device->common.dev, "current_desc %x\n", current_desc);
394 mv_xor_clean_completed_slots(mv_chan);
396 /* free completed slots from the chain starting with
397 * the oldest descriptor
400 list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
403 prefetch(&_iter->async_tx);
405 /* do not advance past the current descriptor loaded into the
406 * hardware channel, subsequent descriptors are either in
407 * process or have not been submitted
412 /* stop the search if we reach the current descriptor and the
415 if (iter->async_tx.phys == current_desc) {
421 cookie = mv_xor_run_tx_complete_actions(iter, mv_chan, cookie);
423 if (mv_xor_clean_slot(iter, mv_chan))
427 if ((busy == 0) && !list_empty(&mv_chan->chain)) {
428 struct mv_xor_desc_slot *chain_head;
429 chain_head = list_entry(mv_chan->chain.next,
430 struct mv_xor_desc_slot,
433 mv_xor_start_new_chain(mv_chan, chain_head);
437 mv_chan->completed_cookie = cookie;
441 mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
443 spin_lock_bh(&mv_chan->lock);
444 __mv_xor_slot_cleanup(mv_chan);
445 spin_unlock_bh(&mv_chan->lock);
448 static void mv_xor_tasklet(unsigned long data)
450 struct mv_xor_chan *chan = (struct mv_xor_chan *) data;
451 __mv_xor_slot_cleanup(chan);
454 static struct mv_xor_desc_slot *
455 mv_xor_alloc_slots(struct mv_xor_chan *mv_chan, int num_slots,
458 struct mv_xor_desc_slot *iter, *_iter, *alloc_start = NULL;
460 int slots_found, retry = 0;
462 /* start search from the last allocated descrtiptor
463 * if a contiguous allocation can not be found start searching
464 * from the beginning of the list
469 iter = mv_chan->last_used;
471 iter = list_entry(&mv_chan->all_slots,
472 struct mv_xor_desc_slot,
475 list_for_each_entry_safe_continue(
476 iter, _iter, &mv_chan->all_slots, slot_node) {
478 prefetch(&_iter->async_tx);
479 if (iter->slots_per_op) {
480 /* give up after finding the first busy slot
481 * on the second pass through the list
490 /* start the allocation if the slot is correctly aligned */
494 if (slots_found == num_slots) {
495 struct mv_xor_desc_slot *alloc_tail = NULL;
496 struct mv_xor_desc_slot *last_used = NULL;
501 /* pre-ack all but the last descriptor */
502 async_tx_ack(&iter->async_tx);
504 list_add_tail(&iter->chain_node, &chain);
506 iter->async_tx.cookie = 0;
507 iter->slot_cnt = num_slots;
508 iter->xor_check_result = NULL;
509 for (i = 0; i < slots_per_op; i++) {
510 iter->slots_per_op = slots_per_op - i;
512 iter = list_entry(iter->slot_node.next,
513 struct mv_xor_desc_slot,
516 num_slots -= slots_per_op;
518 alloc_tail->group_head = alloc_start;
519 alloc_tail->async_tx.cookie = -EBUSY;
520 list_splice(&chain, &alloc_tail->async_tx.tx_list);
521 mv_chan->last_used = last_used;
522 mv_desc_clear_next_desc(alloc_start);
523 mv_desc_clear_next_desc(alloc_tail);
530 /* try to free some slots if the allocation fails */
531 tasklet_schedule(&mv_chan->irq_tasklet);
537 mv_desc_assign_cookie(struct mv_xor_chan *mv_chan,
538 struct mv_xor_desc_slot *desc)
540 dma_cookie_t cookie = mv_chan->common.cookie;
544 mv_chan->common.cookie = desc->async_tx.cookie = cookie;
548 /************************ DMA engine API functions ****************************/
550 mv_xor_tx_submit(struct dma_async_tx_descriptor *tx)
552 struct mv_xor_desc_slot *sw_desc = to_mv_xor_slot(tx);
553 struct mv_xor_chan *mv_chan = to_mv_xor_chan(tx->chan);
554 struct mv_xor_desc_slot *grp_start, *old_chain_tail;
556 int new_hw_chain = 1;
558 dev_dbg(mv_chan->device->common.dev,
559 "%s sw_desc %p: async_tx %p\n",
560 __func__, sw_desc, &sw_desc->async_tx);
562 grp_start = sw_desc->group_head;
564 spin_lock_bh(&mv_chan->lock);
565 cookie = mv_desc_assign_cookie(mv_chan, sw_desc);
567 if (list_empty(&mv_chan->chain))
568 list_splice_init(&sw_desc->async_tx.tx_list, &mv_chan->chain);
572 old_chain_tail = list_entry(mv_chan->chain.prev,
573 struct mv_xor_desc_slot,
575 list_splice_init(&grp_start->async_tx.tx_list,
576 &old_chain_tail->chain_node);
578 if (!mv_can_chain(grp_start))
581 dev_dbg(mv_chan->device->common.dev, "Append to last desc %x\n",
582 old_chain_tail->async_tx.phys);
584 /* fix up the hardware chain */
585 mv_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
587 /* if the channel is not busy */
588 if (!mv_chan_is_busy(mv_chan)) {
589 u32 current_desc = mv_chan_get_current_desc(mv_chan);
591 * and the curren desc is the end of the chain before
592 * the append, then we need to start the channel
594 if (current_desc == old_chain_tail->async_tx.phys)
600 mv_xor_start_new_chain(mv_chan, grp_start);
603 spin_unlock_bh(&mv_chan->lock);
608 /* returns the number of allocated descriptors */
609 static int mv_xor_alloc_chan_resources(struct dma_chan *chan)
613 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
614 struct mv_xor_desc_slot *slot = NULL;
615 struct mv_xor_platform_data *plat_data =
616 mv_chan->device->pdev->dev.platform_data;
617 int num_descs_in_pool = plat_data->pool_size/MV_XOR_SLOT_SIZE;
619 /* Allocate descriptor slots */
620 idx = mv_chan->slots_allocated;
621 while (idx < num_descs_in_pool) {
622 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
624 printk(KERN_INFO "MV XOR Channel only initialized"
625 " %d descriptor slots", idx);
628 hw_desc = (char *) mv_chan->device->dma_desc_pool_virt;
629 slot->hw_desc = (void *) &hw_desc[idx * MV_XOR_SLOT_SIZE];
631 dma_async_tx_descriptor_init(&slot->async_tx, chan);
632 slot->async_tx.tx_submit = mv_xor_tx_submit;
633 INIT_LIST_HEAD(&slot->chain_node);
634 INIT_LIST_HEAD(&slot->slot_node);
635 hw_desc = (char *) mv_chan->device->dma_desc_pool;
636 slot->async_tx.phys =
637 (dma_addr_t) &hw_desc[idx * MV_XOR_SLOT_SIZE];
640 spin_lock_bh(&mv_chan->lock);
641 mv_chan->slots_allocated = idx;
642 list_add_tail(&slot->slot_node, &mv_chan->all_slots);
643 spin_unlock_bh(&mv_chan->lock);
646 if (mv_chan->slots_allocated && !mv_chan->last_used)
647 mv_chan->last_used = list_entry(mv_chan->all_slots.next,
648 struct mv_xor_desc_slot,
651 dev_dbg(mv_chan->device->common.dev,
652 "allocated %d descriptor slots last_used: %p\n",
653 mv_chan->slots_allocated, mv_chan->last_used);
655 return mv_chan->slots_allocated ? : -ENOMEM;
658 static struct dma_async_tx_descriptor *
659 mv_xor_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
660 size_t len, unsigned long flags)
662 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
663 struct mv_xor_desc_slot *sw_desc, *grp_start;
666 dev_dbg(mv_chan->device->common.dev,
667 "%s dest: %x src %x len: %u flags: %ld\n",
668 __func__, dest, src, len, flags);
669 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
672 BUG_ON(unlikely(len > MV_XOR_MAX_BYTE_COUNT));
674 spin_lock_bh(&mv_chan->lock);
675 slot_cnt = mv_chan_memcpy_slot_count(len);
676 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
678 sw_desc->type = DMA_MEMCPY;
679 sw_desc->async_tx.flags = flags;
680 grp_start = sw_desc->group_head;
681 mv_desc_init(grp_start, flags);
682 mv_desc_set_byte_count(grp_start, len);
683 mv_desc_set_dest_addr(sw_desc->group_head, dest);
684 mv_desc_set_src_addr(grp_start, 0, src);
685 sw_desc->unmap_src_cnt = 1;
686 sw_desc->unmap_len = len;
688 spin_unlock_bh(&mv_chan->lock);
690 dev_dbg(mv_chan->device->common.dev,
691 "%s sw_desc %p async_tx %p\n",
692 __func__, sw_desc, sw_desc ? &sw_desc->async_tx : 0);
694 return sw_desc ? &sw_desc->async_tx : NULL;
697 static struct dma_async_tx_descriptor *
698 mv_xor_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,
699 size_t len, unsigned long flags)
701 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
702 struct mv_xor_desc_slot *sw_desc, *grp_start;
705 dev_dbg(mv_chan->device->common.dev,
706 "%s dest: %x len: %u flags: %ld\n",
707 __func__, dest, len, flags);
708 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
711 BUG_ON(unlikely(len > MV_XOR_MAX_BYTE_COUNT));
713 spin_lock_bh(&mv_chan->lock);
714 slot_cnt = mv_chan_memset_slot_count(len);
715 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
717 sw_desc->type = DMA_MEMSET;
718 sw_desc->async_tx.flags = flags;
719 grp_start = sw_desc->group_head;
720 mv_desc_init(grp_start, flags);
721 mv_desc_set_byte_count(grp_start, len);
722 mv_desc_set_dest_addr(sw_desc->group_head, dest);
723 mv_desc_set_block_fill_val(grp_start, value);
724 sw_desc->unmap_src_cnt = 1;
725 sw_desc->unmap_len = len;
727 spin_unlock_bh(&mv_chan->lock);
728 dev_dbg(mv_chan->device->common.dev,
729 "%s sw_desc %p async_tx %p \n",
730 __func__, sw_desc, &sw_desc->async_tx);
731 return sw_desc ? &sw_desc->async_tx : NULL;
734 static struct dma_async_tx_descriptor *
735 mv_xor_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
736 unsigned int src_cnt, size_t len, unsigned long flags)
738 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
739 struct mv_xor_desc_slot *sw_desc, *grp_start;
742 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
745 BUG_ON(unlikely(len > MV_XOR_MAX_BYTE_COUNT));
747 dev_dbg(mv_chan->device->common.dev,
748 "%s src_cnt: %d len: dest %x %u flags: %ld\n",
749 __func__, src_cnt, len, dest, flags);
751 spin_lock_bh(&mv_chan->lock);
752 slot_cnt = mv_chan_xor_slot_count(len, src_cnt);
753 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
755 sw_desc->type = DMA_XOR;
756 sw_desc->async_tx.flags = flags;
757 grp_start = sw_desc->group_head;
758 mv_desc_init(grp_start, flags);
759 /* the byte count field is the same as in memcpy desc*/
760 mv_desc_set_byte_count(grp_start, len);
761 mv_desc_set_dest_addr(sw_desc->group_head, dest);
762 sw_desc->unmap_src_cnt = src_cnt;
763 sw_desc->unmap_len = len;
765 mv_desc_set_src_addr(grp_start, src_cnt, src[src_cnt]);
767 spin_unlock_bh(&mv_chan->lock);
768 dev_dbg(mv_chan->device->common.dev,
769 "%s sw_desc %p async_tx %p \n",
770 __func__, sw_desc, &sw_desc->async_tx);
771 return sw_desc ? &sw_desc->async_tx : NULL;
774 static void mv_xor_free_chan_resources(struct dma_chan *chan)
776 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
777 struct mv_xor_desc_slot *iter, *_iter;
778 int in_use_descs = 0;
780 mv_xor_slot_cleanup(mv_chan);
782 spin_lock_bh(&mv_chan->lock);
783 list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
786 list_del(&iter->chain_node);
788 list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
791 list_del(&iter->completed_node);
793 list_for_each_entry_safe_reverse(
794 iter, _iter, &mv_chan->all_slots, slot_node) {
795 list_del(&iter->slot_node);
797 mv_chan->slots_allocated--;
799 mv_chan->last_used = NULL;
801 dev_dbg(mv_chan->device->common.dev, "%s slots_allocated %d\n",
802 __func__, mv_chan->slots_allocated);
803 spin_unlock_bh(&mv_chan->lock);
806 dev_err(mv_chan->device->common.dev,
807 "freeing %d in use descriptors!\n", in_use_descs);
811 * mv_xor_is_complete - poll the status of an XOR transaction
812 * @chan: XOR channel handle
813 * @cookie: XOR transaction identifier
815 static enum dma_status mv_xor_is_complete(struct dma_chan *chan,
820 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
821 dma_cookie_t last_used;
822 dma_cookie_t last_complete;
825 last_used = chan->cookie;
826 last_complete = mv_chan->completed_cookie;
827 mv_chan->is_complete_cookie = cookie;
829 *done = last_complete;
833 ret = dma_async_is_complete(cookie, last_complete, last_used);
834 if (ret == DMA_SUCCESS) {
835 mv_xor_clean_completed_slots(mv_chan);
838 mv_xor_slot_cleanup(mv_chan);
840 last_used = chan->cookie;
841 last_complete = mv_chan->completed_cookie;
844 *done = last_complete;
848 return dma_async_is_complete(cookie, last_complete, last_used);
851 static void mv_dump_xor_regs(struct mv_xor_chan *chan)
855 val = __raw_readl(XOR_CONFIG(chan));
856 dev_printk(KERN_ERR, chan->device->common.dev,
857 "config 0x%08x.\n", val);
859 val = __raw_readl(XOR_ACTIVATION(chan));
860 dev_printk(KERN_ERR, chan->device->common.dev,
861 "activation 0x%08x.\n", val);
863 val = __raw_readl(XOR_INTR_CAUSE(chan));
864 dev_printk(KERN_ERR, chan->device->common.dev,
865 "intr cause 0x%08x.\n", val);
867 val = __raw_readl(XOR_INTR_MASK(chan));
868 dev_printk(KERN_ERR, chan->device->common.dev,
869 "intr mask 0x%08x.\n", val);
871 val = __raw_readl(XOR_ERROR_CAUSE(chan));
872 dev_printk(KERN_ERR, chan->device->common.dev,
873 "error cause 0x%08x.\n", val);
875 val = __raw_readl(XOR_ERROR_ADDR(chan));
876 dev_printk(KERN_ERR, chan->device->common.dev,
877 "error addr 0x%08x.\n", val);
880 static void mv_xor_err_interrupt_handler(struct mv_xor_chan *chan,
883 if (intr_cause & (1 << 4)) {
884 dev_dbg(chan->device->common.dev,
885 "ignore this error\n");
889 dev_printk(KERN_ERR, chan->device->common.dev,
890 "error on chan %d. intr cause 0x%08x.\n",
891 chan->idx, intr_cause);
893 mv_dump_xor_regs(chan);
897 static irqreturn_t mv_xor_interrupt_handler(int irq, void *data)
899 struct mv_xor_chan *chan = data;
900 u32 intr_cause = mv_chan_get_intr_cause(chan);
902 dev_dbg(chan->device->common.dev, "intr cause %x\n", intr_cause);
904 if (mv_is_err_intr(intr_cause))
905 mv_xor_err_interrupt_handler(chan, intr_cause);
907 tasklet_schedule(&chan->irq_tasklet);
909 mv_xor_device_clear_eoc_cause(chan);
914 static void mv_xor_issue_pending(struct dma_chan *chan)
916 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
918 if (mv_chan->pending >= MV_XOR_THRESHOLD) {
919 mv_chan->pending = 0;
920 mv_chan_activate(mv_chan);
925 * Perform a transaction to verify the HW works.
927 #define MV_XOR_TEST_SIZE 2000
929 static int __devinit mv_xor_memcpy_self_test(struct mv_xor_device *device)
933 dma_addr_t src_dma, dest_dma;
934 struct dma_chan *dma_chan;
936 struct dma_async_tx_descriptor *tx;
938 struct mv_xor_chan *mv_chan;
940 src = kmalloc(sizeof(u8) * MV_XOR_TEST_SIZE, GFP_KERNEL);
944 dest = kzalloc(sizeof(u8) * MV_XOR_TEST_SIZE, GFP_KERNEL);
950 /* Fill in src buffer */
951 for (i = 0; i < MV_XOR_TEST_SIZE; i++)
952 ((u8 *) src)[i] = (u8)i;
954 /* Start copy, using first DMA channel */
955 dma_chan = container_of(device->common.channels.next,
958 if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
963 dest_dma = dma_map_single(dma_chan->device->dev, dest,
964 MV_XOR_TEST_SIZE, DMA_FROM_DEVICE);
966 src_dma = dma_map_single(dma_chan->device->dev, src,
967 MV_XOR_TEST_SIZE, DMA_TO_DEVICE);
969 tx = mv_xor_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
970 MV_XOR_TEST_SIZE, 0);
971 cookie = mv_xor_tx_submit(tx);
972 mv_xor_issue_pending(dma_chan);
976 if (mv_xor_is_complete(dma_chan, cookie, NULL, NULL) !=
978 dev_printk(KERN_ERR, dma_chan->device->dev,
979 "Self-test copy timed out, disabling\n");
984 mv_chan = to_mv_xor_chan(dma_chan);
985 dma_sync_single_for_cpu(&mv_chan->device->pdev->dev, dest_dma,
986 MV_XOR_TEST_SIZE, DMA_FROM_DEVICE);
987 if (memcmp(src, dest, MV_XOR_TEST_SIZE)) {
988 dev_printk(KERN_ERR, dma_chan->device->dev,
989 "Self-test copy failed compare, disabling\n");
995 mv_xor_free_chan_resources(dma_chan);
1002 #define MV_XOR_NUM_SRC_TEST 4 /* must be <= 15 */
1003 static int __devinit
1004 mv_xor_xor_self_test(struct mv_xor_device *device)
1008 struct page *xor_srcs[MV_XOR_NUM_SRC_TEST];
1009 dma_addr_t dma_srcs[MV_XOR_NUM_SRC_TEST];
1010 dma_addr_t dest_dma;
1011 struct dma_async_tx_descriptor *tx;
1012 struct dma_chan *dma_chan;
1013 dma_cookie_t cookie;
1017 struct mv_xor_chan *mv_chan;
1019 for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++) {
1020 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
1021 if (!xor_srcs[src_idx]) {
1023 __free_page(xor_srcs[src_idx]);
1028 dest = alloc_page(GFP_KERNEL);
1031 __free_page(xor_srcs[src_idx]);
1035 /* Fill in src buffers */
1036 for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++) {
1037 u8 *ptr = page_address(xor_srcs[src_idx]);
1038 for (i = 0; i < PAGE_SIZE; i++)
1039 ptr[i] = (1 << src_idx);
1042 for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++)
1043 cmp_byte ^= (u8) (1 << src_idx);
1045 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
1046 (cmp_byte << 8) | cmp_byte;
1048 memset(page_address(dest), 0, PAGE_SIZE);
1050 dma_chan = container_of(device->common.channels.next,
1053 if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
1059 dest_dma = dma_map_page(dma_chan->device->dev, dest, 0, PAGE_SIZE,
1062 for (i = 0; i < MV_XOR_NUM_SRC_TEST; i++)
1063 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
1064 0, PAGE_SIZE, DMA_TO_DEVICE);
1066 tx = mv_xor_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
1067 MV_XOR_NUM_SRC_TEST, PAGE_SIZE, 0);
1069 cookie = mv_xor_tx_submit(tx);
1070 mv_xor_issue_pending(dma_chan);
1074 if (mv_xor_is_complete(dma_chan, cookie, NULL, NULL) !=
1076 dev_printk(KERN_ERR, dma_chan->device->dev,
1077 "Self-test xor timed out, disabling\n");
1079 goto free_resources;
1082 mv_chan = to_mv_xor_chan(dma_chan);
1083 dma_sync_single_for_cpu(&mv_chan->device->pdev->dev, dest_dma,
1084 PAGE_SIZE, DMA_FROM_DEVICE);
1085 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1086 u32 *ptr = page_address(dest);
1087 if (ptr[i] != cmp_word) {
1088 dev_printk(KERN_ERR, dma_chan->device->dev,
1089 "Self-test xor failed compare, disabling."
1090 " index %d, data %x, expected %x\n", i,
1093 goto free_resources;
1098 mv_xor_free_chan_resources(dma_chan);
1100 src_idx = MV_XOR_NUM_SRC_TEST;
1102 __free_page(xor_srcs[src_idx]);
1107 static int __devexit mv_xor_remove(struct platform_device *dev)
1109 struct mv_xor_device *device = platform_get_drvdata(dev);
1110 struct dma_chan *chan, *_chan;
1111 struct mv_xor_chan *mv_chan;
1112 struct mv_xor_platform_data *plat_data = dev->dev.platform_data;
1114 dma_async_device_unregister(&device->common);
1116 dma_free_coherent(&dev->dev, plat_data->pool_size,
1117 device->dma_desc_pool_virt, device->dma_desc_pool);
1119 list_for_each_entry_safe(chan, _chan, &device->common.channels,
1121 mv_chan = to_mv_xor_chan(chan);
1122 list_del(&chan->device_node);
1128 static int __devinit mv_xor_probe(struct platform_device *pdev)
1132 struct mv_xor_device *adev;
1133 struct mv_xor_chan *mv_chan;
1134 struct dma_device *dma_dev;
1135 struct mv_xor_platform_data *plat_data = pdev->dev.platform_data;
1138 adev = devm_kzalloc(&pdev->dev, sizeof(*adev), GFP_KERNEL);
1142 dma_dev = &adev->common;
1144 /* allocate coherent memory for hardware descriptors
1145 * note: writecombine gives slightly better performance, but
1146 * requires that we explicitly flush the writes
1148 adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1149 plat_data->pool_size,
1150 &adev->dma_desc_pool,
1152 if (!adev->dma_desc_pool_virt)
1155 adev->id = plat_data->hw_id;
1157 /* discover transaction capabilites from the platform data */
1158 dma_dev->cap_mask = plat_data->cap_mask;
1160 platform_set_drvdata(pdev, adev);
1162 adev->shared = platform_get_drvdata(plat_data->shared);
1164 INIT_LIST_HEAD(&dma_dev->channels);
1166 /* set base routines */
1167 dma_dev->device_alloc_chan_resources = mv_xor_alloc_chan_resources;
1168 dma_dev->device_free_chan_resources = mv_xor_free_chan_resources;
1169 dma_dev->device_is_tx_complete = mv_xor_is_complete;
1170 dma_dev->device_issue_pending = mv_xor_issue_pending;
1171 dma_dev->dev = &pdev->dev;
1173 /* set prep routines based on capability */
1174 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1175 dma_dev->device_prep_dma_memcpy = mv_xor_prep_dma_memcpy;
1176 if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1177 dma_dev->device_prep_dma_memset = mv_xor_prep_dma_memset;
1178 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1179 dma_dev->max_xor = 8; ;
1180 dma_dev->device_prep_dma_xor = mv_xor_prep_dma_xor;
1183 mv_chan = devm_kzalloc(&pdev->dev, sizeof(*mv_chan), GFP_KERNEL);
1188 mv_chan->device = adev;
1189 mv_chan->idx = plat_data->hw_id;
1190 mv_chan->mmr_base = adev->shared->xor_base;
1192 if (!mv_chan->mmr_base) {
1196 tasklet_init(&mv_chan->irq_tasklet, mv_xor_tasklet, (unsigned long)
1199 /* clear errors before enabling interrupts */
1200 mv_xor_device_clear_err_status(mv_chan);
1202 irq = platform_get_irq(pdev, 0);
1207 ret = devm_request_irq(&pdev->dev, irq,
1208 mv_xor_interrupt_handler,
1209 0, dev_name(&pdev->dev), mv_chan);
1213 mv_chan_unmask_interrupts(mv_chan);
1215 mv_set_mode(mv_chan, DMA_MEMCPY);
1217 spin_lock_init(&mv_chan->lock);
1218 INIT_LIST_HEAD(&mv_chan->chain);
1219 INIT_LIST_HEAD(&mv_chan->completed_slots);
1220 INIT_LIST_HEAD(&mv_chan->all_slots);
1221 mv_chan->common.device = dma_dev;
1223 list_add_tail(&mv_chan->common.device_node, &dma_dev->channels);
1225 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1226 ret = mv_xor_memcpy_self_test(adev);
1227 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1232 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1233 ret = mv_xor_xor_self_test(adev);
1234 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1239 dev_printk(KERN_INFO, &pdev->dev, "Marvell XOR: "
1241 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1242 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "",
1243 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1244 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1246 dma_async_device_register(dma_dev);
1250 dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1251 adev->dma_desc_pool_virt, adev->dma_desc_pool);
1257 mv_xor_conf_mbus_windows(struct mv_xor_shared_private *msp,
1258 struct mbus_dram_target_info *dram)
1260 void __iomem *base = msp->xor_base;
1264 for (i = 0; i < 8; i++) {
1265 writel(0, base + WINDOW_BASE(i));
1266 writel(0, base + WINDOW_SIZE(i));
1268 writel(0, base + WINDOW_REMAP_HIGH(i));
1271 for (i = 0; i < dram->num_cs; i++) {
1272 struct mbus_dram_window *cs = dram->cs + i;
1274 writel((cs->base & 0xffff0000) |
1275 (cs->mbus_attr << 8) |
1276 dram->mbus_dram_target_id, base + WINDOW_BASE(i));
1277 writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
1279 win_enable |= (1 << i);
1280 win_enable |= 3 << (16 + (2 * i));
1283 writel(win_enable, base + WINDOW_BAR_ENABLE(0));
1284 writel(win_enable, base + WINDOW_BAR_ENABLE(1));
1287 static struct platform_driver mv_xor_driver = {
1288 .probe = mv_xor_probe,
1289 .remove = __devexit_p(mv_xor_remove),
1291 .owner = THIS_MODULE,
1292 .name = MV_XOR_NAME,
1296 static int mv_xor_shared_probe(struct platform_device *pdev)
1298 struct mv_xor_platform_shared_data *msd = pdev->dev.platform_data;
1299 struct mv_xor_shared_private *msp;
1300 struct resource *res;
1302 dev_printk(KERN_NOTICE, &pdev->dev, "Marvell shared XOR driver\n");
1304 msp = devm_kzalloc(&pdev->dev, sizeof(*msp), GFP_KERNEL);
1308 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1312 msp->xor_base = devm_ioremap(&pdev->dev, res->start,
1313 res->end - res->start + 1);
1317 res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1321 msp->xor_high_base = devm_ioremap(&pdev->dev, res->start,
1322 res->end - res->start + 1);
1323 if (!msp->xor_high_base)
1326 platform_set_drvdata(pdev, msp);
1329 * (Re-)program MBUS remapping windows if we are asked to.
1331 if (msd != NULL && msd->dram != NULL)
1332 mv_xor_conf_mbus_windows(msp, msd->dram);
1337 static int mv_xor_shared_remove(struct platform_device *pdev)
1342 static struct platform_driver mv_xor_shared_driver = {
1343 .probe = mv_xor_shared_probe,
1344 .remove = mv_xor_shared_remove,
1346 .owner = THIS_MODULE,
1347 .name = MV_XOR_SHARED_NAME,
1352 static int __init mv_xor_init(void)
1356 rc = platform_driver_register(&mv_xor_shared_driver);
1358 rc = platform_driver_register(&mv_xor_driver);
1360 platform_driver_unregister(&mv_xor_shared_driver);
1364 module_init(mv_xor_init);
1366 /* it's currently unsafe to unload this module */
1368 static void __exit mv_xor_exit(void)
1370 platform_driver_unregister(&mv_xor_driver);
1371 platform_driver_unregister(&mv_xor_shared_driver);
1375 module_exit(mv_xor_exit);
1378 MODULE_AUTHOR("Saeed Bishara <saeed@marvell.com>");
1379 MODULE_DESCRIPTION("DMA engine driver for Marvell's XOR engine");
1380 MODULE_LICENSE("GPL");