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/slab.h>
22 #include <linux/delay.h>
23 #include <linux/dma-mapping.h>
24 #include <linux/spinlock.h>
25 #include <linux/interrupt.h>
26 #include <linux/platform_device.h>
27 #include <linux/memory.h>
28 #include <plat/mv_xor.h>
31 static void mv_xor_issue_pending(struct dma_chan *chan);
33 #define to_mv_xor_chan(chan) \
34 container_of(chan, struct mv_xor_chan, common)
36 #define to_mv_xor_device(dev) \
37 container_of(dev, struct mv_xor_device, common)
39 #define to_mv_xor_slot(tx) \
40 container_of(tx, struct mv_xor_desc_slot, async_tx)
42 static void mv_desc_init(struct mv_xor_desc_slot *desc, unsigned long flags)
44 struct mv_xor_desc *hw_desc = desc->hw_desc;
46 hw_desc->status = (1 << 31);
47 hw_desc->phy_next_desc = 0;
48 hw_desc->desc_command = (1 << 31);
51 static u32 mv_desc_get_dest_addr(struct mv_xor_desc_slot *desc)
53 struct mv_xor_desc *hw_desc = desc->hw_desc;
54 return hw_desc->phy_dest_addr;
57 static u32 mv_desc_get_src_addr(struct mv_xor_desc_slot *desc,
60 struct mv_xor_desc *hw_desc = desc->hw_desc;
61 return hw_desc->phy_src_addr[src_idx];
65 static void mv_desc_set_byte_count(struct mv_xor_desc_slot *desc,
68 struct mv_xor_desc *hw_desc = desc->hw_desc;
69 hw_desc->byte_count = byte_count;
72 static void mv_desc_set_next_desc(struct mv_xor_desc_slot *desc,
75 struct mv_xor_desc *hw_desc = desc->hw_desc;
76 BUG_ON(hw_desc->phy_next_desc);
77 hw_desc->phy_next_desc = next_desc_addr;
80 static void mv_desc_clear_next_desc(struct mv_xor_desc_slot *desc)
82 struct mv_xor_desc *hw_desc = desc->hw_desc;
83 hw_desc->phy_next_desc = 0;
86 static void mv_desc_set_block_fill_val(struct mv_xor_desc_slot *desc, u32 val)
91 static void mv_desc_set_dest_addr(struct mv_xor_desc_slot *desc,
94 struct mv_xor_desc *hw_desc = desc->hw_desc;
95 hw_desc->phy_dest_addr = addr;
98 static int mv_chan_memset_slot_count(size_t len)
103 #define mv_chan_memcpy_slot_count(c) mv_chan_memset_slot_count(c)
105 static void mv_desc_set_src_addr(struct mv_xor_desc_slot *desc,
106 int index, dma_addr_t addr)
108 struct mv_xor_desc *hw_desc = desc->hw_desc;
109 hw_desc->phy_src_addr[index] = addr;
110 if (desc->type == DMA_XOR)
111 hw_desc->desc_command |= (1 << index);
114 static u32 mv_chan_get_current_desc(struct mv_xor_chan *chan)
116 return __raw_readl(XOR_CURR_DESC(chan));
119 static void mv_chan_set_next_descriptor(struct mv_xor_chan *chan,
122 __raw_writel(next_desc_addr, XOR_NEXT_DESC(chan));
125 static void mv_chan_set_dest_pointer(struct mv_xor_chan *chan, u32 desc_addr)
127 __raw_writel(desc_addr, XOR_DEST_POINTER(chan));
130 static void mv_chan_set_block_size(struct mv_xor_chan *chan, u32 block_size)
132 __raw_writel(block_size, XOR_BLOCK_SIZE(chan));
135 static void mv_chan_set_value(struct mv_xor_chan *chan, u32 value)
137 __raw_writel(value, XOR_INIT_VALUE_LOW(chan));
138 __raw_writel(value, XOR_INIT_VALUE_HIGH(chan));
141 static void mv_chan_unmask_interrupts(struct mv_xor_chan *chan)
143 u32 val = __raw_readl(XOR_INTR_MASK(chan));
144 val |= XOR_INTR_MASK_VALUE << (chan->idx * 16);
145 __raw_writel(val, XOR_INTR_MASK(chan));
148 static u32 mv_chan_get_intr_cause(struct mv_xor_chan *chan)
150 u32 intr_cause = __raw_readl(XOR_INTR_CAUSE(chan));
151 intr_cause = (intr_cause >> (chan->idx * 16)) & 0xFFFF;
155 static int mv_is_err_intr(u32 intr_cause)
157 if (intr_cause & ((1<<4)|(1<<5)|(1<<6)|(1<<7)|(1<<8)|(1<<9)))
163 static void mv_xor_device_clear_eoc_cause(struct mv_xor_chan *chan)
165 u32 val = (1 << (1 + (chan->idx * 16)));
166 dev_dbg(chan->device->common.dev, "%s, val 0x%08x\n", __func__, val);
167 __raw_writel(val, XOR_INTR_CAUSE(chan));
170 static void mv_xor_device_clear_err_status(struct mv_xor_chan *chan)
172 u32 val = 0xFFFF0000 >> (chan->idx * 16);
173 __raw_writel(val, XOR_INTR_CAUSE(chan));
176 static int mv_can_chain(struct mv_xor_desc_slot *desc)
178 struct mv_xor_desc_slot *chain_old_tail = list_entry(
179 desc->chain_node.prev, struct mv_xor_desc_slot, chain_node);
181 if (chain_old_tail->type != desc->type)
183 if (desc->type == DMA_MEMSET)
189 static void mv_set_mode(struct mv_xor_chan *chan,
190 enum dma_transaction_type type)
193 u32 config = __raw_readl(XOR_CONFIG(chan));
197 op_mode = XOR_OPERATION_MODE_XOR;
200 op_mode = XOR_OPERATION_MODE_MEMCPY;
203 op_mode = XOR_OPERATION_MODE_MEMSET;
206 dev_printk(KERN_ERR, chan->device->common.dev,
207 "error: unsupported operation %d.\n",
215 __raw_writel(config, XOR_CONFIG(chan));
216 chan->current_type = type;
219 static void mv_chan_activate(struct mv_xor_chan *chan)
223 dev_dbg(chan->device->common.dev, " activate chan.\n");
224 activation = __raw_readl(XOR_ACTIVATION(chan));
226 __raw_writel(activation, XOR_ACTIVATION(chan));
229 static char mv_chan_is_busy(struct mv_xor_chan *chan)
231 u32 state = __raw_readl(XOR_ACTIVATION(chan));
233 state = (state >> 4) & 0x3;
235 return (state == 1) ? 1 : 0;
238 static int mv_chan_xor_slot_count(size_t len, int src_cnt)
244 * mv_xor_free_slots - flags descriptor slots for reuse
245 * @slot: Slot to free
246 * Caller must hold &mv_chan->lock while calling this function
248 static void mv_xor_free_slots(struct mv_xor_chan *mv_chan,
249 struct mv_xor_desc_slot *slot)
251 dev_dbg(mv_chan->device->common.dev, "%s %d slot %p\n",
252 __func__, __LINE__, slot);
254 slot->slots_per_op = 0;
259 * mv_xor_start_new_chain - program the engine to operate on new chain headed by
261 * Caller must hold &mv_chan->lock while calling this function
263 static void mv_xor_start_new_chain(struct mv_xor_chan *mv_chan,
264 struct mv_xor_desc_slot *sw_desc)
266 dev_dbg(mv_chan->device->common.dev, "%s %d: sw_desc %p\n",
267 __func__, __LINE__, sw_desc);
268 if (sw_desc->type != mv_chan->current_type)
269 mv_set_mode(mv_chan, sw_desc->type);
271 if (sw_desc->type == DMA_MEMSET) {
272 /* for memset requests we need to program the engine, no
275 struct mv_xor_desc *hw_desc = sw_desc->hw_desc;
276 mv_chan_set_dest_pointer(mv_chan, hw_desc->phy_dest_addr);
277 mv_chan_set_block_size(mv_chan, sw_desc->unmap_len);
278 mv_chan_set_value(mv_chan, sw_desc->value);
280 /* set the hardware chain */
281 mv_chan_set_next_descriptor(mv_chan, sw_desc->async_tx.phys);
283 mv_chan->pending += sw_desc->slot_cnt;
284 mv_xor_issue_pending(&mv_chan->common);
288 mv_xor_run_tx_complete_actions(struct mv_xor_desc_slot *desc,
289 struct mv_xor_chan *mv_chan, dma_cookie_t cookie)
291 BUG_ON(desc->async_tx.cookie < 0);
293 if (desc->async_tx.cookie > 0) {
294 cookie = desc->async_tx.cookie;
296 /* call the callback (must not sleep or submit new
297 * operations to this channel)
299 if (desc->async_tx.callback)
300 desc->async_tx.callback(
301 desc->async_tx.callback_param);
303 /* unmap dma addresses
304 * (unmap_single vs unmap_page?)
306 if (desc->group_head && desc->unmap_len) {
307 struct mv_xor_desc_slot *unmap = desc->group_head;
309 &mv_chan->device->pdev->dev;
310 u32 len = unmap->unmap_len;
311 enum dma_ctrl_flags flags = desc->async_tx.flags;
316 src_cnt = unmap->unmap_src_cnt;
317 dest = mv_desc_get_dest_addr(unmap);
318 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
319 enum dma_data_direction dir;
321 if (src_cnt > 1) /* is xor ? */
322 dir = DMA_BIDIRECTIONAL;
324 dir = DMA_FROM_DEVICE;
325 dma_unmap_page(dev, dest, len, dir);
328 if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
330 addr = mv_desc_get_src_addr(unmap,
334 dma_unmap_page(dev, addr, len,
338 desc->group_head = NULL;
342 /* run dependent operations */
343 dma_run_dependencies(&desc->async_tx);
349 mv_xor_clean_completed_slots(struct mv_xor_chan *mv_chan)
351 struct mv_xor_desc_slot *iter, *_iter;
353 dev_dbg(mv_chan->device->common.dev, "%s %d\n", __func__, __LINE__);
354 list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
357 if (async_tx_test_ack(&iter->async_tx)) {
358 list_del(&iter->completed_node);
359 mv_xor_free_slots(mv_chan, iter);
366 mv_xor_clean_slot(struct mv_xor_desc_slot *desc,
367 struct mv_xor_chan *mv_chan)
369 dev_dbg(mv_chan->device->common.dev, "%s %d: desc %p flags %d\n",
370 __func__, __LINE__, desc, desc->async_tx.flags);
371 list_del(&desc->chain_node);
372 /* the client is allowed to attach dependent operations
375 if (!async_tx_test_ack(&desc->async_tx)) {
376 /* move this slot to the completed_slots */
377 list_add_tail(&desc->completed_node, &mv_chan->completed_slots);
381 mv_xor_free_slots(mv_chan, desc);
385 static void __mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
387 struct mv_xor_desc_slot *iter, *_iter;
388 dma_cookie_t cookie = 0;
389 int busy = mv_chan_is_busy(mv_chan);
390 u32 current_desc = mv_chan_get_current_desc(mv_chan);
391 int seen_current = 0;
393 dev_dbg(mv_chan->device->common.dev, "%s %d\n", __func__, __LINE__);
394 dev_dbg(mv_chan->device->common.dev, "current_desc %x\n", current_desc);
395 mv_xor_clean_completed_slots(mv_chan);
397 /* free completed slots from the chain starting with
398 * the oldest descriptor
401 list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
404 prefetch(&_iter->async_tx);
406 /* do not advance past the current descriptor loaded into the
407 * hardware channel, subsequent descriptors are either in
408 * process or have not been submitted
413 /* stop the search if we reach the current descriptor and the
416 if (iter->async_tx.phys == current_desc) {
422 cookie = mv_xor_run_tx_complete_actions(iter, mv_chan, cookie);
424 if (mv_xor_clean_slot(iter, mv_chan))
428 if ((busy == 0) && !list_empty(&mv_chan->chain)) {
429 struct mv_xor_desc_slot *chain_head;
430 chain_head = list_entry(mv_chan->chain.next,
431 struct mv_xor_desc_slot,
434 mv_xor_start_new_chain(mv_chan, chain_head);
438 mv_chan->completed_cookie = cookie;
442 mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
444 spin_lock_bh(&mv_chan->lock);
445 __mv_xor_slot_cleanup(mv_chan);
446 spin_unlock_bh(&mv_chan->lock);
449 static void mv_xor_tasklet(unsigned long data)
451 struct mv_xor_chan *chan = (struct mv_xor_chan *) data;
452 __mv_xor_slot_cleanup(chan);
455 static struct mv_xor_desc_slot *
456 mv_xor_alloc_slots(struct mv_xor_chan *mv_chan, int num_slots,
459 struct mv_xor_desc_slot *iter, *_iter, *alloc_start = NULL;
461 int slots_found, retry = 0;
463 /* start search from the last allocated descrtiptor
464 * if a contiguous allocation can not be found start searching
465 * from the beginning of the list
470 iter = mv_chan->last_used;
472 iter = list_entry(&mv_chan->all_slots,
473 struct mv_xor_desc_slot,
476 list_for_each_entry_safe_continue(
477 iter, _iter, &mv_chan->all_slots, slot_node) {
479 prefetch(&_iter->async_tx);
480 if (iter->slots_per_op) {
481 /* give up after finding the first busy slot
482 * on the second pass through the list
491 /* start the allocation if the slot is correctly aligned */
495 if (slots_found == num_slots) {
496 struct mv_xor_desc_slot *alloc_tail = NULL;
497 struct mv_xor_desc_slot *last_used = NULL;
502 /* pre-ack all but the last descriptor */
503 async_tx_ack(&iter->async_tx);
505 list_add_tail(&iter->chain_node, &chain);
507 iter->async_tx.cookie = 0;
508 iter->slot_cnt = num_slots;
509 iter->xor_check_result = NULL;
510 for (i = 0; i < slots_per_op; i++) {
511 iter->slots_per_op = slots_per_op - i;
513 iter = list_entry(iter->slot_node.next,
514 struct mv_xor_desc_slot,
517 num_slots -= slots_per_op;
519 alloc_tail->group_head = alloc_start;
520 alloc_tail->async_tx.cookie = -EBUSY;
521 list_splice(&chain, &alloc_tail->tx_list);
522 mv_chan->last_used = last_used;
523 mv_desc_clear_next_desc(alloc_start);
524 mv_desc_clear_next_desc(alloc_tail);
531 /* try to free some slots if the allocation fails */
532 tasklet_schedule(&mv_chan->irq_tasklet);
538 mv_desc_assign_cookie(struct mv_xor_chan *mv_chan,
539 struct mv_xor_desc_slot *desc)
541 dma_cookie_t cookie = mv_chan->common.cookie;
545 mv_chan->common.cookie = desc->async_tx.cookie = cookie;
549 /************************ DMA engine API functions ****************************/
551 mv_xor_tx_submit(struct dma_async_tx_descriptor *tx)
553 struct mv_xor_desc_slot *sw_desc = to_mv_xor_slot(tx);
554 struct mv_xor_chan *mv_chan = to_mv_xor_chan(tx->chan);
555 struct mv_xor_desc_slot *grp_start, *old_chain_tail;
557 int new_hw_chain = 1;
559 dev_dbg(mv_chan->device->common.dev,
560 "%s sw_desc %p: async_tx %p\n",
561 __func__, sw_desc, &sw_desc->async_tx);
563 grp_start = sw_desc->group_head;
565 spin_lock_bh(&mv_chan->lock);
566 cookie = mv_desc_assign_cookie(mv_chan, sw_desc);
568 if (list_empty(&mv_chan->chain))
569 list_splice_init(&sw_desc->tx_list, &mv_chan->chain);
573 old_chain_tail = list_entry(mv_chan->chain.prev,
574 struct mv_xor_desc_slot,
576 list_splice_init(&grp_start->tx_list,
577 &old_chain_tail->chain_node);
579 if (!mv_can_chain(grp_start))
582 dev_dbg(mv_chan->device->common.dev, "Append to last desc %x\n",
583 old_chain_tail->async_tx.phys);
585 /* fix up the hardware chain */
586 mv_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
588 /* if the channel is not busy */
589 if (!mv_chan_is_busy(mv_chan)) {
590 u32 current_desc = mv_chan_get_current_desc(mv_chan);
592 * and the curren desc is the end of the chain before
593 * the append, then we need to start the channel
595 if (current_desc == old_chain_tail->async_tx.phys)
601 mv_xor_start_new_chain(mv_chan, grp_start);
604 spin_unlock_bh(&mv_chan->lock);
609 /* returns the number of allocated descriptors */
610 static int mv_xor_alloc_chan_resources(struct dma_chan *chan)
614 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
615 struct mv_xor_desc_slot *slot = NULL;
616 struct mv_xor_platform_data *plat_data =
617 mv_chan->device->pdev->dev.platform_data;
618 int num_descs_in_pool = plat_data->pool_size/MV_XOR_SLOT_SIZE;
620 /* Allocate descriptor slots */
621 idx = mv_chan->slots_allocated;
622 while (idx < num_descs_in_pool) {
623 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
625 printk(KERN_INFO "MV XOR Channel only initialized"
626 " %d descriptor slots", idx);
629 hw_desc = (char *) mv_chan->device->dma_desc_pool_virt;
630 slot->hw_desc = (void *) &hw_desc[idx * MV_XOR_SLOT_SIZE];
632 dma_async_tx_descriptor_init(&slot->async_tx, chan);
633 slot->async_tx.tx_submit = mv_xor_tx_submit;
634 INIT_LIST_HEAD(&slot->chain_node);
635 INIT_LIST_HEAD(&slot->slot_node);
636 INIT_LIST_HEAD(&slot->tx_list);
637 hw_desc = (char *) mv_chan->device->dma_desc_pool;
638 slot->async_tx.phys =
639 (dma_addr_t) &hw_desc[idx * MV_XOR_SLOT_SIZE];
642 spin_lock_bh(&mv_chan->lock);
643 mv_chan->slots_allocated = idx;
644 list_add_tail(&slot->slot_node, &mv_chan->all_slots);
645 spin_unlock_bh(&mv_chan->lock);
648 if (mv_chan->slots_allocated && !mv_chan->last_used)
649 mv_chan->last_used = list_entry(mv_chan->all_slots.next,
650 struct mv_xor_desc_slot,
653 dev_dbg(mv_chan->device->common.dev,
654 "allocated %d descriptor slots last_used: %p\n",
655 mv_chan->slots_allocated, mv_chan->last_used);
657 return mv_chan->slots_allocated ? : -ENOMEM;
660 static struct dma_async_tx_descriptor *
661 mv_xor_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
662 size_t len, unsigned long flags)
664 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
665 struct mv_xor_desc_slot *sw_desc, *grp_start;
668 dev_dbg(mv_chan->device->common.dev,
669 "%s dest: %x src %x len: %u flags: %ld\n",
670 __func__, dest, src, len, flags);
671 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
674 BUG_ON(unlikely(len > MV_XOR_MAX_BYTE_COUNT));
676 spin_lock_bh(&mv_chan->lock);
677 slot_cnt = mv_chan_memcpy_slot_count(len);
678 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
680 sw_desc->type = DMA_MEMCPY;
681 sw_desc->async_tx.flags = flags;
682 grp_start = sw_desc->group_head;
683 mv_desc_init(grp_start, flags);
684 mv_desc_set_byte_count(grp_start, len);
685 mv_desc_set_dest_addr(sw_desc->group_head, dest);
686 mv_desc_set_src_addr(grp_start, 0, src);
687 sw_desc->unmap_src_cnt = 1;
688 sw_desc->unmap_len = len;
690 spin_unlock_bh(&mv_chan->lock);
692 dev_dbg(mv_chan->device->common.dev,
693 "%s sw_desc %p async_tx %p\n",
694 __func__, sw_desc, sw_desc ? &sw_desc->async_tx : 0);
696 return sw_desc ? &sw_desc->async_tx : NULL;
699 static struct dma_async_tx_descriptor *
700 mv_xor_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,
701 size_t len, unsigned long flags)
703 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
704 struct mv_xor_desc_slot *sw_desc, *grp_start;
707 dev_dbg(mv_chan->device->common.dev,
708 "%s dest: %x len: %u flags: %ld\n",
709 __func__, dest, len, flags);
710 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
713 BUG_ON(unlikely(len > MV_XOR_MAX_BYTE_COUNT));
715 spin_lock_bh(&mv_chan->lock);
716 slot_cnt = mv_chan_memset_slot_count(len);
717 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
719 sw_desc->type = DMA_MEMSET;
720 sw_desc->async_tx.flags = flags;
721 grp_start = sw_desc->group_head;
722 mv_desc_init(grp_start, flags);
723 mv_desc_set_byte_count(grp_start, len);
724 mv_desc_set_dest_addr(sw_desc->group_head, dest);
725 mv_desc_set_block_fill_val(grp_start, value);
726 sw_desc->unmap_src_cnt = 1;
727 sw_desc->unmap_len = len;
729 spin_unlock_bh(&mv_chan->lock);
730 dev_dbg(mv_chan->device->common.dev,
731 "%s sw_desc %p async_tx %p \n",
732 __func__, sw_desc, &sw_desc->async_tx);
733 return sw_desc ? &sw_desc->async_tx : NULL;
736 static struct dma_async_tx_descriptor *
737 mv_xor_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
738 unsigned int src_cnt, size_t len, unsigned long flags)
740 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
741 struct mv_xor_desc_slot *sw_desc, *grp_start;
744 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
747 BUG_ON(unlikely(len > MV_XOR_MAX_BYTE_COUNT));
749 dev_dbg(mv_chan->device->common.dev,
750 "%s src_cnt: %d len: dest %x %u flags: %ld\n",
751 __func__, src_cnt, len, dest, flags);
753 spin_lock_bh(&mv_chan->lock);
754 slot_cnt = mv_chan_xor_slot_count(len, src_cnt);
755 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
757 sw_desc->type = DMA_XOR;
758 sw_desc->async_tx.flags = flags;
759 grp_start = sw_desc->group_head;
760 mv_desc_init(grp_start, flags);
761 /* the byte count field is the same as in memcpy desc*/
762 mv_desc_set_byte_count(grp_start, len);
763 mv_desc_set_dest_addr(sw_desc->group_head, dest);
764 sw_desc->unmap_src_cnt = src_cnt;
765 sw_desc->unmap_len = len;
767 mv_desc_set_src_addr(grp_start, src_cnt, src[src_cnt]);
769 spin_unlock_bh(&mv_chan->lock);
770 dev_dbg(mv_chan->device->common.dev,
771 "%s sw_desc %p async_tx %p \n",
772 __func__, sw_desc, &sw_desc->async_tx);
773 return sw_desc ? &sw_desc->async_tx : NULL;
776 static void mv_xor_free_chan_resources(struct dma_chan *chan)
778 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
779 struct mv_xor_desc_slot *iter, *_iter;
780 int in_use_descs = 0;
782 mv_xor_slot_cleanup(mv_chan);
784 spin_lock_bh(&mv_chan->lock);
785 list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
788 list_del(&iter->chain_node);
790 list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
793 list_del(&iter->completed_node);
795 list_for_each_entry_safe_reverse(
796 iter, _iter, &mv_chan->all_slots, slot_node) {
797 list_del(&iter->slot_node);
799 mv_chan->slots_allocated--;
801 mv_chan->last_used = NULL;
803 dev_dbg(mv_chan->device->common.dev, "%s slots_allocated %d\n",
804 __func__, mv_chan->slots_allocated);
805 spin_unlock_bh(&mv_chan->lock);
808 dev_err(mv_chan->device->common.dev,
809 "freeing %d in use descriptors!\n", in_use_descs);
813 * mv_xor_is_complete - poll the status of an XOR transaction
814 * @chan: XOR channel handle
815 * @cookie: XOR transaction identifier
817 static enum dma_status mv_xor_is_complete(struct dma_chan *chan,
822 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
823 dma_cookie_t last_used;
824 dma_cookie_t last_complete;
827 last_used = chan->cookie;
828 last_complete = mv_chan->completed_cookie;
829 mv_chan->is_complete_cookie = cookie;
831 *done = last_complete;
835 ret = dma_async_is_complete(cookie, last_complete, last_used);
836 if (ret == DMA_SUCCESS) {
837 mv_xor_clean_completed_slots(mv_chan);
840 mv_xor_slot_cleanup(mv_chan);
842 last_used = chan->cookie;
843 last_complete = mv_chan->completed_cookie;
846 *done = last_complete;
850 return dma_async_is_complete(cookie, last_complete, last_used);
853 static void mv_dump_xor_regs(struct mv_xor_chan *chan)
857 val = __raw_readl(XOR_CONFIG(chan));
858 dev_printk(KERN_ERR, chan->device->common.dev,
859 "config 0x%08x.\n", val);
861 val = __raw_readl(XOR_ACTIVATION(chan));
862 dev_printk(KERN_ERR, chan->device->common.dev,
863 "activation 0x%08x.\n", val);
865 val = __raw_readl(XOR_INTR_CAUSE(chan));
866 dev_printk(KERN_ERR, chan->device->common.dev,
867 "intr cause 0x%08x.\n", val);
869 val = __raw_readl(XOR_INTR_MASK(chan));
870 dev_printk(KERN_ERR, chan->device->common.dev,
871 "intr mask 0x%08x.\n", val);
873 val = __raw_readl(XOR_ERROR_CAUSE(chan));
874 dev_printk(KERN_ERR, chan->device->common.dev,
875 "error cause 0x%08x.\n", val);
877 val = __raw_readl(XOR_ERROR_ADDR(chan));
878 dev_printk(KERN_ERR, chan->device->common.dev,
879 "error addr 0x%08x.\n", val);
882 static void mv_xor_err_interrupt_handler(struct mv_xor_chan *chan,
885 if (intr_cause & (1 << 4)) {
886 dev_dbg(chan->device->common.dev,
887 "ignore this error\n");
891 dev_printk(KERN_ERR, chan->device->common.dev,
892 "error on chan %d. intr cause 0x%08x.\n",
893 chan->idx, intr_cause);
895 mv_dump_xor_regs(chan);
899 static irqreturn_t mv_xor_interrupt_handler(int irq, void *data)
901 struct mv_xor_chan *chan = data;
902 u32 intr_cause = mv_chan_get_intr_cause(chan);
904 dev_dbg(chan->device->common.dev, "intr cause %x\n", intr_cause);
906 if (mv_is_err_intr(intr_cause))
907 mv_xor_err_interrupt_handler(chan, intr_cause);
909 tasklet_schedule(&chan->irq_tasklet);
911 mv_xor_device_clear_eoc_cause(chan);
916 static void mv_xor_issue_pending(struct dma_chan *chan)
918 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
920 if (mv_chan->pending >= MV_XOR_THRESHOLD) {
921 mv_chan->pending = 0;
922 mv_chan_activate(mv_chan);
927 * Perform a transaction to verify the HW works.
929 #define MV_XOR_TEST_SIZE 2000
931 static int __devinit mv_xor_memcpy_self_test(struct mv_xor_device *device)
935 dma_addr_t src_dma, dest_dma;
936 struct dma_chan *dma_chan;
938 struct dma_async_tx_descriptor *tx;
940 struct mv_xor_chan *mv_chan;
942 src = kmalloc(sizeof(u8) * MV_XOR_TEST_SIZE, GFP_KERNEL);
946 dest = kzalloc(sizeof(u8) * MV_XOR_TEST_SIZE, GFP_KERNEL);
952 /* Fill in src buffer */
953 for (i = 0; i < MV_XOR_TEST_SIZE; i++)
954 ((u8 *) src)[i] = (u8)i;
956 /* Start copy, using first DMA channel */
957 dma_chan = container_of(device->common.channels.next,
960 if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
965 dest_dma = dma_map_single(dma_chan->device->dev, dest,
966 MV_XOR_TEST_SIZE, DMA_FROM_DEVICE);
968 src_dma = dma_map_single(dma_chan->device->dev, src,
969 MV_XOR_TEST_SIZE, DMA_TO_DEVICE);
971 tx = mv_xor_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
972 MV_XOR_TEST_SIZE, 0);
973 cookie = mv_xor_tx_submit(tx);
974 mv_xor_issue_pending(dma_chan);
978 if (mv_xor_is_complete(dma_chan, cookie, NULL, NULL) !=
980 dev_printk(KERN_ERR, dma_chan->device->dev,
981 "Self-test copy timed out, disabling\n");
986 mv_chan = to_mv_xor_chan(dma_chan);
987 dma_sync_single_for_cpu(&mv_chan->device->pdev->dev, dest_dma,
988 MV_XOR_TEST_SIZE, DMA_FROM_DEVICE);
989 if (memcmp(src, dest, MV_XOR_TEST_SIZE)) {
990 dev_printk(KERN_ERR, dma_chan->device->dev,
991 "Self-test copy failed compare, disabling\n");
997 mv_xor_free_chan_resources(dma_chan);
1004 #define MV_XOR_NUM_SRC_TEST 4 /* must be <= 15 */
1005 static int __devinit
1006 mv_xor_xor_self_test(struct mv_xor_device *device)
1010 struct page *xor_srcs[MV_XOR_NUM_SRC_TEST];
1011 dma_addr_t dma_srcs[MV_XOR_NUM_SRC_TEST];
1012 dma_addr_t dest_dma;
1013 struct dma_async_tx_descriptor *tx;
1014 struct dma_chan *dma_chan;
1015 dma_cookie_t cookie;
1019 struct mv_xor_chan *mv_chan;
1021 for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++) {
1022 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
1023 if (!xor_srcs[src_idx]) {
1025 __free_page(xor_srcs[src_idx]);
1030 dest = alloc_page(GFP_KERNEL);
1033 __free_page(xor_srcs[src_idx]);
1037 /* Fill in src buffers */
1038 for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++) {
1039 u8 *ptr = page_address(xor_srcs[src_idx]);
1040 for (i = 0; i < PAGE_SIZE; i++)
1041 ptr[i] = (1 << src_idx);
1044 for (src_idx = 0; src_idx < MV_XOR_NUM_SRC_TEST; src_idx++)
1045 cmp_byte ^= (u8) (1 << src_idx);
1047 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
1048 (cmp_byte << 8) | cmp_byte;
1050 memset(page_address(dest), 0, PAGE_SIZE);
1052 dma_chan = container_of(device->common.channels.next,
1055 if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
1061 dest_dma = dma_map_page(dma_chan->device->dev, dest, 0, PAGE_SIZE,
1064 for (i = 0; i < MV_XOR_NUM_SRC_TEST; i++)
1065 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
1066 0, PAGE_SIZE, DMA_TO_DEVICE);
1068 tx = mv_xor_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
1069 MV_XOR_NUM_SRC_TEST, PAGE_SIZE, 0);
1071 cookie = mv_xor_tx_submit(tx);
1072 mv_xor_issue_pending(dma_chan);
1076 if (mv_xor_is_complete(dma_chan, cookie, NULL, NULL) !=
1078 dev_printk(KERN_ERR, dma_chan->device->dev,
1079 "Self-test xor timed out, disabling\n");
1081 goto free_resources;
1084 mv_chan = to_mv_xor_chan(dma_chan);
1085 dma_sync_single_for_cpu(&mv_chan->device->pdev->dev, dest_dma,
1086 PAGE_SIZE, DMA_FROM_DEVICE);
1087 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1088 u32 *ptr = page_address(dest);
1089 if (ptr[i] != cmp_word) {
1090 dev_printk(KERN_ERR, dma_chan->device->dev,
1091 "Self-test xor failed compare, disabling."
1092 " index %d, data %x, expected %x\n", i,
1095 goto free_resources;
1100 mv_xor_free_chan_resources(dma_chan);
1102 src_idx = MV_XOR_NUM_SRC_TEST;
1104 __free_page(xor_srcs[src_idx]);
1109 static int __devexit mv_xor_remove(struct platform_device *dev)
1111 struct mv_xor_device *device = platform_get_drvdata(dev);
1112 struct dma_chan *chan, *_chan;
1113 struct mv_xor_chan *mv_chan;
1114 struct mv_xor_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 mv_chan = to_mv_xor_chan(chan);
1124 list_del(&chan->device_node);
1130 static int __devinit mv_xor_probe(struct platform_device *pdev)
1134 struct mv_xor_device *adev;
1135 struct mv_xor_chan *mv_chan;
1136 struct dma_device *dma_dev;
1137 struct mv_xor_platform_data *plat_data = pdev->dev.platform_data;
1140 adev = devm_kzalloc(&pdev->dev, sizeof(*adev), GFP_KERNEL);
1144 dma_dev = &adev->common;
1146 /* allocate coherent memory for hardware descriptors
1147 * note: writecombine gives slightly better performance, but
1148 * requires that we explicitly flush the writes
1150 adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1151 plat_data->pool_size,
1152 &adev->dma_desc_pool,
1154 if (!adev->dma_desc_pool_virt)
1157 adev->id = plat_data->hw_id;
1159 /* discover transaction capabilites from the platform data */
1160 dma_dev->cap_mask = plat_data->cap_mask;
1162 platform_set_drvdata(pdev, adev);
1164 adev->shared = platform_get_drvdata(plat_data->shared);
1166 INIT_LIST_HEAD(&dma_dev->channels);
1168 /* set base routines */
1169 dma_dev->device_alloc_chan_resources = mv_xor_alloc_chan_resources;
1170 dma_dev->device_free_chan_resources = mv_xor_free_chan_resources;
1171 dma_dev->device_is_tx_complete = mv_xor_is_complete;
1172 dma_dev->device_issue_pending = mv_xor_issue_pending;
1173 dma_dev->dev = &pdev->dev;
1175 /* set prep routines based on capability */
1176 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1177 dma_dev->device_prep_dma_memcpy = mv_xor_prep_dma_memcpy;
1178 if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1179 dma_dev->device_prep_dma_memset = mv_xor_prep_dma_memset;
1180 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1181 dma_dev->max_xor = 8;
1182 dma_dev->device_prep_dma_xor = mv_xor_prep_dma_xor;
1185 mv_chan = devm_kzalloc(&pdev->dev, sizeof(*mv_chan), GFP_KERNEL);
1190 mv_chan->device = adev;
1191 mv_chan->idx = plat_data->hw_id;
1192 mv_chan->mmr_base = adev->shared->xor_base;
1194 if (!mv_chan->mmr_base) {
1198 tasklet_init(&mv_chan->irq_tasklet, mv_xor_tasklet, (unsigned long)
1201 /* clear errors before enabling interrupts */
1202 mv_xor_device_clear_err_status(mv_chan);
1204 irq = platform_get_irq(pdev, 0);
1209 ret = devm_request_irq(&pdev->dev, irq,
1210 mv_xor_interrupt_handler,
1211 0, dev_name(&pdev->dev), mv_chan);
1215 mv_chan_unmask_interrupts(mv_chan);
1217 mv_set_mode(mv_chan, DMA_MEMCPY);
1219 spin_lock_init(&mv_chan->lock);
1220 INIT_LIST_HEAD(&mv_chan->chain);
1221 INIT_LIST_HEAD(&mv_chan->completed_slots);
1222 INIT_LIST_HEAD(&mv_chan->all_slots);
1223 mv_chan->common.device = dma_dev;
1225 list_add_tail(&mv_chan->common.device_node, &dma_dev->channels);
1227 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1228 ret = mv_xor_memcpy_self_test(adev);
1229 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1234 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1235 ret = mv_xor_xor_self_test(adev);
1236 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1241 dev_printk(KERN_INFO, &pdev->dev, "Marvell XOR: "
1243 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1244 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "",
1245 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1246 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1248 dma_async_device_register(dma_dev);
1252 dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1253 adev->dma_desc_pool_virt, adev->dma_desc_pool);
1259 mv_xor_conf_mbus_windows(struct mv_xor_shared_private *msp,
1260 struct mbus_dram_target_info *dram)
1262 void __iomem *base = msp->xor_base;
1266 for (i = 0; i < 8; i++) {
1267 writel(0, base + WINDOW_BASE(i));
1268 writel(0, base + WINDOW_SIZE(i));
1270 writel(0, base + WINDOW_REMAP_HIGH(i));
1273 for (i = 0; i < dram->num_cs; i++) {
1274 struct mbus_dram_window *cs = dram->cs + i;
1276 writel((cs->base & 0xffff0000) |
1277 (cs->mbus_attr << 8) |
1278 dram->mbus_dram_target_id, base + WINDOW_BASE(i));
1279 writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
1281 win_enable |= (1 << i);
1282 win_enable |= 3 << (16 + (2 * i));
1285 writel(win_enable, base + WINDOW_BAR_ENABLE(0));
1286 writel(win_enable, base + WINDOW_BAR_ENABLE(1));
1289 static struct platform_driver mv_xor_driver = {
1290 .probe = mv_xor_probe,
1291 .remove = __devexit_p(mv_xor_remove),
1293 .owner = THIS_MODULE,
1294 .name = MV_XOR_NAME,
1298 static int mv_xor_shared_probe(struct platform_device *pdev)
1300 struct mv_xor_platform_shared_data *msd = pdev->dev.platform_data;
1301 struct mv_xor_shared_private *msp;
1302 struct resource *res;
1304 dev_printk(KERN_NOTICE, &pdev->dev, "Marvell shared XOR driver\n");
1306 msp = devm_kzalloc(&pdev->dev, sizeof(*msp), GFP_KERNEL);
1310 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1314 msp->xor_base = devm_ioremap(&pdev->dev, res->start,
1315 res->end - res->start + 1);
1319 res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1323 msp->xor_high_base = devm_ioremap(&pdev->dev, res->start,
1324 res->end - res->start + 1);
1325 if (!msp->xor_high_base)
1328 platform_set_drvdata(pdev, msp);
1331 * (Re-)program MBUS remapping windows if we are asked to.
1333 if (msd != NULL && msd->dram != NULL)
1334 mv_xor_conf_mbus_windows(msp, msd->dram);
1339 static int mv_xor_shared_remove(struct platform_device *pdev)
1344 static struct platform_driver mv_xor_shared_driver = {
1345 .probe = mv_xor_shared_probe,
1346 .remove = mv_xor_shared_remove,
1348 .owner = THIS_MODULE,
1349 .name = MV_XOR_SHARED_NAME,
1354 static int __init mv_xor_init(void)
1358 rc = platform_driver_register(&mv_xor_shared_driver);
1360 rc = platform_driver_register(&mv_xor_driver);
1362 platform_driver_unregister(&mv_xor_shared_driver);
1366 module_init(mv_xor_init);
1368 /* it's currently unsafe to unload this module */
1370 static void __exit mv_xor_exit(void)
1372 platform_driver_unregister(&mv_xor_driver);
1373 platform_driver_unregister(&mv_xor_shared_driver);
1377 module_exit(mv_xor_exit);
1380 MODULE_AUTHOR("Saeed Bishara <saeed@marvell.com>");
1381 MODULE_DESCRIPTION("DMA engine driver for Marvell's XOR engine");
1382 MODULE_LICENSE("GPL");