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 <linux/clk.h>
30 #include <linux/of_irq.h>
31 #include <linux/irqdomain.h>
32 #include <linux/platform_data/dma-mv_xor.h>
34 #include "dmaengine.h"
37 static void mv_xor_issue_pending(struct dma_chan *chan);
39 #define to_mv_xor_chan(chan) \
40 container_of(chan, struct mv_xor_chan, dmachan)
42 #define to_mv_xor_slot(tx) \
43 container_of(tx, struct mv_xor_desc_slot, async_tx)
45 #define mv_chan_to_devp(chan) \
48 static void mv_desc_init(struct mv_xor_desc_slot *desc, unsigned long flags)
50 struct mv_xor_desc *hw_desc = desc->hw_desc;
52 hw_desc->status = (1 << 31);
53 hw_desc->phy_next_desc = 0;
54 hw_desc->desc_command = (1 << 31);
57 static void mv_desc_set_byte_count(struct mv_xor_desc_slot *desc,
60 struct mv_xor_desc *hw_desc = desc->hw_desc;
61 hw_desc->byte_count = byte_count;
64 static void mv_desc_set_next_desc(struct mv_xor_desc_slot *desc,
67 struct mv_xor_desc *hw_desc = desc->hw_desc;
68 BUG_ON(hw_desc->phy_next_desc);
69 hw_desc->phy_next_desc = next_desc_addr;
72 static void mv_desc_clear_next_desc(struct mv_xor_desc_slot *desc)
74 struct mv_xor_desc *hw_desc = desc->hw_desc;
75 hw_desc->phy_next_desc = 0;
78 static void mv_desc_set_dest_addr(struct mv_xor_desc_slot *desc,
81 struct mv_xor_desc *hw_desc = desc->hw_desc;
82 hw_desc->phy_dest_addr = addr;
85 static int mv_chan_memset_slot_count(size_t len)
90 #define mv_chan_memcpy_slot_count(c) mv_chan_memset_slot_count(c)
92 static void mv_desc_set_src_addr(struct mv_xor_desc_slot *desc,
93 int index, dma_addr_t addr)
95 struct mv_xor_desc *hw_desc = desc->hw_desc;
96 hw_desc->phy_src_addr[mv_phy_src_idx(index)] = addr;
97 if (desc->type == DMA_XOR)
98 hw_desc->desc_command |= (1 << index);
101 static u32 mv_chan_get_current_desc(struct mv_xor_chan *chan)
103 return readl_relaxed(XOR_CURR_DESC(chan));
106 static void mv_chan_set_next_descriptor(struct mv_xor_chan *chan,
109 writel_relaxed(next_desc_addr, XOR_NEXT_DESC(chan));
112 static void mv_chan_unmask_interrupts(struct mv_xor_chan *chan)
114 u32 val = readl_relaxed(XOR_INTR_MASK(chan));
115 val |= XOR_INTR_MASK_VALUE << (chan->idx * 16);
116 writel_relaxed(val, XOR_INTR_MASK(chan));
119 static u32 mv_chan_get_intr_cause(struct mv_xor_chan *chan)
121 u32 intr_cause = readl_relaxed(XOR_INTR_CAUSE(chan));
122 intr_cause = (intr_cause >> (chan->idx * 16)) & 0xFFFF;
126 static int mv_is_err_intr(u32 intr_cause)
128 if (intr_cause & ((1<<4)|(1<<5)|(1<<6)|(1<<7)|(1<<8)|(1<<9)))
134 static void mv_xor_device_clear_eoc_cause(struct mv_xor_chan *chan)
136 u32 val = ~(1 << (chan->idx * 16));
137 dev_dbg(mv_chan_to_devp(chan), "%s, val 0x%08x\n", __func__, val);
138 writel_relaxed(val, XOR_INTR_CAUSE(chan));
141 static void mv_xor_device_clear_err_status(struct mv_xor_chan *chan)
143 u32 val = 0xFFFF0000 >> (chan->idx * 16);
144 writel_relaxed(val, XOR_INTR_CAUSE(chan));
147 static int mv_can_chain(struct mv_xor_desc_slot *desc)
149 struct mv_xor_desc_slot *chain_old_tail = list_entry(
150 desc->chain_node.prev, struct mv_xor_desc_slot, chain_node);
152 if (chain_old_tail->type != desc->type)
158 static void mv_set_mode(struct mv_xor_chan *chan,
159 enum dma_transaction_type type)
162 u32 config = readl_relaxed(XOR_CONFIG(chan));
166 op_mode = XOR_OPERATION_MODE_XOR;
169 op_mode = XOR_OPERATION_MODE_MEMCPY;
172 dev_err(mv_chan_to_devp(chan),
173 "error: unsupported operation %d\n",
182 #if defined(__BIG_ENDIAN)
183 config |= XOR_DESCRIPTOR_SWAP;
185 config &= ~XOR_DESCRIPTOR_SWAP;
188 writel_relaxed(config, XOR_CONFIG(chan));
189 chan->current_type = type;
192 static void mv_chan_activate(struct mv_xor_chan *chan)
194 dev_dbg(mv_chan_to_devp(chan), " activate chan.\n");
196 /* writel ensures all descriptors are flushed before activation */
197 writel(BIT(0), XOR_ACTIVATION(chan));
200 static char mv_chan_is_busy(struct mv_xor_chan *chan)
202 u32 state = readl_relaxed(XOR_ACTIVATION(chan));
204 state = (state >> 4) & 0x3;
206 return (state == 1) ? 1 : 0;
209 static int mv_chan_xor_slot_count(size_t len, int src_cnt)
215 * mv_xor_free_slots - flags descriptor slots for reuse
216 * @slot: Slot to free
217 * Caller must hold &mv_chan->lock while calling this function
219 static void mv_xor_free_slots(struct mv_xor_chan *mv_chan,
220 struct mv_xor_desc_slot *slot)
222 dev_dbg(mv_chan_to_devp(mv_chan), "%s %d slot %p\n",
223 __func__, __LINE__, slot);
225 slot->slots_per_op = 0;
230 * mv_xor_start_new_chain - program the engine to operate on new chain headed by
232 * Caller must hold &mv_chan->lock while calling this function
234 static void mv_xor_start_new_chain(struct mv_xor_chan *mv_chan,
235 struct mv_xor_desc_slot *sw_desc)
237 dev_dbg(mv_chan_to_devp(mv_chan), "%s %d: sw_desc %p\n",
238 __func__, __LINE__, sw_desc);
239 if (sw_desc->type != mv_chan->current_type)
240 mv_set_mode(mv_chan, sw_desc->type);
242 /* set the hardware chain */
243 mv_chan_set_next_descriptor(mv_chan, sw_desc->async_tx.phys);
245 mv_chan->pending += sw_desc->slot_cnt;
246 mv_xor_issue_pending(&mv_chan->dmachan);
250 mv_xor_run_tx_complete_actions(struct mv_xor_desc_slot *desc,
251 struct mv_xor_chan *mv_chan, dma_cookie_t cookie)
253 BUG_ON(desc->async_tx.cookie < 0);
255 if (desc->async_tx.cookie > 0) {
256 cookie = desc->async_tx.cookie;
258 /* call the callback (must not sleep or submit new
259 * operations to this channel)
261 if (desc->async_tx.callback)
262 desc->async_tx.callback(
263 desc->async_tx.callback_param);
265 dma_descriptor_unmap(&desc->async_tx);
266 if (desc->group_head)
267 desc->group_head = NULL;
270 /* run dependent operations */
271 dma_run_dependencies(&desc->async_tx);
277 mv_xor_clean_completed_slots(struct mv_xor_chan *mv_chan)
279 struct mv_xor_desc_slot *iter, *_iter;
281 dev_dbg(mv_chan_to_devp(mv_chan), "%s %d\n", __func__, __LINE__);
282 list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
285 if (async_tx_test_ack(&iter->async_tx)) {
286 list_del(&iter->completed_node);
287 mv_xor_free_slots(mv_chan, iter);
294 mv_xor_clean_slot(struct mv_xor_desc_slot *desc,
295 struct mv_xor_chan *mv_chan)
297 dev_dbg(mv_chan_to_devp(mv_chan), "%s %d: desc %p flags %d\n",
298 __func__, __LINE__, desc, desc->async_tx.flags);
299 list_del(&desc->chain_node);
300 /* the client is allowed to attach dependent operations
303 if (!async_tx_test_ack(&desc->async_tx)) {
304 /* move this slot to the completed_slots */
305 list_add_tail(&desc->completed_node, &mv_chan->completed_slots);
309 mv_xor_free_slots(mv_chan, desc);
313 static void __mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
315 struct mv_xor_desc_slot *iter, *_iter;
316 dma_cookie_t cookie = 0;
317 int busy = mv_chan_is_busy(mv_chan);
318 u32 current_desc = mv_chan_get_current_desc(mv_chan);
319 int current_cleaned = 0;
320 struct mv_xor_desc *hw_desc;
322 dev_dbg(mv_chan_to_devp(mv_chan), "%s %d\n", __func__, __LINE__);
323 dev_dbg(mv_chan_to_devp(mv_chan), "current_desc %x\n", current_desc);
324 mv_xor_clean_completed_slots(mv_chan);
326 /* free completed slots from the chain starting with
327 * the oldest descriptor
330 list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
333 /* clean finished descriptors */
334 hw_desc = iter->hw_desc;
335 if (hw_desc->status & XOR_DESC_SUCCESS) {
336 cookie = mv_xor_run_tx_complete_actions(iter, mv_chan,
339 /* done processing desc, clean slot */
340 mv_xor_clean_slot(iter, mv_chan);
342 /* break if we did cleaned the current */
343 if (iter->async_tx.phys == current_desc) {
348 if (iter->async_tx.phys == current_desc) {
355 if ((busy == 0) && !list_empty(&mv_chan->chain)) {
356 if (current_cleaned) {
358 * current descriptor cleaned and removed, run
361 iter = list_entry(mv_chan->chain.next,
362 struct mv_xor_desc_slot,
364 mv_xor_start_new_chain(mv_chan, iter);
366 if (!list_is_last(&iter->chain_node, &mv_chan->chain)) {
368 * descriptors are still waiting after
369 * current, trigger them
371 iter = list_entry(iter->chain_node.next,
372 struct mv_xor_desc_slot,
374 mv_xor_start_new_chain(mv_chan, iter);
377 * some descriptors are still waiting
380 tasklet_schedule(&mv_chan->irq_tasklet);
386 mv_chan->dmachan.completed_cookie = cookie;
390 mv_xor_slot_cleanup(struct mv_xor_chan *mv_chan)
392 spin_lock_bh(&mv_chan->lock);
393 __mv_xor_slot_cleanup(mv_chan);
394 spin_unlock_bh(&mv_chan->lock);
397 static void mv_xor_tasklet(unsigned long data)
399 struct mv_xor_chan *chan = (struct mv_xor_chan *) data;
400 mv_xor_slot_cleanup(chan);
403 static struct mv_xor_desc_slot *
404 mv_xor_alloc_slots(struct mv_xor_chan *mv_chan, int num_slots,
407 struct mv_xor_desc_slot *iter, *_iter, *alloc_start = NULL;
409 int slots_found, retry = 0;
411 /* start search from the last allocated descrtiptor
412 * if a contiguous allocation can not be found start searching
413 * from the beginning of the list
418 iter = mv_chan->last_used;
420 iter = list_entry(&mv_chan->all_slots,
421 struct mv_xor_desc_slot,
424 list_for_each_entry_safe_continue(
425 iter, _iter, &mv_chan->all_slots, slot_node) {
427 prefetch(&_iter->async_tx);
428 if (iter->slots_per_op) {
429 /* give up after finding the first busy slot
430 * on the second pass through the list
439 /* start the allocation if the slot is correctly aligned */
443 if (slots_found == num_slots) {
444 struct mv_xor_desc_slot *alloc_tail = NULL;
445 struct mv_xor_desc_slot *last_used = NULL;
450 /* pre-ack all but the last descriptor */
451 async_tx_ack(&iter->async_tx);
453 list_add_tail(&iter->chain_node, &chain);
455 iter->async_tx.cookie = 0;
456 iter->slot_cnt = num_slots;
457 iter->xor_check_result = NULL;
458 for (i = 0; i < slots_per_op; i++) {
459 iter->slots_per_op = slots_per_op - i;
461 iter = list_entry(iter->slot_node.next,
462 struct mv_xor_desc_slot,
465 num_slots -= slots_per_op;
467 alloc_tail->group_head = alloc_start;
468 alloc_tail->async_tx.cookie = -EBUSY;
469 list_splice(&chain, &alloc_tail->tx_list);
470 mv_chan->last_used = last_used;
471 mv_desc_clear_next_desc(alloc_start);
472 mv_desc_clear_next_desc(alloc_tail);
479 /* try to free some slots if the allocation fails */
480 tasklet_schedule(&mv_chan->irq_tasklet);
485 /************************ DMA engine API functions ****************************/
487 mv_xor_tx_submit(struct dma_async_tx_descriptor *tx)
489 struct mv_xor_desc_slot *sw_desc = to_mv_xor_slot(tx);
490 struct mv_xor_chan *mv_chan = to_mv_xor_chan(tx->chan);
491 struct mv_xor_desc_slot *grp_start, *old_chain_tail;
493 int new_hw_chain = 1;
495 dev_dbg(mv_chan_to_devp(mv_chan),
496 "%s sw_desc %p: async_tx %p\n",
497 __func__, sw_desc, &sw_desc->async_tx);
499 grp_start = sw_desc->group_head;
501 spin_lock_bh(&mv_chan->lock);
502 cookie = dma_cookie_assign(tx);
504 if (list_empty(&mv_chan->chain))
505 list_splice_init(&sw_desc->tx_list, &mv_chan->chain);
509 old_chain_tail = list_entry(mv_chan->chain.prev,
510 struct mv_xor_desc_slot,
512 list_splice_init(&grp_start->tx_list,
513 &old_chain_tail->chain_node);
515 if (!mv_can_chain(grp_start))
518 dev_dbg(mv_chan_to_devp(mv_chan), "Append to last desc %pa\n",
519 &old_chain_tail->async_tx.phys);
521 /* fix up the hardware chain */
522 mv_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
524 /* if the channel is not busy */
525 if (!mv_chan_is_busy(mv_chan)) {
526 u32 current_desc = mv_chan_get_current_desc(mv_chan);
528 * and the curren desc is the end of the chain before
529 * the append, then we need to start the channel
531 if (current_desc == old_chain_tail->async_tx.phys)
537 mv_xor_start_new_chain(mv_chan, grp_start);
540 spin_unlock_bh(&mv_chan->lock);
545 /* returns the number of allocated descriptors */
546 static int mv_xor_alloc_chan_resources(struct dma_chan *chan)
551 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
552 struct mv_xor_desc_slot *slot = NULL;
553 int num_descs_in_pool = MV_XOR_POOL_SIZE/MV_XOR_SLOT_SIZE;
555 /* Allocate descriptor slots */
556 idx = mv_chan->slots_allocated;
557 while (idx < num_descs_in_pool) {
558 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
560 printk(KERN_INFO "MV XOR Channel only initialized"
561 " %d descriptor slots", idx);
564 virt_desc = mv_chan->dma_desc_pool_virt;
565 slot->hw_desc = virt_desc + idx * MV_XOR_SLOT_SIZE;
567 dma_async_tx_descriptor_init(&slot->async_tx, chan);
568 slot->async_tx.tx_submit = mv_xor_tx_submit;
569 INIT_LIST_HEAD(&slot->chain_node);
570 INIT_LIST_HEAD(&slot->slot_node);
571 INIT_LIST_HEAD(&slot->tx_list);
572 dma_desc = mv_chan->dma_desc_pool;
573 slot->async_tx.phys = dma_desc + idx * MV_XOR_SLOT_SIZE;
576 spin_lock_bh(&mv_chan->lock);
577 mv_chan->slots_allocated = idx;
578 list_add_tail(&slot->slot_node, &mv_chan->all_slots);
579 spin_unlock_bh(&mv_chan->lock);
582 if (mv_chan->slots_allocated && !mv_chan->last_used)
583 mv_chan->last_used = list_entry(mv_chan->all_slots.next,
584 struct mv_xor_desc_slot,
587 dev_dbg(mv_chan_to_devp(mv_chan),
588 "allocated %d descriptor slots last_used: %p\n",
589 mv_chan->slots_allocated, mv_chan->last_used);
591 return mv_chan->slots_allocated ? : -ENOMEM;
594 static struct dma_async_tx_descriptor *
595 mv_xor_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
596 size_t len, unsigned long flags)
598 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
599 struct mv_xor_desc_slot *sw_desc, *grp_start;
602 dev_dbg(mv_chan_to_devp(mv_chan),
603 "%s dest: %pad src %pad len: %u flags: %ld\n",
604 __func__, &dest, &src, len, flags);
605 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
608 BUG_ON(len > MV_XOR_MAX_BYTE_COUNT);
610 spin_lock_bh(&mv_chan->lock);
611 slot_cnt = mv_chan_memcpy_slot_count(len);
612 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
614 sw_desc->type = DMA_MEMCPY;
615 sw_desc->async_tx.flags = flags;
616 grp_start = sw_desc->group_head;
617 mv_desc_init(grp_start, flags);
618 mv_desc_set_byte_count(grp_start, len);
619 mv_desc_set_dest_addr(sw_desc->group_head, dest);
620 mv_desc_set_src_addr(grp_start, 0, src);
621 sw_desc->unmap_src_cnt = 1;
622 sw_desc->unmap_len = len;
624 spin_unlock_bh(&mv_chan->lock);
626 dev_dbg(mv_chan_to_devp(mv_chan),
627 "%s sw_desc %p async_tx %p\n",
628 __func__, sw_desc, sw_desc ? &sw_desc->async_tx : NULL);
630 return sw_desc ? &sw_desc->async_tx : NULL;
633 static struct dma_async_tx_descriptor *
634 mv_xor_prep_dma_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
635 unsigned int src_cnt, size_t len, unsigned long flags)
637 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
638 struct mv_xor_desc_slot *sw_desc, *grp_start;
641 if (unlikely(len < MV_XOR_MIN_BYTE_COUNT))
644 BUG_ON(len > MV_XOR_MAX_BYTE_COUNT);
646 dev_dbg(mv_chan_to_devp(mv_chan),
647 "%s src_cnt: %d len: %u dest %pad flags: %ld\n",
648 __func__, src_cnt, len, &dest, flags);
650 spin_lock_bh(&mv_chan->lock);
651 slot_cnt = mv_chan_xor_slot_count(len, src_cnt);
652 sw_desc = mv_xor_alloc_slots(mv_chan, slot_cnt, 1);
654 sw_desc->type = DMA_XOR;
655 sw_desc->async_tx.flags = flags;
656 grp_start = sw_desc->group_head;
657 mv_desc_init(grp_start, flags);
658 /* the byte count field is the same as in memcpy desc*/
659 mv_desc_set_byte_count(grp_start, len);
660 mv_desc_set_dest_addr(sw_desc->group_head, dest);
661 sw_desc->unmap_src_cnt = src_cnt;
662 sw_desc->unmap_len = len;
664 mv_desc_set_src_addr(grp_start, src_cnt, src[src_cnt]);
666 spin_unlock_bh(&mv_chan->lock);
667 dev_dbg(mv_chan_to_devp(mv_chan),
668 "%s sw_desc %p async_tx %p \n",
669 __func__, sw_desc, &sw_desc->async_tx);
670 return sw_desc ? &sw_desc->async_tx : NULL;
673 static void mv_xor_free_chan_resources(struct dma_chan *chan)
675 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
676 struct mv_xor_desc_slot *iter, *_iter;
677 int in_use_descs = 0;
679 mv_xor_slot_cleanup(mv_chan);
681 spin_lock_bh(&mv_chan->lock);
682 list_for_each_entry_safe(iter, _iter, &mv_chan->chain,
685 list_del(&iter->chain_node);
687 list_for_each_entry_safe(iter, _iter, &mv_chan->completed_slots,
690 list_del(&iter->completed_node);
692 list_for_each_entry_safe_reverse(
693 iter, _iter, &mv_chan->all_slots, slot_node) {
694 list_del(&iter->slot_node);
696 mv_chan->slots_allocated--;
698 mv_chan->last_used = NULL;
700 dev_dbg(mv_chan_to_devp(mv_chan), "%s slots_allocated %d\n",
701 __func__, mv_chan->slots_allocated);
702 spin_unlock_bh(&mv_chan->lock);
705 dev_err(mv_chan_to_devp(mv_chan),
706 "freeing %d in use descriptors!\n", in_use_descs);
710 * mv_xor_status - poll the status of an XOR transaction
711 * @chan: XOR channel handle
712 * @cookie: XOR transaction identifier
713 * @txstate: XOR transactions state holder (or NULL)
715 static enum dma_status mv_xor_status(struct dma_chan *chan,
717 struct dma_tx_state *txstate)
719 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
722 ret = dma_cookie_status(chan, cookie, txstate);
723 if (ret == DMA_COMPLETE) {
724 mv_xor_clean_completed_slots(mv_chan);
727 mv_xor_slot_cleanup(mv_chan);
729 return dma_cookie_status(chan, cookie, txstate);
732 static void mv_dump_xor_regs(struct mv_xor_chan *chan)
736 val = readl_relaxed(XOR_CONFIG(chan));
737 dev_err(mv_chan_to_devp(chan), "config 0x%08x\n", val);
739 val = readl_relaxed(XOR_ACTIVATION(chan));
740 dev_err(mv_chan_to_devp(chan), "activation 0x%08x\n", val);
742 val = readl_relaxed(XOR_INTR_CAUSE(chan));
743 dev_err(mv_chan_to_devp(chan), "intr cause 0x%08x\n", val);
745 val = readl_relaxed(XOR_INTR_MASK(chan));
746 dev_err(mv_chan_to_devp(chan), "intr mask 0x%08x\n", val);
748 val = readl_relaxed(XOR_ERROR_CAUSE(chan));
749 dev_err(mv_chan_to_devp(chan), "error cause 0x%08x\n", val);
751 val = readl_relaxed(XOR_ERROR_ADDR(chan));
752 dev_err(mv_chan_to_devp(chan), "error addr 0x%08x\n", val);
755 static void mv_xor_err_interrupt_handler(struct mv_xor_chan *chan,
758 if (intr_cause & (1 << 4)) {
759 dev_dbg(mv_chan_to_devp(chan),
760 "ignore this error\n");
764 dev_err(mv_chan_to_devp(chan),
765 "error on chan %d. intr cause 0x%08x\n",
766 chan->idx, intr_cause);
768 mv_dump_xor_regs(chan);
772 static irqreturn_t mv_xor_interrupt_handler(int irq, void *data)
774 struct mv_xor_chan *chan = data;
775 u32 intr_cause = mv_chan_get_intr_cause(chan);
777 dev_dbg(mv_chan_to_devp(chan), "intr cause %x\n", intr_cause);
779 if (mv_is_err_intr(intr_cause))
780 mv_xor_err_interrupt_handler(chan, intr_cause);
782 tasklet_schedule(&chan->irq_tasklet);
784 mv_xor_device_clear_eoc_cause(chan);
789 static void mv_xor_issue_pending(struct dma_chan *chan)
791 struct mv_xor_chan *mv_chan = to_mv_xor_chan(chan);
793 if (mv_chan->pending >= MV_XOR_THRESHOLD) {
794 mv_chan->pending = 0;
795 mv_chan_activate(mv_chan);
800 * Perform a transaction to verify the HW works.
803 static int mv_xor_memcpy_self_test(struct mv_xor_chan *mv_chan)
807 dma_addr_t src_dma, dest_dma;
808 struct dma_chan *dma_chan;
810 struct dma_async_tx_descriptor *tx;
811 struct dmaengine_unmap_data *unmap;
814 src = kmalloc(sizeof(u8) * PAGE_SIZE, GFP_KERNEL);
818 dest = kzalloc(sizeof(u8) * PAGE_SIZE, GFP_KERNEL);
824 /* Fill in src buffer */
825 for (i = 0; i < PAGE_SIZE; i++)
826 ((u8 *) src)[i] = (u8)i;
828 dma_chan = &mv_chan->dmachan;
829 if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
834 unmap = dmaengine_get_unmap_data(dma_chan->device->dev, 2, GFP_KERNEL);
840 src_dma = dma_map_page(dma_chan->device->dev, virt_to_page(src), 0,
841 PAGE_SIZE, DMA_TO_DEVICE);
843 unmap->addr[0] = src_dma;
845 dest_dma = dma_map_page(dma_chan->device->dev, virt_to_page(dest), 0,
846 PAGE_SIZE, DMA_FROM_DEVICE);
848 unmap->addr[1] = dest_dma;
850 unmap->len = PAGE_SIZE;
852 tx = mv_xor_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
854 cookie = mv_xor_tx_submit(tx);
855 mv_xor_issue_pending(dma_chan);
859 if (mv_xor_status(dma_chan, cookie, NULL) !=
861 dev_err(dma_chan->device->dev,
862 "Self-test copy timed out, disabling\n");
867 dma_sync_single_for_cpu(dma_chan->device->dev, dest_dma,
868 PAGE_SIZE, DMA_FROM_DEVICE);
869 if (memcmp(src, dest, PAGE_SIZE)) {
870 dev_err(dma_chan->device->dev,
871 "Self-test copy failed compare, disabling\n");
877 dmaengine_unmap_put(unmap);
878 mv_xor_free_chan_resources(dma_chan);
885 #define MV_XOR_NUM_SRC_TEST 4 /* must be <= 15 */
887 mv_xor_xor_self_test(struct mv_xor_chan *mv_chan)
891 struct page *xor_srcs[MV_XOR_NUM_SRC_TEST];
892 dma_addr_t dma_srcs[MV_XOR_NUM_SRC_TEST];
894 struct dma_async_tx_descriptor *tx;
895 struct dmaengine_unmap_data *unmap;
896 struct dma_chan *dma_chan;
901 int src_count = MV_XOR_NUM_SRC_TEST;
903 for (src_idx = 0; src_idx < src_count; src_idx++) {
904 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
905 if (!xor_srcs[src_idx]) {
907 __free_page(xor_srcs[src_idx]);
912 dest = alloc_page(GFP_KERNEL);
915 __free_page(xor_srcs[src_idx]);
919 /* Fill in src buffers */
920 for (src_idx = 0; src_idx < src_count; src_idx++) {
921 u8 *ptr = page_address(xor_srcs[src_idx]);
922 for (i = 0; i < PAGE_SIZE; i++)
923 ptr[i] = (1 << src_idx);
926 for (src_idx = 0; src_idx < src_count; src_idx++)
927 cmp_byte ^= (u8) (1 << src_idx);
929 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
930 (cmp_byte << 8) | cmp_byte;
932 memset(page_address(dest), 0, PAGE_SIZE);
934 dma_chan = &mv_chan->dmachan;
935 if (mv_xor_alloc_chan_resources(dma_chan) < 1) {
940 unmap = dmaengine_get_unmap_data(dma_chan->device->dev, src_count + 1,
948 for (i = 0; i < src_count; i++) {
949 unmap->addr[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
950 0, PAGE_SIZE, DMA_TO_DEVICE);
951 dma_srcs[i] = unmap->addr[i];
955 unmap->addr[src_count] = dma_map_page(dma_chan->device->dev, dest, 0, PAGE_SIZE,
957 dest_dma = unmap->addr[src_count];
959 unmap->len = PAGE_SIZE;
961 tx = mv_xor_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
962 src_count, PAGE_SIZE, 0);
964 cookie = mv_xor_tx_submit(tx);
965 mv_xor_issue_pending(dma_chan);
969 if (mv_xor_status(dma_chan, cookie, NULL) !=
971 dev_err(dma_chan->device->dev,
972 "Self-test xor timed out, disabling\n");
977 dma_sync_single_for_cpu(dma_chan->device->dev, dest_dma,
978 PAGE_SIZE, DMA_FROM_DEVICE);
979 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
980 u32 *ptr = page_address(dest);
981 if (ptr[i] != cmp_word) {
982 dev_err(dma_chan->device->dev,
983 "Self-test xor failed compare, disabling. index %d, data %x, expected %x\n",
984 i, ptr[i], cmp_word);
991 dmaengine_unmap_put(unmap);
992 mv_xor_free_chan_resources(dma_chan);
996 __free_page(xor_srcs[src_idx]);
1001 /* This driver does not implement any of the optional DMA operations. */
1003 mv_xor_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
1009 static int mv_xor_channel_remove(struct mv_xor_chan *mv_chan)
1011 struct dma_chan *chan, *_chan;
1012 struct device *dev = mv_chan->dmadev.dev;
1014 dma_async_device_unregister(&mv_chan->dmadev);
1016 dma_free_coherent(dev, MV_XOR_POOL_SIZE,
1017 mv_chan->dma_desc_pool_virt, mv_chan->dma_desc_pool);
1019 list_for_each_entry_safe(chan, _chan, &mv_chan->dmadev.channels,
1021 list_del(&chan->device_node);
1024 free_irq(mv_chan->irq, mv_chan);
1029 static struct mv_xor_chan *
1030 mv_xor_channel_add(struct mv_xor_device *xordev,
1031 struct platform_device *pdev,
1032 int idx, dma_cap_mask_t cap_mask, int irq)
1035 struct mv_xor_chan *mv_chan;
1036 struct dma_device *dma_dev;
1038 mv_chan = devm_kzalloc(&pdev->dev, sizeof(*mv_chan), GFP_KERNEL);
1040 return ERR_PTR(-ENOMEM);
1045 dma_dev = &mv_chan->dmadev;
1047 /* allocate coherent memory for hardware descriptors
1048 * note: writecombine gives slightly better performance, but
1049 * requires that we explicitly flush the writes
1051 mv_chan->dma_desc_pool_virt =
1052 dma_alloc_writecombine(&pdev->dev, MV_XOR_POOL_SIZE,
1053 &mv_chan->dma_desc_pool, GFP_KERNEL);
1054 if (!mv_chan->dma_desc_pool_virt)
1055 return ERR_PTR(-ENOMEM);
1057 /* discover transaction capabilites from the platform data */
1058 dma_dev->cap_mask = cap_mask;
1060 INIT_LIST_HEAD(&dma_dev->channels);
1062 /* set base routines */
1063 dma_dev->device_alloc_chan_resources = mv_xor_alloc_chan_resources;
1064 dma_dev->device_free_chan_resources = mv_xor_free_chan_resources;
1065 dma_dev->device_tx_status = mv_xor_status;
1066 dma_dev->device_issue_pending = mv_xor_issue_pending;
1067 dma_dev->device_control = mv_xor_control;
1068 dma_dev->dev = &pdev->dev;
1070 /* set prep routines based on capability */
1071 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1072 dma_dev->device_prep_dma_memcpy = mv_xor_prep_dma_memcpy;
1073 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1074 dma_dev->max_xor = 8;
1075 dma_dev->device_prep_dma_xor = mv_xor_prep_dma_xor;
1078 mv_chan->mmr_base = xordev->xor_base;
1079 mv_chan->mmr_high_base = xordev->xor_high_base;
1080 tasklet_init(&mv_chan->irq_tasklet, mv_xor_tasklet, (unsigned long)
1083 /* clear errors before enabling interrupts */
1084 mv_xor_device_clear_err_status(mv_chan);
1086 ret = request_irq(mv_chan->irq, mv_xor_interrupt_handler,
1087 0, dev_name(&pdev->dev), mv_chan);
1091 mv_chan_unmask_interrupts(mv_chan);
1093 mv_set_mode(mv_chan, DMA_MEMCPY);
1095 spin_lock_init(&mv_chan->lock);
1096 INIT_LIST_HEAD(&mv_chan->chain);
1097 INIT_LIST_HEAD(&mv_chan->completed_slots);
1098 INIT_LIST_HEAD(&mv_chan->all_slots);
1099 mv_chan->dmachan.device = dma_dev;
1100 dma_cookie_init(&mv_chan->dmachan);
1102 list_add_tail(&mv_chan->dmachan.device_node, &dma_dev->channels);
1104 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1105 ret = mv_xor_memcpy_self_test(mv_chan);
1106 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1111 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1112 ret = mv_xor_xor_self_test(mv_chan);
1113 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1118 dev_info(&pdev->dev, "Marvell XOR: ( %s%s%s)\n",
1119 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1120 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1121 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1123 dma_async_device_register(dma_dev);
1127 free_irq(mv_chan->irq, mv_chan);
1129 dma_free_coherent(&pdev->dev, MV_XOR_POOL_SIZE,
1130 mv_chan->dma_desc_pool_virt, mv_chan->dma_desc_pool);
1131 return ERR_PTR(ret);
1135 mv_xor_conf_mbus_windows(struct mv_xor_device *xordev,
1136 const struct mbus_dram_target_info *dram)
1138 void __iomem *base = xordev->xor_high_base;
1142 for (i = 0; i < 8; i++) {
1143 writel(0, base + WINDOW_BASE(i));
1144 writel(0, base + WINDOW_SIZE(i));
1146 writel(0, base + WINDOW_REMAP_HIGH(i));
1149 for (i = 0; i < dram->num_cs; i++) {
1150 const struct mbus_dram_window *cs = dram->cs + i;
1152 writel((cs->base & 0xffff0000) |
1153 (cs->mbus_attr << 8) |
1154 dram->mbus_dram_target_id, base + WINDOW_BASE(i));
1155 writel((cs->size - 1) & 0xffff0000, base + WINDOW_SIZE(i));
1157 win_enable |= (1 << i);
1158 win_enable |= 3 << (16 + (2 * i));
1161 writel(win_enable, base + WINDOW_BAR_ENABLE(0));
1162 writel(win_enable, base + WINDOW_BAR_ENABLE(1));
1163 writel(0, base + WINDOW_OVERRIDE_CTRL(0));
1164 writel(0, base + WINDOW_OVERRIDE_CTRL(1));
1167 static int mv_xor_probe(struct platform_device *pdev)
1169 const struct mbus_dram_target_info *dram;
1170 struct mv_xor_device *xordev;
1171 struct mv_xor_platform_data *pdata = dev_get_platdata(&pdev->dev);
1172 struct resource *res;
1175 dev_notice(&pdev->dev, "Marvell shared XOR driver\n");
1177 xordev = devm_kzalloc(&pdev->dev, sizeof(*xordev), GFP_KERNEL);
1181 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1185 xordev->xor_base = devm_ioremap(&pdev->dev, res->start,
1186 resource_size(res));
1187 if (!xordev->xor_base)
1190 res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1194 xordev->xor_high_base = devm_ioremap(&pdev->dev, res->start,
1195 resource_size(res));
1196 if (!xordev->xor_high_base)
1199 platform_set_drvdata(pdev, xordev);
1202 * (Re-)program MBUS remapping windows if we are asked to.
1204 dram = mv_mbus_dram_info();
1206 mv_xor_conf_mbus_windows(xordev, dram);
1208 /* Not all platforms can gate the clock, so it is not
1209 * an error if the clock does not exists.
1211 xordev->clk = clk_get(&pdev->dev, NULL);
1212 if (!IS_ERR(xordev->clk))
1213 clk_prepare_enable(xordev->clk);
1215 if (pdev->dev.of_node) {
1216 struct device_node *np;
1219 for_each_child_of_node(pdev->dev.of_node, np) {
1220 struct mv_xor_chan *chan;
1221 dma_cap_mask_t cap_mask;
1224 dma_cap_zero(cap_mask);
1225 if (of_property_read_bool(np, "dmacap,memcpy"))
1226 dma_cap_set(DMA_MEMCPY, cap_mask);
1227 if (of_property_read_bool(np, "dmacap,xor"))
1228 dma_cap_set(DMA_XOR, cap_mask);
1229 if (of_property_read_bool(np, "dmacap,interrupt"))
1230 dma_cap_set(DMA_INTERRUPT, cap_mask);
1232 irq = irq_of_parse_and_map(np, 0);
1235 goto err_channel_add;
1238 chan = mv_xor_channel_add(xordev, pdev, i,
1241 ret = PTR_ERR(chan);
1242 irq_dispose_mapping(irq);
1243 goto err_channel_add;
1246 xordev->channels[i] = chan;
1249 } else if (pdata && pdata->channels) {
1250 for (i = 0; i < MV_XOR_MAX_CHANNELS; i++) {
1251 struct mv_xor_channel_data *cd;
1252 struct mv_xor_chan *chan;
1255 cd = &pdata->channels[i];
1258 goto err_channel_add;
1261 irq = platform_get_irq(pdev, i);
1264 goto err_channel_add;
1267 chan = mv_xor_channel_add(xordev, pdev, i,
1270 ret = PTR_ERR(chan);
1271 goto err_channel_add;
1274 xordev->channels[i] = chan;
1281 for (i = 0; i < MV_XOR_MAX_CHANNELS; i++)
1282 if (xordev->channels[i]) {
1283 mv_xor_channel_remove(xordev->channels[i]);
1284 if (pdev->dev.of_node)
1285 irq_dispose_mapping(xordev->channels[i]->irq);
1288 if (!IS_ERR(xordev->clk)) {
1289 clk_disable_unprepare(xordev->clk);
1290 clk_put(xordev->clk);
1296 static int mv_xor_remove(struct platform_device *pdev)
1298 struct mv_xor_device *xordev = platform_get_drvdata(pdev);
1301 for (i = 0; i < MV_XOR_MAX_CHANNELS; i++) {
1302 if (xordev->channels[i])
1303 mv_xor_channel_remove(xordev->channels[i]);
1306 if (!IS_ERR(xordev->clk)) {
1307 clk_disable_unprepare(xordev->clk);
1308 clk_put(xordev->clk);
1315 static struct of_device_id mv_xor_dt_ids[] = {
1316 { .compatible = "marvell,orion-xor", },
1319 MODULE_DEVICE_TABLE(of, mv_xor_dt_ids);
1322 static struct platform_driver mv_xor_driver = {
1323 .probe = mv_xor_probe,
1324 .remove = mv_xor_remove,
1326 .owner = THIS_MODULE,
1327 .name = MV_XOR_NAME,
1328 .of_match_table = of_match_ptr(mv_xor_dt_ids),
1333 static int __init mv_xor_init(void)
1335 return platform_driver_register(&mv_xor_driver);
1337 module_init(mv_xor_init);
1339 /* it's currently unsafe to unload this module */
1341 static void __exit mv_xor_exit(void)
1343 platform_driver_unregister(&mv_xor_driver);
1347 module_exit(mv_xor_exit);
1350 MODULE_AUTHOR("Saeed Bishara <saeed@marvell.com>");
1351 MODULE_DESCRIPTION("DMA engine driver for Marvell's XOR engine");
1352 MODULE_LICENSE("GPL");