2 * Renesas SuperH DMA Engine support
4 * base is drivers/dma/flsdma.c
6 * Copyright (C) 2009 Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
7 * Copyright (C) 2009 Renesas Solutions, Inc. All rights reserved.
8 * Copyright (C) 2007 Freescale Semiconductor, Inc. All rights reserved.
10 * This is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * - DMA of SuperH does not have Hardware DMA chain mode.
16 * - MAX DMA size is 16MB.
20 #include <linux/init.h>
21 #include <linux/module.h>
22 #include <linux/slab.h>
23 #include <linux/interrupt.h>
24 #include <linux/dmaengine.h>
25 #include <linux/delay.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/platform_device.h>
28 #include <linux/pm_runtime.h>
29 #include <linux/sh_dma.h>
30 #include <linux/notifier.h>
31 #include <linux/kdebug.h>
32 #include <linux/spinlock.h>
33 #include <linux/rculist.h>
36 /* DMA descriptor control */
37 enum sh_dmae_desc_status {
41 DESC_COMPLETED, /* completed, have to call callback */
42 DESC_WAITING, /* callback called, waiting for ack / re-submit */
45 #define NR_DESCS_PER_CHANNEL 32
46 /* Default MEMCPY transfer size = 2^2 = 4 bytes */
47 #define LOG2_DEFAULT_XFER_SIZE 2
50 * Used for write-side mutual exclusion for the global device list,
51 * read-side synchronization by way of RCU, and per-controller data.
53 static DEFINE_SPINLOCK(sh_dmae_lock);
54 static LIST_HEAD(sh_dmae_devices);
56 /* A bitmask with bits enough for enum sh_dmae_slave_chan_id */
57 static unsigned long sh_dmae_slave_used[BITS_TO_LONGS(SH_DMA_SLAVE_NUMBER)];
59 static void sh_dmae_chan_ld_cleanup(struct sh_dmae_chan *sh_chan, bool all);
61 static void sh_dmae_writel(struct sh_dmae_chan *sh_dc, u32 data, u32 reg)
63 __raw_writel(data, sh_dc->base + reg / sizeof(u32));
66 static u32 sh_dmae_readl(struct sh_dmae_chan *sh_dc, u32 reg)
68 return __raw_readl(sh_dc->base + reg / sizeof(u32));
71 static u16 dmaor_read(struct sh_dmae_device *shdev)
73 return __raw_readw(shdev->chan_reg + DMAOR / sizeof(u32));
76 static void dmaor_write(struct sh_dmae_device *shdev, u16 data)
78 __raw_writew(data, shdev->chan_reg + DMAOR / sizeof(u32));
81 static void chcr_write(struct sh_dmae_chan *sh_dc, u32 data)
83 struct sh_dmae_device *shdev = to_sh_dev(sh_dc);
85 __raw_writel(data, sh_dc->base + shdev->chcr_offset / sizeof(u32));
88 static u32 chcr_read(struct sh_dmae_chan *sh_dc)
90 struct sh_dmae_device *shdev = to_sh_dev(sh_dc);
92 return __raw_readl(sh_dc->base + shdev->chcr_offset / sizeof(u32));
96 * Reset DMA controller
98 * SH7780 has two DMAOR register
100 static void sh_dmae_ctl_stop(struct sh_dmae_device *shdev)
102 unsigned short dmaor;
105 spin_lock_irqsave(&sh_dmae_lock, flags);
107 dmaor = dmaor_read(shdev);
108 dmaor_write(shdev, dmaor & ~(DMAOR_NMIF | DMAOR_AE | DMAOR_DME));
110 spin_unlock_irqrestore(&sh_dmae_lock, flags);
113 static int sh_dmae_rst(struct sh_dmae_device *shdev)
115 unsigned short dmaor;
118 spin_lock_irqsave(&sh_dmae_lock, flags);
120 dmaor = dmaor_read(shdev) & ~(DMAOR_NMIF | DMAOR_AE | DMAOR_DME);
122 dmaor_write(shdev, dmaor | shdev->pdata->dmaor_init);
124 dmaor = dmaor_read(shdev);
126 spin_unlock_irqrestore(&sh_dmae_lock, flags);
128 if (dmaor & (DMAOR_AE | DMAOR_NMIF)) {
129 dev_warn(shdev->common.dev, "Can't initialize DMAOR.\n");
135 static bool dmae_is_busy(struct sh_dmae_chan *sh_chan)
137 u32 chcr = chcr_read(sh_chan);
139 if ((chcr & (CHCR_DE | CHCR_TE)) == CHCR_DE)
140 return true; /* working */
142 return false; /* waiting */
145 static unsigned int calc_xmit_shift(struct sh_dmae_chan *sh_chan, u32 chcr)
147 struct sh_dmae_device *shdev = to_sh_dev(sh_chan);
148 struct sh_dmae_pdata *pdata = shdev->pdata;
149 int cnt = ((chcr & pdata->ts_low_mask) >> pdata->ts_low_shift) |
150 ((chcr & pdata->ts_high_mask) >> pdata->ts_high_shift);
152 if (cnt >= pdata->ts_shift_num)
155 return pdata->ts_shift[cnt];
158 static u32 log2size_to_chcr(struct sh_dmae_chan *sh_chan, int l2size)
160 struct sh_dmae_device *shdev = to_sh_dev(sh_chan);
161 struct sh_dmae_pdata *pdata = shdev->pdata;
164 for (i = 0; i < pdata->ts_shift_num; i++)
165 if (pdata->ts_shift[i] == l2size)
168 if (i == pdata->ts_shift_num)
171 return ((i << pdata->ts_low_shift) & pdata->ts_low_mask) |
172 ((i << pdata->ts_high_shift) & pdata->ts_high_mask);
175 static void dmae_set_reg(struct sh_dmae_chan *sh_chan, struct sh_dmae_regs *hw)
177 sh_dmae_writel(sh_chan, hw->sar, SAR);
178 sh_dmae_writel(sh_chan, hw->dar, DAR);
179 sh_dmae_writel(sh_chan, hw->tcr >> sh_chan->xmit_shift, TCR);
182 static void dmae_start(struct sh_dmae_chan *sh_chan)
184 u32 chcr = chcr_read(sh_chan);
186 chcr |= CHCR_DE | CHCR_IE;
187 chcr_write(sh_chan, chcr & ~CHCR_TE);
190 static void dmae_halt(struct sh_dmae_chan *sh_chan)
192 u32 chcr = chcr_read(sh_chan);
194 chcr &= ~(CHCR_DE | CHCR_TE | CHCR_IE);
195 chcr_write(sh_chan, chcr);
198 static void dmae_init(struct sh_dmae_chan *sh_chan)
201 * Default configuration for dual address memory-memory transfer.
202 * 0x400 represents auto-request.
204 u32 chcr = DM_INC | SM_INC | 0x400 | log2size_to_chcr(sh_chan,
205 LOG2_DEFAULT_XFER_SIZE);
206 sh_chan->xmit_shift = calc_xmit_shift(sh_chan, chcr);
207 chcr_write(sh_chan, chcr);
210 static int dmae_set_chcr(struct sh_dmae_chan *sh_chan, u32 val)
212 /* If DMA is active, cannot set CHCR. TODO: remove this superfluous check */
213 if (dmae_is_busy(sh_chan))
216 sh_chan->xmit_shift = calc_xmit_shift(sh_chan, val);
217 chcr_write(sh_chan, val);
222 static int dmae_set_dmars(struct sh_dmae_chan *sh_chan, u16 val)
224 struct sh_dmae_device *shdev = to_sh_dev(sh_chan);
225 struct sh_dmae_pdata *pdata = shdev->pdata;
226 const struct sh_dmae_channel *chan_pdata = &pdata->channel[sh_chan->id];
227 u16 __iomem *addr = shdev->dmars;
228 unsigned int shift = chan_pdata->dmars_bit;
230 if (dmae_is_busy(sh_chan))
233 /* in the case of a missing DMARS resource use first memory window */
235 addr = (u16 __iomem *)shdev->chan_reg;
236 addr += chan_pdata->dmars / sizeof(u16);
238 __raw_writew((__raw_readw(addr) & (0xff00 >> shift)) | (val << shift),
244 static dma_cookie_t sh_dmae_tx_submit(struct dma_async_tx_descriptor *tx)
246 struct sh_desc *desc = tx_to_sh_desc(tx), *chunk, *last = desc, *c;
247 struct sh_dmae_chan *sh_chan = to_sh_chan(tx->chan);
248 dma_async_tx_callback callback = tx->callback;
251 spin_lock_bh(&sh_chan->desc_lock);
253 cookie = sh_chan->common.cookie;
258 sh_chan->common.cookie = cookie;
261 /* Mark all chunks of this descriptor as submitted, move to the queue */
262 list_for_each_entry_safe(chunk, c, desc->node.prev, node) {
264 * All chunks are on the global ld_free, so, we have to find
265 * the end of the chain ourselves
267 if (chunk != desc && (chunk->mark == DESC_IDLE ||
268 chunk->async_tx.cookie > 0 ||
269 chunk->async_tx.cookie == -EBUSY ||
270 &chunk->node == &sh_chan->ld_free))
272 chunk->mark = DESC_SUBMITTED;
273 /* Callback goes to the last chunk */
274 chunk->async_tx.callback = NULL;
275 chunk->cookie = cookie;
276 list_move_tail(&chunk->node, &sh_chan->ld_queue);
280 last->async_tx.callback = callback;
281 last->async_tx.callback_param = tx->callback_param;
283 dev_dbg(sh_chan->dev, "submit #%d@%p on %d: %x[%d] -> %x\n",
284 tx->cookie, &last->async_tx, sh_chan->id,
285 desc->hw.sar, desc->hw.tcr, desc->hw.dar);
287 spin_unlock_bh(&sh_chan->desc_lock);
292 /* Called with desc_lock held */
293 static struct sh_desc *sh_dmae_get_desc(struct sh_dmae_chan *sh_chan)
295 struct sh_desc *desc;
297 list_for_each_entry(desc, &sh_chan->ld_free, node)
298 if (desc->mark != DESC_PREPARED) {
299 BUG_ON(desc->mark != DESC_IDLE);
300 list_del(&desc->node);
307 static const struct sh_dmae_slave_config *sh_dmae_find_slave(
308 struct sh_dmae_chan *sh_chan, struct sh_dmae_slave *param)
310 struct sh_dmae_device *shdev = to_sh_dev(sh_chan);
311 struct sh_dmae_pdata *pdata = shdev->pdata;
314 if (param->slave_id >= SH_DMA_SLAVE_NUMBER)
317 for (i = 0; i < pdata->slave_num; i++)
318 if (pdata->slave[i].slave_id == param->slave_id)
319 return pdata->slave + i;
324 static int sh_dmae_alloc_chan_resources(struct dma_chan *chan)
326 struct sh_dmae_chan *sh_chan = to_sh_chan(chan);
327 struct sh_desc *desc;
328 struct sh_dmae_slave *param = chan->private;
331 pm_runtime_get_sync(sh_chan->dev);
334 * This relies on the guarantee from dmaengine that alloc_chan_resources
335 * never runs concurrently with itself or free_chan_resources.
338 const struct sh_dmae_slave_config *cfg;
340 cfg = sh_dmae_find_slave(sh_chan, param);
346 if (test_and_set_bit(param->slave_id, sh_dmae_slave_used)) {
353 dmae_set_dmars(sh_chan, cfg->mid_rid);
354 dmae_set_chcr(sh_chan, cfg->chcr);
359 spin_lock_bh(&sh_chan->desc_lock);
360 while (sh_chan->descs_allocated < NR_DESCS_PER_CHANNEL) {
361 spin_unlock_bh(&sh_chan->desc_lock);
362 desc = kzalloc(sizeof(struct sh_desc), GFP_KERNEL);
364 spin_lock_bh(&sh_chan->desc_lock);
367 dma_async_tx_descriptor_init(&desc->async_tx,
369 desc->async_tx.tx_submit = sh_dmae_tx_submit;
370 desc->mark = DESC_IDLE;
372 spin_lock_bh(&sh_chan->desc_lock);
373 list_add(&desc->node, &sh_chan->ld_free);
374 sh_chan->descs_allocated++;
376 spin_unlock_bh(&sh_chan->desc_lock);
378 if (!sh_chan->descs_allocated) {
383 return sh_chan->descs_allocated;
387 clear_bit(param->slave_id, sh_dmae_slave_used);
390 pm_runtime_put(sh_chan->dev);
395 * sh_dma_free_chan_resources - Free all resources of the channel.
397 static void sh_dmae_free_chan_resources(struct dma_chan *chan)
399 struct sh_dmae_chan *sh_chan = to_sh_chan(chan);
400 struct sh_desc *desc, *_desc;
402 int descs = sh_chan->descs_allocated;
404 /* Protect against ISR */
405 spin_lock_irq(&sh_chan->desc_lock);
407 spin_unlock_irq(&sh_chan->desc_lock);
409 /* Now no new interrupts will occur */
411 /* Prepared and not submitted descriptors can still be on the queue */
412 if (!list_empty(&sh_chan->ld_queue))
413 sh_dmae_chan_ld_cleanup(sh_chan, true);
416 /* The caller is holding dma_list_mutex */
417 struct sh_dmae_slave *param = chan->private;
418 clear_bit(param->slave_id, sh_dmae_slave_used);
419 chan->private = NULL;
422 spin_lock_bh(&sh_chan->desc_lock);
424 list_splice_init(&sh_chan->ld_free, &list);
425 sh_chan->descs_allocated = 0;
427 spin_unlock_bh(&sh_chan->desc_lock);
430 pm_runtime_put(sh_chan->dev);
432 list_for_each_entry_safe(desc, _desc, &list, node)
437 * sh_dmae_add_desc - get, set up and return one transfer descriptor
438 * @sh_chan: DMA channel
439 * @flags: DMA transfer flags
440 * @dest: destination DMA address, incremented when direction equals
441 * DMA_FROM_DEVICE or DMA_BIDIRECTIONAL
442 * @src: source DMA address, incremented when direction equals
443 * DMA_TO_DEVICE or DMA_BIDIRECTIONAL
444 * @len: DMA transfer length
445 * @first: if NULL, set to the current descriptor and cookie set to -EBUSY
446 * @direction: needed for slave DMA to decide which address to keep constant,
447 * equals DMA_BIDIRECTIONAL for MEMCPY
448 * Returns 0 or an error
449 * Locks: called with desc_lock held
451 static struct sh_desc *sh_dmae_add_desc(struct sh_dmae_chan *sh_chan,
452 unsigned long flags, dma_addr_t *dest, dma_addr_t *src, size_t *len,
453 struct sh_desc **first, enum dma_data_direction direction)
461 /* Allocate the link descriptor from the free list */
462 new = sh_dmae_get_desc(sh_chan);
464 dev_err(sh_chan->dev, "No free link descriptor available\n");
468 copy_size = min(*len, (size_t)SH_DMA_TCR_MAX + 1);
472 new->hw.tcr = copy_size;
476 new->async_tx.cookie = -EBUSY;
479 /* Other desc - invisible to the user */
480 new->async_tx.cookie = -EINVAL;
483 dev_dbg(sh_chan->dev,
484 "chaining (%u/%u)@%x -> %x with %p, cookie %d, shift %d\n",
485 copy_size, *len, *src, *dest, &new->async_tx,
486 new->async_tx.cookie, sh_chan->xmit_shift);
488 new->mark = DESC_PREPARED;
489 new->async_tx.flags = flags;
490 new->direction = direction;
493 if (direction == DMA_BIDIRECTIONAL || direction == DMA_TO_DEVICE)
495 if (direction == DMA_BIDIRECTIONAL || direction == DMA_FROM_DEVICE)
502 * sh_dmae_prep_sg - prepare transfer descriptors from an SG list
504 * Common routine for public (MEMCPY) and slave DMA. The MEMCPY case is also
505 * converted to scatter-gather to guarantee consistent locking and a correct
506 * list manipulation. For slave DMA direction carries the usual meaning, and,
507 * logically, the SG list is RAM and the addr variable contains slave address,
508 * e.g., the FIFO I/O register. For MEMCPY direction equals DMA_BIDIRECTIONAL
509 * and the SG list contains only one element and points at the source buffer.
511 static struct dma_async_tx_descriptor *sh_dmae_prep_sg(struct sh_dmae_chan *sh_chan,
512 struct scatterlist *sgl, unsigned int sg_len, dma_addr_t *addr,
513 enum dma_data_direction direction, unsigned long flags)
515 struct scatterlist *sg;
516 struct sh_desc *first = NULL, *new = NULL /* compiler... */;
524 for_each_sg(sgl, sg, sg_len, i)
525 chunks += (sg_dma_len(sg) + SH_DMA_TCR_MAX) /
526 (SH_DMA_TCR_MAX + 1);
528 /* Have to lock the whole loop to protect against concurrent release */
529 spin_lock_bh(&sh_chan->desc_lock);
533 * first descriptor is what user is dealing with in all API calls, its
534 * cookie is at first set to -EBUSY, at tx-submit to a positive
536 * if more than one chunk is needed further chunks have cookie = -EINVAL
537 * the last chunk, if not equal to the first, has cookie = -ENOSPC
538 * all chunks are linked onto the tx_list head with their .node heads
539 * only during this function, then they are immediately spliced
540 * back onto the free list in form of a chain
542 for_each_sg(sgl, sg, sg_len, i) {
543 dma_addr_t sg_addr = sg_dma_address(sg);
544 size_t len = sg_dma_len(sg);
550 dev_dbg(sh_chan->dev, "Add SG #%d@%p[%d], dma %llx\n",
551 i, sg, len, (unsigned long long)sg_addr);
553 if (direction == DMA_FROM_DEVICE)
554 new = sh_dmae_add_desc(sh_chan, flags,
555 &sg_addr, addr, &len, &first,
558 new = sh_dmae_add_desc(sh_chan, flags,
559 addr, &sg_addr, &len, &first,
564 new->chunks = chunks--;
565 list_add_tail(&new->node, &tx_list);
570 new->async_tx.cookie = -ENOSPC;
572 /* Put them back on the free list, so, they don't get lost */
573 list_splice_tail(&tx_list, &sh_chan->ld_free);
575 spin_unlock_bh(&sh_chan->desc_lock);
577 return &first->async_tx;
580 list_for_each_entry(new, &tx_list, node)
581 new->mark = DESC_IDLE;
582 list_splice(&tx_list, &sh_chan->ld_free);
584 spin_unlock_bh(&sh_chan->desc_lock);
589 static struct dma_async_tx_descriptor *sh_dmae_prep_memcpy(
590 struct dma_chan *chan, dma_addr_t dma_dest, dma_addr_t dma_src,
591 size_t len, unsigned long flags)
593 struct sh_dmae_chan *sh_chan;
594 struct scatterlist sg;
599 sh_chan = to_sh_chan(chan);
601 sg_init_table(&sg, 1);
602 sg_set_page(&sg, pfn_to_page(PFN_DOWN(dma_src)), len,
603 offset_in_page(dma_src));
604 sg_dma_address(&sg) = dma_src;
605 sg_dma_len(&sg) = len;
607 return sh_dmae_prep_sg(sh_chan, &sg, 1, &dma_dest, DMA_BIDIRECTIONAL,
611 static struct dma_async_tx_descriptor *sh_dmae_prep_slave_sg(
612 struct dma_chan *chan, struct scatterlist *sgl, unsigned int sg_len,
613 enum dma_data_direction direction, unsigned long flags)
615 struct sh_dmae_slave *param;
616 struct sh_dmae_chan *sh_chan;
617 dma_addr_t slave_addr;
622 sh_chan = to_sh_chan(chan);
623 param = chan->private;
625 /* Someone calling slave DMA on a public channel? */
626 if (!param || !sg_len) {
627 dev_warn(sh_chan->dev, "%s: bad parameter: %p, %d, %d\n",
628 __func__, param, sg_len, param ? param->slave_id : -1);
632 slave_addr = param->config->addr;
635 * if (param != NULL), this is a successfully requested slave channel,
636 * therefore param->config != NULL too.
638 return sh_dmae_prep_sg(sh_chan, sgl, sg_len, &slave_addr,
642 static int sh_dmae_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
645 struct sh_dmae_chan *sh_chan = to_sh_chan(chan);
647 /* Only supports DMA_TERMINATE_ALL */
648 if (cmd != DMA_TERMINATE_ALL)
654 spin_lock_bh(&sh_chan->desc_lock);
657 if (!list_empty(&sh_chan->ld_queue)) {
658 /* Record partial transfer */
659 struct sh_desc *desc = list_entry(sh_chan->ld_queue.next,
660 struct sh_desc, node);
661 desc->partial = (desc->hw.tcr - sh_dmae_readl(sh_chan, TCR)) <<
665 spin_unlock_bh(&sh_chan->desc_lock);
667 sh_dmae_chan_ld_cleanup(sh_chan, true);
672 static dma_async_tx_callback __ld_cleanup(struct sh_dmae_chan *sh_chan, bool all)
674 struct sh_desc *desc, *_desc;
675 /* Is the "exposed" head of a chain acked? */
676 bool head_acked = false;
677 dma_cookie_t cookie = 0;
678 dma_async_tx_callback callback = NULL;
681 spin_lock_bh(&sh_chan->desc_lock);
682 list_for_each_entry_safe(desc, _desc, &sh_chan->ld_queue, node) {
683 struct dma_async_tx_descriptor *tx = &desc->async_tx;
685 BUG_ON(tx->cookie > 0 && tx->cookie != desc->cookie);
686 BUG_ON(desc->mark != DESC_SUBMITTED &&
687 desc->mark != DESC_COMPLETED &&
688 desc->mark != DESC_WAITING);
691 * queue is ordered, and we use this loop to (1) clean up all
692 * completed descriptors, and to (2) update descriptor flags of
693 * any chunks in a (partially) completed chain
695 if (!all && desc->mark == DESC_SUBMITTED &&
696 desc->cookie != cookie)
702 if (desc->mark == DESC_COMPLETED && desc->chunks == 1) {
703 if (sh_chan->completed_cookie != desc->cookie - 1)
704 dev_dbg(sh_chan->dev,
705 "Completing cookie %d, expected %d\n",
707 sh_chan->completed_cookie + 1);
708 sh_chan->completed_cookie = desc->cookie;
711 /* Call callback on the last chunk */
712 if (desc->mark == DESC_COMPLETED && tx->callback) {
713 desc->mark = DESC_WAITING;
714 callback = tx->callback;
715 param = tx->callback_param;
716 dev_dbg(sh_chan->dev, "descriptor #%d@%p on %d callback\n",
717 tx->cookie, tx, sh_chan->id);
718 BUG_ON(desc->chunks != 1);
722 if (tx->cookie > 0 || tx->cookie == -EBUSY) {
723 if (desc->mark == DESC_COMPLETED) {
724 BUG_ON(tx->cookie < 0);
725 desc->mark = DESC_WAITING;
727 head_acked = async_tx_test_ack(tx);
729 switch (desc->mark) {
731 desc->mark = DESC_WAITING;
735 async_tx_ack(&desc->async_tx);
739 dev_dbg(sh_chan->dev, "descriptor %p #%d completed.\n",
742 if (((desc->mark == DESC_COMPLETED ||
743 desc->mark == DESC_WAITING) &&
744 async_tx_test_ack(&desc->async_tx)) || all) {
745 /* Remove from ld_queue list */
746 desc->mark = DESC_IDLE;
747 list_move(&desc->node, &sh_chan->ld_free);
751 if (all && !callback)
753 * Terminating and the loop completed normally: forgive
754 * uncompleted cookies
756 sh_chan->completed_cookie = sh_chan->common.cookie;
758 spin_unlock_bh(&sh_chan->desc_lock);
767 * sh_chan_ld_cleanup - Clean up link descriptors
769 * This function cleans up the ld_queue of DMA channel.
771 static void sh_dmae_chan_ld_cleanup(struct sh_dmae_chan *sh_chan, bool all)
773 while (__ld_cleanup(sh_chan, all))
777 static void sh_chan_xfer_ld_queue(struct sh_dmae_chan *sh_chan)
779 struct sh_desc *desc;
781 spin_lock_bh(&sh_chan->desc_lock);
783 if (dmae_is_busy(sh_chan))
784 goto sh_chan_xfer_ld_queue_end;
786 /* Find the first not transferred descriptor */
787 list_for_each_entry(desc, &sh_chan->ld_queue, node)
788 if (desc->mark == DESC_SUBMITTED) {
789 dev_dbg(sh_chan->dev, "Queue #%d to %d: %u@%x -> %x\n",
790 desc->async_tx.cookie, sh_chan->id,
791 desc->hw.tcr, desc->hw.sar, desc->hw.dar);
792 /* Get the ld start address from ld_queue */
793 dmae_set_reg(sh_chan, &desc->hw);
798 sh_chan_xfer_ld_queue_end:
799 spin_unlock_bh(&sh_chan->desc_lock);
802 static void sh_dmae_memcpy_issue_pending(struct dma_chan *chan)
804 struct sh_dmae_chan *sh_chan = to_sh_chan(chan);
805 sh_chan_xfer_ld_queue(sh_chan);
808 static enum dma_status sh_dmae_tx_status(struct dma_chan *chan,
810 struct dma_tx_state *txstate)
812 struct sh_dmae_chan *sh_chan = to_sh_chan(chan);
813 dma_cookie_t last_used;
814 dma_cookie_t last_complete;
815 enum dma_status status;
817 sh_dmae_chan_ld_cleanup(sh_chan, false);
819 /* First read completed cookie to avoid a skew */
820 last_complete = sh_chan->completed_cookie;
822 last_used = chan->cookie;
823 BUG_ON(last_complete < 0);
824 dma_set_tx_state(txstate, last_complete, last_used, 0);
826 spin_lock_bh(&sh_chan->desc_lock);
828 status = dma_async_is_complete(cookie, last_complete, last_used);
831 * If we don't find cookie on the queue, it has been aborted and we have
834 if (status != DMA_SUCCESS) {
835 struct sh_desc *desc;
837 list_for_each_entry(desc, &sh_chan->ld_queue, node)
838 if (desc->cookie == cookie) {
839 status = DMA_IN_PROGRESS;
844 spin_unlock_bh(&sh_chan->desc_lock);
849 static irqreturn_t sh_dmae_interrupt(int irq, void *data)
851 irqreturn_t ret = IRQ_NONE;
852 struct sh_dmae_chan *sh_chan = data;
855 spin_lock(&sh_chan->desc_lock);
857 chcr = chcr_read(sh_chan);
859 if (chcr & CHCR_TE) {
864 tasklet_schedule(&sh_chan->tasklet);
867 spin_unlock(&sh_chan->desc_lock);
872 /* Called from error IRQ or NMI */
873 static bool sh_dmae_reset(struct sh_dmae_device *shdev)
875 unsigned int handled = 0;
878 /* halt the dma controller */
879 sh_dmae_ctl_stop(shdev);
881 /* We cannot detect, which channel caused the error, have to reset all */
882 for (i = 0; i < SH_DMAC_MAX_CHANNELS; i++) {
883 struct sh_dmae_chan *sh_chan = shdev->chan[i];
884 struct sh_desc *desc;
890 spin_lock(&sh_chan->desc_lock);
892 /* Stop the channel */
895 list_splice_init(&sh_chan->ld_queue, &dl);
897 spin_unlock(&sh_chan->desc_lock);
900 list_for_each_entry(desc, &dl, node) {
901 struct dma_async_tx_descriptor *tx = &desc->async_tx;
902 desc->mark = DESC_IDLE;
904 tx->callback(tx->callback_param);
907 spin_lock(&sh_chan->desc_lock);
908 list_splice(&dl, &sh_chan->ld_free);
909 spin_unlock(&sh_chan->desc_lock);
919 static irqreturn_t sh_dmae_err(int irq, void *data)
921 struct sh_dmae_device *shdev = data;
923 if (!(dmaor_read(shdev) & DMAOR_AE))
930 static void dmae_do_tasklet(unsigned long data)
932 struct sh_dmae_chan *sh_chan = (struct sh_dmae_chan *)data;
933 struct sh_desc *desc;
934 u32 sar_buf = sh_dmae_readl(sh_chan, SAR);
935 u32 dar_buf = sh_dmae_readl(sh_chan, DAR);
937 spin_lock(&sh_chan->desc_lock);
938 list_for_each_entry(desc, &sh_chan->ld_queue, node) {
939 if (desc->mark == DESC_SUBMITTED &&
940 ((desc->direction == DMA_FROM_DEVICE &&
941 (desc->hw.dar + desc->hw.tcr) == dar_buf) ||
942 (desc->hw.sar + desc->hw.tcr) == sar_buf)) {
943 dev_dbg(sh_chan->dev, "done #%d@%p dst %u\n",
944 desc->async_tx.cookie, &desc->async_tx,
946 desc->mark = DESC_COMPLETED;
950 spin_unlock(&sh_chan->desc_lock);
953 sh_chan_xfer_ld_queue(sh_chan);
954 sh_dmae_chan_ld_cleanup(sh_chan, false);
957 static bool sh_dmae_nmi_notify(struct sh_dmae_device *shdev)
959 /* Fast path out if NMIF is not asserted for this controller */
960 if ((dmaor_read(shdev) & DMAOR_NMIF) == 0)
963 return sh_dmae_reset(shdev);
966 static int sh_dmae_nmi_handler(struct notifier_block *self,
967 unsigned long cmd, void *data)
969 struct sh_dmae_device *shdev;
970 int ret = NOTIFY_DONE;
974 * Only concern ourselves with NMI events.
976 * Normally we would check the die chain value, but as this needs
977 * to be architecture independent, check for NMI context instead.
983 list_for_each_entry_rcu(shdev, &sh_dmae_devices, node) {
985 * Only stop if one of the controllers has NMIF asserted,
986 * we do not want to interfere with regular address error
987 * handling or NMI events that don't concern the DMACs.
989 triggered = sh_dmae_nmi_notify(shdev);
990 if (triggered == true)
998 static struct notifier_block sh_dmae_nmi_notifier __read_mostly = {
999 .notifier_call = sh_dmae_nmi_handler,
1001 /* Run before NMI debug handler and KGDB */
1005 static int __devinit sh_dmae_chan_probe(struct sh_dmae_device *shdev, int id,
1006 int irq, unsigned long flags)
1009 const struct sh_dmae_channel *chan_pdata = &shdev->pdata->channel[id];
1010 struct platform_device *pdev = to_platform_device(shdev->common.dev);
1011 struct sh_dmae_chan *new_sh_chan;
1014 new_sh_chan = kzalloc(sizeof(struct sh_dmae_chan), GFP_KERNEL);
1016 dev_err(shdev->common.dev,
1017 "No free memory for allocating dma channels!\n");
1021 /* copy struct dma_device */
1022 new_sh_chan->common.device = &shdev->common;
1024 new_sh_chan->dev = shdev->common.dev;
1025 new_sh_chan->id = id;
1026 new_sh_chan->irq = irq;
1027 new_sh_chan->base = shdev->chan_reg + chan_pdata->offset / sizeof(u32);
1029 /* Init DMA tasklet */
1030 tasklet_init(&new_sh_chan->tasklet, dmae_do_tasklet,
1031 (unsigned long)new_sh_chan);
1033 spin_lock_init(&new_sh_chan->desc_lock);
1035 /* Init descripter manage list */
1036 INIT_LIST_HEAD(&new_sh_chan->ld_queue);
1037 INIT_LIST_HEAD(&new_sh_chan->ld_free);
1039 /* Add the channel to DMA device channel list */
1040 list_add_tail(&new_sh_chan->common.device_node,
1041 &shdev->common.channels);
1042 shdev->common.chancnt++;
1045 snprintf(new_sh_chan->dev_id, sizeof(new_sh_chan->dev_id),
1046 "sh-dmae%d.%d", pdev->id, new_sh_chan->id);
1048 snprintf(new_sh_chan->dev_id, sizeof(new_sh_chan->dev_id),
1049 "sh-dma%d", new_sh_chan->id);
1051 /* set up channel irq */
1052 err = request_irq(irq, &sh_dmae_interrupt, flags,
1053 new_sh_chan->dev_id, new_sh_chan);
1055 dev_err(shdev->common.dev, "DMA channel %d request_irq error "
1056 "with return %d\n", id, err);
1060 shdev->chan[id] = new_sh_chan;
1064 /* remove from dmaengine device node */
1065 list_del(&new_sh_chan->common.device_node);
1070 static void sh_dmae_chan_remove(struct sh_dmae_device *shdev)
1074 for (i = shdev->common.chancnt - 1 ; i >= 0 ; i--) {
1075 if (shdev->chan[i]) {
1076 struct sh_dmae_chan *sh_chan = shdev->chan[i];
1078 free_irq(sh_chan->irq, sh_chan);
1080 list_del(&sh_chan->common.device_node);
1082 shdev->chan[i] = NULL;
1085 shdev->common.chancnt = 0;
1088 static int __init sh_dmae_probe(struct platform_device *pdev)
1090 struct sh_dmae_pdata *pdata = pdev->dev.platform_data;
1091 unsigned long irqflags = IRQF_DISABLED,
1092 chan_flag[SH_DMAC_MAX_CHANNELS] = {};
1093 int errirq, chan_irq[SH_DMAC_MAX_CHANNELS];
1094 int err, i, irq_cnt = 0, irqres = 0, irq_cap = 0;
1095 struct sh_dmae_device *shdev;
1096 struct resource *chan, *dmars, *errirq_res, *chanirq_res;
1098 /* get platform data */
1099 if (!pdata || !pdata->channel_num)
1102 chan = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1103 /* DMARS area is optional */
1104 dmars = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1107 * 1. there always must be at least one IRQ IO-resource. On SH4 it is
1108 * the error IRQ, in which case it is the only IRQ in this resource:
1109 * start == end. If it is the only IRQ resource, all channels also
1111 * 2. DMA channel IRQ resources can be specified one per resource or in
1112 * ranges (start != end)
1113 * 3. iff all events (channels and, optionally, error) on this
1114 * controller use the same IRQ, only one IRQ resource can be
1115 * specified, otherwise there must be one IRQ per channel, even if
1116 * some of them are equal
1117 * 4. if all IRQs on this controller are equal or if some specific IRQs
1118 * specify IORESOURCE_IRQ_SHAREABLE in their resources, they will be
1119 * requested with the IRQF_SHARED flag
1121 errirq_res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
1122 if (!chan || !errirq_res)
1125 if (!request_mem_region(chan->start, resource_size(chan), pdev->name)) {
1126 dev_err(&pdev->dev, "DMAC register region already claimed\n");
1130 if (dmars && !request_mem_region(dmars->start, resource_size(dmars), pdev->name)) {
1131 dev_err(&pdev->dev, "DMAC DMARS region already claimed\n");
1137 shdev = kzalloc(sizeof(struct sh_dmae_device), GFP_KERNEL);
1139 dev_err(&pdev->dev, "Not enough memory\n");
1143 shdev->chan_reg = ioremap(chan->start, resource_size(chan));
1144 if (!shdev->chan_reg)
1147 shdev->dmars = ioremap(dmars->start, resource_size(dmars));
1153 shdev->pdata = pdata;
1155 if (pdata->chcr_offset)
1156 shdev->chcr_offset = pdata->chcr_offset;
1158 shdev->chcr_offset = CHCR;
1160 platform_set_drvdata(pdev, shdev);
1162 pm_runtime_enable(&pdev->dev);
1163 pm_runtime_get_sync(&pdev->dev);
1165 spin_lock_irq(&sh_dmae_lock);
1166 list_add_tail_rcu(&shdev->node, &sh_dmae_devices);
1167 spin_unlock_irq(&sh_dmae_lock);
1169 /* reset dma controller - only needed as a test */
1170 err = sh_dmae_rst(shdev);
1174 INIT_LIST_HEAD(&shdev->common.channels);
1176 dma_cap_set(DMA_MEMCPY, shdev->common.cap_mask);
1177 if (pdata->slave && pdata->slave_num)
1178 dma_cap_set(DMA_SLAVE, shdev->common.cap_mask);
1180 shdev->common.device_alloc_chan_resources
1181 = sh_dmae_alloc_chan_resources;
1182 shdev->common.device_free_chan_resources = sh_dmae_free_chan_resources;
1183 shdev->common.device_prep_dma_memcpy = sh_dmae_prep_memcpy;
1184 shdev->common.device_tx_status = sh_dmae_tx_status;
1185 shdev->common.device_issue_pending = sh_dmae_memcpy_issue_pending;
1187 /* Compulsory for DMA_SLAVE fields */
1188 shdev->common.device_prep_slave_sg = sh_dmae_prep_slave_sg;
1189 shdev->common.device_control = sh_dmae_control;
1191 shdev->common.dev = &pdev->dev;
1192 /* Default transfer size of 32 bytes requires 32-byte alignment */
1193 shdev->common.copy_align = LOG2_DEFAULT_XFER_SIZE;
1195 #if defined(CONFIG_CPU_SH4) || defined(CONFIG_ARCH_SHMOBILE)
1196 chanirq_res = platform_get_resource(pdev, IORESOURCE_IRQ, 1);
1199 chanirq_res = errirq_res;
1203 if (chanirq_res == errirq_res ||
1204 (errirq_res->flags & IORESOURCE_BITS) == IORESOURCE_IRQ_SHAREABLE)
1205 irqflags = IRQF_SHARED;
1207 errirq = errirq_res->start;
1209 err = request_irq(errirq, sh_dmae_err, irqflags,
1210 "DMAC Address Error", shdev);
1213 "DMA failed requesting irq #%d, error %d\n",
1219 chanirq_res = errirq_res;
1220 #endif /* CONFIG_CPU_SH4 || CONFIG_ARCH_SHMOBILE */
1222 if (chanirq_res->start == chanirq_res->end &&
1223 !platform_get_resource(pdev, IORESOURCE_IRQ, 1)) {
1224 /* Special case - all multiplexed */
1225 for (; irq_cnt < pdata->channel_num; irq_cnt++) {
1226 if (irq_cnt < SH_DMAC_MAX_CHANNELS) {
1227 chan_irq[irq_cnt] = chanirq_res->start;
1228 chan_flag[irq_cnt] = IRQF_SHARED;
1236 for (i = chanirq_res->start; i <= chanirq_res->end; i++) {
1237 if ((errirq_res->flags & IORESOURCE_BITS) ==
1238 IORESOURCE_IRQ_SHAREABLE)
1239 chan_flag[irq_cnt] = IRQF_SHARED;
1241 chan_flag[irq_cnt] = IRQF_DISABLED;
1243 "Found IRQ %d for channel %d\n",
1245 chan_irq[irq_cnt++] = i;
1247 if (irq_cnt >= SH_DMAC_MAX_CHANNELS)
1251 if (irq_cnt >= SH_DMAC_MAX_CHANNELS) {
1255 chanirq_res = platform_get_resource(pdev,
1256 IORESOURCE_IRQ, ++irqres);
1257 } while (irq_cnt < pdata->channel_num && chanirq_res);
1260 /* Create DMA Channel */
1261 for (i = 0; i < irq_cnt; i++) {
1262 err = sh_dmae_chan_probe(shdev, i, chan_irq[i], chan_flag[i]);
1264 goto chan_probe_err;
1268 dev_notice(&pdev->dev, "Attempting to register %d DMA "
1269 "channels when a maximum of %d are supported.\n",
1270 pdata->channel_num, SH_DMAC_MAX_CHANNELS);
1272 pm_runtime_put(&pdev->dev);
1274 dma_async_device_register(&shdev->common);
1279 sh_dmae_chan_remove(shdev);
1281 #if defined(CONFIG_CPU_SH4) || defined(CONFIG_ARCH_SHMOBILE)
1282 free_irq(errirq, shdev);
1286 spin_lock_irq(&sh_dmae_lock);
1287 list_del_rcu(&shdev->node);
1288 spin_unlock_irq(&sh_dmae_lock);
1290 pm_runtime_put(&pdev->dev);
1291 pm_runtime_disable(&pdev->dev);
1294 iounmap(shdev->dmars);
1296 platform_set_drvdata(pdev, NULL);
1298 iounmap(shdev->chan_reg);
1304 release_mem_region(dmars->start, resource_size(dmars));
1306 release_mem_region(chan->start, resource_size(chan));
1311 static int __exit sh_dmae_remove(struct platform_device *pdev)
1313 struct sh_dmae_device *shdev = platform_get_drvdata(pdev);
1314 struct resource *res;
1315 int errirq = platform_get_irq(pdev, 0);
1317 dma_async_device_unregister(&shdev->common);
1320 free_irq(errirq, shdev);
1322 spin_lock_irq(&sh_dmae_lock);
1323 list_del_rcu(&shdev->node);
1324 spin_unlock_irq(&sh_dmae_lock);
1326 /* channel data remove */
1327 sh_dmae_chan_remove(shdev);
1329 pm_runtime_disable(&pdev->dev);
1332 iounmap(shdev->dmars);
1333 iounmap(shdev->chan_reg);
1335 platform_set_drvdata(pdev, NULL);
1340 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1342 release_mem_region(res->start, resource_size(res));
1343 res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1345 release_mem_region(res->start, resource_size(res));
1350 static void sh_dmae_shutdown(struct platform_device *pdev)
1352 struct sh_dmae_device *shdev = platform_get_drvdata(pdev);
1353 sh_dmae_ctl_stop(shdev);
1356 static int sh_dmae_runtime_suspend(struct device *dev)
1361 static int sh_dmae_runtime_resume(struct device *dev)
1363 struct sh_dmae_device *shdev = dev_get_drvdata(dev);
1365 return sh_dmae_rst(shdev);
1369 static int sh_dmae_suspend(struct device *dev)
1371 struct sh_dmae_device *shdev = dev_get_drvdata(dev);
1374 for (i = 0; i < shdev->pdata->channel_num; i++) {
1375 struct sh_dmae_chan *sh_chan = shdev->chan[i];
1376 if (sh_chan->descs_allocated)
1377 sh_chan->pm_error = pm_runtime_put_sync(dev);
1383 static int sh_dmae_resume(struct device *dev)
1385 struct sh_dmae_device *shdev = dev_get_drvdata(dev);
1388 for (i = 0; i < shdev->pdata->channel_num; i++) {
1389 struct sh_dmae_chan *sh_chan = shdev->chan[i];
1390 struct sh_dmae_slave *param = sh_chan->common.private;
1392 if (!sh_chan->descs_allocated)
1395 if (!sh_chan->pm_error)
1396 pm_runtime_get_sync(dev);
1399 const struct sh_dmae_slave_config *cfg = param->config;
1400 dmae_set_dmars(sh_chan, cfg->mid_rid);
1401 dmae_set_chcr(sh_chan, cfg->chcr);
1410 #define sh_dmae_suspend NULL
1411 #define sh_dmae_resume NULL
1414 const struct dev_pm_ops sh_dmae_pm = {
1415 .suspend = sh_dmae_suspend,
1416 .resume = sh_dmae_resume,
1417 .runtime_suspend = sh_dmae_runtime_suspend,
1418 .runtime_resume = sh_dmae_runtime_resume,
1421 static struct platform_driver sh_dmae_driver = {
1422 .remove = __exit_p(sh_dmae_remove),
1423 .shutdown = sh_dmae_shutdown,
1425 .owner = THIS_MODULE,
1426 .name = "sh-dma-engine",
1431 static int __init sh_dmae_init(void)
1433 /* Wire up NMI handling */
1434 int err = register_die_notifier(&sh_dmae_nmi_notifier);
1438 return platform_driver_probe(&sh_dmae_driver, sh_dmae_probe);
1440 module_init(sh_dmae_init);
1442 static void __exit sh_dmae_exit(void)
1444 platform_driver_unregister(&sh_dmae_driver);
1446 unregister_die_notifier(&sh_dmae_nmi_notifier);
1448 module_exit(sh_dmae_exit);
1450 MODULE_AUTHOR("Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>");
1451 MODULE_DESCRIPTION("Renesas SH DMA Engine driver");
1452 MODULE_LICENSE("GPL");
1453 MODULE_ALIAS("platform:sh-dma-engine");