4 * Copyright (c) 2013 Heiko Stuebner <heiko@sntech.de>
6 * based on amba-pl08x.c
8 * Copyright (c) 2006 ARM Ltd.
9 * Copyright (c) 2010 ST-Ericsson SA
11 * Author: Peter Pearse <peter.pearse@arm.com>
12 * Author: Linus Walleij <linus.walleij@stericsson.com>
14 * This program is free software; you can redistribute it and/or modify it
15 * under the terms of the GNU General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option)
19 * The DMA controllers in S3C24XX SoCs have a varying number of DMA signals
20 * that can be routed to any of the 4 to 8 hardware-channels.
22 * Therefore on these DMA controllers the number of channels
23 * and the number of incoming DMA signals are two totally different things.
24 * It is usually not possible to theoretically handle all physical signals,
25 * so a multiplexing scheme with possible denial of use is necessary.
31 #include <linux/platform_device.h>
32 #include <linux/types.h>
33 #include <linux/dmaengine.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/interrupt.h>
36 #include <linux/clk.h>
37 #include <linux/module.h>
38 #include <linux/slab.h>
39 #include <linux/platform_data/dma-s3c24xx.h>
41 #include "dmaengine.h"
44 #define MAX_DMA_CHANNELS 8
46 #define S3C24XX_DISRC 0x00
47 #define S3C24XX_DISRCC 0x04
48 #define S3C24XX_DISRCC_INC_INCREMENT 0
49 #define S3C24XX_DISRCC_INC_FIXED BIT(0)
50 #define S3C24XX_DISRCC_LOC_AHB 0
51 #define S3C24XX_DISRCC_LOC_APB BIT(1)
53 #define S3C24XX_DIDST 0x08
54 #define S3C24XX_DIDSTC 0x0c
55 #define S3C24XX_DIDSTC_INC_INCREMENT 0
56 #define S3C24XX_DIDSTC_INC_FIXED BIT(0)
57 #define S3C24XX_DIDSTC_LOC_AHB 0
58 #define S3C24XX_DIDSTC_LOC_APB BIT(1)
59 #define S3C24XX_DIDSTC_INT_TC0 0
60 #define S3C24XX_DIDSTC_INT_RELOAD BIT(2)
62 #define S3C24XX_DCON 0x10
64 #define S3C24XX_DCON_TC_MASK 0xfffff
65 #define S3C24XX_DCON_DSZ_BYTE (0 << 20)
66 #define S3C24XX_DCON_DSZ_HALFWORD (1 << 20)
67 #define S3C24XX_DCON_DSZ_WORD (2 << 20)
68 #define S3C24XX_DCON_DSZ_MASK (3 << 20)
69 #define S3C24XX_DCON_DSZ_SHIFT 20
70 #define S3C24XX_DCON_AUTORELOAD 0
71 #define S3C24XX_DCON_NORELOAD BIT(22)
72 #define S3C24XX_DCON_HWTRIG BIT(23)
73 #define S3C24XX_DCON_HWSRC_SHIFT 24
74 #define S3C24XX_DCON_SERV_SINGLE 0
75 #define S3C24XX_DCON_SERV_WHOLE BIT(27)
76 #define S3C24XX_DCON_TSZ_UNIT 0
77 #define S3C24XX_DCON_TSZ_BURST4 BIT(28)
78 #define S3C24XX_DCON_INT BIT(29)
79 #define S3C24XX_DCON_SYNC_PCLK 0
80 #define S3C24XX_DCON_SYNC_HCLK BIT(30)
81 #define S3C24XX_DCON_DEMAND 0
82 #define S3C24XX_DCON_HANDSHAKE BIT(31)
84 #define S3C24XX_DSTAT 0x14
85 #define S3C24XX_DSTAT_STAT_BUSY BIT(20)
86 #define S3C24XX_DSTAT_CURRTC_MASK 0xfffff
88 #define S3C24XX_DMASKTRIG 0x20
89 #define S3C24XX_DMASKTRIG_SWTRIG BIT(0)
90 #define S3C24XX_DMASKTRIG_ON BIT(1)
91 #define S3C24XX_DMASKTRIG_STOP BIT(2)
93 #define S3C24XX_DMAREQSEL 0x24
94 #define S3C24XX_DMAREQSEL_HW BIT(0)
97 * S3C2410, S3C2440 and S3C2442 SoCs cannot select any physical channel
98 * for a DMA source. Instead only specific channels are valid.
99 * All of these SoCs have 4 physical channels and the number of request
100 * source bits is 3. Additionally we also need 1 bit to mark the channel
102 * Therefore we separate the chansel element of the channel data into 4
103 * parts of 4 bits each, to hold the information if the channel is valid
104 * and the hw request source to use.
107 * SDI is valid on channels 0, 2 and 3 - with varying hw request sources.
108 * For it the chansel field would look like
110 * ((BIT(3) | 1) << 3 * 4) | // channel 3, with request source 1
111 * ((BIT(3) | 2) << 2 * 4) | // channel 2, with request source 2
112 * ((BIT(3) | 2) << 0 * 4) // channel 0, with request source 2
114 #define S3C24XX_CHANSEL_WIDTH 4
115 #define S3C24XX_CHANSEL_VALID BIT(3)
116 #define S3C24XX_CHANSEL_REQ_MASK 7
119 * struct soc_data - vendor-specific config parameters for individual SoCs
120 * @stride: spacing between the registers of each channel
121 * @has_reqsel: does the controller use the newer requestselection mechanism
122 * @has_clocks: are controllable dma-clocks present
131 * enum s3c24xx_dma_chan_state - holds the virtual channel states
132 * @S3C24XX_DMA_CHAN_IDLE: the channel is idle
133 * @S3C24XX_DMA_CHAN_RUNNING: the channel has allocated a physical transport
134 * channel and is running a transfer on it
135 * @S3C24XX_DMA_CHAN_WAITING: the channel is waiting for a physical transport
136 * channel to become available (only pertains to memcpy channels)
138 enum s3c24xx_dma_chan_state {
139 S3C24XX_DMA_CHAN_IDLE,
140 S3C24XX_DMA_CHAN_RUNNING,
141 S3C24XX_DMA_CHAN_WAITING,
145 * struct s3c24xx_sg - structure containing data per sg
146 * @src_addr: src address of sg
147 * @dst_addr: dst address of sg
148 * @len: transfer len in bytes
149 * @node: node for txd's dsg_list
155 struct list_head node;
159 * struct s3c24xx_txd - wrapper for struct dma_async_tx_descriptor
160 * @vd: virtual DMA descriptor
161 * @dsg_list: list of children sg's
162 * @at: sg currently being transfered
163 * @width: transfer width
164 * @disrcc: value for source control register
165 * @didstc: value for destination control register
166 * @dcon: base value for dcon register
167 * @cyclic: indicate cyclic transfer
170 struct virt_dma_desc vd;
171 struct list_head dsg_list;
172 struct list_head *at;
180 struct s3c24xx_dma_chan;
183 * struct s3c24xx_dma_phy - holder for the physical channels
184 * @id: physical index to this channel
185 * @valid: does the channel have all required elements
186 * @base: virtual memory base (remapped) for the this channel
187 * @irq: interrupt for this channel
188 * @clk: clock for this channel
189 * @lock: a lock to use when altering an instance of this struct
190 * @serving: virtual channel currently being served by this physicalchannel
191 * @host: a pointer to the host (internal use)
193 struct s3c24xx_dma_phy {
200 struct s3c24xx_dma_chan *serving;
201 struct s3c24xx_dma_engine *host;
205 * struct s3c24xx_dma_chan - this structure wraps a DMA ENGINE channel
206 * @id: the id of the channel
207 * @name: name of the channel
208 * @vc: wrappped virtual channel
209 * @phy: the physical channel utilized by this channel, if there is one
210 * @runtime_addr: address for RX/TX according to the runtime config
211 * @at: active transaction on this channel
212 * @lock: a lock for this channel data
213 * @host: a pointer to the host (internal use)
214 * @state: whether the channel is idle, running etc
215 * @slave: whether this channel is a device (slave) or for memcpy
217 struct s3c24xx_dma_chan {
220 struct virt_dma_chan vc;
221 struct s3c24xx_dma_phy *phy;
222 struct dma_slave_config cfg;
223 struct s3c24xx_txd *at;
224 struct s3c24xx_dma_engine *host;
225 enum s3c24xx_dma_chan_state state;
230 * struct s3c24xx_dma_engine - the local state holder for the S3C24XX
231 * @pdev: the corresponding platform device
232 * @pdata: platform data passed in from the platform/machine
233 * @base: virtual memory base (remapped)
234 * @slave: slave engine for this instance
235 * @memcpy: memcpy engine for this instance
236 * @phy_chans: array of data for the physical channels
238 struct s3c24xx_dma_engine {
239 struct platform_device *pdev;
240 const struct s3c24xx_dma_platdata *pdata;
241 struct soc_data *sdata;
243 struct dma_device slave;
244 struct dma_device memcpy;
245 struct s3c24xx_dma_phy *phy_chans;
249 * Physical channel handling
253 * Check whether a certain channel is busy or not.
255 static int s3c24xx_dma_phy_busy(struct s3c24xx_dma_phy *phy)
257 unsigned int val = readl(phy->base + S3C24XX_DSTAT);
258 return val & S3C24XX_DSTAT_STAT_BUSY;
261 static bool s3c24xx_dma_phy_valid(struct s3c24xx_dma_chan *s3cchan,
262 struct s3c24xx_dma_phy *phy)
264 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
265 const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
266 struct s3c24xx_dma_channel *cdata = &pdata->channels[s3cchan->id];
269 /* every phy is valid for memcopy channels */
273 /* On newer variants all phys can be used for all virtual channels */
274 if (s3cdma->sdata->has_reqsel)
277 phyvalid = (cdata->chansel >> (phy->id * S3C24XX_CHANSEL_WIDTH));
278 return (phyvalid & S3C24XX_CHANSEL_VALID) ? true : false;
282 * Allocate a physical channel for a virtual channel
284 * Try to locate a physical channel to be used for this transfer. If all
285 * are taken return NULL and the requester will have to cope by using
286 * some fallback PIO mode or retrying later.
289 struct s3c24xx_dma_phy *s3c24xx_dma_get_phy(struct s3c24xx_dma_chan *s3cchan)
291 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
292 struct s3c24xx_dma_phy *phy = NULL;
297 for (i = 0; i < s3cdma->pdata->num_phy_channels; i++) {
298 phy = &s3cdma->phy_chans[i];
303 if (!s3c24xx_dma_phy_valid(s3cchan, phy))
306 spin_lock_irqsave(&phy->lock, flags);
309 phy->serving = s3cchan;
310 spin_unlock_irqrestore(&phy->lock, flags);
314 spin_unlock_irqrestore(&phy->lock, flags);
317 /* No physical channel available, cope with it */
318 if (i == s3cdma->pdata->num_phy_channels) {
319 dev_warn(&s3cdma->pdev->dev, "no phy channel available\n");
323 /* start the phy clock */
324 if (s3cdma->sdata->has_clocks) {
325 ret = clk_enable(phy->clk);
327 dev_err(&s3cdma->pdev->dev, "could not enable clock for channel %d, err %d\n",
338 * Mark the physical channel as free.
340 * This drops the link between the physical and virtual channel.
342 static inline void s3c24xx_dma_put_phy(struct s3c24xx_dma_phy *phy)
344 struct s3c24xx_dma_engine *s3cdma = phy->host;
346 if (s3cdma->sdata->has_clocks)
347 clk_disable(phy->clk);
353 * Stops the channel by writing the stop bit.
354 * This should not be used for an on-going transfer, but as a method of
355 * shutting down a channel (eg, when it's no longer used) or terminating a
358 static void s3c24xx_dma_terminate_phy(struct s3c24xx_dma_phy *phy)
360 writel(S3C24XX_DMASKTRIG_STOP, phy->base + S3C24XX_DMASKTRIG);
364 * Virtual channel handling
368 struct s3c24xx_dma_chan *to_s3c24xx_dma_chan(struct dma_chan *chan)
370 return container_of(chan, struct s3c24xx_dma_chan, vc.chan);
373 static u32 s3c24xx_dma_getbytes_chan(struct s3c24xx_dma_chan *s3cchan)
375 struct s3c24xx_dma_phy *phy = s3cchan->phy;
376 struct s3c24xx_txd *txd = s3cchan->at;
377 u32 tc = readl(phy->base + S3C24XX_DSTAT) & S3C24XX_DSTAT_CURRTC_MASK;
379 return tc * txd->width;
382 static int s3c24xx_dma_set_runtime_config(struct dma_chan *chan,
383 struct dma_slave_config *config)
385 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
389 /* Reject definitely invalid configurations */
390 if (config->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
391 config->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
394 spin_lock_irqsave(&s3cchan->vc.lock, flags);
396 if (!s3cchan->slave) {
401 s3cchan->cfg = *config;
404 spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
413 struct s3c24xx_txd *to_s3c24xx_txd(struct dma_async_tx_descriptor *tx)
415 return container_of(tx, struct s3c24xx_txd, vd.tx);
418 static struct s3c24xx_txd *s3c24xx_dma_get_txd(void)
420 struct s3c24xx_txd *txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
423 INIT_LIST_HEAD(&txd->dsg_list);
424 txd->dcon = S3C24XX_DCON_INT | S3C24XX_DCON_NORELOAD;
430 static void s3c24xx_dma_free_txd(struct s3c24xx_txd *txd)
432 struct s3c24xx_sg *dsg, *_dsg;
434 list_for_each_entry_safe(dsg, _dsg, &txd->dsg_list, node) {
435 list_del(&dsg->node);
442 static void s3c24xx_dma_start_next_sg(struct s3c24xx_dma_chan *s3cchan,
443 struct s3c24xx_txd *txd)
445 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
446 struct s3c24xx_dma_phy *phy = s3cchan->phy;
447 const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
448 struct s3c24xx_sg *dsg = list_entry(txd->at, struct s3c24xx_sg, node);
449 u32 dcon = txd->dcon;
452 /* transfer-size and -count from len and width */
453 switch (txd->width) {
455 dcon |= S3C24XX_DCON_DSZ_BYTE | dsg->len;
458 dcon |= S3C24XX_DCON_DSZ_HALFWORD | (dsg->len / 2);
461 dcon |= S3C24XX_DCON_DSZ_WORD | (dsg->len / 4);
465 if (s3cchan->slave) {
466 struct s3c24xx_dma_channel *cdata =
467 &pdata->channels[s3cchan->id];
469 if (s3cdma->sdata->has_reqsel) {
470 writel_relaxed((cdata->chansel << 1) |
471 S3C24XX_DMAREQSEL_HW,
472 phy->base + S3C24XX_DMAREQSEL);
474 int csel = cdata->chansel >> (phy->id *
475 S3C24XX_CHANSEL_WIDTH);
477 csel &= S3C24XX_CHANSEL_REQ_MASK;
478 dcon |= csel << S3C24XX_DCON_HWSRC_SHIFT;
479 dcon |= S3C24XX_DCON_HWTRIG;
482 if (s3cdma->sdata->has_reqsel)
483 writel_relaxed(0, phy->base + S3C24XX_DMAREQSEL);
486 writel_relaxed(dsg->src_addr, phy->base + S3C24XX_DISRC);
487 writel_relaxed(txd->disrcc, phy->base + S3C24XX_DISRCC);
488 writel_relaxed(dsg->dst_addr, phy->base + S3C24XX_DIDST);
489 writel_relaxed(txd->didstc, phy->base + S3C24XX_DIDSTC);
490 writel_relaxed(dcon, phy->base + S3C24XX_DCON);
492 val = readl_relaxed(phy->base + S3C24XX_DMASKTRIG);
493 val &= ~S3C24XX_DMASKTRIG_STOP;
494 val |= S3C24XX_DMASKTRIG_ON;
496 /* trigger the dma operation for memcpy transfers */
498 val |= S3C24XX_DMASKTRIG_SWTRIG;
500 writel(val, phy->base + S3C24XX_DMASKTRIG);
504 * Set the initial DMA register values and start first sg.
506 static void s3c24xx_dma_start_next_txd(struct s3c24xx_dma_chan *s3cchan)
508 struct s3c24xx_dma_phy *phy = s3cchan->phy;
509 struct virt_dma_desc *vd = vchan_next_desc(&s3cchan->vc);
510 struct s3c24xx_txd *txd = to_s3c24xx_txd(&vd->tx);
512 list_del(&txd->vd.node);
516 /* Wait for channel inactive */
517 while (s3c24xx_dma_phy_busy(phy))
520 /* point to the first element of the sg list */
521 txd->at = txd->dsg_list.next;
522 s3c24xx_dma_start_next_sg(s3cchan, txd);
525 static void s3c24xx_dma_free_txd_list(struct s3c24xx_dma_engine *s3cdma,
526 struct s3c24xx_dma_chan *s3cchan)
530 vchan_get_all_descriptors(&s3cchan->vc, &head);
531 vchan_dma_desc_free_list(&s3cchan->vc, &head);
535 * Try to allocate a physical channel. When successful, assign it to
536 * this virtual channel, and initiate the next descriptor. The
537 * virtual channel lock must be held at this point.
539 static void s3c24xx_dma_phy_alloc_and_start(struct s3c24xx_dma_chan *s3cchan)
541 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
542 struct s3c24xx_dma_phy *phy;
544 phy = s3c24xx_dma_get_phy(s3cchan);
546 dev_dbg(&s3cdma->pdev->dev, "no physical channel available for xfer on %s\n",
548 s3cchan->state = S3C24XX_DMA_CHAN_WAITING;
552 dev_dbg(&s3cdma->pdev->dev, "allocated physical channel %d for xfer on %s\n",
553 phy->id, s3cchan->name);
556 s3cchan->state = S3C24XX_DMA_CHAN_RUNNING;
558 s3c24xx_dma_start_next_txd(s3cchan);
561 static void s3c24xx_dma_phy_reassign_start(struct s3c24xx_dma_phy *phy,
562 struct s3c24xx_dma_chan *s3cchan)
564 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
566 dev_dbg(&s3cdma->pdev->dev, "reassigned physical channel %d for xfer on %s\n",
567 phy->id, s3cchan->name);
570 * We do this without taking the lock; we're really only concerned
571 * about whether this pointer is NULL or not, and we're guaranteed
572 * that this will only be called when it _already_ is non-NULL.
574 phy->serving = s3cchan;
576 s3cchan->state = S3C24XX_DMA_CHAN_RUNNING;
577 s3c24xx_dma_start_next_txd(s3cchan);
581 * Free a physical DMA channel, potentially reallocating it to another
582 * virtual channel if we have any pending.
584 static void s3c24xx_dma_phy_free(struct s3c24xx_dma_chan *s3cchan)
586 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
587 struct s3c24xx_dma_chan *p, *next;
592 /* Find a waiting virtual channel for the next transfer. */
593 list_for_each_entry(p, &s3cdma->memcpy.channels, vc.chan.device_node)
594 if (p->state == S3C24XX_DMA_CHAN_WAITING) {
600 list_for_each_entry(p, &s3cdma->slave.channels,
602 if (p->state == S3C24XX_DMA_CHAN_WAITING &&
603 s3c24xx_dma_phy_valid(p, s3cchan->phy)) {
609 /* Ensure that the physical channel is stopped */
610 s3c24xx_dma_terminate_phy(s3cchan->phy);
616 * Eww. We know this isn't going to deadlock
617 * but lockdep probably doesn't.
619 spin_lock(&next->vc.lock);
620 /* Re-check the state now that we have the lock */
621 success = next->state == S3C24XX_DMA_CHAN_WAITING;
623 s3c24xx_dma_phy_reassign_start(s3cchan->phy, next);
624 spin_unlock(&next->vc.lock);
626 /* If the state changed, try to find another channel */
630 /* No more jobs, so free up the physical channel */
631 s3c24xx_dma_put_phy(s3cchan->phy);
635 s3cchan->state = S3C24XX_DMA_CHAN_IDLE;
638 static void s3c24xx_dma_desc_free(struct virt_dma_desc *vd)
640 struct s3c24xx_txd *txd = to_s3c24xx_txd(&vd->tx);
641 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(vd->tx.chan);
644 dma_descriptor_unmap(&vd->tx);
646 s3c24xx_dma_free_txd(txd);
649 static irqreturn_t s3c24xx_dma_irq(int irq, void *data)
651 struct s3c24xx_dma_phy *phy = data;
652 struct s3c24xx_dma_chan *s3cchan = phy->serving;
653 struct s3c24xx_txd *txd;
655 dev_dbg(&phy->host->pdev->dev, "interrupt on channel %d\n", phy->id);
658 * Interrupts happen to notify the completion of a transfer and the
659 * channel should have moved into its stop state already on its own.
660 * Therefore interrupts on channels not bound to a virtual channel
661 * should never happen. Nevertheless send a terminate command to the
662 * channel if the unlikely case happens.
664 if (unlikely(!s3cchan)) {
665 dev_err(&phy->host->pdev->dev, "interrupt on unused channel %d\n",
668 s3c24xx_dma_terminate_phy(phy);
673 spin_lock(&s3cchan->vc.lock);
676 /* when more sg's are in this txd, start the next one */
677 if (!list_is_last(txd->at, &txd->dsg_list)) {
678 txd->at = txd->at->next;
680 vchan_cyclic_callback(&txd->vd);
681 s3c24xx_dma_start_next_sg(s3cchan, txd);
682 } else if (!txd->cyclic) {
684 vchan_cookie_complete(&txd->vd);
687 * And start the next descriptor (if any),
688 * otherwise free this channel.
690 if (vchan_next_desc(&s3cchan->vc))
691 s3c24xx_dma_start_next_txd(s3cchan);
693 s3c24xx_dma_phy_free(s3cchan);
695 vchan_cyclic_callback(&txd->vd);
697 /* Cyclic: reset at beginning */
698 txd->at = txd->dsg_list.next;
699 s3c24xx_dma_start_next_sg(s3cchan, txd);
702 spin_unlock(&s3cchan->vc.lock);
711 static int s3c24xx_dma_terminate_all(struct dma_chan *chan)
713 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
714 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
718 spin_lock_irqsave(&s3cchan->vc.lock, flags);
720 if (!s3cchan->phy && !s3cchan->at) {
721 dev_err(&s3cdma->pdev->dev, "trying to terminate already stopped channel %d\n",
727 s3cchan->state = S3C24XX_DMA_CHAN_IDLE;
729 /* Mark physical channel as free */
731 s3c24xx_dma_phy_free(s3cchan);
733 /* Dequeue current job */
735 s3c24xx_dma_desc_free(&s3cchan->at->vd);
739 /* Dequeue jobs not yet fired as well */
740 s3c24xx_dma_free_txd_list(s3cdma, s3cchan);
742 spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
747 static void s3c24xx_dma_free_chan_resources(struct dma_chan *chan)
749 /* Ensure all queued descriptors are freed */
750 vchan_free_chan_resources(to_virt_chan(chan));
753 static enum dma_status s3c24xx_dma_tx_status(struct dma_chan *chan,
754 dma_cookie_t cookie, struct dma_tx_state *txstate)
756 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
757 struct s3c24xx_txd *txd;
758 struct s3c24xx_sg *dsg;
759 struct virt_dma_desc *vd;
764 spin_lock_irqsave(&s3cchan->vc.lock, flags);
765 ret = dma_cookie_status(chan, cookie, txstate);
768 * There's no point calculating the residue if there's
769 * no txstate to store the value.
771 if (ret == DMA_COMPLETE || !txstate) {
772 spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
776 vd = vchan_find_desc(&s3cchan->vc, cookie);
778 /* On the issued list, so hasn't been processed yet */
779 txd = to_s3c24xx_txd(&vd->tx);
781 list_for_each_entry(dsg, &txd->dsg_list, node)
785 * Currently running, so sum over the pending sg's and
786 * the currently active one.
790 dsg = list_entry(txd->at, struct s3c24xx_sg, node);
791 list_for_each_entry_from(dsg, &txd->dsg_list, node)
794 bytes += s3c24xx_dma_getbytes_chan(s3cchan);
796 spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
799 * This cookie not complete yet
800 * Get number of bytes left in the active transactions and queue
802 dma_set_residue(txstate, bytes);
804 /* Whether waiting or running, we're in progress */
809 * Initialize a descriptor to be used by memcpy submit
811 static struct dma_async_tx_descriptor *s3c24xx_dma_prep_memcpy(
812 struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
813 size_t len, unsigned long flags)
815 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
816 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
817 struct s3c24xx_txd *txd;
818 struct s3c24xx_sg *dsg;
819 int src_mod, dest_mod;
821 dev_dbg(&s3cdma->pdev->dev, "prepare memcpy of %zu bytes from %s\n",
824 if ((len & S3C24XX_DCON_TC_MASK) != len) {
825 dev_err(&s3cdma->pdev->dev, "memcpy size %zu to large\n", len);
829 txd = s3c24xx_dma_get_txd();
833 dsg = kzalloc(sizeof(*dsg), GFP_NOWAIT);
835 s3c24xx_dma_free_txd(txd);
838 list_add_tail(&dsg->node, &txd->dsg_list);
841 dsg->dst_addr = dest;
845 * Determine a suitable transfer width.
846 * The DMA controller cannot fetch/store information which is not
847 * naturally aligned on the bus, i.e., a 4 byte fetch must start at
848 * an address divisible by 4 - more generally addr % width must be 0.
854 txd->width = (src_mod == 0 && dest_mod == 0) ? 4 : 1;
857 txd->width = ((src_mod == 2 || src_mod == 0) &&
858 (dest_mod == 2 || dest_mod == 0)) ? 2 : 1;
865 txd->disrcc = S3C24XX_DISRCC_LOC_AHB | S3C24XX_DISRCC_INC_INCREMENT;
866 txd->didstc = S3C24XX_DIDSTC_LOC_AHB | S3C24XX_DIDSTC_INC_INCREMENT;
867 txd->dcon |= S3C24XX_DCON_DEMAND | S3C24XX_DCON_SYNC_HCLK |
868 S3C24XX_DCON_SERV_WHOLE;
870 return vchan_tx_prep(&s3cchan->vc, &txd->vd, flags);
873 static struct dma_async_tx_descriptor *s3c24xx_dma_prep_dma_cyclic(
874 struct dma_chan *chan, dma_addr_t addr, size_t size, size_t period,
875 enum dma_transfer_direction direction, unsigned long flags)
877 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
878 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
879 const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
880 struct s3c24xx_dma_channel *cdata = &pdata->channels[s3cchan->id];
881 struct s3c24xx_txd *txd;
882 struct s3c24xx_sg *dsg;
884 dma_addr_t slave_addr;
888 dev_dbg(&s3cdma->pdev->dev,
889 "prepare cyclic transaction of %zu bytes with period %zu from %s\n",
890 size, period, s3cchan->name);
892 if (!is_slave_direction(direction)) {
893 dev_err(&s3cdma->pdev->dev,
894 "direction %d unsupported\n", direction);
898 txd = s3c24xx_dma_get_txd();
904 if (cdata->handshake)
905 txd->dcon |= S3C24XX_DCON_HANDSHAKE;
907 switch (cdata->bus) {
908 case S3C24XX_DMA_APB:
909 txd->dcon |= S3C24XX_DCON_SYNC_PCLK;
910 hwcfg |= S3C24XX_DISRCC_LOC_APB;
912 case S3C24XX_DMA_AHB:
913 txd->dcon |= S3C24XX_DCON_SYNC_HCLK;
914 hwcfg |= S3C24XX_DISRCC_LOC_AHB;
919 * Always assume our peripheral desintation is a fixed
922 hwcfg |= S3C24XX_DISRCC_INC_FIXED;
925 * Individual dma operations are requested by the slave,
926 * so serve only single atomic operations (S3C24XX_DCON_SERV_SINGLE).
928 txd->dcon |= S3C24XX_DCON_SERV_SINGLE;
930 if (direction == DMA_MEM_TO_DEV) {
931 txd->disrcc = S3C24XX_DISRCC_LOC_AHB |
932 S3C24XX_DISRCC_INC_INCREMENT;
934 slave_addr = s3cchan->cfg.dst_addr;
935 txd->width = s3cchan->cfg.dst_addr_width;
938 txd->didstc = S3C24XX_DIDSTC_LOC_AHB |
939 S3C24XX_DIDSTC_INC_INCREMENT;
940 slave_addr = s3cchan->cfg.src_addr;
941 txd->width = s3cchan->cfg.src_addr_width;
944 sg_len = size / period;
946 for (i = 0; i < sg_len; i++) {
947 dsg = kzalloc(sizeof(*dsg), GFP_NOWAIT);
949 s3c24xx_dma_free_txd(txd);
952 list_add_tail(&dsg->node, &txd->dsg_list);
955 /* Check last period length */
957 dsg->len = size - period * i;
958 if (direction == DMA_MEM_TO_DEV) {
959 dsg->src_addr = addr + period * i;
960 dsg->dst_addr = slave_addr;
961 } else { /* DMA_DEV_TO_MEM */
962 dsg->src_addr = slave_addr;
963 dsg->dst_addr = addr + period * i;
967 return vchan_tx_prep(&s3cchan->vc, &txd->vd, flags);
970 static struct dma_async_tx_descriptor *s3c24xx_dma_prep_slave_sg(
971 struct dma_chan *chan, struct scatterlist *sgl,
972 unsigned int sg_len, enum dma_transfer_direction direction,
973 unsigned long flags, void *context)
975 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
976 struct s3c24xx_dma_engine *s3cdma = s3cchan->host;
977 const struct s3c24xx_dma_platdata *pdata = s3cdma->pdata;
978 struct s3c24xx_dma_channel *cdata = &pdata->channels[s3cchan->id];
979 struct s3c24xx_txd *txd;
980 struct s3c24xx_sg *dsg;
981 struct scatterlist *sg;
982 dma_addr_t slave_addr;
986 dev_dbg(&s3cdma->pdev->dev, "prepare transaction of %d bytes from %s\n",
987 sg_dma_len(sgl), s3cchan->name);
989 txd = s3c24xx_dma_get_txd();
993 if (cdata->handshake)
994 txd->dcon |= S3C24XX_DCON_HANDSHAKE;
996 switch (cdata->bus) {
997 case S3C24XX_DMA_APB:
998 txd->dcon |= S3C24XX_DCON_SYNC_PCLK;
999 hwcfg |= S3C24XX_DISRCC_LOC_APB;
1001 case S3C24XX_DMA_AHB:
1002 txd->dcon |= S3C24XX_DCON_SYNC_HCLK;
1003 hwcfg |= S3C24XX_DISRCC_LOC_AHB;
1008 * Always assume our peripheral desintation is a fixed
1009 * address in memory.
1011 hwcfg |= S3C24XX_DISRCC_INC_FIXED;
1014 * Individual dma operations are requested by the slave,
1015 * so serve only single atomic operations (S3C24XX_DCON_SERV_SINGLE).
1017 txd->dcon |= S3C24XX_DCON_SERV_SINGLE;
1019 if (direction == DMA_MEM_TO_DEV) {
1020 txd->disrcc = S3C24XX_DISRCC_LOC_AHB |
1021 S3C24XX_DISRCC_INC_INCREMENT;
1022 txd->didstc = hwcfg;
1023 slave_addr = s3cchan->cfg.dst_addr;
1024 txd->width = s3cchan->cfg.dst_addr_width;
1025 } else if (direction == DMA_DEV_TO_MEM) {
1026 txd->disrcc = hwcfg;
1027 txd->didstc = S3C24XX_DIDSTC_LOC_AHB |
1028 S3C24XX_DIDSTC_INC_INCREMENT;
1029 slave_addr = s3cchan->cfg.src_addr;
1030 txd->width = s3cchan->cfg.src_addr_width;
1032 s3c24xx_dma_free_txd(txd);
1033 dev_err(&s3cdma->pdev->dev,
1034 "direction %d unsupported\n", direction);
1038 for_each_sg(sgl, sg, sg_len, tmp) {
1039 dsg = kzalloc(sizeof(*dsg), GFP_NOWAIT);
1041 s3c24xx_dma_free_txd(txd);
1044 list_add_tail(&dsg->node, &txd->dsg_list);
1046 dsg->len = sg_dma_len(sg);
1047 if (direction == DMA_MEM_TO_DEV) {
1048 dsg->src_addr = sg_dma_address(sg);
1049 dsg->dst_addr = slave_addr;
1050 } else { /* DMA_DEV_TO_MEM */
1051 dsg->src_addr = slave_addr;
1052 dsg->dst_addr = sg_dma_address(sg);
1056 return vchan_tx_prep(&s3cchan->vc, &txd->vd, flags);
1060 * Slave transactions callback to the slave device to allow
1061 * synchronization of slave DMA signals with the DMAC enable
1063 static void s3c24xx_dma_issue_pending(struct dma_chan *chan)
1065 struct s3c24xx_dma_chan *s3cchan = to_s3c24xx_dma_chan(chan);
1066 unsigned long flags;
1068 spin_lock_irqsave(&s3cchan->vc.lock, flags);
1069 if (vchan_issue_pending(&s3cchan->vc)) {
1070 if (!s3cchan->phy && s3cchan->state != S3C24XX_DMA_CHAN_WAITING)
1071 s3c24xx_dma_phy_alloc_and_start(s3cchan);
1073 spin_unlock_irqrestore(&s3cchan->vc.lock, flags);
1077 * Bringup and teardown
1081 * Initialise the DMAC memcpy/slave channels.
1082 * Make a local wrapper to hold required data
1084 static int s3c24xx_dma_init_virtual_channels(struct s3c24xx_dma_engine *s3cdma,
1085 struct dma_device *dmadev, unsigned int channels, bool slave)
1087 struct s3c24xx_dma_chan *chan;
1090 INIT_LIST_HEAD(&dmadev->channels);
1093 * Register as many many memcpy as we have physical channels,
1094 * we won't always be able to use all but the code will have
1095 * to cope with that situation.
1097 for (i = 0; i < channels; i++) {
1098 chan = devm_kzalloc(dmadev->dev, sizeof(*chan), GFP_KERNEL);
1103 chan->host = s3cdma;
1104 chan->state = S3C24XX_DMA_CHAN_IDLE;
1108 chan->name = kasprintf(GFP_KERNEL, "slave%d", i);
1112 chan->name = kasprintf(GFP_KERNEL, "memcpy%d", i);
1116 dev_dbg(dmadev->dev,
1117 "initialize virtual channel \"%s\"\n",
1120 chan->vc.desc_free = s3c24xx_dma_desc_free;
1121 vchan_init(&chan->vc, dmadev);
1123 dev_info(dmadev->dev, "initialized %d virtual %s channels\n",
1124 i, slave ? "slave" : "memcpy");
1128 static void s3c24xx_dma_free_virtual_channels(struct dma_device *dmadev)
1130 struct s3c24xx_dma_chan *chan = NULL;
1131 struct s3c24xx_dma_chan *next;
1133 list_for_each_entry_safe(chan,
1134 next, &dmadev->channels, vc.chan.device_node) {
1135 list_del(&chan->vc.chan.device_node);
1136 tasklet_kill(&chan->vc.task);
1140 /* s3c2410, s3c2440 and s3c2442 have a 0x40 stride without separate clocks */
1141 static struct soc_data soc_s3c2410 = {
1143 .has_reqsel = false,
1144 .has_clocks = false,
1147 /* s3c2412 and s3c2413 have a 0x40 stride and dmareqsel mechanism */
1148 static struct soc_data soc_s3c2412 = {
1154 /* s3c2443 and following have a 0x100 stride and dmareqsel mechanism */
1155 static struct soc_data soc_s3c2443 = {
1161 static const struct platform_device_id s3c24xx_dma_driver_ids[] = {
1163 .name = "s3c2410-dma",
1164 .driver_data = (kernel_ulong_t)&soc_s3c2410,
1166 .name = "s3c2412-dma",
1167 .driver_data = (kernel_ulong_t)&soc_s3c2412,
1169 .name = "s3c2443-dma",
1170 .driver_data = (kernel_ulong_t)&soc_s3c2443,
1175 static struct soc_data *s3c24xx_dma_get_soc_data(struct platform_device *pdev)
1177 return (struct soc_data *)
1178 platform_get_device_id(pdev)->driver_data;
1181 static int s3c24xx_dma_probe(struct platform_device *pdev)
1183 const struct s3c24xx_dma_platdata *pdata = dev_get_platdata(&pdev->dev);
1184 struct s3c24xx_dma_engine *s3cdma;
1185 struct soc_data *sdata;
1186 struct resource *res;
1191 dev_err(&pdev->dev, "platform data missing\n");
1195 /* Basic sanity check */
1196 if (pdata->num_phy_channels > MAX_DMA_CHANNELS) {
1197 dev_err(&pdev->dev, "to many dma channels %d, max %d\n",
1198 pdata->num_phy_channels, MAX_DMA_CHANNELS);
1202 sdata = s3c24xx_dma_get_soc_data(pdev);
1206 s3cdma = devm_kzalloc(&pdev->dev, sizeof(*s3cdma), GFP_KERNEL);
1210 s3cdma->pdev = pdev;
1211 s3cdma->pdata = pdata;
1212 s3cdma->sdata = sdata;
1214 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1215 s3cdma->base = devm_ioremap_resource(&pdev->dev, res);
1216 if (IS_ERR(s3cdma->base))
1217 return PTR_ERR(s3cdma->base);
1219 s3cdma->phy_chans = devm_kzalloc(&pdev->dev,
1220 sizeof(struct s3c24xx_dma_phy) *
1221 pdata->num_phy_channels,
1223 if (!s3cdma->phy_chans)
1226 /* acquire irqs and clocks for all physical channels */
1227 for (i = 0; i < pdata->num_phy_channels; i++) {
1228 struct s3c24xx_dma_phy *phy = &s3cdma->phy_chans[i];
1232 phy->base = s3cdma->base + (i * sdata->stride);
1235 phy->irq = platform_get_irq(pdev, i);
1237 dev_err(&pdev->dev, "failed to get irq %d, err %d\n",
1242 ret = devm_request_irq(&pdev->dev, phy->irq, s3c24xx_dma_irq,
1243 0, pdev->name, phy);
1245 dev_err(&pdev->dev, "Unable to request irq for channel %d, error %d\n",
1250 if (sdata->has_clocks) {
1251 sprintf(clk_name, "dma.%d", i);
1252 phy->clk = devm_clk_get(&pdev->dev, clk_name);
1253 if (IS_ERR(phy->clk) && sdata->has_clocks) {
1254 dev_err(&pdev->dev, "unable to acquire clock for channel %d, error %lu\n",
1255 i, PTR_ERR(phy->clk));
1259 ret = clk_prepare(phy->clk);
1261 dev_err(&pdev->dev, "clock for phy %d failed, error %d\n",
1267 spin_lock_init(&phy->lock);
1270 dev_dbg(&pdev->dev, "physical channel %d is %s\n",
1271 i, s3c24xx_dma_phy_busy(phy) ? "BUSY" : "FREE");
1274 /* Initialize memcpy engine */
1275 dma_cap_set(DMA_MEMCPY, s3cdma->memcpy.cap_mask);
1276 dma_cap_set(DMA_PRIVATE, s3cdma->memcpy.cap_mask);
1277 s3cdma->memcpy.dev = &pdev->dev;
1278 s3cdma->memcpy.device_free_chan_resources =
1279 s3c24xx_dma_free_chan_resources;
1280 s3cdma->memcpy.device_prep_dma_memcpy = s3c24xx_dma_prep_memcpy;
1281 s3cdma->memcpy.device_tx_status = s3c24xx_dma_tx_status;
1282 s3cdma->memcpy.device_issue_pending = s3c24xx_dma_issue_pending;
1283 s3cdma->memcpy.device_config = s3c24xx_dma_set_runtime_config;
1284 s3cdma->memcpy.device_terminate_all = s3c24xx_dma_terminate_all;
1286 /* Initialize slave engine for SoC internal dedicated peripherals */
1287 dma_cap_set(DMA_SLAVE, s3cdma->slave.cap_mask);
1288 dma_cap_set(DMA_CYCLIC, s3cdma->slave.cap_mask);
1289 dma_cap_set(DMA_PRIVATE, s3cdma->slave.cap_mask);
1290 s3cdma->slave.dev = &pdev->dev;
1291 s3cdma->slave.device_free_chan_resources =
1292 s3c24xx_dma_free_chan_resources;
1293 s3cdma->slave.device_tx_status = s3c24xx_dma_tx_status;
1294 s3cdma->slave.device_issue_pending = s3c24xx_dma_issue_pending;
1295 s3cdma->slave.device_prep_slave_sg = s3c24xx_dma_prep_slave_sg;
1296 s3cdma->slave.device_prep_dma_cyclic = s3c24xx_dma_prep_dma_cyclic;
1297 s3cdma->slave.device_config = s3c24xx_dma_set_runtime_config;
1298 s3cdma->slave.device_terminate_all = s3c24xx_dma_terminate_all;
1299 s3cdma->slave.filter.map = pdata->slave_map;
1300 s3cdma->slave.filter.mapcnt = pdata->slavecnt;
1301 s3cdma->slave.filter.fn = s3c24xx_dma_filter;
1303 /* Register as many memcpy channels as there are physical channels */
1304 ret = s3c24xx_dma_init_virtual_channels(s3cdma, &s3cdma->memcpy,
1305 pdata->num_phy_channels, false);
1307 dev_warn(&pdev->dev,
1308 "%s failed to enumerate memcpy channels - %d\n",
1313 /* Register slave channels */
1314 ret = s3c24xx_dma_init_virtual_channels(s3cdma, &s3cdma->slave,
1315 pdata->num_channels, true);
1317 dev_warn(&pdev->dev,
1318 "%s failed to enumerate slave channels - %d\n",
1323 ret = dma_async_device_register(&s3cdma->memcpy);
1325 dev_warn(&pdev->dev,
1326 "%s failed to register memcpy as an async device - %d\n",
1328 goto err_memcpy_reg;
1331 ret = dma_async_device_register(&s3cdma->slave);
1333 dev_warn(&pdev->dev,
1334 "%s failed to register slave as an async device - %d\n",
1339 platform_set_drvdata(pdev, s3cdma);
1340 dev_info(&pdev->dev, "Loaded dma driver with %d physical channels\n",
1341 pdata->num_phy_channels);
1346 dma_async_device_unregister(&s3cdma->memcpy);
1348 s3c24xx_dma_free_virtual_channels(&s3cdma->slave);
1350 s3c24xx_dma_free_virtual_channels(&s3cdma->memcpy);
1352 if (sdata->has_clocks)
1353 for (i = 0; i < pdata->num_phy_channels; i++) {
1354 struct s3c24xx_dma_phy *phy = &s3cdma->phy_chans[i];
1356 clk_unprepare(phy->clk);
1362 static void s3c24xx_dma_free_irq(struct platform_device *pdev,
1363 struct s3c24xx_dma_engine *s3cdma)
1367 for (i = 0; i < s3cdma->pdata->num_phy_channels; i++) {
1368 struct s3c24xx_dma_phy *phy = &s3cdma->phy_chans[i];
1370 devm_free_irq(&pdev->dev, phy->irq, phy);
1374 static int s3c24xx_dma_remove(struct platform_device *pdev)
1376 const struct s3c24xx_dma_platdata *pdata = dev_get_platdata(&pdev->dev);
1377 struct s3c24xx_dma_engine *s3cdma = platform_get_drvdata(pdev);
1378 struct soc_data *sdata = s3c24xx_dma_get_soc_data(pdev);
1381 dma_async_device_unregister(&s3cdma->slave);
1382 dma_async_device_unregister(&s3cdma->memcpy);
1384 s3c24xx_dma_free_irq(pdev, s3cdma);
1386 s3c24xx_dma_free_virtual_channels(&s3cdma->slave);
1387 s3c24xx_dma_free_virtual_channels(&s3cdma->memcpy);
1389 if (sdata->has_clocks)
1390 for (i = 0; i < pdata->num_phy_channels; i++) {
1391 struct s3c24xx_dma_phy *phy = &s3cdma->phy_chans[i];
1393 clk_unprepare(phy->clk);
1399 static struct platform_driver s3c24xx_dma_driver = {
1401 .name = "s3c24xx-dma",
1403 .id_table = s3c24xx_dma_driver_ids,
1404 .probe = s3c24xx_dma_probe,
1405 .remove = s3c24xx_dma_remove,
1408 module_platform_driver(s3c24xx_dma_driver);
1410 bool s3c24xx_dma_filter(struct dma_chan *chan, void *param)
1412 struct s3c24xx_dma_chan *s3cchan;
1414 if (chan->device->dev->driver != &s3c24xx_dma_driver.driver)
1417 s3cchan = to_s3c24xx_dma_chan(chan);
1419 return s3cchan->id == (uintptr_t)param;
1421 EXPORT_SYMBOL(s3c24xx_dma_filter);
1423 MODULE_DESCRIPTION("S3C24XX DMA Driver");
1424 MODULE_AUTHOR("Heiko Stuebner");
1425 MODULE_LICENSE("GPL v2");